Scholarly article on topic 'Monitoring Single-point Dressers Using Fuzzy Models'

Monitoring Single-point Dressers Using Fuzzy Models Academic research paper on "Mechanical engineering"

CC BY-NC-ND
0
0
Share paper
Academic journal
Procedia CIRP
OECD Field of science
Keywords
{Dressing / "Fuzzy logic" / Wear / Grinding / Acoustic / Vibration}

Abstract of research paper on Mechanical engineering, author of scientific article — H.I. Miranda, C.A. Rocha, P. Oliveira, C. Martins, P.R. Aguiar, et al.

Abstract Grinding causes progressive dulling and glazing of the grinding wheel grains and clogging of the voids on the wheel's surface with ground metal dust particles, which gradually increases the grinding forces. The condition of the grains at the periphery of a grinding wheel strongly influences the damage induced in a ground workpiece. Therefore, truing and dressing must be carried out frequently. Dressing is the process of conditioning the grinding wheel surface to reshape the wheel when it has lost its original shape through wear, giving the tool its original condition of efficiency. Despite the very broad range of dressing tools available today, the single-point diamond dresser is still the most widely used dressing tool due to its great versatility. The aim of this work is to predict the wear level of the single-point dresser based on acoustic emission and vibration signals used as input variables for fuzzy models. Experimental tests were performed with synthetic diamond dressers on a surface-grinding machine equipped with an aluminum oxide grinding wheel. Acoustic emission and vibration sensors were attached to the tool holder and the signals were captured at 2MHz. During the tests, the wear of the diamond tip was measured every 20 passes using a microscope with 10 to 100 X magnification. A study was conducted of the frequency content of the signals, choosing the frequency bands that best correlate with the diamond's wear. Digital band-pass filters were applied to the raw signals, after which two statistics were calculated to serve as the inputs for the fuzzy models. The results indicate that the fuzzy models using the aforementioned signal statistics are highly effective for predicting the wear level of the dresser.

Academic research paper on topic "Monitoring Single-point Dressers Using Fuzzy Models"

Available online at www.sciencedirect.com

ScienceDirect

Procedía CIRP 33 (2015) 281 - 286

www.elsevier.com/locate/procedia

9th CIRP Conference on Intelligent Computation in Manufacturing Engineering - CIRP ICME '14

Monitoring single-point dressers using fuzzy models

Miranda, H. I.a, Rocha, C. A.a, Oliveira Jr, P.a, Martins, C.a, Aguiar, P. R.a*, Bianchi, E. C.b

aUniv. Estadual Paulista - UNESP - Faculty of Engineering, Department of Electrical Engineering, 17033-360, Bauru, SP, Brazil bUniv. Estadual Paulista - UNESP - Faculty of Engineering, Department of Mechanical Engineering, 17033-360, Bauru, SP, Brazil

* Corresponding author. Tel.:+55 14 3103-6115 E-mail address: aguiarpr@feb.unesp.br

Abstract

Grinding causes progressive dulling and glazing of the grinding wheel grains and clogging of the voids on the wheel's surface with ground metal dust particles, which gradually increases the grinding forces. The condition of the grains at the periphery of a grinding wheel strongly influences the damage induced in a ground workpiece. Therefore, truing and dressing must be carried out frequently. Dressing is the process of conditioning the grinding wheel surface to reshape the wheel when it has lost its original shape through wear, giving the tool its original condition of efficiency. Despite the very broad range of dressing tools available today, the single-point diamond dresser is still the most widely used dressing tool due to its great versatility. The aim of this work is to predict the wear level of the single-point dresser based on acoustic emission and vibration signals used as input variables for fuzzy models. Experimental tests were performed with synthetic diamond dressers on a surface-grinding machine equipped with an aluminum oxide grinding wheel. Acoustic emission and vibration sensors were attached to the tool holder and the signals were captured at 2MHz. During the tests, the wear of the diamond tip was measured every 20 passes using a microscope with 10 to 100 X magnification. A study was conducted of the frequency content of the signals, choosing the frequency bands that best correlate with the diamond's wear. Digital band-pass filters were applied to the raw signals, after which two statistics were calculated to serve as the inputs for the fuzzy models. The results indicate that the fuzzy models using the aforementioned signal statistics are highly effective for predicting the wear level of the dresser.

© 2014 The Authors. Published by Elsevier B.V.This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

Selection and peer-review under responsibility of the International Scientific Committee of "9th CIRP ICME Conference" Keywords: Dressing; Fuzzy logic; Wear; Grinding; Acoustic; Vibration

1. Introduction

Increasing interest has focused on the study and application of the grinding process, because it is the finishing process most widely used in the metal and mechanical industry to machine workpieces with better surface quality and tight tolerances. Moreover, the ever expanding use of CNC grinding machines in the industrial sector makes it increasingly necessary to study the process as it pertains to automation capacity.

Despite the advances achieved in research to eliminate problems that affect the normal operation of grinding machines, which generally cause stoppages and require corrections to be made by an experienced operator, usually manually, many problems still need solutions. This is due mainly to the high complexity of the process, which results from the numerous variables involved.

Dressing, on the other hand, is the operation of conditioning the surface of the grinding wheel to restore its original profile and reestablish the characteristics of sharpness that were lost due to wear during the grinding process. This operation is responsible for producing a satisfactory topography on the grinding wheel, which has a significant impact on the total grinding force, energy, temperatures, wheel wear and finish of ground surfaces [1].

However, dressing tool (dresser) wear has a stronger effect on the surface roughness of the workpiece than do other parameters of the grinding process, such as the depth of cut and feed velocity [1]. According to Habrat in the writings of Martins et al. [2], monitoring the wear of the tip of the dresser will allow for adequate control of the grinding process in terms of required workpiece roughness.

The sensors most commonly used to monitor machining processes are force, power, acoustic emission (AE) and

2212-8271 © 2014 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

Selection and peer-review under responsibility of the International Scientific Committee of "9th CIRP ICME Conference" doi:10.1016/j.procir.2015.06.050

vibration sensors, which have become increasingly present in manufacturing plants that strive to improve their quality and productivity. When handled properly, the data collected by sensors can provide information required for the accurate diagnosis of problems in machines and processes, as well as real analyses of the productivity and status of production, for example.

Artificial intelligence models have been designed to estimate or predict important parameters of machining processes, using as inputs signals collected by one or more sensors and parameters derived from these signals. In the case of fuzzy logic, according to Pham and Pham [3], these models reflect the qualitative and inexact nature of human judgment, and thus make specialized systems more flexible. In fuzzy logic, the precise value of a variable is replaced by a linguistic description represented by a fuzzy set, and inferences are made based on this representation.

This article contributes by presenting a methodology to estimate the wear of single-point dressers based on acoustic emission and vibration signals and fuzzy models. It is an extension of the work of Martins et al. [2], which used only acoustic emission signals and neural models for estimating dresser wear.

One of the reasons for the development of this work was the possibility of creating a model more closely resembling the language used by machining operators to refer to monitored wheel dresser wear. This motivation is based on the work of Shaw in Ali and Zhang [4], who report that a production operation does not require an absolute model that can deliver high accuracy, since such a model is not reproduced in practice. What is needed, according to the authors, is a "relative" model that can provide general guidelines for the user about what to do and how to do it. The reason for this is that there will always be some degree of trial and error on the shop floor, so a relative model serves as a good starting point.

2. Monitoring the Dressing Operation

Dressing conditions can strongly influence the performance of the grinding operation. To exemplify this influence, suffice it to say that grinding forces can vary by about 500% simply by varying the dressing conditions in the same type of operation [5].

The efficiency of the grinding operation is highly dependent on the grinding wheel surface. However, Badger [6] reported that thermal damage during the grinding process can be reduced significantly by adopting correct operating parameters and dressing strategies.

The dressing process can be monitored to produce consistent grinding wheel surface quality. The service life of the grinding wheel can be defined by monitoring changes in the characteristic amplitude and frequency of the acoustic emission signal [7]. However, the market still lacks devices that can evaluate the grinding wheel surface during the grinding operation in a manufacturing environment.

The use of acoustic emission to monitor the grinding process has been a subject of research since 1984. One of the first conclusions of these researches was the high sensitivity

of the root mean square (RMS) value of the AE signal in detecting contact between the grinding wheel and workpiece [8]. Other parameters and/or statistics of the raw AE signal are reported in the literature. A statistic that has been employed successfully to estimate dresser wear is the ratio of power (ROP) of the AE signal [2], which is defined by equation (1).

II XK\

k=" ~Ñ-1

II Xrf

where Xk is the k-th discrete Fourier transform, n1 and n2 are values that define the frequency range to be analyzed, and N is the size of the AE data block. The denominator of the equation eliminates the effect of local power.

On the other hand, vibration (acceleration) signals are rarely used to monitor the process, perhaps due to the lack of researchers interested in exploring the characteristics of this signal in the grinding and dressing processes, or the mistaken belief that the vibration signal will always be influenced by frequencies deriving from ambient noise and other processes. The vibration sensor captures the vibrations produced by cyclic variations in the dynamic components of cutting forces. Metal cutting processes emit free, forced, periodic and random vibration signals. It is difficult to measure vibration directly because the vibration mode is frequency dependent. Therefore, measurements are taken by means of vibration-related parameters, such as the rate at which the dynamic forces change per unit time (acceleration), and the characteristics of the vibration are extracted from the patterns thus obtained [9].

Several studies have focused on correlating the vibration signal with the characteristics of machining processes. For example, Hassui et al. [10] showed that the RMS signal of workpiece vibration is better correlated with grinding wheel wear than the acoustic emission signal. Furthermore, the sensitivity of the vibration sensor is as good as that of the acoustic emission sensor in detecting the moment of contact between the grinding wheel and workpiece and the end of spark-out. Hassui and Diniz [11] examined the ability of the vibration signal to perceive variations in workpiece roughness and roundness, and hence, the feasibility of its use in defining the moment of dressing. Their findings indicated a good correlation between vibration signals and the roughness and condition of the grinding wheel.

Other studies have been published about monitoring the dressing operation, but only a few studies have focused on the wear of single-point dressers. Shin and Zhang [12] studied the wear properties of single-point dressers in laser-assisted truing and dressing of vitrified CBN grinding wheels. Their experimental results indicated that dresser wear rates depend on laser power, dressing depth, and feed rate. The authors also concluded that the heating history influences dresser wear to some extent. The work of Habrat and Porzycki [13] describes an optical monitoring system for measuring the equivalent radius of the diamond dresser, and makes inferences about its wear. Martins et al.'s work [14] describes a study to classify

single-point dresser wear by means of AE RMS signals, neural networks and Kohonen maps. Their results showed that the Kohonen map produced a better classification rate, but at a higher computational cost. Martins et al. [2] presented a method to characterize the wear condition of single-point dressers based on acoustic emission signals. Investigations have also focused on neural models whose inputs were the RMS values of AE obtained from preselected frequencies. The results demonstrated that certain models are very good at classifying dresser wear.

3. Fuzzy Modeling

According to Ali and Zhang [4] a fuzzy model can be considered a specialized system based on rules with the added benefits of the fuzzy set theory. This theory, which was introduced by Zadeh (1965) and is built upon well established concepts of classical logic, is aimed at providing a mathematical methodology for treating imprecise or vague information. Zadeh proposed a broader characterization, generalizing the characteristic function typical of classical logic, so that it could assume an infinite number of values in the [0,1] interval. For example, when the input values do not correspond exactly to one of the input conditions (assumptions) of the existing rules, a conventional specialist system cannot trigger any rule, and thus fails to provide any output (consequence). A fuzzy system can overcome this disadvantage, since it will trigger at least one rule for any set of input values regardless of the completeness or accuracy of the values, and will work even in the absence of some input values.

The fuzzy sets for a given universe are given names, which are known as linguistic variables. The main function of the linguistic variables is to provide a systematic way for the approximate characterization of phenomena that are usually imprecise, as in the case of dresser wear. The use of this type of linguistic description used by humans, instead of quantitative variables, allows for the treatment of situations that are too complex to be analyzed by means of conventional mathematical terms. These characteristics are obtained by a simple mechanism, i.e., by membership functions, which are fuzzy sets.

4. Materials and Methods

4.1. Experimental tests

The test bench was designed to allow for the analysis of the wear conditions of the dressing tool from raw acoustic emission (AE) and acceleration signals collected by a Yokogawa DL850 oscilloscope, at a rate of 2 million samples per second.

The tests were performed for the dressing operation of a conventional type 38220KVS aluminum oxide grinding wheel with dimensions of 127 x 12.7 x 355.6 mm manufactured by NORTON. The operating parameters were controlled carefully to ensure the same conditions for all the dressers.

A single-point synthetic diamond dressing tool (Chemical Vapor Deposition - CVD) was used in the dressing tests for

the study its wear. Measurements of the diamond tip were taken at intervals of 20 passes in each test, using an optical system and CAD software to calculate the worn area. The dressing test consisted of a number of passes of the diamond on the surface of the grinding wheel until the dresser reached its end of life, i.e., until metallic material was visibly in contact with the material of the grinding wheel. The dressing tests were carried out with three replications. The dresser speed was kept constant at 3.45 mm/s, with a dressing depth of 40 ^m, and without the use of cutting fluid in order to induce faster wear.

The raw AE signals were captured in real time, using an acoustic emission system comprising a fixed sensor with 0 to 400 kHz spectrum and a Sensis DM42 signal processing module. The acceleration signals were captured with a model 353B03 fixed sensor and a model 482B11 signal conditioning module, both from PCB Piezotronics, Inc. The two sensors were attached to the dresser holder.

4.2. Signal Spectrum and Input and Output Vectors

The raw AE and vibration signals were initially analyzed in the frequency domain in an attempt to identify frequency bands more related with dresser wear. Frequency spectra were obtained for three different stages of dresser wear (new, half-life, and worn). The discrete Fourier transform was used to obtain the spectra, implemented in Matlab by the fast Fourier transform (FFT) algorithm with a set of 8192 samples and the Hanning window. Figure 1 shows the spectrum of the AE signal, while Figure 2 depicts the spectrum of the vibration signal spectrum, where the crosshatched areas represent the frequency ranges chosen for this work, and k is a scale constant. In these ranges, the signals do not overlap significantly and one can see a considerable difference between the three stages of wear. Thus, for this study, the frequency ranges of 0 to 13 kHz and 210 to 260 kHz were selected for the AE signal, and the 0 to 13 kHz range for the vibration signal.

Fig. 1. Spectrum of the AE signal

Fifth-order Butterworth bandpass digital filters were applied to the raw AE and vibration signals in the previously selected frequencies to obtain the RMS and ROP statistics. The input vectors consisted of 4,200 samples of the values of these statistics for each statistic and frequency band,

representing the behavior of the amplitudes of the respective signals for all the dressing passes. Figure 3 illustrates the AE RMS vector obtained for the frequency range of 210 to 260 kHz.

axis corresponds to the amplitude values of the statistic in k*Volts and, as can be seen, the greater the amplitude the greater the wear level in the fuzzified set.

Fig. 2. Spectrum of the vibration signal

Fig. 3. AE RMS vector for 210 to 260 kHz

A clustering process [15] was applied to each numerical input, which consisted of determining four subclusters. The purpose of this clustering was to divide the data of the input vectors into four major groups representing the wear levels of the dresser. Each given group contained a cluster center, i.e., a point that best represents the group. In the case under analysis, the cluster centers were very close to the mean values of each group. Therefore, the point that best represents a wear group in this work is the actual average value of the range of points of a wear level. These wear levels or conditions are classified in the fuzzy models as "New," "Semi-New," "Half-Life" and "Worn." The data vectors were then normalized with values between 0 and 1, and the center of the cluster was assigned the value of 1 for each wear condition of the dresser, demonstrating that this amplitude value of a given statistic strongly represents the wear levels considered. This amplitude value 1 decreases as the values of the statistic deviate from that cluster point. Thus, the further away from the cluster center the lower the relevance of the information and the lower the value assigned in the data normalization.

Figure 4 illustrates the AE RMS statistic (210 to 260 kHz) fuzzified by means of the membership functions. The abscissa

Fig. 4. Fuzzified input of the EA RMS signal

A single output vector was then created according to the following dresser wear levels: New, Semi-new, Half-life, and Worn. A graph similar to that of the previous figure was thus obtained for the output, in which the membership functions are Gaussian, the axis of the abscissas represents the percentage of wear of the dresser from 0 to 100, and the ordinate axis represents the fuzzified values of the wear levels normalized from 0 to 1.

4.3. Fuzzy Models

Based on the fuzzified inputs and outputs, an inference system was established for the two fuzzy models. The first model consisted of three inputs: AE RMS of 210 to 260 kHz, ROP of 0 to 13 kHz, and vibration RMS of 0 to 13 kHz. The second model was based on the first model, but with the elimination of one input (AE ROP) to simplify the predictive model.

The extracted rule set for the models was obtained based on the data of the input signals (AE and vibration) and on the measured area of wear, as well as on the experience of the research group involved (specialists), characterized a typical fuzzy system. Table 1 shows the rule set for model 1 while Table 2 describes the rule set for model 2.

5. Results and Discussion

The fuzzy models were implemented in the Fuzzy Logic toolbox of MATLAB using the Mamdani method [16]. The results in the form of response surfaces and simulation examples for the models are presented below.

5.1. Model 1

Based on the rules previously obtained for model 1 and inserted in Matlab, 3D surfaces were created to allow one to visualize the effects of the inputs and outputs of the fuzzy model. If the surfaces present disproportions or a gross irregularity, this means that the rule base is not completely consistent.

Table 1. Rule set for Model 1

AE RMS AE ROP VIB RMS Output AE RMS AE ROP VIB RMS Output

(210-260 kHz) (0-13 kHz) (0-13 kHz) (210-260 kHz) (0-13 kHz) (0-13 kHz)

New New New New Half-life New New Seminew

New New Seminew New Half-life New Seminew Seminew

New New Half-life Seminew Half-life New Half-life Half-life

New New Worn Half-life Half-life New Worn Half-life

New Seminew New New Half-life Seminew New Seminew

New Seminew Seminew New Half-life Seminew Seminew Seminew

New Seminew Half-life Seminew Half-life Seminew Half-life Half-life

New Seminew Worn Seminew Half-life Seminew Worn Half-life

New Half-life New Seminew Half-life Half-life New Seminew

New Half-life Seminew Seminew Half-life Half-life Seminew Half-life

New Half-life Half-life Half-life Half-life Half-life Half-life Half-life

New Half-life Worn Half-life Half-life Half-life Worn Half-life

New Worn New Seminew Half-life Worn New Half-life

New Worn Seminew Seminew Half-life Worn Seminew Worn

New Worn Half-life Half-life Half-life Worn Half-life Worn

New Worn Worn Worn Half-life Worn Worn Worn

Seminew New New New Worn New New New

Seminew New Seminew Seminew Worn New Seminew Seminew

Seminew New Half-life Seminew Worn New Half-life Seminew

Seminew New Worn Half-life Worn New Worn Half-life

Seminew Seminew New Seminew Worn Seminew New Seminew

Seminew Seminew Seminew Seminew Worn Seminew Seminew Half-life

Seminew Seminew Half-life Seminew Worn Seminew Half-life Worn

Seminew Seminew Worn Half-life Worn Seminew Worn Worn

Seminew Half-life New Seminew Worn Half-life New Half-life

Seminew Half-life Seminew Seminew Worn Half-life Seminew Half-life

Seminew Half-life Half-life Half-life Worn Half-life Half-life Worn

Seminew Half-life Worn Half-life Worn Half-life Worn Worn

Seminew Worn New Seminew Worn Worn New Half-life

Seminew Worn Seminew Seminew Worn Worn Seminew Worn

Seminew Worn Half-life Half-life Worn Worn Half-life Worn

Seminew Worn Worn Worn Worn Worn Worn Worn

Table 2. Rule set for Model 2

EA RMS (210-260 kHz)

S W S s New Seminew Half-life Worn

New New Seminew Seminew Half-life

Seminew New Seminew Seminew Half-life

s A > a Half-life Seminew Seminew Half-life Worn

Worn Seminew Half-life Half-life Worn

Figures 5 and 6 show the 3D surfaces generated for model 1. These surfaces exhibit characteristics of continuity, transition and reasonable symmetry. They portray the simulations of the combinations of two inputs and the output level. Note that the combined inputs result in different influences on the final prediction. According to the surface of rules, if the statistical values are high, a percentage of wear is estimated on the ordinate axis. Using the Matlab toolbox, one can also simulate input values and obtain the wear value of the output. For example, considering one of the simulations performed, the input values were set to high levels of wear, with an AE RMS signal amplitude of 0.128 (k*Volts), AE ROP amplitude of 0.422 (k*Volts), and vibration RMS input of 0.224 (k*Volts). In this configuration, there was an output indicating the percentage of 80.6% of dresser wear. This model showed good results; however, the three inputs generated a total of 64 rules. The advantage of this model is that if any input value indicates an error, such as a reading error, this will not substantially affect the model's predictive ability.

Model 2 was designed to simplify the first model, and therefore only two inputs were used: vibration (0-13 kHz) and AE RMS (210-260 kHz). Figure 7 depicts the 3D

surface of this model, showing smooth transitions at different levels of wear, in addition to a good surface

symmetry.

0 0 AE RN

Fig. 5. The Predicted Wear of Model 1

Fig. 6. The Predicted Wear of Model 1

However, note that, as the amplitude of AE increases, the wear increases up to about the middle of the range of values, becoming fairly stable from this point on. As for the vibration signal, the amplitude increases up to approximately % of the range of values, whereupon it stabilizes. Thus, it is clear that, in Model 2, the AE signal exerts a greater influence in predicting wear than does the vibration signal. Nevertheless, if any of the inputs fails, the prediction of wear will be inaccurate.

The same simulation as that performed for Model 1 was done for Model 2, using the same values, i.e., with an amplitude of the AE RMS signal of 0.128 (k*Volts), and a vibration RMS input of 0.224 (k*Volts). The result of the wear prediction output was the same as that of Model 1 and equal to 80.6%, demonstrating the effectiveness of this model despite having one less input variable. Thus, the reduction from 48 to 16 rules of Model 1 did not result in loss of the wear prediction capacity.

Fig. 7. The Predicted Wear of Model 2

6. Conclusions

This research presented a method to evaluate the wear of single-point dressers based on a frequency study of raw acoustic emission and vibration signals, applying digital filters to obtain the RMS and ROP statistics for selected frequencies, and fuzzy models for predicting the wear. The root mean square (RMS) value and the ratio of power (ROP) of the signals were determined for each of the selected frequency ranges, using digital bandpass filters. Two fuzzy models were developed and implemented in Matlab, the first model with three inputs: AE (0 to 13 kHz), AE (210 to 260 kHz) and vibration (0 to 13 kHz), and the second model with only two inputs: AE (210 to 260 kHz) and vibration (0 to 13 kHz). The two models yielded good

results for predicting dresser wear. With three inputs, Model 1 proved to be more robust against possible errors in some of the inputs, albeit more complex, with 64 rules. Model 2, with only two inputs, 16 rules, a simplified version of Model 1, proved to be effective and easier to implement. The models may be attractive for technicians and engineers responsible for the operation of grinding machines, who are interested in obtaining fast responses for intelligent control or optimization. The results of this research are limited to the CVD single-point dresser and aluminum oxide grinding wheel employed here, but can be easily extended to other types of dressing tools and grinding wheels.

Acknowledgements

The authors gratefully acknowledge the Brazilian research funding agencies CNPq, CAPES and FAPESP for their financial support. Thanks also go to the NORTON Company and to IPEN (Institute for Energy and Nuclear Research) for donating the wheels and dressers.

References

[1] Xue L, Naghdy F, Cook C. Monitoring of wheel dressing operations for precision grinding. IEEE ICIT'02 - Int Conf Ind Technol 2002;1296-9.

[2] Martins C, Aguiar P, Frech A, Bianchi EC. Tool Condition Monitoring of Single-Point Dresser Using Acoustic Emission and Neural Networks Models. IEEE Trans Instrum Meas 2013;1—13.

[3] Pham D, Pham PT. Artificial intelligence in engineering. Int J Mach Tools Manuf 1999;39:937-49.

[4] Ali YM, Zhang LC. A fuzzy model for predicting burns in surface grinding of steel. Int J Mach Tools Manuf 2004;44:563-71.

[5] Oliveira JFG. Analysis of dressing macro-effect of grinding wheels in grinding process. EESC- University of Sao Paulo, 1988.

[6] Badger BYJA, Torrance A. Burn awareness - understanding the causes of grinding burn helps alleviate the problem. Cut Tools Eng Mag 2000;52.

[7] Inasaki 1 Okamura K. Monitoring of dressing and grinding processes with acoustic emission signals. CIRP Ann Technol 1985;34:277-80.

[8] Oliveira JJFG, Dornfeld D. Application of AE contact sensing in reliable grinding monitoring. CIRP Ann Technol 2001;50:217-20.

[9] Dimla, Snr. DE. The Correlation of Vibration Signal Features to Cutting Tool Wear in a Metal Turning Operation. Int J Adv Manuf Technol 2002;19:705-13.

[10] Hassui A, Diniz AE, Oliveira JFG, Felipe J, Gomes JJF. Experimental evaluation on grinding wheel wear through vibration and acoustic emission. Wear 1998;217:7-14.

[11] Hassui A, Diniz AE. Correlating surface roughness and vibration on plunge cylindrical grinding of steel. Int J Mach Tools Manuf 2003;43:855-62.

[12] Zhang C, Shin YC. Wear of diamond dresser in laser assisted truing and dressing of vitrified CBN wheels. Int J Mach Tools Manuf 2003;43:41-9.

[13] Habrat W, Batsch A, Porzycki J. Monitoring of the single-point diamond dresser wear. Arch Civ Mech Eng 2005;V:13-8.

[14] Martins CH, Aguiar PR, Bianchi EC, Junior AF, Ruzzi RS. Application of MLP and Kohonen Networks for Recognition of Wear Patterns of Single-Point Dressers. IASTED Int. Conf. Artif. Intell. Appl. (AIA 2013), Innsbruck, Austria: Actapress; 2013, p. 70-4.

[15] Ge P., Liu W., Liu Z., Lu C., Li J. Fuzzy clustering analysis of the grinding burn damage level of a workpiece surface layer. J Mater Process Technol 2002;129:373-6.

[16] Mamdani E, Assilian S. An experiment in linguistic synthesis with a fuzzy logic controller. Int J Man Mach Stud 1975;7:1-13.