Monitoring of flank wear of coated tools in high speed machining with a neural network ART2

doi:10.1016/j.ijmachtools.2004.04.021

International Journal of Machine Tools and Manufacture

Volume 44, Issues 12–13, October 2004, Pages 1311-1318

https://doi.org/10.1016/j.ijmachtools.2004.04.021 Get rights and content

Abstract

A monitoring system for classifying the levels of the tool flank wear of coated tools into some categories has been developed using an unsupervised and self-organizing artificial neural network, ART2. The input pattern used for the ART2 was an array of normalized mean wavelet coefficients of the feed force, which was affected by not only the flank wear but also the severe crater wear observed in high speed machining. The outputs of ART2 were classified into four or five categories of wear levels: the incipient stage, one or two intermediate stages, final stage and hazardous stage. For two apparently different series of input data obtained under the same cutting conditions, which are often experienced in the experiment, the ART2 neural network showed very similar classification of tool wear levels from the beginning to the end of cutting. Further study proved that this monitoring system detected the excessive wear in the hazardous stage for different cutting speeds 5–7 m/s and different feed rates 0.10–0.20 mm/rev.

Introduction

The tool condition monitoring increases the reliability of machining by detecting the symptoms of possible hazards, such as excessive tool wear, tool breakage and/or chatter vibration [1]. It also improves the economics of machining by changing the machining conditions based on the evaluation of tool life [2], [3]. Therefore, many researches of tool condition monitoring have been reported based on various signals and their processing methods. Cutting forces, tool vibration and acoustic emission from the tool and/or workpiece are signals frequently used for the monitoring. The Fourier transform and wavelet transform are the main signal processing methods. In indirect monitoring, the threshold method and artificial neural networks are usually applied to the recognition of tool conditions [4]. The threshold method is a robust classifier into two categories, e.g., safe or hazardous condition. Artificial neural networks have so high ability of learning that they have been applied to the classifications of more complicated conditions.

Recently, an unsupervised and self-organizing neural network ART2 [5] based on the adaptive resonance theory has been noted for its excellent performance in tool condition monitoring [6], [7], [8], [9]. The number of categories, into which signals are classified, is automatically determined by specifying the value of vigilance. Thus, the learning of ART2 neural networks is straightforward. Lots of sample patterns of known classes do not have to be prepared for learning; over-learning by repeated learning of sample patterns, which must be avoided in back-propagation method, may not be taken into consideration.

Generally, a tool condition monitoring system is developed for a particular machining operation. The parameters of the system for recognizing tool conditions are adjusted to the operation. Hence, it is quite difficult to apply the developed system to the monitoring of other machining operations. This is why the general-purpose system of tool condition monitoring has not been developed and the advanced machine tools with various monitoring sensors have not been used widely yet.

In this paper, a monitoring system for identifying four or five categorized levels of tool flank wear and detecting the excessive wear with a neural network ART2 is described. The wavelet coefficients of feed force were used for comprising the feature pattern of wear. The developed system was applied to high speed turning of a carbon steel under different cutting conditions: cutting speeds 5–7 m/s, feed rates 0.10–0.20 mm/rev and depths of cut 1.5–2.0 mm.

Section snippets

Tool wear monitoring system

A tool wear monitoring system developed in this study is schematically shown in Fig. 1. In the developed system, the categorized levels of the flank wear were monitored indirectly by processing the feed force in a personal computer. The feed force was measured with piezoelectric dynamometers at a sampling rate of f_s=32 kHz and a set of 4096 data for 128 ms were transferred to the computer repeatedly. A discrete wavelet transform of a set of feed force data was conducted offline to the eighth

Cutting experiment

One of the P10 coated carbide tool inserts SNMG120408SH with thick layers of TiCN/Al₂O₃/TiN were used in cutting experiments. The type of tool holder was PSBNR2525M12. The work material was a 0.45%C carbon steel S45C, which was turned with a CNC lathe without cutting fluid. The recommended cutting speeds of the insert are 3.3–5 m/s for machining carbon steels. However, in order to investigate the performance of the monitoring system in high speed machining, the cutting speed V was set at 5, 6

Wavelet coefficients and input feature patterns

Figs. 4(a)–(c) show the changes in measured feed forces and their wavelet coefficients for 128 ms when flank wear VB=0.167, 0.320 and 0.388 mm, respectively. Cutting conditions were V=6 m/s, f=0.15 mm/rev and d=1.5 mm. In the upper part of each figure, the sizes of eight resolution levels of wavelet coefficients w_ij (i=1,2,…,8) corresponding to eight frequency bands (f_s/2ⁱ)−(f_s/2ⁱ⁻¹) are expressed with a gray scale. It is seen that the wavelet coefficients of lower frequencies around 500 Hz

Conclusions

A monitoring system for classifying the levels of tool flank wear in high speed machining into categories has been developed using an unsupervised and self-organizing artificial neural network, ART2. The feature pattern of the ART2 comprised normalized wavelet coefficients of feed force. The ART2 neural network indicated quite similar categories to the development of flank wear for two different series of force data obtained under the same cutting conditions. In all the cutting conditions, the

Acknowledgements

The authors would like to thank Mr. H. Nakamoto for his contribution to the research work.

References (11)

G. Byrne et al.
Tool condition monitoring (TCM)—the status of research and industrial application
Annals of the CIRP
(1995)
T. Obikawa et al.
Tool wear monitoring for optimizing cutting conditions
Journal of Materials Processing Technology
(1996)
T. Matsumura et al.
Autonomous turning operation planning with adaptive prediction of tool wear and surface roughness
Journal of Manufacturing Systems
(1993)
I. Tansel et al.
Micro-end-milling III. Wear estimation and tool breakage detection using acoustic emission signals
International Journal of Machine Tools and Manufacture
(1998)
Y.S. Tarng et al.
Adaptive pattern recognition of drilling chatter
Journal of Materials Processing Technology
(1995)

There are more references available in the full text version of this article.

Cited by (40)

Tool wear monitoring in turning using image processing techniques
2021, Materials Today: Proceedings
Citation Excerpt :
A primary goal of tool wear research has develop methods of predicting tool life from tool failure mechanisms. It has been observed that many researchers [1–4] studied on developing the techniques by which tool life can be predicted. Predicting rate of tool wear is possible based on the past data or an experimentation, but it is difficult to correlate this information to tool life.
Observing the tool wear is very important in metal cutting industry as it affects the dimensional accuracy and surface quality of work part. Generally, tool wear is measured directly using an optical microscope which is an offline and time-consuming technique. In the indirect technique, the parameter such as force, surface finish, temperature etc. that is affected by tool wear is measured and correlated with tool life. It has been observed that measurement of tool wear using image processing techniques has a lot of scope. Digital image processing techniques automate the task of measurement and monitoring of tool wear. A novel on-line tool wear measuring algorithm is proposed to monitor tool wear using machine vision. This work is focused on implementing digital image processing techniques to automate the task of two dimensional flank wear measurement. The tool wear images are acquired by a camera, and the wear boundaries are recognized by image pre-processing, threshold segmentation and edge detection to measure the wear on tool. The wear values obtained by proposed method are compared with manual measurement. The maximum flank wear with proposed algorithm and manual measurement is found as 730 and 718 µm respectively. It is found that the maximum variation in the results obtained by algorithm and manual measurement is 4.98%. This shows the higher detection accuracy for on-line tool wear monitoring by the proposed algorithm.
Tool life predictions in milling using spindle power with the neural network technique
2016, Journal of Manufacturing Processes
Tool wear is an important limitation to machining productivity and part quality. In this paper, remaining useful life (RUL) prediction of tools is demonstrated based on the machine spindle power values using the neural network (NN) technique. End milling tests were performed on a stainless steel workpiece at different spindle speeds and spindle power was recorded. The NN curve fitting approach with different MATLAB™ training functions was applied to the root mean square power (P_rms) values. Sample P_rms growth curves were generated to take into account uncertainty. The P_rms value in the time domain was found to be sensitive to tool wear. Results show a good agreement between the predicted and true RUL of tools. The proposed method takes into account the uncertainty in tool life and the percentage increase in nominal P_rms value during the RUL prediction. Using MATLAB™ on an Intel i7 processor, the computation takes 0.5 s Thus, the method is computationally inexpensive and can be incorporated for real time RUL predictions during machining.
An improved ART2 neural network: Resisting pattern drifting through generalized similarity and confidence measures
2015, Neurocomputing
ART2 network is a kind of non-supervised neural network based on the adaptive resonance theory, and has been widely used in real-time classification because of its rapid response and real-time learning. There are two problems in the traditional ART2 network: the indistinguishable of the different samples with similar phase and the pattern drifting caused by the gradual changing data. In this paper, we propose an improvement version of the ART2 network based on the generalized similarity and confidence measures, named GSC-ART2 (Generalized Similarity Confidence ART2) network. In this neural network, the similarity detection mechanism based on the generalized similarity measure is proposed to solve the indistinguishable problem of the different samples with similar phase. Furthermore, the updating method of the connection weights considering both the generalized similarity and the confidence measures is proposed to inhibit pattern drifting problem. The stimulation data is created to evaluate the proposed GSC-ART2 network, and the outcomes approved that the performance of the GSC-ART2 network is better than traditional ART2 network about the classification and the inhibiting pattern drifting. The GSC-ART2 network would become a universal solution to the pattern drifting problem in various applications.
Use of digital image processing techniques for evaluating wear of cemented carbide bits in rotary drilling
2014, Automation in Construction
This paper presents the use of image processing techniques in monitoring of bit wear, especially, WC/Co cemented carbide bits that are commonly used in rotary drilling in mining, civil and petroleum engineering. Image of the bits was acquired using a CCD camera. The background was subtracted from the image to reduce noise effects. A Laplacian filter has been used to enhance edge contrasts. Structural elements have been applied to dilate, erode and close boundary edges. Edge detecting was conducted using a canny edge detector. Image processing approaches; first order surface metrics, gray level co-occurrence matrix (GLCM) based texture analysis and minimum distance based classifiers have been used to estimate wear of tricone drill bits in rock drilling. A digital balance was used to obtain weight loss of the bits and also wear of their heel row and gage row (dimension loss) was measured using a micrometer in different directions. Results showed that, of the surface metrics, bit area and major length axis could be good measures for bit wear estimation. The entropy and contrast features of the GLCM method showed good correlations with bit weight loss. Of the minimum distance based classifiers just Euclidean, City block and Chebychev distances had reliable correlations with weight loss and heel row wear rather than other features.
Tool wear monitoring by machine learning techniques and singular spectrum analysis
2011, Mechanical Systems and Signal Processing
Citation Excerpt :
He used the average wavelets coefficients and the maximum of peaks in entropy spectrum. Obikawa and Shinozuka [15] also made use of wavelets coefficients of cutting force signals. In his review of indirect tools monitoring methods, Dimla [9,16] cites several works in which correlation between tool wear and vibrations is analyzed.
This paper explores the use of data mining techniques for tool condition monitoring in metal cutting. Pseudo-local singular spectrum analysis (SSA) is performed on vibration signals measured on the toolholder. This is coupled to a band-pass filter to allow definition and extraction of features which are sensitive to tool wear. These features are defined, in some frequency bands, from sums of Fourier coefficients of reconstructed and residual signals obtained by SSA. This study highlights two important aspects: strong relevance of information in high frequency vibration components and benefits of the combination of SSA and band-pass filtering to get rid of useless components (noise).
Wavelet analysis of sensor signals for tool condition monitoring: A review and some new results
2009, International Journal of Machine Tools and Manufacture
This paper reviews the state-of-the-art of wavelet analysis for tool condition monitoring (TCM). Wavelet analysis has been the most important non-stationary signal processing tool today, and popular in machining sensor signal analysis. Based on the nature of monitored signals, wavelet approaches are introduced and the superiorities of wavelet analysis to Fourier methods are discussed for TCM. According to the multiresolution, sparsity and localization properties of wavelet transform, literatures are reviewed in five categories in TCM: time–frequency analysis of machining signal, signal denoising, feature extraction, singularity analysis for tool state estimation, and density estimation for tool wear classification. This review provides a comprehensive survey of the current work on wavelet approaches to TCM and also proposes two new prospects for future studies in this area.

View all citing articles on Scopus

View full text