key: cord-0319235-ylkmp5g6 authors: Nath, Chandra title: Integrated Tool Condition Monitoring Systems and Their Applications: A Comprehensive Review date: 2020-12-31 journal: Procedia Manufacturing DOI: 10.1016/j.promfg.2020.05.123 sha: 7ada9d10d8cfa6505a906cf76c943490f762038d doc_id: 319235 cord_uid: ylkmp5g6 Abstract In conventional metal cutting, different tool wear modes, and their individual deterioration rates play vital roles in overall production performance. For a given tool (i.e., geometry or materials), many shop floors still follow a standard rule by pre-setting a tool life, which is conservative but not realistic. Premature failure of a tool can cause unexpected machine downtime and material losses, while another tool could serve beyond that pre-set life. As a result, optimized tool life and productivity cannot be achieved. Moreover, nowadays, there is an increased demand of process monitoring and optimization on the unmanned and the semi-automated shop floors. Tool condition monitoring (TCM) systems for process improvement and optimization have been in research for several decades. Both offline and online TCM systems are invented and discussed. A wide range of original publications are reported focusing on different sub-topics, e.g., specific machining process-based TCM methods, measurement or signal acquisition methods, processing methods, and classifiers. With the recent evolution of smart sensors in the era of Industry 4.0, development of online TCM systems received much attention to the researchers. Accordingly, research on some sub-topics also gets motivated into different directions, such as, feasibility of power or current sensors, machine vision technique, and combination of multi-sensors. Thus, from the industrial viewpoint, the current state of implementation of the proposed TCM systems for (near) real-time process monitoring and control needs to be clear. This paper presents the state-of-the-art of the TCM systems covering three major machining operations, discusses their application feasibility in industry environments, and states some current TCMS implementations. Challenges being faced by the industry are concluded, along with direction and suggestions for future researches. Conventional machining processes including turning, milling, drilling, grinding, and so on involve obvious tool wear due to plastic deformation during metal cutting. Flank wear, corner wear, rake wear, crater wear, notch wear, breakage/failure, chipping, and edge wear are commonly reported in the literature [1] [2] [3] [4] [5] [6] . Cutting tools used in any of the such metal cutting processes can vary in a wide range, in terms of shape and geometries, materials, coatings, surface finishing, etc. Not only these, tool life and product quality also but depend on many other variables like machining parameters, cooling and lubrication methods, machines, toolwork settings, to mention a few. All this make the tool wear phenomenon very complex. The wear modes that are well observed in different machining operations are listed in Table 1 . Of these, flank wear is the most critical one that directly impacts part surface finish and dimensional accuracy, thus requiring the obvious tool change. Though some industry standards are set by the standardization institutes, like ANSI/ASME B94.55M [7] regarding the flank wear level for changing a new tool, which is usually between 0.3-0.6 mm, many different industries set their own standards to meet the customer requirements. However, besides flank wear, other failure modes , such as, chipping, corner wear, notch wear, crater wear, chisel wear also highly impact the turning and the drilling operations. In many cases of these operations, if producing continuous long chips, they start entangling and jamming around the tool, scratch parts, accelerate tool wear. These cause premature tool Conventional machining processes including turning, milling, drilling, grinding, and so on involve obvious tool wear due to plastic deformation during metal cutting. Flank wear, corner wear, rake wear, crater wear, notch wear, breakage/failure, chipping, and edge wear are commonly reported in the literature [1] [2] [3] [4] [5] [6] . Cutting tools used in any of the such metal cutting processes can vary in a wide range, in terms of shape and geometries, materials, coatings, surface finishing, etc. Not only these, tool life and product quality also but depend on many other variables like machining parameters, cooling and lubrication methods, machines, toolwork settings, to mention a few. All this make the tool wear phenomenon very complex. The wear modes that are well observed in different machining operations are listed in Table 1 . Of these, flank wear is the most critical one that directly impacts part surface finish and dimensional accuracy, thus requiring the obvious tool change. Though some industry standards are set by the standardization institutes, like ANSI/ASME B94.55M [7] regarding the flank wear level for changing a new tool, which is usually between 0.3-0.6 mm, many different industries set their own standards to meet the customer requirements. However, besides flank wear, other failure modes , such as, chipping, corner wear, notch wear, crater wear, chisel wear also highly impact the turning and the drilling operations. In many cases of these operations, if producing continuous long chips, they start entangling and jamming around the tool, scratch parts, accelerate tool wear. These cause premature tool Conventional machining processes including turning, milling, drilling, grinding, and so on involve obvious tool wear due to plastic deformation during metal cutting. Flank wear, corner wear, rake wear, crater wear, notch wear, breakage/failure, chipping, and edge wear are commonly reported in the literature [1] [2] [3] [4] [5] [6] . Cutting tools used in any of the such metal cutting processes can vary in a wide range, in terms of shape and geometries, materials, coatings, surface finishing, etc. Not only these, tool life and product quality also but depend on many other variables like machining parameters, cooling and lubrication methods, machines, toolwork settings, to mention a few. All this make the tool wear phenomenon very complex. The wear modes that are well observed in different machining operations are listed in Table 1 . Of these, flank wear is the most critical one that directly impacts part surface finish and dimensional accuracy, thus requiring the obvious tool change. Though some industry standards are set by the standardization institutes, like ANSI/ASME B94.55M [7] regarding the flank wear level for changing a new tool, which is usually between 0.3-0.6 mm, many different industries set their own standards to meet the customer requirements. However, besides flank wear, other failure modes , such as, chipping, corner wear, notch wear, crater wear, chisel wear also highly impact the turning and the drilling operations. In many cases of these operations, if producing continuous long chips, they start entangling and jamming around the tool, scratch parts, accelerate tool wear. These cause premature tool Conventional machining processes including turning, milling, drilling, grinding, and so on involve obvious tool wear due to plastic deformation during metal cutting. Flank wear, corner wear, rake wear, crater wear, notch wear, breakage/failure, chipping, and edge wear are commonly reported in the literature [1] [2] [3] [4] [5] [6] . Cutting tools used in any of the such metal cutting processes can vary in a wide range, in terms of shape and geometries, materials, coatings, surface finishing, etc. Not only these, tool life and product quality also but depend on many other variables like machining parameters, cooling and lubrication methods, machines, toolwork settings, to mention a few. All this make the tool wear phenomenon very complex. The wear modes that are well observed in different machining operations are listed in Table 1 . Of these, flank wear is the most critical one that directly impacts part surface finish and dimensional accuracy, thus requiring the obvious tool change. Though some industry standards are set by the standardization institutes, like ANSI/ASME B94.55M [7] regarding the flank wear level for changing a new tool, which is usually between 0.3-0.6 mm, many different industries set their own standards to meet the customer requirements. However, besides flank wear, other failure modes , such as, chipping, corner wear, notch wear, crater wear, chisel wear also highly impact the turning and the drilling operations. In many cases of these operations, if producing continuous long chips, they start entangling and jamming around the tool, scratch parts, accelerate tool wear. These cause premature tool Conventional machining processes including turning, milling, drilling, grinding, and so on involve obvious tool wear due to plastic deformation during metal cutting. Flank wear, corner wear, rake wear, crater wear, notch wear, breakage/failure, chipping, and edge wear are commonly reported in the literature [1] [2] [3] [4] [5] [6] . Cutting tools used in any of the such metal cutting processes can vary in a wide range, in terms of shape and geometries, materials, coatings, surface finishing, etc. Not only these, tool life and product quality also but depend on many other variables like machining parameters, cooling and lubrication methods, machines, toolwork settings, to mention a few. All this make the tool wear phenomenon very complex. The wear modes that are well observed in different machining operations are listed in Table 1 . Of these, flank wear is the most critical one that directly impacts part surface finish and dimensional accuracy, thus requiring the obvious tool change. Though some industry standards are set by the standardization institutes, like ANSI/ASME B94.55M [7] regarding the flank wear level for changing a new tool, which is usually between 0.3-0.6 mm, many different industries set their own standards to meet the customer requirements. However, besides flank wear, other failure modes , such as, chipping, corner wear, notch wear, crater wear, chisel wear also highly impact the turning and the drilling operations. In many cases of these operations, if producing continuous long chips, they start entangling and jamming around the tool, scratch parts, accelerate tool wear. These cause premature tool 48th SME North American Manufacturing Research Conference, NAMRC 48 (Cancelled due to breakage or crush, and demand for early stage change of the tool. The issue becomes even more worse when machining hard-to-cut materials like Ti-and Ni-based alloys, hardened steels, aluminum alloys, etc. [3, 4] . As such, the tool that fails before the predetermined life can cause part defects, material loss, and unavoidable machine downtime. On the other hand, some tools could survive beyond that pre-set life due to less tool wear rate. Thus, the standard practice of the predetermined tool life for the machining processes is not an effective and realistic method, and the optimum tool life along with minimum machine down time cannot be achieved. Turning Flank wear [1] [2] [3] 8] ; Corner wear; Rake/crater wear [1] [2] [3] ; Notch wear [2] ; Chipping [3, 5] ; Fracture [1, 3] ; Breakage/failure [5] Milling Flank wear [4, 5, 9] ; Chipping [5, 9] ; Tip/corner breakage [5, [9] [10] [11] [12] Drilling Flank wear [5, 6] ; Corner wear [6] ; Crater wear [6] ; Margin wear [6] ; Chisel edge wear [6] ; Chipping [6] ; Premature tool breakage [13] [14] [15] ; Breakage [6] Nowadays, due to highly competitive market pressure, cost cut-down is an important factor in the production floors. Considering an industry scenario, a product may involve more than one machining process. Suppose, in production lines, about thirty-forty processes are required by several machines to complete an automotive engine part, and one machine is assigned to perform five to ten machining processes combining turning, milling, and drilling, etc. In such a case, premature failure or breakage of one or more tools causes huge material loss, often machine stoppage and down time [5] , which impacts productivity and quality. In the era of current industry 4.0, application of digital technology, such as, smart sensors, signal acquisition, data analytics, robotics and automation, the internet of things (IoTs), and controller, etc. has significantly been increasing to encounter the market issues and customer satisfactions [8, 16] . Unmanned machining with the help of automatic monitoring technology for all possible machining operations at a time is determined to be one of the critical factors to reduce the production costs in industries. A dependable tool condition monitoring system (TCMS) can accurately perform detecting tool wear progress, machining chatter, chip jamming, cutting temperature, and part surface finish and dimensional error, and predicting tool premature wear/breakage and remaining useful life (RUL). Over the past and the present three decades, with the advancement of digital technology, numerous researches had been performed on the TCMS development [5, 8, 9, 17] . While considering TCMS functions related to all individual tool wear characteristics, several hundreds of publications can be found. Most original and review papers available in the literature are limited to tool flank wear monitoring (refer Table 1 ). However, as stated above, some other wear modes also negatively impact the metal cutting process, especially in the turning and the drilling operations, and could damage machine tools or produce disqualified parts. Some review papers covered only single operations (e.g., milling or turning), or specific data acquisition and processing techniques, e.g, TCMS image processing [31] . However, in today's market, a complete, efficient, cheap, and plug-andplay TCMS, which could reliably perform all the functions is rarely available. Also, most practical parts require multiple processes in a single machine, but discussion on TCMSs that can closely meet such demand is very limited. This paper aims to present a comprehensive review on the advancement of TCMSs for different tool wear modes under the stated three major machining operations. It also emphasizes on the recent technology trend, and the current implementation state of the TCMS in industry environments. Some industry reviews and tech reviews are briefly included. Based on the reviews, challenges being faced by the industries are also summarized. As presented in Figure 1 , a typical TCMS includes sensor signal/data acquisition, signal processing, feature extraction, and decision making or classifiers. Data or signal acquisition methods are performed by hardware, while the rest of the TCMS are executed by software. This section discusses their functions with advantages, applicability, and limitations. In TCMS study, a variety of data (or signal) acquisition methods are used to predict the states of tool. They can be divided into either offline and online, or direct and indirect methods. In some past reviews [8, 18] , the direct methods are categorized as offline, while the indirect methods are defined as online TCMS. However, with continuous efforts on TCMS developments, both offline and online methods now can offer direct and indirect data acquisition; thus, they are redefined into four groups as shown in Figure 2 . With the offline acquisition methods, the tool or the work is taken off from the machine to estimate and map tool wear and pattern. Based on that, an adjustment is made to tool setting or process parameters for machining the next parts. If the wear data is obtained directly (by measurement) or indirectly (by signal) but on machine, and further data/signal processing is performed online so that zero or minimum process time is required, then it can be termed as direct/ indirect online measurement method. With indirect methods, some relevant output parameters are correlated with the tool wear. Direct offline measurement methods are established based on: i) optical measurement, ii) radioactivity on the tool or chips, and iii) electrical resistance between the tool and the workpiece. Of these, conventional optical microscopic measurement method has been a very popular choice. Optical microscope facilitates to map the flank wear land and the other wear characteristics, such as, chipping, breakage, notch, cater, etc., and offers very accurate results. However, to characterize and measure the wear with this off-machine practice, the tool inspection needs to be performed with special measurement setup, e.g., microscope. This inspection process takes long time, and the machine goes to 'idle state' that causes high down time. Also, the prior toolwork setting reference is lost due to repositioning of the tool after inspection. These two major drawbacks with this method depreciate both productivity and product quality [19] . But the off-machine optical measurement method of tool wear has been in laboratory practice over decades, mainly for studying machining science, but remain impractical in commercial production lines. The radioactive and the electrical resistance methods are also tested before 1990 (Table 2) . These methods though are found accurate, but seem to be impractical due to safety concern, slow response or processing time, and/or misleading results with cutting force variation [8, 20] . As an alternative to the direct inspection methods, offline indirect methods offer a quick check of relevant error in machined parts, in terms of dimensions, surface textures, roughness, etc. Different dimensional gages, such as, slide/Vernier caliper, GO/NOGO gage, 3-pin gage, air gage, surface roughness profilometer, roundness profilometer, optical microscope, CMM, edge-burr gaging, and so on are used to correlate with any error from the tool wear or the machine settings [21, 22] . Based on the error(s), corrective actions (e.g., tool offset or reject, etc.) are taken accordingly. In many manned and semi-automatic shop floors of part making manufacturing industries, the offline indirect methods are still a common practice for machinists and production engineers. Though the offline indirect methods are found to be better than the offline direct methods; however, such TCMS cannot deal the demand of higher productivity in modern unmanned automated industries. Recently, due to rapid progress of digital camera technology, the machine vision technique is receiving much attention from researchers and engineers [23] [24] [25] [26] [27] [28] [29] [30] [31] [32] . The method involves high-speed and high-resolution camera, digital image processing, image descriptors, data/file communication system, etc. Under special arrangement, a high-quality optical camera sensor system, e.g., a charge-coupled device (CCD) with proper illumination can be oriented in some remote zone(s) within the machine chamber to avoid interaction with cutting fluids [25, 33, 34] . Estimated accuracy of flank wear, fracture, built-up edge (BUE), and chipping is reported to be about 97% [32] , and deviation is reported to be less than 1% [30] . At any unloading-loading interval, the tool image(s) is captured and processed online with a proper artificial intelligence (AI) system for extracting tool wear information, and predicting remaining useful life (RUL). In the direction of developing [35] reviewed some articles on the machine vision technique, and stated that the global thresholding technique used by researchers is not yet efficient. More efficient algorithms and standardizations are necessary for covering different toolwork combinations for any given machining conditions and process types (turning, milling, and drilling). Some researches proposed geometrical descriptors obtained from the image analysis for tool wear monitoring [23, 24, 31] . It was found that the descriptor-based methods would increase productivity as the tool can be used beyond the standard tool life defined by ISO 3685 (turning) and 8688 (milling) [36] . The wear state signal can be incorporated with the NC program on CNC machines as to make decision within a few seconds, if continuing or not with the current tool [30, 31] . This TCM methodology makes it possible to predict the on-machine tool state with possibly no process cycle delay. So, it can be considered as an online direct (in-process or in-situ) tool wear measurement technique, which has high potential to be implemented in industry. As seen the list in Figure 2 , this group involves measurement of process output responses, such as, cutting force, vibration, acoustic emission (AE), current or power, sound, temperature, surface roughness, part dimensions, etc., which are correlated with tool wear characteristics. Such output data can be directly collected online. A calibration procedure is performed to estimate the flank wear and forecast the other wear characteristics. These methods offer ease of data collection and (near) real time wear estimation with rare or no issue from cutting fluid interactions. However, the amount of computational efforts involved in the calibration procedure is high. Most published and currently in practice TCMSs can be referred to these methods. Number of articles reported in terms of process output parameters are presented in Table 2 . Historically, the force and vibration signal measurement methods are studied by most researchers for fundamental understanding of machining processes and tool wear prediction, as compared to other indirect methods. Cutting force and torque: Cutting force and torque (for drilling and milling), both have a direct but nonlinear complex relationship with the tool wear mechanism. A force sensor (or dynamometer) is used to collect force and torque data. This measurement method is the most common in research labs. The collected data can be applied in TCMS algorithm to monitor tool wear (flank, crater, notch, and nose wear, catastrophic failure/breakage), chip formation mechanism, machinability observation, elastic deformation of work material, and change of cutting conditions/ parameters, [13, 18, 38] . However, the static forces results vary with the joints and couplings of machine elements, and the minute changes in cutting conditions, which can cause chatter [8] . Moreover, considering the industry scenario, where multiple tools (turning, milling, drilling, etc.) arranged on a CNC turret are used one by one for completing a part, setting up sensors with special arrangement for individual tools may not be feasible [38] . In addition, if several machines are in active operation in production lines, then various sources of noises and vibrations can interfere original cutting force and torque signals, which can make the TCM system design very complex. Also, most force sensors are sensitive to cutting fluids , which cause signal drifting and fluctuations, thus are not sustainable in lines . Vibration: Rubbing at the tool-work-chip interaction zone can deteriorate the tool, thus the rate of wear evolution with time can be easily captured online by monitoring vibration via accelerometers or piezo-sensors [39] . They can be used to detect tool wear and breakage, chip formation type, chatter, part surface profile, surface roughness, and process anomalies during machining [18, [40] [41] [42] . However, autonomous algorithms and paradigms are required for performing feature extraction and making decision without involvement of the operators. Freyer et al. [43] proposed a concept of self-sensing piezoelectric actuators to carry out condition monitoring of the cutting tools. The concept was verified in a simulated environment; however, its application for the industrial TCMS is yet to be realized. By reviewing [44] [45] [46] [47] [48] [49] , Dan and Mathew [50] concluded that vibration monitoring techniques may be more practical and cost-effective. Vibration measurement signals may be easy to capture, but highly depend on workpiece material, cutting conditions, and machine structure [51] . Moreover, as the signals are very sensitive, the vibrations producing from many different sources in industrial settings may not be easy to distinguish. Acoustic Emission: Acoustic emission (AE) is the intuitive transient elastic wave or energy emitted within materials due to plastic deformation or disorder at any location, such as, the tool-work interface. The signals are captured by AE sensors. AE signals can be continuous or transient type [52] . Continuous AE signals are produced from continuous interaction between the tool and the workpiece or the continuous chips (e.g., turning or drilling), while transient (or burst) type AE signals can be produced by milling operation, discontinuous or broken chips in any machining operations, and catastrophic tool failure [8] . The AE devices can be found, as both wired and wireless, and in form of thin film piezo sensors [5, [53] [54] [55] . They can detect most machining related signals including tool wear (flank and crater wear) and breakage, material defects, surface roughness and integrity, chip formation and breakage, [18, 40, 52, 56, 57] , thus trusted to be a part of a reliable TCMS for detecting any machining disorders. However, acquiring proper tool anomaly data from AE sensors is very challenging due to various reasons, especially, placing location(s), ambient noise, sensor alignment, fouling from chip and coolant, sensitivity to change in parameters [5, 8, [58] [59] [60] . Considering such difficulties, Dimla [58] suggested to use the AE as an additional sensing agent for increased reliability. As AE sensors can be connected wireless from remote locations, able offer all sensitive and useful information about tool-work-chip interactions, cheap, portable, and small in size, this signal/data acquisition method has high potential in industry applications. However, efforts should be invested on developing precise AE sensors that are robust enough to handle a TCMS. In machining, audible sound generation is a common consequence due to friction between the tool, the workpiece, and the flowing chips. Unlike AEs, this sound is transmitted externally via air media, and can be captured by microphones. However, in production floors, there are many other sources of sound, e.g., neighbor machines and machining processes, robots, part loading-unloading, air blow, etc. Directional microphone is proposed to deal this issue [5] , but not thoroughly tested. Thus, the use of audible sound signals for a reliable TCMS seems still impractical [8] . Power/current: Motors used in axes drives and spindles in a machining system are driven with power or current supply. Machining power required during metal cutting is directly proportional to the cutting force, thus shows strong correlation with tool wear growth [13, 61] . Since the power data source is available at the outside of the machine, such as, the motor, the power panel, or CNC monitor/control box, the hardware can be easily plugged in to collect machining power/current data [4, 13] . This acquisition method was described for online TCMS in turning [62, 63] , milling [4] , and drilling [13] . Siddhpura [8] stated that this method is relatively simple, but less sensitive for flank wear as compared to force or vibration measurement. However, with an aim of online TCMS development by this author [13] , a recent investigation on power and force data comparison during drilling operation reveals that the error range in tool wear prediction with both power and force data is almost same (0.8-18% vs. 0.44-17.94%, respectively). Not only the tool flank wear state, power data but is also able to monitor catastrophic tool breakage during Inconel drilling. This method also is also able to predict the early state of tool wear before sudden failure, even without any concern in change in spindle speed [13, 64] . In last few years, power data collection and analysis received much attention to the researchers. In conclusion, power or current signal can be easily obtained without interfering individual processes inside the machine, needs minimum hardware with no special and expensive arrangement, is reliable, and is useful for monitoring the tool wear state and forecasting early failure of tool. Thus, a TCMS with power data seems to be more feasible, practical, and has high potential for unmanned production environment. Temperature: In metal cutting, temperature is another considerable outcome due to friction between the tool and the workpiece. As the cutting progresses, with the increase in cutting temperature, the chemical dissolution of the tool material increases. Thus, there is a good relationship between the mechanical abrasion/friction, tool wear, chemical dissolution, and cutting temperature, which can be established by developing mathematical models [65] [66] [67] . The models may be able to estimate wear rate and temperature, but it is hard to predict the wear length and requires a database of thermochemical properties of specific tool-workpiece combination. In order to measure the cutting temperature, as reviewed by this author in [68] , several techniques including the tool-work thermocouple, the inserted thermocouple, the spectral radiation thermography, and the recently proposed thin film thermal sensor are investigated by researchers. The tool tip is a very small zone (usually, 0.5-2 mm depending on parameters), where the maximum temperature is produced. It is hard to place a sensor in this tiny zone. The other techniques need special arrangement. Due to this, the measurement of actual temperature from the active zone is almost impossible. The data or signals received by the above stated techniques are calibrated to estimate the flank wear land over cutting time. However, these techniques cannot provide any information about tool chipping, breakage and catastrophic failure, and tool life. Thus, the temperature data acquisition method is not useful for a TCMS in manufacturing production systems. Surface quality and part error: As discussed before, tool wear, especially at the corner and the flank face, has direct relationship with the surface roughness and part dimensions. Thus, the measurement can be quickly performed on machine (in-situ) in order to map tool wear and predict RUL. Some successful research results for tool flank wear estimation are reported. Kassim et al. [25] and Dutta et al. [31] proposed geometrical descriptors obtained from the surface texture analysis for online tool wear prediction. Though not in real time, this process can be completed within the part unloadingloading cycle. Cakan [72] used a photo electronic sensor to map the changes in diameter of workpiece in order to carry out flank wear monitoring task. However, other than flank wear, this process can be limited to other forms of tool wear. For example, rake wear has most relationship with the chip flow. Moreover, lack of proper illumination over the whole machined part within the machine and complex part geometry may limit this technique to be applied in real production lines . From all individual online indirect methods described above, it seems that a TCMS system using a single signal acquisition method is not yet reliable enough for production environments. Moreover, a product often needs more than one cutting operations, where application of different suitable sensors for one or more machines at the same time is a requirement. Considering such scenario, some researchers started paying attention to developing a reliable online TCMS by combining more than one signal acquisition method for one or more operations. Based on test results on tool wear, chipping, and breakage detection, Cakir and Isik [73] observed that the force data are more sensitive than the vibration and the motor current data. Also, the combination of force and vibration features correlated well with tool wear. Another study on combined force and AE signals in a feature fusion model reported that the system is able to predict tool wear automatically [74] . Jemielniak et al. [75, 76] developed a sensor fusion model by combining AE, vibration, force sensors, and machine vision technique. They concluded that the combination of vibration and AE signal features offers the optimum result as compared any other combinations. An online TCMS developed by Dornfeld and DeVries [62] , by combining spindle current with AE and force signals, showed that the system works well over a wide range of cutting conditions. Zhou and Xue [77] successfully investigated force, acceleration and AE sensors during milling. Silva et al. [61] also monitored spindle current along with force, vibration, and sound signals to develop an online tool wear monitoring system. Rehorn [5] developed a TCMS by combining accelerometers , cutting force, AE, power data; and found that the system can detect and classify tool wear with at least 80% confidence level. Ding and He [51] successfully developed a reliability model for individual tool wear states based on tool vibration, AE, spindle and servo motor current signals, and video microscopy technique. In summary, signal or data acquisition methods can be categorized into four groups: 1) offline direct, 2) offline indirect, 3) online direct, and 4) online indirect methods. Of these, the direct tool wear inspection (Groups 1 and 3) with optical camera system can offer the most accurate tool wear measurement and the other wear characteristics. However, the offline direct tool inspection method is limited to fundamental study in research labs, and not feasible for production floors due to longer inspection time and tool repositioning error. The offline methods with indirect part quality check (Group 2) at some predefined loading-unloading intervals offer improved productivity and quality as the tool is not required to be taken off. The machine or the tool-machine settings are readjusted with the tool corner offset or the tool is rejected based on the part dimensions. Based on literature and online reports, it is evident that these methods are still practical in manned or semi-automatic industries. However, such methods are difficult to adopt with online database or TCMS. To improve productivity and quality, especially in large and unmanned production plants, an online TCMS is suitable in terms of information collection, data storage, data base, recordkeeping, mapping current wear state and predicting the RUL. Direct measurement of tool within the machine chamber (Group #3) is feasible, but it may not be performed in 'real time' due to a requirement of a special setup in order to avoid coolant and chip interactions with the camera system. If the measurement process can be performed during the interval of the loading-unloading cycle or with negligible time delay, then this process could be the best TCMS for real industry environment. This in-situ process would not compromise part quality, and the tool wear state monitoring process can be quickly performed with online data receiving from the highspeed high-resolution camera with zero or minimum tolerance of productivity. However, it would be difficult to arrange individual wear inspection systems for parts that are shaped by a number of tools. The online indirect measurement methods are a big group (Group 4), and they are mostly studied in the current TCMS development. Indirect online methods can continuously produce data during cutting processes, thus offer real time TCM demands and optimum productivity. In industry environment, vibration, AE, and power/current signals are reported to be more viable than other measurement methods. The relevant hardware is also found energy-efficient, cheaper, easy to use, and exhibits consistent performance [78] . Thus, for parts requiring different operations and in unmanned industries, this group can offer the most suitable and reliable TCMS. Raw data received by the hardware systems can be either analog or digital. Analog data are recorded from offline (offmachine) measurement methods with human involvement. The data are noted in some specific standard datasheets and/or via software system installed within the corresponding measuring hardware (e.g., profilometer, microscopes, etc.). As discussed before, raw data logged by human involvement in paper sheets or digital devices is still a common practice in semi-automatic production systems in many small and medium industries. Such data processing for understanding of wear and its characteristics delays the part processing time and slows down the productivity. An ideal and reliable TCMS means that the system has to be practically (near) online for data collection and processing in order to help detecting tool condition with wear characteristics and predicting RUL, protecting the machine tools from any immediate crushing events, alerting shop floor (operators and managers) about any process anomaly with possible suggestionsall to improve productivity and part quality. Online (i.e., on machine) measurement methods, with no or minimum possible process delay, can satisfy the requirement of a digital TCMS for obtaining optimum productivity and part quality. Any data received via these methods can be stored in only one form, such as, voltage signals. Online data or signal are nonlinear time-variant in nature, and can be mixed up with many noises and disturbances resulting from different machines, processes, and other sources. Due to this, the raw data are passed through preprocessing (filtration and linearization) and extraction processes for obtaining meaningful feature of tool failure. In the data extraction stage, the preprocessed data are extracted via different domains including time, frequency, timefrequency, and statistics. Many researches are published to justify the appropriate domain for specific requirements in machining systems as presented in Table 3 . Different domains with advantages and limitations are reviewed in this section. Time domain: This is the mostly applied domain for extracting force signals in terms of magnitude, RMS value, or force ratio [8, 18, 73, [79] [80] [81] [82] [83] [84] [85] [86] [87] [88] . It was reported that the time domain can well correlate with the tool wear with forces, AE signals, and surface roughness data [59, [89] [90] [91] . Time domain features are comparatively simple in terms of extraction, but are prone to disturbances [8] . Also, they are required to be tested for data receiving from different online measurement methods, such as, power and vibration, which are more practical signal measurement methods in industry scenario. The fast Fourier transform (FFT) method is mostly used in this domain. Like time domain, frequency domain is also widely studied, but mainly based on vibration, sound, and force signals. Researchers reported that the frequency domain can correlate well tool wear with such signals [45, 47, 49, 93, 94] . Using sound signals, Yamamoto et al. [94] was able to predict tool wear in early stage. This domain was also successfully investigated for tool condition monitoring in the cases of AE signals [95, 96] , force signals [97] , and roughness data [93] . In the frequency domain, the amplitude of typical frequency signals was found to increase monotonically with the increase in tool wear. The signals fall sharply as soon as the tool wear is observed to be in the tertiary zone [97] . The low-frequency AE signals were found highly dependent on the tool-workpiece and the tool-chip interfacial contact conditions. On the other hand, the highfrequency AE signals were related with cracking that takes place at and below the tool surface [96] . Though the frequency domain can successfully detect different tool wear characteristics and predict early stage of tool failure, it is however sometimes difficult to identify the characteristic spectral bands that are sensitive to tool wear. Moreover, it is not yet well understood why some specific frequencies are dominated by tool wear [8] . This domain is used to extract features from non-stationary signals. It is performed mainly based on wavelet transforms that can provide useful information about singularity (i.e., localization) of a signal in both the time and the frequency domains at the same time. Wavelet transforms can be of three typescontinuous, discrete, and stationary. One or more are applied to extract tool condition features of machining signals including force [98] , vibration [99, 100] , AE [52, 100] . As the metal cutting is a dynamic phenomenon, it is important to identify the most stationary part of the signals originating from neighbor sources. In their study, Scheffer et al. [98] observed that the time-frequency analysis with spectrograms can detect the most stationary parts of force signals. Wang et al. [99] found that the extracted features from discrete wavelet coefficients along with HMM can accurately predict the tool wear. In their review of various applications of wavelet analysis in TCMSs, Zhu et al. [101] concluded that, due to its sparsity and localization properties, this analysis method is very effective in accurately analyzing non-stationary machining sensor signals than any other time-frequency methods. This method requires less processing time, but it is difficult to determine exact contribution of a specific frequency at a given time because of time variant nature of wavelet transforms. Overall, this method offers much better performance in TCM as compared to individual time or frequency domains. Although wavelet transform in the time-frequency domain has great potential for TCM, more efforts are needed to prove its superiority among all other techniques. Statistical domain: In this domain, from a set of experimental data for a random process, a relationship between the amount of tool wear and sensor data is constructed. This method works based on the probability distribution, and the output features are described in terms of mean, variance, standard deviation, skew, kurtosis, and standard coefficients of time series signals, such as, autoregression (AR) [62] , moving average (MA), and combined ARMA [47, 62, 80] . Statistical analysis method has also been studied to investigate tool wear from different output signals, including force [102] [103] [104] , vibration [61, 103] , sound [61] , AE [57, 104] , wear images [105] , surface roughness [106] . The skewness and the kurtosis of AE signals generated during gradual tool flank wear occurrence were found to be very sensitive to the degree of wear [57] . Oraby and Hayhurst [102] developed an analytical tool wear model, and performed nonlinear regression using force data. Silva et al. [61] estimated absolute deviation, mean, skew, and kurtosis for extracting sound and vibration signals; however, they exhibited limited correlation with tool flank wear. When applied neural networks along with all the parameters and Taylor's tool life model, more accurate tool wear was predicted. Choudhury and Srinivas [107] developed a tool wear regression model, and was able to correlate the cutting parameters with tool wear, wear diffusion coefficient, and tool hardness. With surface roughness signals, Özel and Karpat [106] were able to estimate tool wear accurately by using analysis of variance (ANOVA) and neural network models. In some researches [103, 104, 108, 109] , polynomial and empirical models are also developed and found to be in good correlation with tool wear. The statistical domain requires less computational efforts. However, some of its features mainly depend upon the sample size, otherwise can provide misleading results. Moreover, for machining hard-to-cut materials like titanium or Inconel, where the catastrophic tool failure occurs more often, this method is still questionable in regard to the ability of early stage tool wear or breakage information. Moreover, the relationship function developed by the statistical method is often a linear model, and is unable to explain non-linearity between the independent and the dependent variables. As such, in current autonomous production system, application of this method for a TCMS would not be that useful. In practical industry environments, a product requires one or more cutting processes that are further surrounded by many other random variables and noises, making the individual processes and systems stochastic. Thus, accurate determination of the output parameters like tool wear and breakage, and the relationship between the input and the output parameters are very challenging. For offline TCMS, tool condition is inspected mostly based on the predetermined intervals. The data received from inspection helps the iteration process to adjust the input parameters. In modern automatic or even semi-automatic production system, an intelligent online TCMS needs to identify process anomalies and initiate corrective action with no human intervention. Based on acquired and processed signals, the system should be able to state the present tool condition and estimate the RUL. The artificial intelligence (AI) techniques that work based on classifiers play a very important role in determining such important decision. Artificial neural networks (ANN), fuzzy logic, neuro-fuzzy, generic algorithms, and support vector machines, among others, are in high practice as viable options for online TCMS development. Table 4 presents publications of the AI algorithms or classifiers. In this section, they are briefly overviewed and their viability for practical applications is justified with advantages and limitations . This classifier is highly studied among all the AI techniques, mainly because of its data-driven nature, nonlinear and complex functionality, high fault tolerance and adaptability, and noise suppression capability [8, 13, 110, 111] . Instead of experimental data used in polynomial fitting that offers model trend, NNs work based on experience from these data for establishing an input-output relationship. With experience, NNs can implicitly determine complex nonlinear functions between dependent and independent variables, and their compound interactions [13] . Once a NN is trained with at least one data set, it is able to approximate the future output of the process, like machining. However, as NNs approximate the forecasting results based on the limited data size, the error on the predicted results is observed to be higher in the beginning. More experience with continuous feeding of the data helps better prediction of the [82, 86, 88, 98, 103, 104, 110, [111] [112] [113] [114] [115] [116] , AE [110, 104] , wear image [117, 118] , vibration [103, 86, 114] , power [4, 13] , sound [25, 114, 118] , and temperature [119] . Dornfeld and DeVries [62] studied multilayer perception (MLP) with backpropagation algorithm to perform sensor fusion for AE, force, and current signals. They reported that relatively small-sized NNs correlate well with tool wear estimation. However, they did not find it appropriate for the unsupervised networks due to the lack of self-organizing capability of the backpropagation technique. In an effort, this author [13] also successfully applied a MLP structure for verifying the applicability and reliability of spindle power by correlating with secondary cutting force data and on-machine microscopic images of actual tool wear progress and breakage prediction. With twenty six (26) output wear data for training by five combinations of parameters during drilling of Inconel 625, the average error in prediction was found to be about 6.86% and 6.09% with power and force, respectively. A NN-based TCMS developed by Lee [100] showed that the tool wear can be predicted at 97% accuracy, which can be further improved by increasing the number of hidden layers and nodes. In conclusion, NNs show superior performance for nonlinear complex processes like machining. However, explicit limitations of NNs include training requirement with data, computational burden with software use (e.g., MATLAB, C/C++), proneness to over-fitting, and empirical nature of model development [13] . Also, if the process environment changes, for example, noise, new coolant, or cutting conditions change, data need be reproduced and loaded for the retraining, at least for one new set to start the tool wear prediction process [13, 82] . This means that the TCMS will experience larger error in the beginning. Also, the training is reported to be a time-consuming process. In TCM system development and study, fuzzy logic (FL) has been another popular choice, next to NNs. Fuzzy logic is capable of modeling complex system behavior with fuzziness or a fully defined system with realistic approximation [120] . The basic difference with NNs is that fuzzy logic is capable of direct encoding of structured knowledge in a numerical framework, and can estimate the system functions with even a partial system behavior [121] . The decision making with FL system is faster than NNs due to its simplicity, but it suffers from difficulties in establishing a functional relationship between inputs and outputs, which needs to be done by an expert. Fuzzy logic is applied in machining TCMS for tool wear and failure monitoring from signals including force [120, 122] , and vibration [122] . With fuzzy vertical clustering method, Chen et al. [122] was able to extract information from a large number of data characteristics for turning. Using the fuzzy Taguchi deduction optimization method, a parameter optimization approach was proposed to improve tool wear performance [123] . Ren et al. [120] proposed a TSK fuzzy model, which was found to be effective for tool wear monitoring, though it hinders the ability to estimate the error of approximation. Neuro-fuzzy: Neuro-fuzzy inference techniques, also known as fuzzy neural networks (FNNs), combine the models of neural networks and fuzzy logic systems. They aim to take benefits of both the techniques by achieving the simplicity of modeling from NNs and by providing structured knowledge for complex system behavior offered by fuzzy logic systems [125] . Sharma et al. [126] developed an adaptive neuro-fuzzy inference system (ANFIS) for predicting tool wear from the measured signals by force, vibration, and AE sensors during turning. The overall accuracy was estimated to be 82.9 %. Different neuro-fuzzy approaches are attempted for TCM system development in [125] , which include ANFIS, dynamic evolving neuro-fuzzy inference system (DENFIS), and transductive weighted neuro-fuzzy inference system (TWNFIS). Experiments show that the transductive methods are better than the inductive methods , because incremental online learning with data updates the knowledge in model. The FNN techniques are comparatively new since 2000s; thus, a very few studies are performed so far [87, 88, 125, 126] . For tool wear estimation, researchers paid more attentions rather to realize the advantage of FNN techniques over both NN and fuzzy logics (FL). With processing force signals, Balazinski et al. [87] concluded that the tool wear prediction accuracy is almost the same by these three classifiers. However, FNN perform better, while NNs are more time-consuming model with the training duration and FL system models require some degree of skill and expert knowledge. Rao and Srikant [88] also studied comparison between NN, FL, and FNN classifiers for tool wear prediction with radial cutting force. They concluded that FNN can estimate better results than the other two techniques, but argued that the Kohonen's SOM and FL could be applied for shop floors due to an advantage of less processing time. With vibration signals, Zhang et al. [128] compared these three techniques, and concluded that FNN technique is the best in accuracy (R 2 = 99.2%), followed by NN (R 2 = 98.5%), while FL is the least (R 2 = 73.7%). In summary, although the average tool wear and the error prediction accuracy offered by NNs is highly acceptable, the training duration was found to be higher. Due to this, practical application of NNs in production floors is a bit difficult. The processing time with fuzzy logic system is lower. However, practical use of FL system needs expertise of the operator to analyze and correlate the wear and the input signals. Neurofuzzy (FNN) techniques proved to be more effective due to their better or at least the same wear prediction ability by avoiding the burdens of less processing time and expertise. A drawback is that the calibrated model in FNN techniques can predict tool wear only for the specific cutting parameters for which it is trained. So, if the parameters are modified, the model has to be retrained [126] . Though FNN techniques are suggested to be more viable, only a few researches are done [87, 88, 125, 126] . Morever, they are only based on cutting forces, thus not that practical in the factory floors. However, a recent effort by this author in [13] suggest that, due to close prediction accuracy, the force data are replaceable with power data, which can be received from the spindle motor directly from the outside of the machine tools. Thus, FNN approaches should be further tested and confirmed for such viable signal options, like spindle power, AE, vibration, or in-situ wear image signals in order to implement in unmanned or semiautomatic production floors. The materials discussed above are based on reported studies in research labs. However, successful application of a TCMS in industrial environment is a big challenge. Effectiveness and reliability of a TCMS are evaluated by prediction accuracy of tool wear/failure, repeatability, and sustainability for continuous production with almost no human intervention. This section aims to state some examples for realizing the current state of TCMS applications in industries, and the associated challenges. By collaboration with an industry partner Schivo Precision Ltd located in Waterford, Ireland, Downey et al. [76] tested a TCMS system on a CNC turning center for 160 hours in their live production floor with no human interference. They applied multisensors, including acoustic emission, cutting force, vibration, and automatic image acquisition method of tool wear, and concluded that the combined system can detect tool wear. Some commercial companies offering TCMS application package include Waites Wireless (Kentucky, USA) Vibralign (Virginia, USA), TechSolve Inc. (Ohio, USA), Blum Novotest GmbH (Germany), DXP (Texas, USA), Dynapar (Illionois, USA). In most cases, these companies prefer applying vibration and acceleration sensors as data acquisition methods. With the rapid advancement of technology, optical camera-based machine vision and power/current data show great potential, they are still limited to lab studies. In modern autonomous and semi-autonomous machining production floors, a comprehensive, efficient, reliable, and sustainable tool condition monitoring system (TCMS) is one of the highest interests. Quick and accurate detection of tool wear and failure/breakage, and any process anomaly is very crucial for taking take necessary correction steps . With increased interests, different data/signal acquisition methods with sensors, data extraction and noise removal methods, and critical decision-making AI techniques are explored. In some cases, research direction is also switched due to availability and feasibility of relevant hardware and software systems. This article attempts to review the published researches in those hardware and software systems for realizing TCMS options for industry applications, especially where different machining operations are needed to receive a final product. Conclusions and future suggestions are stated as follows: • For signal or data acquisition, measurement by optical microscope or sensors like CCD camera, force, vibration, AE, and power sensors are promising techniques. • Measurements obtained by optical imaging techniques (tools or parts) are found to be the most accurate (~97%) and repeatable in tool wear prediction. Signal acquisitions with vibration and AE sensors are also simple and viable options for industry settings, but less accurate. • Of online indirect measurement methods, force sensors were well tested and found to be the most accurate. However, their applications are limited to research labs. Requirement of force sensors for multiple operations, higher investments and maintenance costs, and downtime are the major limits for their industry applications. • Power or current measurement method is more straightforward. It received high attention from researchers to develop a simple and cheap, but reliable TCMS. Initial results are found to be comparable with the force data. However, it needs more studies for industry applications. • Offline tools or parts measurements with optical microscopes or other handy instruments at predetermined intervals are still common in manned and semi-automatic industries, but it delays production time. • Recently, the on-machine/online optical measurement method (called 'machine vision') has recently received considerable attention from researchers. However, online imaging needs some special arrangement with lighting and camera, and may be questionable for compact machines as to avoid fluid intervention during machining. More researches are required in this direction. • For feature extraction, time-frequency domain (wavelet analyses) seems to be the best choice, followed by frequency domain, and then time domain. The wavelet transform in the time-frequency domain has a great future for TCM, but more research efforts are needed to prove its superiority over the other domains. • For decision making with AI algorithms, neural networks received the highest attention from the beginning and still seem to be a popular choice, followed by fuzzy logic systems. Neuro-fuzzy technique combines the advantages of the former two techniques , and overcomes all their limitations when tested on force signals. Comprehensive evidences are required for different signal acquisitions like vibration, AE, and power signals. • Some examples of TCMS application in industry reveal that vibration and acceleration sensors are preferable. Reliability, industrial applicability, and sustainability may be the big issues behind the limited application of other DAQ methods. A technique, which is applicable to industrial environment, should be realized and developed. Optical image-based machine vision, power/current and AE data seem highly promising, but more comprehensive studies are needed. A brief review: acoustic emission method for tool wear monitoring during turning The effect of addition of hBN nanoparticles to nanofluid-MQL on tool wear patterns, tool life, roughness and temperature in turning of Ni-based Inconel 625 Effect of machining parameters in ultrasonic vi bration cutting Kur fess TR. Tool life predictions in milling using spindle power with the neural network technique State-of-the-art methods and results in tool condition monitoring: a review Statistical characteristics of drill wear and drill life for the standardized performance tests Tool life testing with single-point turning tools. B94.55M-1985 A review of flank wear prediction methods for tool condition monitoring in a turning process Review of tool condition monitoring methods in milling processes Monitoring and control of manu facturing processes: a review Micro-end milling -III. Wear estimation and tool breakage detection using acoustic emission signals Tool wear monitoring and breakage detection based on intelligent filtering Kur fess TR. Study of spindle power data with neural network for predicting real-time tool wear/breakage during drilling Obstruction-type chip breakers for controllable chips and improved cooling/lubrication during drilling -a feasibility study Detection of tool breakage in microdrilling operation with RCE neural networks Kur fess TR . Intelligent and accessible data flow architectures for manu facturing system optimization State-of-the-art in methods applied to tool condition monitoring (TCM) in un manned machining operations: a review Review of sensor applications in tool condition monitoring in machining Optical flank wear monitoring of cutting tools by image processing Tool wear sensors More than just measure ment: dimensional inspection FAQs On -line tool wear monitoring using geometric descriptors from digital images Use of descriptors based on mo ments fro m digital images for tool wear monitoring Te xture analysis methods for tool condition monitoring Images acquisition of a high -speed boring cutter for tool condition monitoring purposes Surface textural analysis using acousto optic emission-and vision-based 3D surface topography-a base for online tool condition monitoring in face turning Co mputer vision -based approach to end mill tool monitoring Analysis of tool wear using ma chine vision system Using the machine vision method to develop an onma chine insert condition monitoring system for co mputer nu merical control turning machine tools Application of digital image processing in tool condition monitoring: a review Automatic identification of tool wear based on convolutional neural network in face milling process New approach in tool wear me asuring technique using CCD vision system Applications of surface activation in metal cutting A review of machine vision sensors for t ool condition monitoring Tool-life testing with single-point turning tools Flank wear measurement by a threshold independent method with sub-pixel accuracy Je mielniak K. R eal ti me monitoring of the CNC process in a production environment -the data collection & analysis phase Vibration as a para meter for monitoring the health of precision machine tools Advanced monitoring of machining operations Analysis of sound signal generation due to flank wear in turning Analysis of the structure o f vibration signals for tool wear detection Simulated active control of tool vibrations and simultaneous tool condition monitoring Tool wear monitoring through the dynamics o f stable turning What sound can be expected from worn tool? An approach to online measure ment of tool wear by spectrum analysis A data dependent system strategy of on -line tool wear sensing Variation in friction coefficient with tool wear In-process monitoring of tool wear stage by the frequency band-energy method Tool wear and failure monitoring techniques for turning -a review Cutting tool wear monitoring for reliability analysis using proportional hazards model Acoustic emission method for tool condition monitoring based on wavelet analysis A new sensor for real-time milling tool condition monitoring Acoustic emission sensing of tool wear in face milling Real-ti me tool condition sensing with a new class of sensor-transducer system Advanced monitoring of tool wear and cutting states in CNC turning process by utilizing sensor fusion A study of tool wear using statistical analysis of metal-cutting acoustic emission Sensor signals for tool-wear monitoring in metal cutting operations-a review of methods A new approach to automatic detection of life o f coated tool based on acoustic emission measure ment A study on acoustic emission from the orthogonal cutting process Tool wear monitoring of turning operations by neural network and expert system classification of a feature set generated from multiple sensors Neural network sensor fusion for tool condition monitoring Development of a monitoring technique for tool change purpose in turning operations Balakrishnan S P redicting tool flank wear using spindle speed change A co mprehensive tool wear model Tool life in oblique cutting as a function of computed flank contact temperature Flank wear prediction of cera mic tools in hard turning On te mperature measure ment during titanium ma chining with the atomization-based cutting fluid (ACF) spray system Re al-time monitoring of flank wear behavior of cera mic cutting tool in turning hardened steels Detecting tool breakage in turning AISI 1050 steel using coated and uncoated cutting tools The application of force and acoustic emission sensors for detecting tool damage in turning processes Tool condition monitoring based on numerous signal features Automatic multiple sensor data acquisition system in a real-time production environment A multisensor fusion method for tool condition monitoring in milling Development of a miniature abrasion-detecting device for a s mall precision lathe Online flank wear estimation using an adaptive observer and co mputer vision, part 2: experi ment. Tr ans ASME Modelling of tool wear based on cutting forces in turning On-line metal cutting tool condition monitoring: I: force and vibration analyses Neural network modeling of flank wear for tool condition monitoring in orthogonal cutting of hardened steels Flank wear estimation under varying cutting conditions The development o f ma chine-tool force reconstruction for wear identification Relationship between tool flank wear area and component forces in single point turning On-line metal cutting tool condition monitoring.: II: tool-state classification using multi-layer perceptron neural networks Tool condition moni toring using artificial intelligence methods Tool wear monitoring -an intelligent approach Acoustic emission fro m single point machining: source mechanisms and signal changes with tool wear Real ti me monitoring of sur face roughness by acoustic emissions in CNC turning A machine vision approach to tool wear monitoring based on the image of workpiece surface texture Assessment of the suitability of industrial robots for the machining of carbon-fiber rein forced polymers (CF RP s) E ffect o f differ ent tool edge conditions on wear detection by vibration spectrum analysis in turning operation Study on metal cutting monitor using microphone signal Detection of tool flank wear using acoustic signature analysis Correlation between acoustic e mission and wear o f multi-layer ceramic coated carbide tools On the correlation between dynamic cutting force and tool wear Development o f a tool wearmonitoring system for hard turning Hidden Markov model-based tool wear monitoring in turning Lee S Tool condition monitoring system in turning operation utilizing wavelet signal processing and multi-learning ANNs algorithm methodology Wavelet analysis of sensor signals for tool condition monitoring: a review and some new results Development of models for tool wear force relationships in metal cutting Development o f a cutting tool condition monitoring system for high speed turning operation by vibration and strain analysis On modeling of tool wear using sensor fusion and polynomial classifiers Tool condition monitoring using laser scatter pattern P redictive modeling of surface roughness and tool wear in hard turning using regression and neural networks Tool wear prediction in turning Application of response surface methodology for predicting flank wear in turning operation Sugumar an V. Studies on Bayes classifier for condition monitoring of single point carbide tipped tool based on statistical and histogram features Sensor integration using n eural networks for intelligent tool condition monitoring On-line wear estimation using neural networks Force para meters for on -line tool wear estimation: a neural network approach On-line monitoring of flank wear in turning with multilayered feed-forward neural network Development of a system for monitoring tool wear using artificial intelligence techniques Design of neural network-based estimator for tool wear modeling in hard turning Analyzing the perfor mance o f a rtificial neural network (ANN)-, fuzzy logic (FL)-, and least square (LS)-based models for online tool condition monitoring Estimation of cutting tool life by processing tool image data with neural network Application of i mage and sound analysis techniques to monitor the condition of cutting tools Online prediction of diffusion wear on the flank through tool tip temperature in turning using artificial neural networks TSK fuzzy modeling for tool wear condition in turning processes: an experimental study Development of a wear monitoring system for turning tools using artificial intelligence. P h.D. thesis, Univ . of P retoria Turning tool wear monitoring based on fuzzy cluster analysis Tool wear optimization for general CNC turning using fuzzy deduction Optimization of Machining P arameters Using Fuzzy Taguchi Method for Reducing Tool Wear Tool wear monitoring using neuro-fuzzy techniques: a comparative study in a turning process Cutting tool wear estimation for turning Neuro -fuzzy predictive model for surfa ce roughness and cutting force o f machined Al -20 Mg2Si-2Cu metal matrix co mposite using additives Tool condition monitoring and remaining useful life prognostic based on a wireless sensor in dry milling operations