threshold a ‘decision’ is made that the tool is worn. The success of this strategy depends upon the degree to which the mean value of the sensor output actually represents the state (and progress) of tool wear. 1.2.4.2 Sensor Fusion With a specific focus for the monitoring in mind, researchers have developed over the years a wide variety of sensors and sensing strategies, each attempting to pre- dict or detect a specific phenomenon during the operation of the process and in the presence of noise and other environmental contaminants. A good number of these sensing techniques applicable to manufacturing have been reviewed in the early part of this chapter. Although able to accomplish the task for a narrow set of conditions, these specific techniques have almost uniformly failed to be reliable enough to work over the range of operating conditions and environments com- monly available in manufacturing facilities. Therefore, researchers have begun to look at ways to collect the maximum amount of information about the state of a process from a number of different sensors (each of which is able to provide an output related to the phenomenon of interest although at varying reliability). The strategy of integrating the information from a variety of sensors with the expecta- tion that this will ‘increase the accuracy and . resolve ambiguities in the knowl- edge about the environment’ (Chiu et al. [14]) is called sensor fusion. Sensor fusion is able to provide data for the decision-making process that has a low uncertainty owing to the inherent randomness or noise in the sensor signals, includes significant features covering a broader range of operating conditions, and accommodates changes in the operating characteristics of the individual sensors (due to calibration, drift, etc.) because of redundancy. In fact, perhaps the most advantageous aspect of sensor fusion is the richness of information available to the signal processing/feature extraction and decision-making methodology em- ployed as part of the sensor system. Sensor fusion is best defined in terms of the ‘intelligent’ sensor as introduced in [15] since that sensor system is structured to utilize many of the same elements needed for sensor fusion. The objective of sensor fusion is to increase the reliability of the information so that a decision on the state of the process is reached. This tends to make fusion techniques closely coupled with feature extraction methodologies and pattern rec- ognition techniques. The problem here is to establish the relationship between the measured parameter and the process parameter. There are two principal ways to encode this relationship (Rangwala [13]): · theoretical – the relationship between a phenomenon and the measured param- eters of the process (say tool wear and the process); and · empirical – experimental data is used to tune parameters of a proposed model. As mentioned earlier, reliable theoretical models relating sensor output and pro- cess characteristics are often difficult to develop because of the complexity and variability of the process and the problems associated with incorporating large numbers of variables in the model. As a result, empirical methods which can use 1.2 Principles of Sensors in Manufacturing 21 sensor data to tune unknown parameters of a proposed relation are very attrac- tive. These types of approaches can be implemented by either (a) proposing a rela- tionship between a particular process characteristic and sensor outputs and then using experimental data to tune unknown parameters of a model, or (b) associat- ing patterns of sensor data with an appropriate decision on the process state with- out consideration of any model relating sensor data to the state. The second approach is generally referred to as pattern recognition and involves three critical stages (Ahmed and Rao [16]): · sampling of input signal to acquire the measurement vector; · feature selection and extraction; · classification in the feature space to permit a decision on the process state. The pattern recognition approach provides a framework for machine learning and knowledge synthesis in a manufacturing environment by observation of sensor data and with minimal human intervention. More important, such an approach allows for integration of information from multiple sources (such as different sen- sors) which is our principal interest here. Sata et al. [17, 18] were among the first researchers to propose the application of pattern recognition techniques to machine process monitoring. They attempted to recognize chip breakage, formation of built-up edge and the presence of chatter in a turning operation using the features of the spectrum of the cutting force in the 0–150 Hz range. Dornfeld and Pan [19] used the event rate of the rms energy of an acoustic emission signal along with feed rate and cutting velocity in order to provide a decision on the chip formation produced during a turning operation. Emel and Kannatey-Asibu [20] used spectral features of the acoustic emission sig- nal in order to classify fresh and worn cutting tools. Balakrishnan et al. [21] use a linear discriminant function technique to combine cutting force and acoustic emission information for cutting tool monitoring. The manufacturing process may be monitored by a variety of sensors and, typi- cally, the sensor output is a digitized time-domain waveform. The signal can then be either processed in the time domain (eg, extract the time series parameters of the signal) or in the frequency domain (power spectrum representation). The ef- fect of this is to convert the original time-domain record into a measurement vec- tor. In most cases, this mapping does not preserve information in the original sig- nal. Usually, the dimension of the measurement vector is very high and it be- comes necessary to reduce this dimension due to computational considerations. There are two prevalent approaches at this stage: select only those components of the measurement vector which maximize the signal-to-noise ratio or map the measurement vector into a lower dimensional space through a suitable transfor- mation (feature extraction). The outcome of the feature selection/extraction stage is a lower dimensional feature vector. These features are used in pattern recogni- tion techniques and as inputs to sensor fusion methodologies. This was illus- trated in Figure 1.2-6. 1 Fundamentals22 1.2.5 Summary The subject of sensors for manufacturing processes is well covered in other chapters of this book. The material in this chapter serves to acquaint the reader with the clas- sification of sensor systems and some of the measurands that are associated with these sensors. How these sensor types and measurands map on to the various man- ufacturing processes will be the subject of the rest of the text. One important factor in the implementation of sensors in manufacturing is clearly the rapid growth of silicon micro-sensors based on MEMS technology. This technology already allows the integration of traditional and novel new sensing methodologies on to miniatur- ized platforms, providing in hardware the reality of multi-sensor systems. Further, since these sensors are easily integrated with the electronics for signal processing and data handling, on the same chip, sophisticated signal analysis including feature extraction and intelligent processing will be straightforward (and inexpensive). This bodes well for the vision of the intelligent factory with rapid feedback of vital infor- mation to all levels of the operation from machine control to process planning. 1.2 Principles of Sensors in Manufacturing 23 1.2.6 References 1 Sze, S.M. (ed.) Semiconductor Sensors; New York: Wiley, 1994. 2 Allocca, J. A., Stuart, A., Transducers: Theory and Applications; Reston, VA: Re- ston Publishing, 1984. 3 Bray, D. E., McBride, D. (eds.) Nondes- tructive Testing Techniques; New York: Wi- ley, 1992. 4 Webster’s Third New International Diction- ary; Springfield, MA: G. C. Merriam, 1971. 5 Usher, M. J., Sensors and Transducers; New Hampshire: Macmillan, 1985. 6 Middlehoek, S., Audet, S. A., Silicon Sensors; New York: Academic Press, 1989. 7 White, R. M., IEEE Trans. Ultrason. Fero- elect. Freq. Contr. UFFC-34 (1987) 124. 8 Shiraishi, M., Precision Eng. 10(4) (1988) 179–189. 9 Shiraishi, M., Precision Eng. 11(1) (1989) 27–37. 10 Shiraishi, M., Precision Eng. 11(1) (1989) 39–47. 11 Byrne, G., Dornfeld, D., Inasaki, I., Kettler, G., König, W., Teti, R., Ann. CIRP 44(2) (1995) 541–567. 12 Goch, G., Schmitz, B., Karpuschewski, B., Geerkins, J., Reigel, M., Sprongl, P., Ritter, R., Precision Eng. 23 (1999) 9–33. 13 Rangwala, S., PhD Thesis; Department of Mechanical Engineering, University of Ca- lifornia, Berkeley, CA, 1988. 14 Chiu, S. L., Morley, D. J., Martin, J. F., in: Proceedings of 1987 IEEE International Conference on Robotics and Automation; Ra- leigh, NC: IEEE, 1987, pp. 1629–1633. 15 Dornfeld, D. A., Ann. CIRP 39 (1990) 16 Ahmed, N., Rao, K. K., Orthogonal Trans- forms for Digital Signal Processing; New York: Springer, 1975. 17 Sata, T., Matsushima, K., Nagakura, T., Kono, E., Ann. CIRP 22(1) (1973) 41–42. 18 Matsushima, K., Sata, T., J. Fac. Eng. Univ. Tokyo (B) 35(3) (1980) 395–405. 19 Dornfeld, D.A., Pan, C.S., in: Proceedings of 13th North American Manufacturing Re- search Conference, University of California, Berkeley, CA: SME, 1985, pp. 285–303. 20 Emel, E., Kannatey-Asibu, E., Jr., in: Pro- ceedings of 14th North American Manufactur- ing Research Conference, University of Min- nesota, MN: SME, 1986, pp. 266–272. 21 Balakrishnan, P., Trabelsi, H., Kanna- tey-Asibu, Jr., E., Emel, E., in: Proceedings of 15th NSF Conference on Production Re- search and Technology, University of Cali- fornia, Berkeley, CA: SME, 1989, pp. 101– 108. 1.3 Sensors in Mechanical Manufacturing – Requirements, Demands, Boundary Conditions, Signal Processing, Communication Techniques, and Man-Machine Interfaces T. Moriwaki, Kobe University, Kobe, Japan 1.3.1 Introduction The role of sensor systems for mechanical manufacturing is generally composed of sensing, transformation/conversion, signal processing, and decision making, as shown in Figure 1.3-1. The output of the sensor system is either given to the op- erator via a human-machine interface or directly utilized to control the machine. Objectives, requirements, demands, boundary conditions, signal processing, com- munication techniques, and the human-machine interface of the sensor system are described in this section. 1.3.2 Role of Sensors and Objectives of Sensing An automated manufacturing system, in particular a machining system, such as a cutting or grinding system, is basically composed of controller, machine tool and machining process, as illustrated schematically in Figure 1.3-2. The machining command is transformed into the control command of the actuators by the CNC 1 Fundamentals24 Fig. 1.3-1 Basic composition of sensor system for mechanical manufacturing Fig. 1.3-2 Role of sensors in automated machining system Sensors in Manufacturing. Edited by H.K. Tönshoff, I. Inasaki Copyright © 2001 Wiley-VCH Verlag GmbH ISBNs: 3-527-29558-5 (Hardcover); 3-527-60002-7 (Electronic) controller, which controls the motion of the actuators and generates the actual machining motion of the machine tool. The motion of the actuator, or the ma- chining motion of the machine tool, is fed back to the controller so as to ensure that the relative motion between the tool and the work follows exactly the prede- termined command motion. Motion sensors, such as an encoder, tacho-generator or linear scale, are generally employed for this purpose. The machining process is generally carried out beyond this loop, where fin- ished surfaces of the work are actually generated. Most conventional CNC ma- chine tools currently available on the market are operated under the assumption that the machining process normally takes place once the tool work-relative mo- tion is correctly given. Some advanced machine tools equipped with an AC (adap- tive control) function utilize the feedback information of the machining process, such as the cutting force, to optimize the machining conditions or to stop the ma- chine tool in case of an abnormal state such as tool breakage. The machining process normally takes place under extreme conditions, such as high stress, high strain rate, and high temperature. Further, the machining pro- cess and the machine tool itself are exposed to various kinds of external distur- bances including heat, vibration, and deformation. In order to keep the machin- ing process normal and to guarantee the accuracy and quality of the work, it is necessary to monitor the machining process and control the machine tool based on the sensed information. The objectives and the items to be sensed and monitored for general mechani- cal manufacturing are summarized in Table 1.3-1 together with the direct pur- poses of sensing and monitoring. Some items can be directly sensed with proper sensors, but they can be utilized to estimate other properties at the same time. For instance, the cutting force is sensed with a tool dynamometer to monitor the cutting state, but its information can be utilized to estimate the wear of the cut- ting tool simultaneously. Almost all kinds of machining processes require sensing and monitoring to maintain high reliability of machining and to avoid abnormal states. Table 1.3-2 gives a summary of the answers to a questionnaire to machine tool users asking about the machining processes which require monitoring [1]. It is understood that monitoring is imperative especially when weak tools are used, such as in tapping, drilling, and end milling. 1.3 Sensors in Mechanical Manufacturing 25 1 Fundamentals26 Tab. 1.3-1 Objects, items, and purposes of sensing Object of sensing and monitoring Items to be sensed Purpose of sensing and monitoring Work State of work clamping Geometrical and dimensional accuracy Surface roughness Surface quality Maintain high quality Avoid damage and loss of work Machining process Force (torque, thrust) Heat generation Temperature Vibration Noise and sound State of chip Maintain normal machining process Predict and avoid abnormal state Tool Tool edge position Wear Damage including chipping, breakage, and others Manage tool changing time, including dressing Avoid damage or deterioration of work Machine tool, and auxiliary facility Malfunction Vibration Deformation (elastic, thermal) Maintain normal condition of ma- chine tool and assure high accu- racy Environment Ambient temperature change External vibration Condition of cutting fluid Minimize environmental effect Tab. 1.3-2 Machining processes which require sensing Kind of machining Number of answers Percentage Tapping Drilling End milling Internal turning External turning Face milling Parting Thread cutting Others* Total 67 66 55 51 30 25 17 13 15 338 19.8 19.2 16.8 15.1 8.9 7.4 5.0 3.9 4.4 100 * Grinding, reaming, deep hole boring, etc. 1.3.3 Requirements for Sensors and Sensing Systems The most important and basic part of the sensor is the transducer, which trans- forms the physical or sometimes chemical properties of the object into another physical quantity such as electric voltage that is easily processed. The properties of the object to be sensed are either one-dimensional, such as force and tempera- ture, or multi-dimensional, such as image and distribution of the physical proper- ties. The multi-dimensional properties are treated either as plural signals or a time series of signals after scanning. The basic requirements for the transducers and sensor systems for mechanical manufacturing are summarized in Table 1.3-3. Figure 1.3-3 shows a schematic il- lustration of the characteristics of a typical transducer, such as a force transducer. 1.3 Sensors in Mechanical Manufacturing 27 Tab. 1.3-3 Basic requirements for transducers and sensing systems Performance/ accuracy Reliability Adaptability Economy Sensitivity Resolution Exactness Precision Linearity Hysteresis Repeatability Signal-to-noise ratio Dynamic range Dynamic response Frequency response Cross talk Low drift Thermal stability Stability against environment, such as cutting, fluid and heat Low deterioration Long life Fail safe Low emission of noise Compact in size Light in weight Easy operation Easy to be installed Low effect of ma- chining process and machine tool Safety Good connectivity to other equipment Low cost Easy to manufacture Easy to purchase Low power requirement Easy to calibrate Easy maintenance Fig. 1.3-3 Typical input-output relation of transducer Nonlinear range The figure represents the relation between the change in a property of the object, or the input and the output of the transducer. It is desirable that the transducer output represents the property of the object as exactly and precisely as possible. It is also essential for a transducer to output the same value at any time when the same amount of input is given. This characteristic is called repeatability. In most cases, the output increases or decreases in proportion to the input in the linear range, and then gradually saturates and becomes almost constant. When the amount of input exceeds the limit of sensing, the transducer becomes normally malfunctioning. The measurable range of the input is called the dynamic range of the sensor. The ratio of output to input is called the sensitivity, and it is desirable that the sensitivity is high and the linear range of sensing is wide. The input-output rela- tion is sometimes nonlinear depending on the principle of the transducer, as in the case of capacitive type proximeter (see Figure 1.3-4). Only a small range of lin- ear input-output relation can be used in such a case when the accuracy require- ment of sensing is high. When the nonlinear input-output relation is known ex- actly by calibration or by other methods in advance, the nonlinearity can be com- pensated afterwards by calculation. The nonlinear characteristics of thermocouples are well known, and the compensation circuits are installed in most thermo- meters for different types of thermocouples. The input-output relation sometimes differs when the amount of input is in- creased and decreased, as shown in Figure 1.3-5. Such a characteristic is called hysteresis, and is sometimes encountered when a strain gage sensor is employed to measure the strain or the force. It is almost impossible to compensate for the hysteresis of the transducer, hence it is recommended to select transducers with small hysteresis. The property of the object to be sensed in mechanical manufacturing is gener- ally time varying or dynamic. The measurable dynamic range of the transducer is generally limited by the maximum velocity and acceleration of the output signal 1 Fundamentals28 Fig. 1.3-4 Nonlinear input-output relation + +– – and also by the maximum frequency to which the change in the input property can be exactly transformed to the output. Figure 1.3-6 shows typical frequency characteristics of the transducers in terms of the frequency response. The vertical axis shows the gain or the ratio of the magnitudes of the output to the input, and also the phase or the delay of the output signal to the input. Some transducers show resonance characteristics, and the gain in terms of out- put/input becomes relatively larger at the resonant frequency. It should be noted that the phase is shifted for about k/2 at the resonant frequency. The phase shift in the output signal cannot be avoided generally even with well-damped type or non-resonant type transducers, as shown in the figure. The sinusoidal wave forms of the input and the output at some typical frequen- cies are shown in Figure 1.3-7 to illustrate the changes in the gain and the phase. When the phase information is essential to identify the state of the object, it is important to select a transducer with resonant frequency high enough compared with the frequency range of the phenomenon to be sensed. 1.3 Sensors in Mechanical Manufacturing 29 Fig. 1.3-5 Hysteresis in input-output relation Fig. 1.3-6 Frequency response of typical transducers + +– – –p As was mentioned before, the machining process normally takes place under high-stress, high-strain rate and high-temperature conditions with various kinds of external disturbances including the cutting and grinding fluids. It is therefore understood that high reliability and stability against various kinds of disturbances are the most important requirements for the sensors in addition to the basic per- formance and accuracy of the transducers. According to the answers given by in- dustry engineers to the questionnaire concerning tool condition monitoring [2], the importance of technical criteria in selecting the sensors is in the order (1) reli- ability against malfunctioning, (2) reliability in signal transmission, (3) ease of in- stallation, (4) life of the sensor, and (5) wear resistance of the sensor. The importance of items in evaluating the monitoring system is also given in the order (1) reliability against malfunctions, (2) performance to cost ratio, (3) in- formation obtained by the sensor, (4) speed of diagnosis, (5) adaptability to changes of process, (6) usable period, (7) ease of maintenance and repair, (8) level of automation, (9) ease of installation, (10) standard interface, (11) standardized user interface, (12) completeness of manuals, and (13) possibility of additional functions. Table 1.3-4 summarizes items to be considered generally in selecting transdu- cers and the sensors. It is basically desirable to implement on-line, in-process, continuous, non-contact, and direct sensing, but it is generally difficult to satisfy all of these requirements. The property of the object is directly sensed in the case of direct sensing, whereas in the case of indirect sensing it is estimated indirectly from other properties which can be easily measured and are related to the prop- erty to be measured. It should be noted that the property of object to be estimated indirectly must have a good correlation with the property to be measured. Indirect sensing is useful and is widely adopted when direct sensing is difficult. 1 Fundamentals30 Fig. 1.3-7 Relation of input and output at some typical frequencies . described in this section. 1.3.2 Role of Sensors and Objectives of Sensing An automated manufacturing system, in particular a machining system, such as a cutting. manufacturing Fig. 1.3-2 Role of sensors in automated machining system Sensors in Manufacturing. Edited by H.K. Tönshoff, I. Inasaki Copyright © 2001 Wiley-VCH