AN ABSTRACT OF THE THESIS OF

Yimin Zenci for the degree of Master of Science in

Industrial and Manufacturing Enciineerinci presented on

MaY 5, 1993

Title: An Expert System For Softwood Lumber Grading

Abstract approved:
Dr. Sabah Randhawa

The focus of this research is to develop a prototype

expert system for softwood lumber grading. The grading

rules used in the knowledge base of the system are based on

Western Lumber Grading Rules 88 published by the Western

Wood Products Association. The system includes 27 grades

in Dimension, Select/Finish, and Boards categories.

The system is designed to be interactive and menu-

driven. The user input to the system consists of lumber

size, grade category, and type, location and size of

defects for each face. The system then infers the grade

corresponding to each face, and an overall grade for the

lumber. The system provides limited explanation

capabilities.

Evaluation of the system was performed using 85

samples of pre-graded Siberian larch 2x4x12s in Structural

Light Framing category. The initial evaluation was

performed using the two wide faces of boards. Results

indicated a 60 percent match between the grade assigned by



the human expert and the system. The largest cause of

deviation was exclusion of defects on the two narrow faces.

The knowledge base was expanded to include the two narrow

faces; the match rate improved to 76.5 percent.

Evaluations for other grading categories need to be

conducted in the future to assess the adequacy of the

knowledge base.

The prototype development concentrates on selected

defect characteristics for each grade. These

characteristics are clearly defined and described in the

rule book, and are usually the most frequently encountered

defects on softwood lumber. The knowledge base needs to be

refined and expanded if additional factors such as knot

positions relative to each other, warp, manufacturing

imperfections and clustering of defects are to be

considered.



(C) Copyright by Yimin Zeng
May 5, 1993

All Rights Reserved



An Expert System For Softwood Lumber Grading

by

Yimin Zeng

A THESIS

submitted to

Oregon State University

in partial fulfillment of
the requirements for the

degree of

Master of Science

Completed May 5, 1993

Commencement June 1993



APPROVED:

Professor of Industrial and Manufacturing Engineering
in charge of major

Head of Department of Industrial and
Manufacturing Engineering

Dean of Gradua Sè'hdol

Date thesis is presented May 5. 1993

Typed by Yimin Zeng for Yimin Zenci



ACIOWLEDGEMENTB

I wish to express my deepest gratitude and appreciation

to my major professor, Dr. Sabah Randhawa, for his

guidance, encouragement and support throughout the course

of my graduate study. His tremendous patience and

judicious suggestions made possible the successful

completion of this thesis.

I wish to thank Dr. Kenneth Funk and Dr. Eldon Olsen

for contributing their time and expertise while serving

on my graduate committee.

My very special appreciations go to Dr. James Funk. He

spent tremendous amount of time to teach me about lumber

production and grading, and to help me to evaluate the

system using real lumber.

I would also like to thank Dr. Robert L. Ethington and

Dr. Rakesh Gupta for providing me with lumber samples.

I am grateful to my friend Mr. Johannes B. Forrer.

The discussions with him always inspired me.

I am deeply indebted to my wife, Yuan Zhong, for her

love, understanding, encouragement and help in all aspects

of my daily life and graduate studies.

I am deeply indebted to my parents and sisters for

their deep and eternal love, understanding, encouragement

and support.



TABLE OP CONTENTS

CHAPTER1INTRODUCTION . . 1

CHAPTER 2 RESEARCH OBJECTIVES ............ 7
CHAPTER 3 LITERATURE REVIEW ....... . . . . . . 8

3.1 Programs Using Algorithmic Approach . . . 9
3.2 Programs Using Expert System Approaches . 12

3.2.1 Structure of Expert Systems . . 12
3.2.2 Expert Systems in Lumber Grading . 15

3.3 Systems Developed in Industry ....... 17
CHAPTER4BYSTEMDEVELOPMENT ............ 19

4.1 ProblemDomain .............. 19
4.1.1 Softwood Lumber Grading Practices 19
4.1.2 Grading Categories ....... 2].

4.2 Knowledge Acquisition and Representation 31
4.2.1 Knowledge Acquisition ....... 31
4.2.2 Knowledge Representation ...... 32

4.3 Inference Mechanism ............ 38
4.4 Tool Selection .............. 43
4.5 Prototype Development ........... 48

4.5.1 Prototype Components ........ 48
4.5.2 Knowledge Base Implementation . 51
4.5.3 Grading Process, Input, and Output . 57

CHAPTER 5 EVALUATION AND ENHANCEMENTS ....... 67
5.1 System Evaluation ............ 67
5.2 System Improvement ............ 69

CHAPTER6CONCLUSIONS ................ 73
BIBLIOGRAPHY .................... 76
APPENDICES

Appendix A. Source Code for Structural Light Framing
Category ................ 7 8

Appendix B. Test Results Using Generated Data . . . 93



LIST OF FIGURES

Figure Page

1.1. Flow chart of a typical sawmill ........ 2
1.2. Flow chart of a typical planer mill ...... 3
3.1. Basic structure of an expert system ...... 13
4.1. Grading categories, sub-categories, and

grades included in this project ........ 22
4.2. Basic structure of the prototype ....... 49
4.3. Gradingprocess ............... 58
4.4. Opening window of the run-time version

ofthesystein ................ 61
4.5. Dialog box for selecting input method ..... 62
4.6. Dialog box for entering lumber thickness . . . 62

4.7. Dialog box for confirming lumber dimension . . 62

4.8. Dialog box for selecting grade category . . . . 63

4.9. Dialog box for selecting dimension sub-
categorr ................... 63

4.10. Dialog box for selecting lumber face ..... 64
4.11. Dialog box with partial defect list ...... 64
4.12. Dialog box with another partial defect list . . 65

4.13. Dialog box for entering knot type and
location ................... 65

4.14. Dialog box for entering knot diameter ..... 66
4.15. System output display ............. 66
5.1. Evaluation results considering all

factors .................... 68
5.2. Evaluation results when warp, manufacturing

imperfections, and measurement bias were not
considered................... 71



LIST OF TABLES

Table Page

4.1. Grading categories .............. 21
4.2. Grade characteristics considered by

the project for grades in Select/Finish's
Selectsub-category .............. 24

4.3. Grade characteristics considered by
the project for grades in Select/Finish's
Finish sub-category .............. 25

4.4. Grade characteristics considered by
the project for grades in Boards' Commons
sub-category ................. 26

4.5. Grade characteristics considered by
the project for grades in Boards' Alternate
sub-category ................. 27

4.6. Grade characteristics considered by
the project for grades in Dimensions'
Light Framing sub-category ........... 28

4.7. Grade characteristics considered by
the project for grades in Dimensions'
Structural Light Framing sub-category ..... 29

4.8. Grade characteristics considered by
the project for grades in Dimensions'
Studsub-category ............... 30

4.9. Characteristics permitted and limiting
provisions for CONSTRUCTION grade in Light
Framing sub-category of dimension lumber . . . 35

4.10. Grading rules for the CONSTRUCTION grade
represented using production rules ...... 36

4.11. Source files and their contents ........ 50



LIST OF APPENDICES TABLES

Table

Bi. Specifications of grades for a piece of

Page

lumber in the Dimension category (lumber
size=

94

B2. Test results of grading the lumber in the
Dimensioncategory ............... 95

B3. Specifications of grades for a piece of
lumber in the Select/Finish category
(lumber size = l"x8"x12') .......... 96

B4. Test results of grading the lumber in
the Select/Finish category ........... 97

B5. Specifications of grades for a piece of
lumber in the Boards category
(lumber size = l"x8"x12') ........... 98

B6. Test results of grading the lumber in
theBoardscategory .............. 99



AN EXPERT SYSTEM FOR SOFTWOOD LUMBER GRADING

CHAPTER 2

INTRODUCTION

Lumber sawiuilling is one of the oldest industries in

America. In mills, logs are converted into lumber by

sawing, edging and trimming operations (Williston, 1988).

At the sawing station, logs are broken down into rough,

unedged and untrimmed lumber. The rough lumber is then cut

into green lumber in specified width and length at edging

and trimming stations. Rough green lumber is then dried

and surfaced into finished products. Figures 1.1 and 1.2

show the primary steps in sawmill and planer mills for

lumber manufacturing. Configuration of different mills may

vary depending upon the log supply and finished products.

One of the most important issues concerning the lumber

manufacturing industry is the efficiency of raw material

usage. Lumber recovery, a term used to describe how much

finished lumber is produced from a log or a given

population of logs, is used as the primary measure for the

performance of a sawmill. The trend is to use computer-

based technology to optimize lumber recovery. Trimming and

edging stations first used computer technology to obtain

optimum cutting patterns of unedged and untrimmed lumber.

Later, optimization techniques were used at log breakdown

stations where the optimum log breakdown patterns are

determined.



2

Raw Material Storage

Debark! Buck

Log Breakdown

Edging

Trimming

Rough Green Lumber
Grading/Sorting/Stacking

Rough Green Lumber
Storage

Drying

Dry Storage

Figure 1.1. Flow chart of a typical sawmill



3

Unstacking

Planing

Grading and Trimming

Sorting

Stacking and Packing

Storage

Ship

Figure 1.2. Flow chart of a typical planer mill

The use of optimization techniques at the log

breakdown stations has evolved from initially considering

only one geometric dimension of a log to where three

dimensional features of the log are now being considered.



4

However, currently available models make their "optimum"

decisions based only on the log's or the board's

geometrical dimensions. An important aspect ignored by the

optimization models is the effect of log internal defect

type, size, and location on final lumber grade (i.e.,

recovery and quality). The presence of defects needs to be

considered in all manufacturing steps, especially in the

log breakdown, edging and triimning operations. More value

recovery is expected if an optimization system can combine

both geometrical and quality information, and predict

lumber grade when making cutting decisions.

Natural log characteristics and manufacturing

imperfections have a great impact on lumber quality. To

provide specified quality levels for end-use, lumber must

be graded according to its quality characteristics. Lumber

grading is performed in a sawmill or a sawmill and planing

mill, and is monitored by authorized agencies who publish

grading rules which conform to the American Softwood Lumber

Standards (ASLS, 1970). Traditionally, lumber grading is

performed by trained human graders through visual

inspection. Although computer aided manufacturing

techniques have been widely used in modern sawmills,

development of an automated lumber grading system was not

initiated until recently because of the complexity of

grading rules and the difficulties in the detection and

recognition of defects using a machine vision system.



5

A lumber grading software package can be invaluable to

the lumber manufacturing industry in the following aspects:

1. It can be linked to a log breakdown optimization

program, such as SAW3D (Zeng, 1991), to improve lumber

recovery. As mentioned before, lumber breakdown

optimization programs have been used extensively in

sawmills. After receiving log scanning data, the

optimization system can determine optimum sawing, edging

and trimming patterns to find the best lumber recovery from

the given log. Nevertheless, such optimization systems do

not consider external and internal defects of the log. As

such, the resulting solution is not optimal. From a value

stand point, an advanced log breakdown optimization system

that considers log defects in addition to other

characteristics will be able to provide better solutions.

The lumber grading program can be embedded in a log

breakdown program (such as SAW3D).

2. It can be coupled with a machine vision system to

make an on-line automatic lumber grading system. An on-

line automatic lumber grading system consists of two basic

components, a machine vision system which performs as human

eyes to capture the size and quality characteristics of

lumber, and a grading software program which uses the

information provided by the machine vision system to

determine the grade of lumber. An integrated system should

be able to provide more reliable and accurate grading



[.1

results than a human grader, and it should be more

economical than using .a human grader.

3. It can be used as an instruction tool to aid in

training novices and upgrading knowledge of grading

personnel.



7

CHAPTER 2

RESEARCH OBJECTIVES

The objective of this research was to develop an

expert system for softwood lumber grading. The primary

objective of developing this system was to link it to a log

breakdown optimization system, SAW3D (Zeng, 1991), so that

the integrated system can consider both log geometric and

quality information when making sawing, edging, and

trimming decisions to optimize lumber recovery.

Grading rules used in this system were based on

"Western Lumber Grading Rules 88" published by the Western

Wood Products Association. The research focused on

feasibility analysis, and the conceptual and structure

design of the system. Therefore, only selected grades and

wood characteristics were included in the system. When

SAW3D makes optimum log breakdown decisions, some defects

such as manufacturing imperfections do not exist. Such

defects were not included in the grading system.



8

CHAPTER 3

LITERATURE REVIEW

Although the significance of computer lumber grading

programs has been recognized and some researchers developed

programs for lumber grading in the early 1970's (Hallock

and Galiger, 1971), such programs have not been extensively

used in industry. One possible reason for lack of use is

that a prerequisite for a defect detection system is a

machine vision system for lumber grading. Developing a

reliable machine vision system is a difficult task. As

research on machine vision has evolved, activity in the

development of lumber grading programs has also increased.

Some prototype automatic grading systems have been recently

developed in the sawmill industry.

Currently available computer grading programs have

been developed for hardwood lumber grading, and use grading

rules established by the National Hardwood Lumber

Association (NHLA). There are two basic problems that

exist while grading lumber: 1) the recognition of the type

and seriousness of each defect, and 2) the calculation of

usable areas. In grading hardwood lumber, all grades are

based on the recognition of defects and the calculation of

usable area.

The focus of this research is softwood lumber, for

which little has been developed in way of automatic grading



systems. In contrast to hardwood lumber, softwood lumber

grading is primarily based on the recognition of defects;

only grades of Factory Lumber need to calculate usable

area. Since grading problems for hardwood lumber and

softwood lumber are very similar except for the differences

in using different grading rules and grading procedures,

it is worthwhile to review some of the research for

hardwood lumber grading.

3.1 Programs Using Algorithmic Approach

Most of the early programs were developed using an

algorithmic programming approach. Although the expert

system approach has an obvious advantage over the

algorithmic approach, lack of good development tools and

the requirement of expensive computing facilities prevented

use of the expert system approach until recently.

Hallock and Gauger (1971) developed a computer

program for grading hardwood lumber. The program analyzes

the board with its defects, finds the number of cuttings of

the sizes permitted, and assigns a grade according to NHL

rules. It classifies lumber according to six major

classifications. The board and all its defects are

described by rectangles in a Cartesian coordinate system;

the X-Y coordinates of those rectangles are entered on a

series of cards. Once the program receives the input data,



10

it begins the grading process by assigning the board a

potential grade based on the board size. This is the

highest possible grade for the board without considering

defects. Then the defects are checked and the board is

assigned a potentially lower grade because of the impact of

defects. Next, sizes and number of usable areas are

calculated, and the board is assigned the final grade. The

program was developed in FORTRAN. The authors believed the

program's accuracy to be well within NHLA tolerances.

However, some limitations exist. For example, grading rules

and procedures were all coded into the program algorithms.

This programming practice makes modification and expansion

of grading rules difficult. Additionally, the algorithms

considered only one face of the board, whereas both faces

of a board are evaluated in the actual grading process.

Klinkhachorn et al. (1988) created a hardwood lumber

grading program as part of an automated lumber process

system. The program was implemented using FORTRAN 77 for an

IBM PC or compatible. After reading in the board

dimensions and defect types and their location on each

face, the program assigns a potential grade to each face.

Next the program calculates usable area to determine the

final grade assigned to the board. This program improved

upon the previous efforts by considering both faces of the

board and applying a more flexible grading procedure which

provides the option of admitting or excluding certain types



11

of defects when searching for cuttings. In addition, the

program uses the BLOCK DATA feature of FORTRAN 77 to

logically separate grading rules from the program logical

code so that reconfiguration of the grading rules is

easier. When grading rules need to be modified, only the

data specified in the BLOCK DATA segment need to be

changed.

Klinkhachorn et al. (1989) presented a computer-aided

instruction tool, named HaLT (Hardwood Lumber Training) for

training hardwood lumber graders. The HaLT system

incorporates a graphic user interface and explanation

facility in the hardwood lumber grading program discussed

above. The interface provides graphic views of the board

including each side of the entire board with defects or the

expanded view giving the detailed section of the board,

asks questions, compares answers, and provides reasoning.

The program runs on an IBM PC or compatible. The objective

is to provide the lumber industry with an interactive

learning tool to educate and train their graders on-site.

Klinkhachorn et al. (1991) improved their previous

program (Klinkhachorn et al, 1988) by converting the code

from FORTRAN 77 to C, and restructuring it. The grading

rules for various species, the mechanical dimensions of a

board, the violations of the various rules, and the grade

of the board are all mathematically represented in terms of

program variables in data structure so that the program



12

data are logically separated from the program code. This

enhances the program's capabilities, its adaptability and

portability, and makes the restructuring of grading rules

easier.

3.2 Programs Using Expert System Approaches

3.2.1 Structure of Expert Systems

An expert system is a computer program that uses

knowledge, facts, and reasoning techniques to solve

problems that normally require the abilities of human

experts. Expert system architectures include different

components. However, certain components are common to most

expert systems. These include a user interface, a

knowledge base, and an inference engine. Figure 3.1 shows

the basic structure of an expert system.

The interface provides the communication between the

user and the system. It allows the user to query the

system, enter facts about a specific situation that is

relevant to the system's subject domain, and receive

advice, consultation and explanation. Many expert systems

also accept new knowledge through the user interface.

The knowledge base contains expert-level knowledge on

the system's subject domain. The knowledge can be obtained

from books and publications, and/or from one or more human

experts, and is stored using a knowledge representation



13

scheme.

The inference engine provides the reasoning

capabilities of the system. It uses the domain specific

knowledge in the knowledge base and facts provided by the

user to infer new facts that may be conclusions to goals

and subgoals as required by the system. The inference

method is built into the inference engine, while the domain

specific knowledge is contained in the knowledge base. The

general problem solving knowledge and the domain specific

User

User Interface

Inference Engine

Knowledge Base

Figure 3.1. Basic structure of an expert system



14

knowledge are separated, thus the same inference engine can

be used to reason with different knowledge bases.

In addition to those basic components, many expert

systems have various supplemental components, including

knowledge acquisition facilities, knowledge-base

maintenance and update facilities, explanation facilities,

self-training facilities, and validation tools.

Expert system techniques are very appropriate to

problems for which human expertise is required, which. need

symbolic reasoning but cannot be easily handled numerically

or algorithmically, or for which information required for

a solution is incomplete or uncertain.

The lumber grading problem is declarative rather than

algorithmic in nature. The grading process involves

symbolic rather than numeric reasoning. Maintaining and

updating grading rules is required as criteria change over

time, and from one mill to another. In some cases, data

about a defect may be uncertain. If used as an instruction

tool, explanation on the actions and conclusion would be

necessary. All these facts indicate that while an

algorithmic approach is possible, the expert system

approach is more appropriate. Using the expert system

approach for lumber grading has the following advantages

over the algorithmic approach:

1. In knowledge-based systems, lumber data and grading

rules are stored in a knowledge base which is decoupled



15

from the program logic. Hence, modification and expansion

of grading rules can be made easily without affecting the

program logic.

2. When used as a computer-aided instruction (CAl)

tool, knowledge-based systems are more appropriate because

they provide consultation and explanation capabilities

required in training situations.

3. Rules-of-thumb used by grading experts can also be

incorporated in the knowledge base to supplement the rules

from publications, and to speed up the grading process as

these may result in reducing the search space. Furthermore,

due to the separation of the knowledge base from the

program logic, customization of such rules can be easier.

3.2.2 Expert Systems in Lumber Grading

Huang and sparrow (1989) developed a CAl tool for

hardwood lumber grading using the expert system approach.

The expert system consists of five main components: a

knowledge base, an inference engine, a procedure for

calculating the usable area, a user interface facility, and

an explanation subsystem. The knowledge base contains facts

and rules. Facts include lumber data, current data, and

information derived from the rules. Rules are the

conditions and restrictions for judging the grades of the

boards. Knowledge is represented by rules and working

memory elements. A forward-chaining inference mechanism is



16

used. The system was implemented using the LISP language

and runs on the VAX 11/780 computer. The program

facilitates the training of professional hardwood lumber

graders by providing consultation and explanation of lumber

grading procedure in an interactive question-answer

process. Lumber data, including lumber dimensions, types

and sizes of defects, need to be entered by the user. The

program instructs the user in three steps: 1) to judge

grades of the two faces based on a combination of defect

types and sizes; 2) to judge grades of the two faces based

on calculation of usable areas on both faces; and 3) to

judge the overall grade of the board. At each step, it

asks the user to determine a tentative grade, and then

explains the user's mistakes, if any. Otherwise, it

provides an explanation of the process of reaching the

correct grade. The program reduces the training costs by

decreasing the need to rely solely on an expensive human

instructor, and increases the speed and retention rates of

students because it offers a flexible time table that makes

it more convenient for the student to work at his/her own

pace. The knowledge base is limited in the system. Among

over 100 types of hardwood lumber, only two types (walnut

and poplar) were chosen for this initial work.

Softwood lumber grading programs using the expert

system approach have not been developed.



17

3.3 Systems Developed in Industry

As mentioned before, lumber recovery optimization

systems based on dimension scanning have been widely used

in sawmills. While more complete geometric information is

included in recent three-dimensional scanning and

optimization systems, higher recovery is to be expected if

the defect information along with the dimensional data is

used in making cutting decisions. Though some corporations

have been developing grade scanning and optimization

systems which combine both dimension and quality

information for logs and lumber in making optimum

decisions, there is little information available in

published literature. In addition to the difficulty of

accessing information from industry, available papers focus

on defect detection systems, as compared to implementing

grading rules in their programs. This is not surprising

since the development of real time defect detection systems

has been a long time barrier in this area.

VisionSmart Inc. of Canada developed a Grade-based

Optimizer system which uses a camera to scan all features

related to dimension and wane, and an X-ray sensor array to

detect knots and other density related features such as

compression wood and pitch pockets (Flatman and Bodell,

1989; Flatinan and Kenway, 1991; Kenway, 1990; Aune,

1991). Other features such as shakes, checks, holes, and



18

stains are detected by cameras operating under different

lighting than the lighting for the camera used for

dimension and wane. The system combines both profile and

defect data and then determines optimum cutting solutions

for maximum value return. The system can be used either

for green lumber or for dry lumber.

A Grade-Based Edger Optimizer was installed in

MacMillan Bloedel's Alberni Pacific Division sawmill in

March, 1989. The Edger Optimizer scans wet green lumber

for grading features and based on the scanning data, uses

a dynamic programming algorithm to determine an optimum

edging solution. The Edger Optimizer showed a $160,000

return per month, as measured by the mill. A Grade-Based

Trimmer Optimizer was also installed in CANFOR'S Fort St.

John Mill in May 1990, where dry lumber is scanned after

planing and the optimum trimming solutions calculated.

The Coe Manufacturing Company developed the D*TEC

GradeScan system that is able to detect knots, splits,

pockets, stain and decay, as well as wane and dimensions1.

The edger, trimmer and planer mill applications

manufactured by the company have been equipped with the

GradeScan system so that both defects and shape can be

considered when optimizing for value recovery.

The Coe Manufacturing Company. 1991. Company
brochure. Portland, Oregon.



19

CHAPTER 4

SYSTEM DEVELOPMENT

4.1 Problem Domain

4.1.1 Softwood Lumber Grading Practices

Lumber grading is performed in a sawmill or a planer

mill by human graders. The graders evaluate each piece of

lumber and make their decision according to grading rules

as specified by regional association, proprietary, or

national grading rules. The following çiescriptions are

given in Western Lumber Grading Rules 88 (1988):

The grading of lumber cannot be considered an

exact science because it is based on either a

visual inspection of each piece and the judgment

of the grader or on such visual inspection and

judgment combined with the results of method of

mechanically determining the stiffness

characteristics of structural lumber. The

results are, however, sufficiently explicit to

establish a maximum of 5 percent below grade as

a reasonable variation between graders.

In general, American Standard softwood lumber is

classified according to use, manufacturing, and size



20

classifications. Grading of lumber is based upon a visual

inspection and is primarily a judgment of end-use strength

or appearance. Natural log characteristics which have an

effect upon strength are, taken into account along with

manufacturing imperfections. In most cases these include

limiting provisions on characteristics such as stain,

checks, grain, skip, cup, crook, wane, twist, knots, holes,

pitch, pocket, split, steaks and patches, shakes, decay,

and cutout. Definitions and detailed descriptions of these

characteristics can be found in the Western Lumber Grading

Rules 88 (1988).

Grading rules applied to softwood lumber for domestic

market are governed by National Bureau of Standards

Voluntary Product Standard PS 20-70 (American Softwood

Lumber Standard, 1970). All of the grading rule books

which are published by grading organizations are based on

this standard. The two largest grading organizations on

the West Coast are the Western Wood Products Association

(WWPA) and the West Coast Lumber Inspection Bureau (WCLIB).

In 1986, nearly 35 percent of the nation's softwood lumber

production was graded under WWPA supervision. This

research project will use the grading rules published by

the WWPA as described in the Western Lumber Grading Rules

88 (1988).



21

4.1.2 Grading Categories

According to the WWPA, lumber is classified into one

of the following five categories (Table 4.1) based on the

size and end-use: Select and Finish lumber, Boards,

Dimension, Timbers, and Factory lumber and related

products.

Table 4.1. Gradinci cateaories

Name of

category

Nominal Size Judgment
on Use
EmphasisThickness Width

Selects
and
Finish______________

3/8" - 16/4" >= 2" appearance

Boards 3/4" - 16/4" >= 2" appearance

Dimension 2" - 4" >= 2" strength

Timber >= 5" >= 5" strength

Factory any any remanufac-
turing

Each major category is further grouped into many sub-

categories and each sub-category includes many grades.

Since the focus of this project is prototype development,

it considers only grading rules for Select/Finish, Boards,

and Dimension categories. Figure 4.1 show these categories

and their sub-categories, and the grades selected for this

project implementation.



22

Grading Category 8ub-categorv Grade

B&BTR - l&2 CLEAR
Selects C SELECT

D SELECT
Selects and Finish

SUPERIOR FINISH
h PRIME FINISH

E FINISH

Boards

Dimension

1 COMMON
2 COMMON

Commons 3 COMMON
4 COMMON
5 COMMON

SELECI MERCHJNTABLE
Alternate CONSTRUCTION
Board STANDARD
Grades UTILITY

ECONOMY

CONSTRUCTION
Light STANDARD
Framing UTILITY

ECONOMY

SELECT STRUCTURAL
Structural NO. 1
Light NO. 2
Framing NO. 3

ECONOMY

Stud STUD
I ECONOMY

Figure 4.1. Grading categories, sub-categories, and grades
included in this project



23

Characteristics and limiting provisions for each grade

include wood defects and manufacturing imperfections, and

their type, size, location and distribution. Tables 4.2 to

4.8 give the characteristics for each grade as described by

the rule book, and show the characteristics included in the

scope of this project. There are two primary reasons for

focusing on a subset of the grading rules. First, in many

applications, not all characteristics listed in the rule

book appear on lumber. At this initial development stage,

only the most frequently occurring characteristics are

included in the system. The focus was on conceptual design

and testing the feasibility of the expert system approach.

The system design is such that the knowledge base can be

easily refined and extended. Second, the system is planned

to be used as a component embedded into SAW3D (Zeng, 1991),

a log breakdown optimization system. When SAW3D makes

optimum breakdown decisions, manufacturing imperfections

are not included. For example, warp, such as bow, crook,

cup and twist, are not included.



24

Table 4.2. Grade characteristics considered by the project
for grades in Select/Finish's Select sub-
cateciory

Character-
istics

Grade (Selects/Finish - Selects)

B&BTR - l&2
CLEAR

C SELECT D SELECT

Stain Y Y Y

Checks Y Y Y

Splits N/A - Y N/A - Y Y

Torn Grain N N N

Skips N N N

Cup N N N

Crook N N N

Wane Y Y Y

Twist N N N

Knots Y Y Y

Holes N/A-Y Y Y
Pitch Y Y Y

Pockets Y Y Y

Pitch
Streak_____________

N/A N N

Cutout N/A N/A N

Y Listed in the rule book and considered by the
project

N : Listed in the rule book, but not considered by
the project

N/A : Not listed in the rule book and not considered
by the project

N/A - Y : Not listed in the rule book, but considered by
the project



25

Table 4.3. Grade characteristics considered by the project
for grades in Select/Finish's Finish sub-
cateciory

Character-
istics

Grade (Selects/Finish - Finish)

SUPERIOR
FINISH

PRIME
FINISH

E FINISH

Stain Y Y Y

Checks Y Y Y

Splits Y N/A - Y Y

Torn Grain N N N

Skips N N N

Cup N N N

Crook N N N

Wane Y Y Y

Twist N N N

Knots Y Y Y

Holes N/A-Y Y Y
Pitch Y Y Y

Pockets Y Y Y

Pitch
Streak_____________

N N N

White
Speck____________

N/A N N

Cutout N N N

Chip Marks N N N

Y : Listed in the rule book and considered by the
project

N : Listed in the rule book, but not considered by
the project

N/A : Not listed in the rule book and not considered
by the project

N/A - Y : Not listed in the rule book, but considered by
the project



Table 4.4. Grade characteristics considered by the project
for qrades in Boards' Commons sub-cateqory

Character-
istics

Grade_(Boards - Commons)

1
COMMON

2

COMMON
3

COMMON
4

COMMON
5

COMMON

Stain Y Y Y Y Y

Checks Y y y y y

Torn/Raised
Grain_______

N N N N N

Skips N N N N N
Cup N N N N N

Crook N N N N N
Wane Y Y Y Y Y

Twist N N N N N

Splits y y y y y

Streaks
and Patches

N/A N N N N

Pitch Y Y Y Y y

Pockets Y Y Y Y Y

White Speck
Honeycomb

N/A N/A N N Y

Unsound N/A N/A N N Y

Shake N/A N N N N
Roller
Checks

N/A N N/A N N

Pith Y Y N/A N/A Y
Knots Y Y Y Y Y

Holes N/A-Y Y Y Y Y
Y : Listed in the rule book and considered by the

project
N : Listed in the rule book, but not considered by

the project
N/A : Not listed in the rule book and not considered

by the project
N/A - Y : Not listed in the rule book, but considered by

the project



27

Table 4.5. Grade characteristics considered by the project
for grades in Boards' Alternate sub-category

Character-
istics

Grade (Boards - Alternate)

SEL-
MERCH

CONST STAND UTIL ECONOMY

Stain Y Y Y Y Y

Checks Y Y Y Y Y

Torn/Raised
Grain_______

N N N N N

Skips N N N N N

Cup N N N N N

Crook N N N N N

Wane Y Y Y Y Y

Twist N N N N N

Splits Y Y Y Y Y

Streaks and
Patches

N N N N N

Pitch Y N/A N/A N/A Y

Pockets Y Y y y y

White Speck
Honeycomb________

N N/A N N Y

Unsound
Wood

N N/A N/A N Y

Shake N/A N/A N N N

Roller
Checks

N N/A N/A N/A N

Pith N/A N/A N/A N/A Y

Knots Y Y Y Y Y

Holes Y Y Y Y Y

Burls N/A N/A N/A N N
Y : Listed in the rule book and considered by the

proj ect
N : Listed in the

the project
N/A : Not listed in

by the project

rule book, but not considered by

the rule book and not considered



28

Table 4.6. Grade characteristics considered by the project
for grades in Dimensions' Light Framing sub-
cateaorv

Character-
istics
______________

Grade (Dimension - Light Framing)

CONST STAND UTIL ECONOMY

Checks Y Y Y y

Knots Y Y Y Y

Holes Y Y Y Y

Manufacture N N N N

Pitch Y Y Y Y

Pitch Streaks N N N N

Pockets Y Y y y

Shake N N N N

Skips N N N Y

Slopeof y y y y

Splits Y Y Y Y

Stain N N N N

Wane Y Y Y Y

Warp N N N N

Unsound Wood N/A -Y Y I Y

White Speck N/A N N N

Y : Listed in the rule book and considered by the
project

N : Listed in the rule book, but not considered by
the project

N/A : Not listed in the rule book and not considered
by the project

N/A - Y : Not listed in the rule book, but considered by
the project



29

Table 4.7. Grade characteristics considered by the project
for grades in Dimensions' Structural Light
Framing sub-category

Character-
istics

Grade (Dimension - Structural Light
Framing)

SEL-
STR

NO.1 NO.2 NO.3 ECONOMY

Checks Y Y Y Y Y

Grain N N N N/A N

Knots Y Y Y Y Y

Holes Y Y Y Y Y

Manufacture N N N N N

Pitch Y Y Y Y Y

Pitch
Streaks

N N N N N

Pockets Y Y Y Y Y

Shake N N N N N

Skips N N N N N

Slopeof
Grain_______

Y Y Y Y Y

Splits Y Y Y Y Y

Stain N N N Y Y

Wane Y Y Y Y Y

Warp N N N N N

Unsound
Wood

N/A-Y N/A-Y Y Y Y

White Speck N/A N/A N N N

Y : Listed in the rule book and considered by the
project

N : Listed in the rule book, but not considered by
the project

N/A : Not listed in the rule book and not considered
by the project

N/A - Y : Not listed in the rule book, but considered by
the project



30

Table 4.8. Grade characteristics considered by the project
for grades in Dimensions' Stud sub-category

Character-
istics

Grade (Dimension - Stud)

STUD ECONOMY

Checks Y Y

Grain N/A N/A

Knots Y Y

Holes Y Y

Manufacture N N

Pitch Y Y

Pitch Streaks N N

Pockets Y Y

Shake N N

Skips N N

Slope of
Grain

Y Y

Splits y y

Stain N Y

Wane Y Y

Warp N N

Unsound Wood Y y

White Speck N N

Y : Listed in the rule book and considered by the
project

N : Listed in the rule book, but not considered by
the project

N/A : Not listed in the rule book and not considered
by the project

N/A - Y : Not listed in the rule book, but considered by
the project



31

4.2 Knowledge Acquisition and Representation

4.2.1 Knowledge Acquisition

Knowledge acquisition deals with the issue of

acquiring information about the problem domain. Knowledge

used in problem solving may be based on experiences of one

or more human experts, case histories, reference sources,

etc. The key source of knowledge for this project is the

grading rules. These are based upon the "Western Lumber

Grading Rules 88" published by the Western Wood Products

Association (1988).

Compared with many other expert systems, the knowledge

acquisition task for this project is relatively simple

because most grading rules are well defined in the rule

book. However, the rule book cannot give all the

information required to perform the grading task accurately

and efficiently. First, some of the rules are not

explicitly defined. For example, certain grades allow

"several" as the number of certain defects. However,

"several" is not precisely defined. Second, the rule book

just stipulates characteristics and limiting provisions,

but not the grading procedures. In practice, grading

procedures affect the speed with which lumber is graded.

For instance, knots are the most frequently encountered

defects. Looking for frequently encountered defects

accelerates the grading process. Third, as mentioned



32

before, grading is more of an art than a science, and the

results depend on the inspection and judgement of the

grader. Five percent variation from standards is accepted

as a reasonable difference between graders. Some mills and

markets may use their local rules, and take advantage of

the allowed variations. This may involve specific

knowledge not documented in the rule book. Thus, the

knowledge base using grading rules from the rule book can

be refined based on actual grading practices.

4.2.2 Knowledge Representation

The domain knowledge must be well organized and

structured in the knowledge base, thus allowing the

inference mechanism of the expert system to operate on it

effectively and efficiently. There are many ways to

represent domain knowledge, including production rules,

frames, semantic networks, and first-order logic. Perhaps

the most frequently used representation scheme is

production rules.

Production rules represent knowledge in the following

form:

IF (list of antecedents)

AND

OR

NOT

THEN (list of consequents or actions).



33

This format is almost identical to the description of

grading rules stipulated in the rule book. Thus, the

scheme of production rules is intuitively the best

representation scheme for the domain knowledge for this

project.

The grading rules are of the following form:

IF (list of characteristics permitted and limiting

provisions)

THEN Grade x.

For example, consider the CONSTRUCTION grade in the Light

Framing sub-category of dimension lumber. The

characteristics permitted and limiting provisions for this

grade are summarized in Table 4.9. These can be listed as

the condition part of a production rule; the conclusion of

the rule will be grade CONSTRUCTION if all the conditions

are satisfied. However, such a single rule containing a

description of all the characteristics will be inefficient

because the inference process will be long. A better

representation mechanism is to break down a single rule

into several smaller rules where each component rule

focuses on a distinct characteristic. Since the example in

Table 4.9 has eleven components, the 11 rules shown in

Table 4.10 may be used to represent the CONSTRUCTION grade.

Finally, Rule 12 integrates the conclusions derived from



34

the individual sub-components.

In this illustrative example, not all characteristics

are included in the rules. For instance, spike knots and

stains are not considered. Also, streaks and pockets are

not listed because they have no impact on the grading

result. Some rules, such as Rules 5, 7, and 11, need more

details or other higher level rules to provide the

antecedents.



35

Table 4.9. Characteristics permitted and limiting
provisions for CONSTRUCTION grade in Light
Framing sub-category of dimension lumber

Character- Provisions
istics

Checks Surface seasoning checks not limited.
Through checks at ends are limited as
splits.

Knots Sound, firm, encased and pith, must
be tight and are permitted in the
following sizes or their equivalent
displacement:

Unsound or
Nominal Anywhere Loose Knots
Width On Wide Face and Holes

2" 3/4" 5/8"
3" l&]./4" 3/4"
4" l&l/2" 1"

Knots spiked entirely across the wide
face are limited to a displacement of
approximately 1/4 the cross section.

Nanufacture Standard "E".

Pitch and Not limited.
pitch steaks

Pockets, pitch Not limited.
or bark

Shake Several heart shakes up to 2' long,
similar to seasoning checks, none
through.

Skips Hit and miss on 10% of the pieces.

Slope of grain 1 in 6.

Splits Equal in length to the width of the
piece.

Stain Stained sapwood. Firm heart stain or
firm red heart.

Wane 1/4 the thickness, 1/4 the width. 5%
of the pieces may have wane up to 1/2
the thickness and 1/3 the width for
1/4 the length.

Warp 1/2 of medium.



Table 4.10. Grading rules for the CONSTRUCTION grade
represented usinq production rules

Rule 1:
IF NO through checks at ends

OR longest through check at end <= lumber width
THEN Checks satisfy grade CONSTRUCTION

Rule 2:
IF NO firm knots on face

OR (lumber width = 2"
AND diameter of largest firm knot <= 0.75")

OR (lumber width = 3"
AND diameter of largest firm knot <= 1.25")

OR (lumber width = 4"
AND diameter of largest firm knot <= 1.5")

THEN firm knots satisfy grade CONSTRUCTION

Rule 3:
IF NO loose knots on face

OR (lumber width = 2"
AND diameter of largest firm knot <= 5/8")

OR (lumber width = 3"
AND diameter of largest firm knot <= 3/4")

OR (lumber width = 4"
AND diameter of largest firm knot <= 1")

THEN loose knots satisfy grade CONSTRUCTION

Rule 4:
IF NO holes on face

OR (lumber width = 2"
AND diameter of largest hole <= 5/8")

OR (lumber width = 3"
AND diameter of largest hole <= 3/4")

OR (lumber width = 4"
AND diameter of largest hole <= 1")

THEN holes satisfy grade CONSTRUCTION

Rule 5:
IF manufacture standard is "E"
THEN manufacture satisfies grade CONSTRUCTION

Rule 6:
IF NO shakes

OR (NO through shakes
AND length of largest shake <= 2')

THEN shakes satisfy grade CONSTRUCTION



37

Table 4.10. (Continued)

Rule 7:
IF NO skips

OR hit and miss on <= 10% pieces
THEN skips satisfy grade CONSTRUCTION

Rule 8:
IF slope of grain <= 1/6"
THEN slope of grain satisfies grade CONSTRUCTION

RULE 9:
IF NO splits

OR length of longest split <= lumber width
THEN splits satisfy grade CONSTRUCTION

RULE 10:
IF NO wane

OR (thickness of wane <= 1/4 lumber thickness
AND width of wane <= 1/4 lumber width)

THEN wane satisfies grade CONSTRUCTION

RULE 1.1:
IF NO warp

OR warp <= 1/2 medium
THEN warp satisfies grade CONSTRUCTION

RULE 12:
IF checks satisfy grade CONSTRUCTION

AND firm knots satisfy grade CONSTRUCTION
AND loose knots satisfy grade CONSTRUCTION
AND holes satisfy grade CONSTRUCTION
AND manufacture satisfies grade CONSTRUCTION
AND shakes satisfy grade CONSTRUCTION
AND skips satisfy grade CONSTRUCTION
AND slope of grain satisfies grade CONSTRUCTION
AND splits satisfy grade CONSTRUCTION
AND wane satisfies grade CONSTRUCTION
AND warp satisfies grade CONSTRUCTION

THEN grade is CONSTRUCTION



38

4e3 Inference Mechanism

Given a knowledge base, an expert system performs

inference or deduction to arrive at conclusions. For rule-

base systems, the logic deduction is based on the following

modus ponens:

Given the rule: IF antecedent 1

AND antecedent 2

AND antecedent N

THEN conclusion 1

and facts: antecedent 1

antecedent 2

antecedent N,

it is possible to infer that conclusion 1 is true.

Facts and rules are basic components of the knowledge base.

The reasoning process involves looking for facts and using

rules to arrive at conclusions.

Inference strategies used in expert systems include

backward chaining, forward chaining, breadth-first search,

depth-first search, heuristic search, problem reduction,

pattern matching, hierarchical control, unification, and

event-driven control. Selection of a proper strategy for



39

an expert system depends on the knowledge representation

scheme used in the system. Matching the task to the

correct combination of knowledge representation and

inference strategy is a complex task. For rule-based

systems using production rules as the representation

scheme, the major inference strategies used are backward

chaining, forward chaining, or a combination of these two

strategies. These strategies are primarily used to specify

how rules contained in a knowledge-base are to be

processed.

Backward chaining is also called goal-driven

inference. It begins with the selection of a goal using

some selection criterion, then tries to prove that the goal

is true. A goal is the conclusion of a rule. The process

usually starts with matching existing facts with

antecedents of that rule. If the required facts cannot be

found, the inference engine then looks for rules that may

have the required facts as their conclusions. This may

direct the search process to sub-goals. If the required

facts still cannot be found, the system may begin a query

process if the facts can be obtained from the user.

Otherwise the system will select another goal and repeat

the process until a solution is found or there are no more

goals to be checked.

To illustrate how this strategy works, assume that

there is a piece of lumber 2 inches in nominal thickness,



40

4 inches in width, and 10 feet long. There are no defects

on its face except for a few well spaced firm knots.

Assune that the largest knot has a diameter of 0.5 inches.

The rules in Table 4.10 show that the grade of this piece

of lumber is CONSTRUCTION in the light framing category of

dimension lumber. The backward chaining process starts by

selecting a grade from the 27 grades listed in Figure 4.1,

and then tries to prove that this grade is the appropriate

one. Assume that the grade selected is CONSTRUCTION. The

system first looks for facts that are antecedents of Rule

12. Since these facts do not exist at this stage, it

looks for rules 1 through 11 which have those facts as

conclusions and tries to prove that rules 1 through 11 are

true. To prove these rules, additional information is

required. This can be obtained from the user, or by a

search for more rules. Once rules 1 through 11 are proven

to be true, all antecedents in Rule 12 are generated a.s new

facts. This in turn proves that Rule 12 is true and the

grading process is completed. If the selected initial

grade is not CONSTRUCTION, the process may fail to find a

solution; another grade may then be selected, and the

process is repeated until a grade is found.

This strategy is best suited for the problems with

many initial facts but relatively few solutions. It has

been successfully used for diagnostic problems. However,

for the lumber grading problem, it is not the appropriate



41

method. One reason is that there are a large number of

characteristics that need to be checked before a grade can

be determined. It can be very inefficient if a wrong grade

(a goal) is assumed. Another reason is that this inference

mechanism is not analogous to the human grading process

where conclusions are reached based on facts.

The forward chaining inference mechanism, also called

data-driven chaining or reasoning, starts with existing

facts and works on deducing new facts that will eventually

lead to the goal. The inference process works in the

following cycle:

1) Matching rule(s): find all rules whose antecedents

are satisfied by working storage (that is, where

existing facts are stored). The collection of

such rules is called the conflict set.

2) selecting rule(s): When more than one rule matches

the facts, apply a conflict resolution strategy

to determine which rule will be selected among

those rules. Various conflict resolution

strategies have been used for forward chaining

systems, including:

Do One: Select the first rule that matches the

facts.

Do All: Apply all the matching rules in one

batch.



42

Do in Seauence: Apply the matching rules one at

a time.

Do the Most Specific: Select the more specific

rule.

Do the Most Recent: Select the one that matched

a most recently derived fact.

3) Executing rule(s): the selected rule is proved,

also called "fire the rule", and the working

storage is updated with fact(s) of the newly

fired rule.

4) The process is repeated until a solution is found

or no more rules can be applied.

For the lumber example used to illustrate backward

chaining, the forward chaining process will start by

collecting all the defect information, and storing it in

the working storage (memory). It then tries to find all

rules whose antecedents are matched with the existing

facts. For the given data, it will conclude that

antecedents of rules 1 through 11 exist in the working

storage. Thus rules 1 through 11 are placed in the

conflict set. Selection and firing of these 11 rules

result in the facts that match the antecedents of Rule 12;

these are then placed in the working storage. Now Rule 12

is also put into the conflict set because all of its

antecedents are satisfied. Finally, firing Rule 12 results



43

in the CONSTRUCTION grade for the specified input data.

The forward chaining procedure is best used to solve

problems in which data is to be used as the starting point

for problem solving. It is intuitively the most

appropriate method for the lumber grading problem; it also

matches the human grader's reasoning process. The

inference used in this implementation is the forward

chaining inference mechanism.

4.4 Tool Selection

After identifying an appropriate knowledge

representation scheme and inference mechanism, the next

phase in the development process is to select the

implementation software. Four options were considered for

this project: conventional programming languages, knowledge

engineering languages, knowledge-based programming tool

kits, and knowledge-based programming shells.

Conventional languages, such as C, C++ and Pascal, can

be used to implement a knowledge-based system. They

support a broad range of machine platforms. Compared with

other options, they are very flexible and efficient in

terms of execution speed and memory usage. The major

disadvantage is that the development time is long requiring

extremely large amount of coding. This option is obviously



44

not recommended for rapid prototyping and system design.

Knowledge engineering languages, such as LISP and

PROLOG, support symbolic manipulation and/or have built-in

inference engines. They are very powerful and efficient

when used on the right platforms such as LISP machines and

work stations. However, the opposite is true on

microcomputers. Since this project selected microcomputers

as the implementation platform, this option also did not

represent the best choice.

Knowledge-based programming toolkits, such as KEE,

OPS5, ART, Clips and Eclipse, are somewhat like high-level

languages. While relieving the programmer from the burden

of tedious coding by providing built-in knowledge

representation schemes and inference engines, they do

require some programming following their syntax so that

maximum flexibility can be achieved. This option was

considered appropriate for this project.

Knowledge-based shells usually provide a complete

development environments. The programmer does not need to

do any real programming. However, they are usually

expensive and have limited flexibility. This option was

prohibitive due to the cost involved and the flexibility

required by the project, particularly interaction with

external systems.

Several features were required of the development tool

to be used. First, the tool should support rule-based



45

knowledge representation and a forward chaining inference

mechanism. Second, the system should interact with C/C++

programs since the intended use of this product is to

integrate it into SAW3D, a log breakdown optimization

program written in C. Therefore, support of links between

this project and any ANSI C/C++ compiler was considered to

be an important feature of the development tool. Third,

the tool should be run on an IBM-PC and support protected

mode memory access. The knowledge base can potentially be

so large that the 640k conventional memory limitation can

become an obstacle. Fourth, the tool should be financially

feasible. Other considerations included user interface and

explanation facilities, ease of learning and use,

documentation, and technical support.

Considering the above factors, Eclipse, a commercial

rule-based programming tool from The Haley Enterprise (The

Haley Enterprise, 1991), was selected as the

implementation tool for this project. Eclipse supports the

features described above, including rule-based knowledge

representation, a forward chaining inference engine,

running under protected mode and integration with other

applications.

The syntax for Eclipse is derived from Inference

Corporation' s ART (for Automatic Reasoning Tool), and is

compatible with NASA'S Clips (for C Language Integrated

Production System). The basic syntax uses LISP-like prefix



46

notation and parentheses. Eclipse uses the Rete Algorithm,

which is one of the most efficient techniques for

performing complex pattern matching of the type used in

rule-based systems (Forgy, 1979, 1982). Although many

other tools also use the Rete Algorithm, some benchmarks

and applications have proven that Eclipse is superior to

many other tools in terms of execution speed and program

size2 (Brooke, 1992). For example, it runs four times as

fast as Clips on 386 machines and requires roughly one half

the memory of Clips. Eclipse supports more advanced

functionality and better integration, especially on

personal computers. It supports more powerful logical

reasoning capabilities and provides more expressive pattern

matching.

One of Eclipse's best features is its integration

capacity. It supports two-way integration, integrating

other applications into Eclipse, and Eclipse into other

applications. Other applications can be integrated into

Eclipse by calling routines written in the C language, in

some cases even Pascal or FORTRAN. Eclipse toolkits are

provided with run-time libraries and source code necessary

to construct applications which embed Eclipse. It allows

declarative domain and procedural control knowledge to be

explicitly separated and codified using rules or

2. The Haley Enterprise, Inc. 1992. Company brochure.
Sewickley, PA.



47

procedures, retaining more of the advantages of a pure

production system even when the requirement to procedurally

control and embed a knowledge-base is implemented within

any real application.

The Microsoft Windows version of Eclipse has a real-

time WYSIWYG (What You See Is What You Get) development

interface. It provides knowledge base monitoring windows

which can be used to simultaneously view the relations

between facts and goals, and the match states of any number

of rule sets, agenda, rules, joins, patterns, templates,

relations, facts, matches, etc. Each of these views is

supported in real time as commands or rules are executed.

In effect, these modeless dialog windows allow full

manipulation and monitoring of all knowledge base

components, scheduling and pattern matching activities.

The windows version also supports dynamic data exchange

(DDE). This support provides more communication methods

between Eclipse and other applications. Using this

feature, Eclipse commands can be passed from other

applications and Eclipse can issue commands to other

applications. Eclipse can request and respond to requests

for data from other applications. Dynamic link libraries

(DLLs) are also provided in the Windows version. This

research uses the Windows version of Eclipse.



48

4.5 Prototype Development

Expert systems are best developed using an incremental

prototyping methodology. A running prototype should be

developed at an early stage. The prototype deals with a

restricted portion of the problem domain; nonetheless, it

will validate the feasibility and conceptual design of the

system. After the prototype has been constructed, it can

be extended into a fully functional system while its basic

structure is preserved.

4.5.3. Prototype Components

The prototype for this project has four basic

components as shown in Figure 4.2. The user interface

collects data from the user and provides the grading

results. The inference engine is a built-in component of

Eclipse. The knowledge base contains selected grading

rules. The explanation facility consists of a set of rules

for tracing back the facts upon which the grading results

are based.

At the beginning of system execution, the user loads

the knowledge base, and then enters data describing lumber

size and defect characteristics via the user interface.

The inference engine tries to map the input data to one or

more proper grades by performing the reasoning process on

the input data and the knowledge base. Finally, the user



49

interface shows the final grades resulting from the

matching process, along with explanations.

The knowledge base is implemented in several source

files. Each file contains grading rules for a sub-category

of a category (see Table 4.11 for explanation). The user

can load one or more of them, depending on the grading

categories to be worked on. The user interface is

implemented using rules and functions, and is contained in

User Interface

Inference Engine
j

Explanation
Facility

Knowledge Base

Figure 4.2. Basic structure of the prototype



50

a separate file. Also, separate files contain output

printing and some supplemental control strategies. Table

4.11 summarizes the file list used in the system.

When using the Eclipse development toolkits, the

development interface allows the system developer to watch

how conclusions are reached and to monitor all facts in

the knowledge base. Thus, an explanation facility is not

necessary. However, the run-time version of the prototype

does not include those knowledge-base debugging facilities.

Table 4.11. Source files and their contents

File Name Content

Alt.clp Grading rules for ALTERMATE sub-
category of BOARD

Coinmons.clp Grading rules for COMMONS sub-
category of BOARD

Define.clp Definition of relations used to
describe facts

Finish.clp Grading rules for FINISH sub-
category of SELECT/FINISH

Lframe.clp Grading rules for LIGHT FRAMING
sub-category of DIMENSION

Query.clp Rules and functions for query
interface.

Select.clp Grading rules for SELECT sub-
category of SELECT/FINISH

Sframe.clp Grading rules for STRUCTURAL
sub-category of DIMENSION

Stud.clp Grading rules for STUD sub-
category of DIMENSION

System.clp Rules for system control,
explanation, and output printing



51

Instead, a simple explanation facility is coded into a set

of rules contained in the system.clp file to explain how

the conclusions are reached. If used as an instruction

tool, the prototype should be expanded by developing more

detailed explanation features.

4.5.2 Knowledge Base Implementation

The domain knowledge is described by objects,

concepts, and relationships between them. The major

objects in the problem domain are those defects considered

by the system, including various types of stains, checks,

splits, wane, knots, holes, pitch, and pockets. The major

concepts are lumber grades.

As discussed earlier, this system is a rule based

system, in which the knowledge base consists of a

collection of rules completely describing the relationships

between the objects and concepts. In Eclipse syntax, the

general format of a rule is as follows:

(defrule rule_name "optional_comment"
(pattern 1)
(pattern 2)

(pattern N)
=>

(action 1)
(action 2)

(action 1.1))



52

The first line is the rule header; "defrule" is a

keyword indicating that this segment of code is a rule.

Zero or more patterns are specified after the rule header.

Each pattern consists of one or more fields. The patterns

before the arrow symbol, "=>", are antecedents of an IF-

THEN rule, and called the left-hand of the rule. The

actions after "=>" are the THEN part of the IF-THEN rule,

called the right-hand of the rule. Eclipse attempts to

match the patterns of rules against existing facts. If all

the patterns of a rule match facts, the rule is activated

and put into an agenda (also called conflict set). The

agenda is a collection of activated rules. Zero or more

rules may be in the agenda. When more than one rule is in

the agenda, Eclipse applies a conflict resolution procedure

to select a rule to execute, called firing a rule. The

actions following the arrow symbol will be executed when

the rule in the agenda fires. Patterns, or antecedents of

a rule are facts that are required to fire the rule. Facts

are usually thought of as propositions, as in proposition

calculus. A propositional fact is a fact that consists of

a relation and one or more arguments following the

relation. The relation is always in the first place in the

fact with arguments following the relation. The relation

describes how the argument relate to each other. For

instance, a fact may be represented as (dia hole 0.5),

where dia (for diameter) is the relation, hole and the



53

value of 0.5 are arguments. This pattern means that the

hole has a diameter of 0.5 (inches).

To illustrates how a rule is implemented in Eclipse,

consider Rule 2 in Table 4.10. The rule is modified to

accommodate knot description. As stated in Table 4.10, the

rule only considers knot type (firm or loose), but not knot

location (at edge or center line). However, some grades in

the light framing sub-category need to consider both knot

type and location. To keep the query interface consistent,

Rule 2 is modified as follows:

Modified Rule 2:

IF NO firm knots at edge
OR (lumber width = 2" AND diameter of largest

firm knot at edge <= 0.75")
OR (lumber width = 3" AND diameter of largest

firm knot at edge <= 1.25")
OR (lumber width = 4" AND diameter of largest

firm knot at edge <= 1.5")
THEN firm knots at edge satisfy grade CONSTRUCTION of

Light Framing

The rule is implemented in Eclipse as follows:

(defrule CONST-LF-firin-edge-knot
(declare (salience 1000))
(face ?)
(category light-framing)
(face_width ?width)
(or (not (symptom firm_knot_at_edge))

(dia firm_knot_at_edge ?firiudia&: ( or
(and (= ?width 2) (<= ?firmdia 0.75))
(and (= ?width 3) (<= ?firmdia 1.25))
(and (= ?width 4) (<= ?firmdia 1.5)))))

=>
(assert (satisfy firm_edge_knot CONST-LF)))

The name of this rule is CONST-LF-firiu-edge-knot. The

patterns before "=>" look more complicated than the general



54

format described above because many Eclipse-defined

functions are used. A function looks like a fact pattern.

The entire function is surrounded by parentheses. The name

of function precedes the arguments. The arguments for the

function are patterns of facts or other functions. Here

"declare", "or", "not", "and", and "assert" are names of

functions which perform the functions implied by their

names. Any field with the symbol ? means that value of

this field is a variable.

Comparing the modified Rule 2 and its Eclipse

implementation, it becomes apparent that the first four

patterns, or antecedents, within the implementation are not

present in the modified Rule 2. These are supplemental

patterns used for process control. The patterns after

these process control patterns are just the Rule 2

expressed using Eclipse's syntax. An explanation of each

component of the rule is given below.

1. (declare (salience 1000)): assigns a priority

number to this rule. "Salience" is a keyword to indicate

the priority of the rule, followed by a priority value.

The system is driven by facts in the working memory. The

system starts with matching the working memory facts with

antecedents of rules. Once all antecedents of a rule are

matched, that rule is put in the agenda. In many case more

than one rule will be in the agenda waiting for firing.

The system then applies a conflict resolution strategy to



55

fire those rules. However, sometimes priority ordering

other than the Eclipse's predefined control strategy is

necessary; "salience" orders the rule execution sequence.

2. (face ?): This has a special purpose. If both

faces have the same defect characteristic described by a

rule, without (face ?) once the first face has used in a

rule, that rule will not be fired again for the second

face, even if all the facts were present. This is due to

the design of the Eclipse system. When grading different

faces, ? in the pattern of (face ?) will have a different

value. Therefore, even if no other facts in the rule

change, the rule will still be activated because (face ?)

pattern has changed.

3. (category light-framing): is used to prevent

other grading categories from using this rule. This fact

will be asserted into working memory during the query

process when the user selects this category. Without using

this pattern, Eclipse will try to match all antecedents of

all grading rules, no matter what grading category the

system is working on. That would be extremely inefficient.

4. (face_width ?width): is used to introduce variable

?width required in other patterns.

5. The patterns after (face_width ?width) and before

"=>" are just statements of the modified Rule 2 in Eclipse

syntax. They are written in compound logical "or", "not",

and "and" format. The symbol "&:" has a special meaning.



56

The variable "?firmdia" before "&:" is tested in the

calculations following "&:". For example, in the fact (dia

firm_edge_knot 0.5), the ?firmdia is 0.5 If the lumber

width is 2 inches, the ?firmdia will be tested to see

whether the diameter of the knot is less than 0.75. The

result will be true, and the system considers that the

antecedent has been matched by the fact. If all the facts

on the left-hand side of the rule are present in the

working memory, the rule will be put on the agenda, and the

new fact (satisfy firm_edge_knot CONST-LF) will be asserted

into the working memory.

6. In the pattern of (satisfy firm_edge_knot CONST-

LF), "satisfy" describes the relationship between the knot

and the grade. Argument firm_edge_knot is an abstracted

object, which is a combination of a knot object and its two

attributes, type and location. Alternatively, the knot

object can be represented using knot-type-value and knot-

location-value triples, or using a knot template structure

with type and location as its attributes. As a matter of

fact, Eclipse has a "deftemplate" representation to support

this kind of structure. Using deftemplate to represent a

knot object was tried during prototyping, but was less

preferable because of some inconvenience encountered in

fact manipulation.

All grading rules in the knowledge base have a similar

structure. The query interface, system output, and system



57

controls are all implemented using such rules. To

illustrate, Appendix A gives the Eclipse implementation of

all grading rules for the Structural Light Framing

category.

4.5.3 Grading Process, Input, and Output

The system is menu-driven and interacts with the user

through menus and a series of dialog boxes. The input

process begins with loading the knowledge base and

selecting the input method (interactively or loading a file

containing facts). If the interactive input method is

selected, a series of dialog boxes will guide the user to

enter lumber size, select grade category and sub-

categories, and specify the face where defects are located

along with the defects' types, locations and sizes. The

system then infers the grade corresponding to this face.

Once the grade has been determined, the system allows the

user to ask for explanation. The explanation provided by

the system includes types of defects on the face, size of

each defect, the grade each defect has satisfied, and the

defects that determine the grade.

The system determines the grade for each face

separately and the overall grade. The flow chart in Figure

4.3 shows the grading process. Figures 4.4 to 4.15 explain

the various menus and dialog boxes in details. Initially,

the prototype only considered the two wide faces. After



58

Start

Load knowledge base

id fact file? Load fact file
son

andInput lumber dimension
print output

Select grade category

End

Select one or
more subcategories

Yes
Work on facel?

No

Yes

Work on face2 2

No

No
Find overall grade? End_D

Reason Enter defect
and characteristics

print results

chrocteristics Wont explona
fect

No

from working

Figure 4.3. Grading process (This prototype has been
improved by including all four faces as
described in Chapter 5)



59

prototype evaluation, the system was enhanced to consider

all four faces (see Chapter 5).

The system's initial window is shown in Figure 4.4.

This window contains several pull-down menus for loading

the knowledge base, cleaning memory space, and controlling

system execution. The user can load a selected knowledge

base into the system. Grading categories can be placed

into one knowledge base, or several different knowledge

bases, depending upon the user's preference. The

development and maintenance of knowledge bases are done

using Eclipse's development tool kits. Once the knowledge

bases are error free, they can be saved as binary files.

The run-time version of the system can only load knowledge

bases in binary files. Binary files are efficient for fast

loading and secure delivery.

The system allows the user to input lumber

characteristics in two ways: either by entering the facts

during the query process, or by loading a fact file at the

beginning of the session (Figure 4.5). Once the user

selects to input facts interactively, a series of dialog

boxes will pop up to guide the user to enter the facts.

First, the user enters lumber thickness (Figure 4.6); this

is followed by lumber width and length. Each input dialog

box is followed by a confirmation dialog box (Figure 4.7)

so that the user has a chance to modify the input data.



60

Next the system asks the user to select a grade

category (Figure 4.8) and sub-categories in the category

(Figure 4.9). Although the same piece of lumber may have

more than one grade in different categories, the current

version of the system does not consider grade overlap

between different categories. However, the system allows

the user to select one or all sub-categories in the

category, that is, grade overlap is allowed between sub-

categories.

After selecting grade sub-categories, the user

specifies the face of lumber to be graded (Figure 4.10) and

then enters defects on this face. After a face has been

graded, the dialog box in Figure 4.10 may again be used to

specify another face. The overall grade is determined

after both faces have been graded.

Figures 4.1]. and 4.12 show the dialog boxes for

entering defects. The defects shown in the figures include

all defect types considered in the system. If the same

defect type occurs more than once, only the worst defect is

input. For example, if there are two holes, 0.1 inches and

0.5 inches in diameter, the larger hole (0.5 inches) is

entered into the system.

Once a defect type has been selected, a dialog box

will pop up to ask for more detailed information about the

defect type and location, as appropriate. Figure 4.13

shows a dialog box for entering information on knots. This



61

dialog box asks for knot type and location. The

information in the dialog box may differ across grade

categories as various categories consider knot types

differently. The last piece of information required for a

defect is its size, as shown in Figure 4.14.

After all the defect data is entered, the user selects

the no_more_defects item in Figure 4.12 to exit the input

mode. The system maps the input data to appropriate

grade(s). Figure 4.15 shows a an example output. The

output includes each face grade based on the defect

distribution on that face, the overall lumber grade, and

the corresponding sub-category. Overlap between sub-

categories is allowed; Figure 4.15 shows the grades for a

piece of dimension lumber.

{ile Lxecute stop help

Figure 4.4. Opening window of the run-time version of the
system



Please select data input method

load_tact_tile

fact_tile_I oa d ed

Figure 4.5. Dialog box for selecting input method

enter lumber nominal thickness in
inches

Figure 4.6. Dialog box for entering lumber thickness

Youve entered the following lumber size:

0 Nominal thIckness: 2 InNominal width: 4 InNomInal length: 8 ft
Want to modify?

I4.

Figure 4.7. Dialog box for confirming lumber dimension



63

determine grade category

sel cctlfinish
jIiiIi4iE1[u1.

boards

Figure 4.8. Dialog box for selecting grade category

determine dimension sub category

light_framing
structural_light_framing
stud

Figure 4.9. Dialog box for selecting Dimension sub-
category



64

select face to work on

Figure 4.10. Dialog box for selecting lumber face

Select defect to enter

holes
wane
splits
checks
grain
pith

Figure 4.11. Dialog box with partial defect list



Select detect to enter

grain
pith
decay
stain
pitch
pockets

Figure 4.12. Dialog box with another partial defect list

select knot type and location

fi rmkn ot_at_ce nte rI inc
loose_knot_at_edge
loose_knot_at_centerline

Figure 4.13. Dialog box for entering knot type and
location



Figure 4.14.

enter diameter of firm_knot_at_edge in
inches

o.1

Dialog box for entering knot diameter

ace: Facel Grade: CONST Category: Light_framing
ace: facel Grade: SEL-STR Category: Structural_light_Framing
ace: Facel Crade: STUD Category: Stud
ace: face2 Grade: NO.3 Category: Structural_light_framing
ace: Face2 Grade: CONST Category: Light_Framing
ace: Face2 Grade: STUD Category: Stud
uerall Grade: CONST Category: Light_fraiinq
uerall Grade: NO.3 Category: Structural_light_f raiiing
uerall Grade: STUD Category: Stud

Figure 4.15. System output display



67

CHAPTER 5

EVALUATION AND ENHANCEMENTS

5.1 System Evaluation

The system was initially tested using generated input

data. Simulated lumber and defect data were entered into

the system. The resulting grades assigned by the system

were compared with those assigned by using the rule book

manually. All rules in the system were checked and 100

percent match was achieved for the simulated data. The

test results are given in Appendix B.

The system was then evaluated using real lumber.

Eighty five samples of Siberian larch 2x4x12 lumber were

used to evaluate the system. The samples had been graded

by a human grader from the West Coast Lumber Inspection

Bureau (WCLIB). All samples belong to the Structural Light

Framing category. Among the five grades in this category,

20 sample pieces were randomly selected from 61 pieces of

Select Structural lumber, 20 from 31 pieces of No.1, 20

from 53 No.2, 20 from 73 pieces of No.3, and 5 from 10

Economy.

To test the performance of the system, defects on

these pieces were measured by the author. These were then

entered into the system. As shown in Figure 5.1, 60

percent of the samples (51 pieces) matched the original



0,

0
E

.4-
0
I-
0)

E
z

68

Matched UM-1 UM-2 UM-3 UM-4 UM-5 UM-6

Before Modification After Modification

Figure 5.1. Evaluation results considering all

factorsgrades as judged by the human grader. A total of 31

pieces (36.5 percent) did not match the original grades

because of limitations of the computer system, primarily

not considering defects on the two narrow faces (edges),

shake, warp, and manufacturing imperfections. An



additional 3.5 percent samples (3 pieces) did not match the

grades assigned because of measurement bias or the mismatch

could not be explained. For example, a firm knot at edge

on a piece was measured as being 1.1 inches in diameter,

which made this piece No.2. However, for some reason the

human grader assigned Select Structural grade to the piece.

Since the human grader was not present, it was difficult to

correlate some of the mismatched observations.

If used as a sub-system of SAW3D, warp and

manufacturing imperfections do not exist when SAW3D making

decisions. If these defects are excluded, the matched rate

between the grade assigned by the human grader and that

assigned by the system is 72.9 percent.

Although some marginal errors are introduced due to

measurement bias, a difference of 40 percent requires

modifications of the system to improve its performance.

5.2 System Improvement

As shown in Figure 5.1, the largest cause of deviation

between the grades assigned by the computer and those

assigned by the human grader was exclusion of defects on

the two narrow faces. Thus the focus of improvement was on

including the narrow faces in the evaluation process.

The system was improved to allow defects on narrow

faces to be input and evaluated. The enhanced model allows



rLi

the user to input defects on all four faces, assigns a

grade to each of the four faces, and then determines the

overall grade.

The eighty five samples tested earlier were again

evaluated using the revised system. Figure 5.1 shows the

grading results after the modification. The percentage of

matched grades increased from 60 to 76.5 percent. The

samples that did not match due to system limitations

decreased from 36.5 to 17.6 percent. Samples which were

assigned a grade different from that assigned by the human

grader because of measurement error and unexplained

differences increased from 3.5 to 59 percent.

All 15 unmatched samples caused by system limitations

are upgraded in the revised system because defects such as

shakes, warp and manufacturing imperfections are not

considered by the system. All five unmatched samples

caused by measurement difference are downgraded because

defect measurements are larger than those permitted by the

original grades.

Although warp and manufacturing imperfections also

contributed to deviations in grade assignment, these are

introduced as a result of the manufacturing process and do

not exist before log breakdown. As such, they are not

considered in a log breakdown model such as SAW3D. If the

system is to be used for a log breakdown optimization

system where warps and manufacturing imperfections are not



7]-

present, the percentage of matched samples increases to

72.9 percent before the modification, and to 89.4 percent

after the modification (Figure 5.2). However, if the

system is to be used as an instruction tool or a component

of an automatic grading system, warp and manufacturing

imperfection are important factors that affect grading

accuracy, and therefore need to be included in the system.

The bias due to measurement error can be eliminated by

having the individual who visually graded the samples also

a,
w
a-
E
Ct
0)
1-
0

-a
E
z

Matched UM-1 UM- UM-O

Before Modification After Modification if no measure bias

Figure 5.2. Evaluation results when warp, manufacturing
imperfections, and measurement bias were not
considered



72

measure the defects. In this case the matched samples

represent 95.3 percent of the total boards (Figure 5.2).

The nuiber of unmatched samples due to shake and not-well-

spaced knots is very small. These may also depend upon

wood species; inclusion of these factors will further

improve the system accuracy.



73

CHAPTER 6

CONCLUSIONS

A prototype expert system for softwood lumber grading

was developed. The prototype can grade lumber for 27

grades in Dimension, Select/Finish, and Boards category.

It can judge each of four faces of lumber individually and

assign an overall grade based on the grades of the

individual faces. The user interface helps to obtain facts

about a piece of lumber, and provides grading results and

explanations about how the conclusions are reached. The

initial prototype considered only two faces of lumber.

After an evaluation using real lumber samples, the

prototype was modified to include all four faces resulting

in significant performance improvement.

The intended use of the system is in conjunction with

SAW3D, a log breakdown optimization system, to determine

optimum log breakdown decisions. SAW3D will provide this

grading system with lumber size and defect characteristics

information. The expert system determines the lumber grade

and will then pass this information back to SAW3D where

SAW3D will determine the lumber value based on the grade,

and use it to arrive at optimum log breakdown decisions.

This application represents a new direction for improving

lumber production efficiency: optimization of cutting

strategies using both lumber geometric and defect



74

information.

The grading system can be used to build and maintain

grading rules independently, and communicate with other

components of the overall system in different ways. It can

be embedded in the overall system, or it can exchange data

with the system using Dynamic Data Exchange facility

supported by MS-WINDOWS. The separation of the grading

system from other parts of the overall system is very

important. The knowledge base can be developed and

maintained without interfering with any other program

logic. This provides greater portability and flexibility

of the grading system, and in turn simplifies the

development and maintenance of the system.

Areas for further research include:

1. The evaluation using real lumber was only performed

for the Structural Light Framing sub-category of the

Dimension. Such a evaluation should be done for all

grading categories.

2. For rapid prototyping, the development concentrates

on selected defect characteristics for each grade. The

selected characteristics are clearly defined and described

in the rule book, and are usually the most frequently

encountered defects on softwood lumber. Once the specific

use of the system is determined, the knowledge base should

be refined and more rules should be added to the knowledge

base with the help of human grading experts. For example,



75

knot positions relative to each other, shake, warp and

manufacturing imperfections need to be considered.

Clustering of defects is another important feature that may

effect a grade, and thus may need to be included in the

knowledge base.

3. As mentioned before, this system is to be used as

a sub-system of SAW3D. The system may also be used as an

instruction or training tool. When used as an instruction

tool, a better user interface should be developed. A help

facility should be added to guide the user at each step,

and more detailed explanations should also be provided at

each step. A graphic display of defect distribution and

grade stamp bitmaps may be very helpful. The system may

need to include some controls allowing the user to grade

the lumber, and then judge the results using the knowledge

base.



76

BIBLIOGRAPHY

1. American Lumber Standards Committee. 1970. American
Softwood Lumber Standard. National Bureau of
Standards, U.S. Department of Commerce, Washington,
D.C.

2. Aune, Jan. 1991. X-Ray edger optimizer makes money at
Macmillan Bloederl's Alberni Pacific Division.
Proceedings of the 4th International Conference on
Scanning Technoloav in the Wood Industry, pp. Aune-1 -
Aune-2, San Francisco, CA.

3. Brooke, Tom. 1992. The art of production systems.
Expert, 7(1):31-35.

4. Flatman, Carl, and Egan Bodell. 1989. Computer-aided
lumber grader-optimizer system. Proceedings of the
Third International Conference on Scanning Technology
in Sawmilling, pp. XIX-1 to XIX-15, San Francisco, CA.

5. Flatman, Carl, and Daniel J. Kenway. 1991. Mill
experience with high speed grade based optimizers.
Proceedings of the 4th International Conference on
Scanning Technolov in the Wood Industry, pp. kenway-1
to kenway-l4, San Francisco, CA.

6. Forgy, L. Charles. 1979. On the efficient
implementation of production systems. Ph.D Thesis.
Carnegie-Mellon University.

7. Forgy, L. Charles. 1982. Rete: a fast algorithm for
the many pattern/many object match problem.
Artificial Intelligence, 19(1) :17-37.

8. Haley Enterprise. 1992. Eclipse Reference Manual.
The Haley Enterprise, Inc., Sewickley, PA.

9. Hallock, H., and L. Gauger. 1971. Grading hardwood
lumber by computer. Res. Paper FPL-157. Forest
Products Laboratory, Madison, WI.

10. Huang, S.S.L., and F.T. Sparrow. 1989. A computer-
aided instruction tool for grading hardwood lumber.
Forest Products Journal, 39(10) :39-42.

11. Kenway, J. Daniel. 1990. A supercomputer-based machine
vision grading system and trimmer optimizer for
dimension lumber. Proceedings of Process
Control/Production Manaaement of Wood Products:
Technoloav for the '90s, pp. 53-65, Athens, Georgia.



77

12. Klin]thachorn, P., J.P. Franklin, C.W. McMillin, R.W.
Conners, and H.A. Huber. 1988. Automated computer
grading of hardwood lumber. Forest Products Journal,
38(3) :67-69.

13. Klinkhachorn, P., C.J. Schwehiu, CW. McMillin, and
H.A. Huber. 1989. HaLT: a computerized training
program for hardwood lumber grading. Forest Products
Journal, 39(2) :38-40.

14. Klinkhachorn, P., D. Yost, R. Kothari, C.W. McMillin,
H.A. Huber. 1991. Computer lumber grading and HaRem
for an automated lumber processing system.
Proceedinas of the First International Conference on
Automated Lumber Processing Systems and Laser
Machining of Wood, pp. 173-181, Michigan State
University, East Lansing, MI.

15. Western Wood Products Association. Western lumber
grading rule 88. 1988. Portland, OR.

16. Williston, Edward M. 1988. Lumber manufacturing - The
design and operation of sawmills and planer mills.
Miller Freeman Publications, Inc.

17. Zeng, Y. 1991. Log breakdown using dynamic programming
and 3-D log shape. M.S. thesis. Oregon State
University, Corvallis, OR.



APPENDICES



78

Appendix A. Source Code for Structural Light Framing
Category

, , , ------------------------------------------------------
; This file contains rules for Structural Light Framing
; category of Dimension lumber grades. This category
; contains the following four grades:

1) SEL-STR (Select Structural, Structural Light
Framing)

; 2) NO.1 (NO. 1, Structural Light Framing)
3) NO.2 (NO. 2, Structural Light Framing)
4) NO.3 (NO. 3, Structural Light Framing)
5) ECONOI1Y-SLF (Economy, Structural Light Framing)

Validation of defects for each grade is checked by the
; rules separated from the grading rule.

SE L-S TR GRADE

;----Rules to check validation of defects for SEL-STR -----

(defrule SEL-STR_firm_edge_knot "Firm knots at edge <=
SEL-STR limit?"

(declare (salience 1000))
(face ?)
(category structural_light_framing)
(face_width ?width)
(or (not (symptom firm_knot_at_edge))

(dia firm_knot_at_edge ?firmdia&:( or
(and (= ?width 2) (<= ?firmdia 0.375))
(and (= ?width 3) (<= ?firmdia 0.5))
(and (= ?width 4) (<= ?firmdia 0.75)))))

=>
(assert (satisfy firm_knot_at_edge SEL-STR)))

(defrule SEL-STR_firm_center_knot "Firm knots at center <=
SEL-STR limit?"

(declare (salience 1000))
(face ?)
(category structural_i ight_f raming)
(face_width ?width)
(or (not (symptom firm_knot_at_centerline))

(dia firm_knot_at_centerline ?firmdia&:( or
(and (= ?width 2) (<= ?firmdia 0.375))
(and (= ?width 3) (<= ?firmdia 0.5))
(and (= ?width 4) (<= ?firmdia 0.875)))))

(assert (satisfy firm_knot_at_centerline SEL-STR)))



79

(defrule SEL-STR_loose_edge_knot "Loose edge knots <=
SEL-STR grade limit?"

(declare (salience 1000))
(face ?)
(category structural_i ight_f raining)
(face_width ?width)
(or (not (symptom loose_knot_at_edge))

(dia loose_knot_at_edge ?loosedia&: ( or
(and (= ?width 2) (<= ?loosedia 0.375))
(and (= ?width 3) (<= ?ioosedia 0.5))
(and (= ?width 4) (<= ?loosedia 0.75)))))

=>
(assert (satisfy loose_knot_at_edge SEL-STR)))

(defrule SEL-STR_loose_center_knot "Loose cnt knots <=
SEL-STR grade limit?"

(declare (salience 1000))
(face ?)
(category structural_i ight_f raining)
(face_width ?width)
(or (not (symptom loose_knot_at_centerline))

(dia loose_knot_at_centerline ?loosedia&: ( or
(and (= ?width 2) (<= ?loosedia 0.375))
(and (= ?width 3) (<= ?loosedia 0.5))
(and (= ?width 4) (<= ?loosedia 0.75)))))

=>
(assert (satisfy loose_knot_at_centerline SEL-STR)))

(defrule SEL-STR_hoie "Holes <= SEL-STR grade limit ?"

(declare (salience 1000))
(face ?)
(category structural_i ight_f raining)
(face_width ?width)
(or (not (symptom holes))

(dia hole ?holedia&:( or
(and (= ?width 2) (<= ?holedia 0.375))
(and (= ?width 3) (<= ?holedia 0.5))
(and (= ?width 4) (<= ?holedia 0.75)))))

=>
(assert (satisfy holes SEL-STR) ))

(defrule SEL-STR_wane "Wanes <= SEL-STR grade limit ?"

(declare (salience 1000))
(face ?)



(category structural_light_framing)
(lumber thickness ?thickness)
(lumber_width ?width)
(or (not (symptom wane))

(and (thickness wane ?thx&: ( <=
4.0)))

(width wane ?wid&: ( <= ?wid
=>
(assert (satisfy wane SEL-STR) ))

?thx (/ ?thickness

(/ ?width 4.0)))))

(defrule SEL-STR_split "Splits <= SEL-STR grade limit ?"
(declare (salience 1000))

(face ?)
(category structural_light framing)
(lumber_width ?width)
(or (not (symptom splits))

(length split ?splitlength&:(<= (- ?splitlength
?width) 0.000001)))

=>
(assert (satisfy splits SEL-STR) ))

I

(defrule SEL-STR_check "Checks <= SEL-STR grade limit ?"
(declare (salience 1000))

(face ?)
(category structural_i ight_f raming)
(lumber_width ?width)
(or (not (symptom through_check_on_end))

(length through_check_on_end
?checklength&: (<= (- ?checklength ?width) 0.000001)))

(assert (satisfy through_check_on_end SEL-STR) ))

(defrule SEL-STR_slope "Slope within SEL-STR grade limit ?"
(declare (salience 1000))
(face ?)
(category structural_i ight_f raining)
(or (not (symptom grain))

(slope_of_grain ?slope&:( <= ?slope (/ 1 12.0))))
=>

(assert (satisfy grain SEL-STR) ))

Grading rule for grade of SEL-STR -----------

(defrule grade_SEL-STR "SEL-STR Grade"
(declare (salience 900))



81

(not (symptom decay))
(satisfy firm_knot_at_edge SEL-STR)
(satisfy firm_knot_at_centerline SEL-STR)
(satisfy loose_knot_at_edge SEL-STR)
(satisfy loose_knot_at_centerline SEL-STR)
(satisfy holes SEL-STR)
(satisfy wane SEL-STR)
(satisfy splits SEL-STR)
(satisfy through_check_on_end SEL-STR)
(satisfy grain SEL-STR)
=>
(assert (grade Structural_light_framing SEL-STR 1)))

NO. ]. GRADE

Rules to check validation of defects for NO.1 ------

(defrule NO.1_firm_edge_knot "Firm knots at edge <= NO.1
grade limit ?"

(declare (salience 800))
(face ?)
(category structural_i ight_framing)
(not (grade Structural_light_framing SEL-STR 1))
(face_width ?width)
(or (not (symptom firm_knot_at_edge))

(dia firm_knot_at_edge ?knotdia&:( or
(and (= ?width 2) (<= ?knotdia 0.5))
(and (= ?width 3) (<= ?knotdia 0.75))
(and (= ?width 4) (<= ?knotdia 1.0)))))

=>
(assert (satisfy firm_knot_at_edge NO.1)))

(defrule NO. 1_firm_center_knot "Firm knots at center <=
NO.1 grade limit ?"

(declare (salience 800))
(face ?)
(category structural_i ight_framing)
(not (grade Structural_light_framing SEL-STR 1))
(face_width ?width)
(or (not (symptom firm_knot_at_centerline))

(dia firm_knot_at_centerline ?knotdia&: ( or
(and (= ?width 2) (<= ?knotdia 0.5))
(and (= ?width 3) (<= ?knotdia 0.75))
(and (= ?width 4) (<= ?knotdia 1.5)))))

=>
(assert (satisfy firm_knot_at_centerline NO.1)))

(defrule NO.1_loose_edge_knot "Loose edge knots <= NO.1



82

grade limit ?"
(declare (salience 800))
(face ?)
(category structural 1 ight_framing)
(not (grade Structural_light_framing SEL-STR 1))
(face width ?width)
(or (not (symptom loos

(dia loose_knot_at
(and (= ?width
(and (= ?width
(and (= ?width

s_knot_at_edge))
edge ?knotdia&: ( or
2) (<= ?knotdia 0.5))
3) (<= ?knotdia 0.75))
4) (<= ?knotdia 1.0)))))

=>
(assert (satisfy loose_knot_at_edge NO.1)))

(defrule NO.1_loose_center_knot "Loose center knots <=
NO.1 grade limit ?"

(declare (salience 800))
(face ?)
(category structural_light_framing)
(not (grade Structural_light_framing SEL-STR 1))
(face_width ?width)
(or (not (symptom loose_knot_at_centerline))

(dia loose_knot_at_centerline ?knotdia&: ( or
(and (= ?width 2) (<= ?knotdia 0.5))
(and (= ?width 3) (<= ?knotdia 0.75))
(and (= ?width 4) (<= ?knotdia 1.0)))))

=>
(assert (satisfy loose_knot_at_centerline NO.1)))

(defrule NO. 1_hole "Holes <= NO.1 grade limit ?"

(declare (salience 800))
(face 7)
(category structural_light_framing)
(not (grade Structural_light_framing SEL-STR 1))

(face_width ?width)
(or (not (symptom holes))

(dia hole ?holedia&:( or
(and (= 7width 2) (<= ?holedia 0.5))
(and (= ?width 3) (<= ?holedia 0.75))
(and (= ?width 4) (<= ?holedia 1.0)))))

(assert (satisfy holes NO.1)))

(defrule NO.1_wane "wanes <= NO.1 grade limit ?"



83

(declare (salience 800))
(face ?)
(category structural_light_framing)
(not (grade Structural_light_framing SEL-STR 1))

(lumber_thickness ?thickness)
(lumber_width ?width)
(or (not (symptom wane))

(and (thickness wane ?thx&: ( <= ?thx (/ ?thickness
4.0)))

(width wane ?wid&:( <= ?wid (/ ?width 4.0)))))
=>
(assert (satisfy wane NO.1)))

(defrule NO.1_split "splits <= NO.1 grade limit ?"

(declare (salience 800))
(face ?)
(category structural_i ight_framing)
(not (grade Structural_light_framing SEL-STR 1))

(lumber_width ?width)
(or (not (symptom splits))

(length split ?splitlength&:(<= (- ?splitlength
?width) 0.000001)))

=>
(assert (satisfy splits NO.1)))

(defrule NO.1_check "checks <= NO.1 grade limit ?"

(declare (salience 800))
(face ?)
(category structural_i ight_framing)
(not (grade Structural_light_framing SEL-STR 1))

(lumber_width ?width)
(or (not (symptom through_check_on_end))

(length through_check_on_end
?checklength&: (<= (- ?checklength ?width)

0.000001)))
=>
(assert (satisfy through_check_on_end NO.1)))

(def rule NO.1_slope "slope is within NO.1 grade limit ?"
(declare (salience 800))
(face ?)
(category structural_light_framing)
(not (grade Structural_light_framing SEL-STR 1))

(or (not (symptom grain))



84

(Slope_of_grain ?slope&:(<= ?slope (1 1 10.0))))
=>
(assert (satisfy grain NO.1)))

Grading rule for grade of NO.]. --------------

(defrule grade NO.1 "NO.1 Grade"
(declare (salience 700))
(not (symptom decay))
(satisfy firm_knot_at_edge NO.].)
(satisfy firm_knot_at_centerline NO.1)
(satisfy loose_knot_at_edge NO.].)
(satisfy loose_knot_at_centerline NO.].)
(satisfy holes NO.].)
(satisfy wane NO.1)
(satisfy splits NO.].)
(satisfy through_check_on_end NO.1)
(satisfy grain NO.].)
=>
(assert (grade Structural_light_framing NO.1 2)))

NO . 2 GRADE
; ------ Rules to check validation of defects for NO.2

(defrule No.2_firm_edge_knot "Firm knots at edge <=
NO.2 grade limit ?"

(declare (salience 600))
(face ?)
(category structural_i ight_framing)
(not (grade Structural_light_framing SEL-STR 1))
(not (grade Structural_light_framing NO.1 2))
(face_width ?width)
(or (not (symptom firm

(dia firm_knot_at_
(and (= ?width
(and (= ?width
(and (= ?width

_knot_at_edge))
dge ?knotdia&:( or
2) (<= ?knotdia 0.625))
3) (<= ?knotdia 0.875))
4) (<= ?knotdia 1.25)))))

=>
(assert (satisfy firm_knot_at_edge NO.2)))

(defrule NO.2_firm_center_knot "Firm knots at center <=
NO.2 grade limit ?"

(declare (salience 600))
(face ?)
(category structural_i ight_framing)
(not (grade Structural_light_framing SEL-STR 1))
(not (grade Structural_light_framing NO.]. 2))
(face_width ?width)
(or (not (symptom firm_knot_at_centerline))



(dia firm_knot_at_
(and (= ?width
(and (= ?width
(and (= ?width

=>
(assert (satisfy firm

enter1ine ?knotdia&:( or
2) (<= ?knotdia 0.625))
3) (<= ?knotdia 0.875))
4) (<= ?knotdia 2.0)))))

_knot_at_centerline NO.2)))

85

(defrule No.2_loose_edge_knot "Loose edge knots <= NO.2
grade limit ?"

(declare (salience 600))
(face ?)
(category structural_i ight_framing)
(not (grade Structural
(not (grade Structural
(face_width ?width)
(or (not (symptom loos

(dia loose_knot_at
(and (= ?width
(and (= ?width
(and (= ?width

light_framing SEL-STR 1))
_light_framing NO.1 2))

s_knot_at_edge))
_edge ?knotdia&:( or
2) (<= ?knotdia 0.625))
3) (<= ?knotdia 0.875))
4) (<= ?knotdia 1.25)))))

(assert (satisfy loose_knot_at_edge NO. 2)))

(defrule NO.2_loose_center_knot "Loose cnt knots <= NO.2
grade limit ?"

(declare (salience 600))
(face ?)
(category structural_light_framing)
(not (grade Structural_light_framing SEL-STR 1))
(not (grade Structural_light_framing NO.1 2))
(face_width ?width)
(or (not (symptom loose_knot_at_centerline))

(dia loose_knot_at_centerline ?knotdia&: ( or
(and (= ?width 2) (<= ?knotdia 0.625))
(and (= ?width 3) (<= ?knotdia 0.875))
(and (= ?width 4) (<= ?knotdia 2.0)))))

=>
(assert (satisfy loose_knot_at_centerline NO. 2)))

(defrule NO.2_hole "Holes <= NO.2 grade limit ?"

(declare (salience 600))
(face ?)
(category structural_i ight_framing)
(not (grade Structural_light_framing SEL-STR 1))

(not (grade Structural_light_framing NO.1 2))



(face_width ?width)
(or (not (symptom holes))

(dia hole ?holedia&:( or
(and (= ?width 2)
(and (= ?width 3)
(and (= ?width 4)

(assert (satisfy holes NO.2)))

I

(defrule NO. 2_wane

I

86

(<= ?holedia 0.625))
(<= ?holedia 0.875))
(<= ?holedia 1.25)))))

"wanes <= NO.2 grade limit ?"

(declare (salience 600))
(face ?)
(category structural_i ight_f raining)
(not (grade Structural_light_framing SEL-STR 1))

(not (grade Structural_light_framing NO.1 2))
(lumber_thickness ?thickness)
(lumber_width ?width)
(or (not (symptom wane))

(and (thickness wane ?thx&: ( <= ?thx (/ ?thickness
3.0)))

(width wane ?wid&: ( <= ?wid (/ ?width 3.0)))))
=>
(assert (satisfy wane NO.2)))

(defrule NO.2_split "splits <= NO.2 grade limit ?"

(declare (salience 600))
(face ?)
(category structural_i ight_f raining)
(not (grade Structural_light_framing SEL-STR 1))

(not (grade Structural_light_framing NO.1 2))
(lumber_width ?width)
(or (not (symptom splits))

(length split ?splitlength&: (<= ?splitlength (* 1.5
?width))))

=>
(assert (satisfy splits NO.2)))

(defrule NO.2_check "checks <= NO.2 grade limit ?"

(declare (salience 600))
(face ?)
(category structural_light_framing)
(not (grade Structural_light_framing SEL-STR 1))

(not (grade Structural_light_framing NO.1 2))



87

(lumber_width ?width)
(or (not (symptom through_check_on_end))

(length through_check_on_end
?checklength&: (<= ?checklength (* 1.5 ?width))))

(assert (satisfy through_check_on_end NO. 2)))

(defrule NO.2_slope "slope is within NO.2 grade limit ?"
(declare (salience 600))
(face ?)
(category structural_light_framing)
(not (grade Structural_light_framing SEL-STR 1))

(not (grade Structural_light_framing NO.1 2))
(or (not (symptom grain))

(slope_of_grain ?slope&:(<= ?slope (/ 1 8.0))))
=>
(assert (satisfy grain NO.2)))

(defrule NO.2_decay "decay <= NO.2 grade limit ?"
(declare (salience 600))
(face ?)
(category structural_i ight_framing)
(not (grade Structural_light_framing SEL-STR 1))

(not (grade Structural_light_framing NO.1 2))
(lumber_thickness ?thickness)
(lumber_width ?width)
(or (not (symptom decay))

(and (test (= ?thickness 2 ))
(thickness decay ?thx&: ( <= ?thx (I ?thickness

3.0)))
(width decay ?wid&: (<= ?wid (I ?width 3.0)))))

(assert (satisfy decay NO.2)))

(defrule gra
(declare
(satisfy
(satisfy
(satisfy
(satisfy
(satisfy
(satisfy
(satisfy

Grading rule for grade of NO.2 -------------

e NO.2 "NO.2 Grade"
(salience 500))
firm_knot_at_edge NO.2)
firm_knot_at_centerline NO.2)
loose_knot_at_edge NO.2)
loose_knot_at_centerline NO.2)
holes NO.2)
wane NO.2)
splits NO.2)

(satisfy through_check_on_end NO.2)
(satisfy grain NO.2)



88

(satisfy decay NO.2)
=>
(assert (grade Structural_light_framing NO.2 3)))

NO. 3 GRADE

Rules to check validation of defects for NO.3

(defrule NO.3_firm_edge_knot "Firm knots at edge <=
NO.3 grade limit ?"

(declare (salience 400))
(face ?)
(category structural_i ight_framing)
(not (grade Structural_light_framing SEL-STR 1))
(not (grade Structural_light_framing NO.1 2))
(not (grade Structural_light_framing NO.2 3))
(face_width ?width)
(or (not (symptom firm_knot_at_edge))

(dia firm_knot_at_edge ?knotdia&:( or
(and (= ?width 2) (<= ?knotdia 0.75))
(and (= ?width 3) (<= ?knotdia 1.25))
(and (= ?width 4) (<= ?knotdia 1.75)))))

=>
(assert (satisfy firm_knot_at_edge NO.3)))

(defrule NO.3_firm_center_knot "Firm knots at center <=
NO.3 grade limit ?"

(declare (salience 400))
(face ?)
(category structural_light_framing)
(not (grade Structural_light_framing SEL-STR 1))
(not (grade Structural_light_framing NO.1 2))
(not (grade Structural_light_framing NO.2 3))
(face_width ?width)
(or (not (symptom firm_knot_at_centerline))

(dia firm_knot_at_centerline ?knotdia&: ( or
(and (= ?width 2) (<= ?knotdia 0.75))
(and (= ?width 3) (<= ?knotdia 1.25))
(and (= ?width 4) (<= ?knotdia 2.5)))))

=>
(assert (satisfy firm_knot_at_centerline NO.3)))

(defrule NO.3_loose_edge_knot "Firm knots at edge <=
NO.3 grade limit ?"

(declare (salience 400))
(face ?)
(category structural_i ight_framing)
(not (grade Structural_light_framing SEL-STR 1))



(not (grade Structural_light_framing NO.1 2))
(not (grade Structural_light_framing NO.2 3))
(face_width ?width)
(or (not (symptom loose_knot_at_edge))

(dia loose_knot_at_edge ?knotdia&: ( or
(and (= ?width 2) (<= ?knotdia 0.75))
(and (= ?width 3) (<= ?knotdia 1.25))
(and (= ?width 4) (<= ?knotdia 1.75)))))

(assert (satisfy loose_knot_at_edge NO.3)))

1

89

(defrule NO.3_loose_center_knot "Firm cntknots at edge <=
NO.3 grade limit?"

(declare (salience 400))
(face ?)
(category structural_i ight_f raining)
(not (grade Structural_light_framing SEL-STR 1))
(not (grade Structural_light_framing NO.1 2))
(not (grade Structural_light_framing NO.2 3))
(face_width ?width)
(or (not (symptom loose_knot_at_centerline))

(dia loose_knot_at_centerline ?knotdia&: ( or
(and (= ?width 2) (<= ?knotdia 0.75))
(and (= ?width 3) (<= ?knotdia 1.25))
(and (= ?width 4) (<= ?knotdia 2.5)))))

(assert (satisfy loose_knot_at_centerline NO.3)))

I

(defrule NO.3_hole "Holes <= NO.3 grade limit ?"
(declare (salience 400))
(face ?)
(category structural_light_framing)
(not (grade Structural_light_framing SEL-STR 1))

(not (grade Structural_light_framing NO.1 2))
(not (grade Structural_light_framing NO.2 3))
(face_width ?width)
(or (not (symptom holes))

(dia hole ?holedia&:( or
(and (= ?width 2) (<= ?holedia 0.75))
(and (= ?width 3) (<= ?holedia 1.25))
(and (= ?width 4) (<= ?holedia 1.75)))))

=>
(assert (satisfy holes NO.3)))

(defrule NO.3_wane "wanes <= NO.3 grade limit ?"
(declare (salience 400))



90

(face ?)
(category structural 1 ight_framing)
(not (grade Structural_light_framing SEL-STR 1))

(not (grade Structural_light_framing NO.1 2))
(not (grade Structural_light_framing NO.2 3))
(lumber_thickness ?thickness)
(lumber width ?width)
(or (not (symptom wane))

(and (thickness wane ?thx&: ( <= ?thx (/ ?thickness
2.0)))

(width wane ?wid&: ( <= ?wid (/ ?width 2.0)))))
=>
(assert (satisfy wane NO.3)))

(defrule NO.3_split "splits <= NO.3 grade limit ?"
(declare (salience 400))
(face ?)
(category structural_light_framing)
(lumber_length ?length)
(not (grade Structural_light_framing SEL-STR 1))

(not (grade Structural_light_framing NO.1 2))
(not (grade Structural_light_framing NO.2 3))
(lumber_width ?width)
(or (not (symptom splits))

(length split
?splitlength&:(<= ?splitlength (/ (* 12

?iength) 6.0))))
=>
(assert (satisfy splits NO.3)))

(defrule NO.3_check "checks <= NO.3 grade limit ?"
(declare (salience 400))
(face ?)
(category structural_i ight_framing)
(lumber_length ?length)
(not (grade Structural_light_framing SEL-STR 1))

(not (grade Structural_light_framing NO.1 2))
(not (grade Structural_light_framing NO.2 3))
(lumber_width ?width)
(or (not (symptom through_check_on_end))

(length through_check_on_end
?checklength&: (<= ?checklength (/ (* 12 ?length)

6.0))))
=>
(assert (satisfy through_check_on_end NO.3)))



(defrule NO.3_slope "slope is within NO.3 grade limit ?"
(declare (salience 400))

(face ?)
(category structural_i ight_framing)
(not (grade Structural_light_framing SEL-STR 1))

(not (grade Structural_light_framing NO.1 2))
(not (grade Structural_light_framing NO.2 3))
(or (not (symptom grain))

(slope_of_grain ?siope&:(<= ?slope (/ 1 4.0))))
=>
(assert (satisfy grain NO.3)))

(defrule NO.3_decay "decay <= NO.3 grade limit ?"
(declare (salience 400))
(face ?)
(category structural_i ight_framing)
(not (grade Structural_light_framing SEL-STR 1))

(not (grade Structural_light_framing NO.1 2))
(not (grade Structural_light_framing NO.2 3))
(lumber_cross_section ?iumberCS)
(or (not (symptom decay))

(cross_section decay ?decaySize&: (<= ?decaySize (I
?lumberCS 3.0))))

=>
(assert (satisfy decay NO.3)))

Grading rule for grade of NO.3 ------------

(defrule gra
(declare
(satisfy
(satisfy
(satisfy
(satisfy
(satisfy
(satisfy
(satisfy

le_NO.3 "NO.3 Grade"
(salience 300))
firm_knot_at_edge NO.3)
firm_knot_at_centerline NO.3)
loose_knot_at_edge NO.3)
loose_knot_at_centerline NO.3)
holes NO.3)
wane NO.3)
splits NO.3)

(satisfy through_check_on_end NO.3)
(satisfy grain NO.3)
(satisfy decay NO.3)
=>
(assert (grade Structural_light_framing NO.3 4)))

ECONOMY -SLF GRADE ==================
Grading rule for grade of ECONOMY-SLF ----------

(defrule grade_ECONOMY-SLF "ECONONY-SLF Grade"
(declare (salience 200))
(face ?)



(category structural_i ight_f:
(not (grade Structurai_iight
(not (grade Structural_light
(not (grade Structurai_iight
(not (grade Structural_light
(not (symptom broken ends ))
=>

raming)
_framing SEL-STR 1))
framing NO.1 2))
framing NO.2 3))
_framing NO.3 4))

92

(assert (grade Structural_light_framing ECONOMY 5)))
;======== END OF STRUCTURAL LIGHT FRAMING GRADES



93

Appendix B. Test Results Using Generated Data

Tables Bi, B3, and B5 show the grade specifications.

Data in each column in the upper section of these tables

represent the limiting provisions on defects; the grades

are listed in the lower section. An X in the lower section

means that data in the column are limiting provisions for

the grade in the row. The grading rules for each grade in

the knowledge base were tested using facts entered through

the interactive query process. These facts represent

realistic data values. Tables B2, B4, and B6 show the

resulting grades. These tables use a format similar to the

rule specification tables (Tables Bi, B3, and B5). Each

column contains data for a face of a lumber sample. Data

in the top half of the table is the generated defect

measurements (facts). Empty cells in a column reflect

defects that do not exist in a sample. As in the

specification tables, column and row intersections mean

that the data (facts) in the column satisfies the grade in

the row. Multiple Xs in a column means that the input

data satisfy more than one grade. For example, a lumber

sample with defects indicated in the first column of Table

B2 satisfies CONST, NO.3, and STUD grades.



94

Table Bi. Specifications of grades for a piece of lumber
in the Dimension category (lumber size =
2x4txlO)

)efect Fype/Location easure imiting Provisions <= inches)
firm-edge a____ 1 . 0.75 T 1

not firm-center a_____ 1 .5_ 2 2.5 0.875 1 .5_ 2 2.5 y 2.5 6
loose-edge d a_____ 1 2 2.5 0.75 1.25 1.75 y 1.75 6
loose-center a_____ 1 2 2.5 0.75

_1
2 2.5 y 2.5 6

lole d a_____ 1.25 1.5 y 0.75
_1

1.25 1.75 1.5
Vane hx___

_1
0.5 2(3 1 0.5

_1
0.5 2/3 1 y 2/3 2

wid 1 4/3 2 1 1 4/3 2 4/3 3
plft ength 4 6 6 4 4 6 20 8 30
heck through-on-end ength 4 6 6 y 4 4 6 20 8 30
)ecay hx none 2/3 area

8/3
none none 2/3 area

8/3
area
8/3wid none 4/3 y none none 4/3 y

3rain slope 1/6 1/41/4 1/12 1/10 1/8 1/4 an_n 1/4
ub-
ategory Grade

GONST
ight STAND X
raming UTIL X

ECONOMY X
SEL-STR X

itructural NO.1 X
ight NO.2 X
raming NO.3 X

ECONOMY X
itud STUD

X
ECONOMY

x

X - Data in the column is the specification for the
grade in the row.

none - Not allowed
any - No limitations
area - Width x length



95

Table B2. Test results of grading the lumber in the
Dimension category

)efect Type/Location Measure fi =inchL
(not

firm-edge die 0.75
____

1 T5
firm-center die 2 0.875 1.5 2.5
loose-edge dia 1 2.5 0.75 1.75 1.75
loose-center dia 2 0.75 2 2.5

lole dia 1 1.25 1.5 0.75 1 1.25 1.75 1.5
Vane thx 0.5 0.6 1 0.5 0.5 0.6 1 0.6

wid 1 1.3 2 1 1 1.3 2 2
pIft length 4 6 6 4 4 6 20 8

)heck through-on-end length 4 6 6 4 4 6 20 8
)ecay thx 0.5 1 0.66 1 1

wid 1 2.6 1.33 2.6 2.6
rain slope 0.16 0.25 0.25 0.08 0.1 0.13 0.25 0.25

pub-
ategory Grade

ight
raming

CONST x r
STAND X X
UTIL X
ECONOMY X X X

tructuraI
ight
raming

SEL-STA X X
NO.1 X
NO2 X
NO.3 X X X X
ECONOMY X X

tud STUD X X X X X XX
ECONOMY X X X

1) X indicates that input data in the column satisfies
the grade in the row.

2) Blank cell means that no such characteristic exists.
3) For the same set of facts, grade overlaps are allowed.

For example, facts in the first column match grades
CONST, NO.3, and STUD.

4) The results satisfy the specification, indicating the
system works well.



Table B3. Specifications of grades for a piece of lumber
in the Select/Finish category (lumber size =
l"x8"x12')

pet ect Type/Location Measure Limiting Provisions (<= inches)

not
tirm number * 2 * 2 any * 2 any any

dia 0.5 0.75 any 0.75 any any
fixed number none none * 4 none 4 6

dia none none 0.75 none 1 1.25
loose number none none none none none none

dia none none none none none none
oIe dia none none none none 1/16 0.75

Vane

on face thx none none none none 0.25 0.75
wid none none none none 1 2
ength none none none none 24 48

on back thx 0.25 0.5 0.75 0.5 any any
wid 0.5 1 2 1 any any
ength 24 36 36 36 any any

plft number none none any any any any
length none none 24 24 24 48

Theck
surface number 3 any any any any any

wid 1/32 1/32 1/32 1/32 1/32 any
length 4 4 4 4 10 any

'itch

light wid none *area
576

any *area
576

any any
length none any any any

medium wid none none *area
768

none *area

768
any

length none none none any
heavy wid none none none none none any

length none none none none none any

'ocket
number *1 *2 *4 *4 4 6
width 1/16 1/16 1/16 1/16 1/16 area

4length 3 3 6 6 8

taln
type light medium medium medium any any
wid area

115.2
area
384

any any any any
length any any any any

pub-
ategory Grade

elect
&BTR-1&2CLEAR

C SELECT X
DSELECT X

inish
SUPERIOR X
PRIME X
EFINISH X

X - Data in the column is the specification for the
grade in the row.

none - Not allowed
any No limitations
area - Width x length
* - Only one of the characteristics with symbol * in

the same column is allowed. For example, grade
B&BTR-1&2CLEAR allows 2 firm knots or 1 pocket.



97

Table B4. Test results of grading the lumber in
the Select/Finish category

)efect Type/Location Measure fact (<= inches)_____

(not
firm number 2 2

dia 0.5 0.75
fixed number 4 6

dia 1 1.25
loose number

dia
dole die 1/16 0.75

vane

on face thx 0.25 0.75
wid 1 2
length 24 48

on back thx 0.25 0.5 0.75 0.5
wid 0.5 1 2 1
length 24 36 36 36

put number 2 1 1 1
length 24 24 24 48

heck
surface number 3 1 2 1 2

wid 1/32 1/32 1/32 1/32 1/32
length 4 4 4 4 10

itch

light wid
length

medium wid 20 20
length 30 30

heavy wid
length

'ocket
number 2 4 6
width 1/16 1/16 1
length 3 8 3

itain
type light medium medium medium
wid 8 15 20 20
length 10 20 30 30

pub-
ategoIy Grade

elect
I&BTR-1&2CLEAR

C SELECT X
DSELECT X X

inish
SUPERIOR X X X
PRIME X X
EFINISH X

1) X indicates that input data in the column satisfies
the grade in the row.

2) Blank cell means that no such characteristic exists.
3) For the same set of facts, grade overlaps are allowed.

For example, facts in the first column match grades
B&BTR-l&2CLEAR and SUPERIOR.

4) The results satisfy the specification, indicating the
system works well.



Table B5. Specifications of grades for a piece of lumber
in the Boards category (lumber size = l"x8"x12')

)efect rype/Location Measure Limiting Provisions (<.. in)
rirm-red dla 2.25 3 3.5 1W5 2.5 T

not firm-black dia 0.75 1.25 7/3 16/3 1.25 2.5 3.5 16/3
loose dia none none 1.5 2.5 none 1.25 1.75 2.5

-lole dia none none 1.5 25 1/16 1.25 1.75 2.5

dane

on face thx none none none 0.5 none 0.5 0.5 <1
wid none none none 1 none 1 4/3 2
length none none none 24 none 24

on back thx 0.5 0.5 2/3 any
wid 1 4/3 2

_y
any any

length 36 48 72 any y_
pUt number 1 1 any any

length 4 8 24 48 8
..!_

8 24 36

heck
surface number 4 2 any any

wld 1/32 1132 1/32 1/32
length 4 10 10 10

ith firm-heart wid 0.25 0.5 any
..!

? ?
.!L
? ? ?

length 24 72 any ? ? 7 ? ?
light wid area

144
any any y_ any !k streak streak

length any any any streak streak streak
pitch medium wid none area

576
any any streak streak streak

length none any y any streak streak streak
heavy wid none none area

576
area none streak streak streak ylength none none 576 none streak streak streak y

'ocket
number 2 3 any ! y any Lwidth 1/16 1/16 1/16 area

4
1/16 1/16

.i. ..!!i.
length 3 6 12 12 12

_L
itain

type light medium medium !' medium ywid area
384

any any any
length any any

_y
any

_!L
)ecay wid

length
none none area none none none samli y
none none 144

_!i
an an none none none small anub-

:ategory Grade
1 COMMON
2 COMMON x

ommons 3 COMMON X
4COMMON X
5 COMMON X
SEL-MERCH

x
CONST

X
lternate STAND

X
UTIL

X
-- ECONOMY x

X - Data in the column is the specification for the
grade in the row.

none - Not allowed
any - No limitations
area - Width x length



Table B6. Test results of grading the lumber in the Boards
category

ii-.

FJF

YA'fli

fli:

I__
------

- --
;I II k'I --

--
li rr --_----

--_

[.i.J --
_________Lull

I

-

1) X indicates that input data in the column satisfies
the grade in the row.

2) Blank cell means that no such characteristic exists.
3) For the same set of facts, grade overlaps are allowed.

For example, facts in the first column match grades 1
COMMON and SEL-MERCH.

4) The results satisfy the specification, indicating the
system works well.