Elsevier

ISA Transactions

Volume 125, June 2022, Pages 514-527
ISA Transactions

Practice article
Effective multi-sensor data fusion for chatter detection in milling process

https://doi.org/10.1016/j.isatra.2021.07.005Get rights and content

Highlights

  • Introducing a cost-effective multi-sensor data fusion for the milling chatter detection.

  • The WPD is utilized and tuned based on the kurtosis and crest factors.

  • The irrelevant features are identified and eliminated utilizing the RFE method.

  • Several machine learning techniques have been adopted to identify chatter vibration.

Abstract

This paper introduces a newly developed multi-sensor data fusion for the milling chatter detection with a cheap and easy implementation compared with traditional chatter detection schemes. The proposed multi-sensor data fusion utilizes microphone and accelerometer sensors to measure the occurrence of chatter during the milling process. It has the advantageous over the dynamometer in terms of easy installation and low cost. In this paper, the wavelet packet decomposition is adopted to analyze both measured sound and vibration signals. However, the parameters of the wavelet packet decomposition require fine-tuning to provide good performance. Hence the result of the developed scheme has been improved by optimizing the selection of the wavelet packet decomposition parameters including the mother wavelet and the decomposition level based on the kurtosis and crest factors. Furthermore, the important chatter features are selected using the recursive feature elimination method, and its performance is compared with metaheuristic algorithms. Finally, several machine learning techniques have been adopted to classify the cutting stabilities based on the selected features. The results confirm that the proposed multi-sensor data fusion scheme can provide an effective chatter detection under industrial conditions, and it has higher accuracy than the traditional schemes.

Introduction

Milling is the main method of manufacturing parts in almost all industries. The current concern of machining in general and milling particularly is to increase the productivity of computer numerical control (CNC) machine tools while maintaining high dimensional accuracy and surface quality. Increasing cutting speeds, feed rates, and cutting depths is the keys to dealing with the demand for industrial products [1], [2]. However, at high cutting speeds, chatter occurs frequently, which leads to a decrease in machining productivity [3], [4]. Because chatter affects dimensional accuracy, surface roughness, tool life, and machine tool life, it is worth identifying the occurrence of chatter and avoiding increasing vibration levels during cutting, especially at the early stage. Furthermore, chatter identification using the traditional schemes is costly and complicated to implement in the industrial environment [5], [6]. Besides, the signal analysis and feature selection require effective techniques before applying any classification method to provide high accuracy for the chatter identification due to the chatter issue [7], [8].

Recently, several different types of signal collection using different sensors have been used to receive cutting signals for chatter identification [9], [10], [11]. For example, vibration signals from accelerometers, sound and audio signals from microphones, and current signals are utilized for chatter identification in milling machines [12]. Among them, the vibration from accelerometers is the most suitable and widely used signal for detecting chatter due to its quick response to changes in cutting conditions [13]. However, using a dynamometer to measure the cutting force is just suitable in the laboratory, because the installation of a tabletop dynamometer in the middle of the workpiece and the workbench and the significant cost of dynamometers/amplifiers are drawbacks in a workshop [14]. Accelerometers also have good sensitivity to changing cutting conditions because it is attached to the machine spindle and reflects the vibration from the cutting tool directly. Like accelerometers, microphones are among the most effective sensors for signal acquisition and chatter detection [15]. It was found that the amplitude of microphone response is very sensitive to the chatter vibration stability [16]. Moreover, both accelerometers and microphones show the potentials for not only laboratory tests but also for practical applications. By implementing the fusion of two sensors and the simultaneous collection of two signals, it will improve the accuracy of the model for detecting chatter. More information related to the cutting vibration could be obtained using multi-sensor signals. The tool condition monitoring system, especially chatter detection, could be enhanced with some developed multi-sensor fusion techniques [17]. Several multi-sensor chatter indicators were applied to force and vibration signals. It was found that the statistical inference multi-sensor chatter indicator could improve the classification accuracy and it is more affordable for the manufacturing environment [18], [19]. However, the analysis of the collected signals requires an effective technique for chatter detection in milling machines [20], [21], [22]. Signal analysis techniques for example fast Fourier transform (FFT), wavelets transform (WT), wavelet packet decomposition (WPD), and Hilbert–Huang transform (HHT) are widely utilized to determine cutting chatter [21], [23]. Signal analysis in the frequency domain does not consider time domain that causes difficulties to perform any information about individual frequency events in the local time [14]. Besides, spectrogram obtained from short-time Fourier transform is also difficult to apply because it loses either frequency resolution or time resolution and requires a large number of sample points. Discrete Fourier Transform (DFT) which transfers a time-domain signal to the frequency domain signal is a simple and fast signal processing tool, but it has several limitations when applying to non-stationary and nonlinear systems [13]. Wavelet Transform (WT), a convolution of a wavelet function with a signal, is known as an effective tool for analyzing and classifying signals [24]. WT can convert the signal into the time–frequency​ domain and allows it to be effectively applied to both nonlinear and non-stationary signals [25]. Because the WT is able to provide high-resolution of both time and frequency domains, it shows advantages when analyzing non-stationary time series for chatter detection [26]. Therefore, the time–frequency analysis would be a great method to identify a representative of chatter features [27], [28]. Energy and entropy from wavelet packets could effectively represent chatter features by applying the wavelet packet transform (WPT) [27], [29], [30]. Nevertheless, the effectiveness of wavelet packets for chatter detection highly relies on choosing wavelet parameters including the mother wavelet and the decomposition level that were not fully considered in previous chatter detection methods. In addition to WT, Hilbert–Huang Transform (HHT) is also a great approach to deal with non-stationary behaviors of cutting signals [31]. HHT is a transform that is used to decompose a signal into empirical modes and then applies the Hilbert transform to the decomposition components. A combination of the wavelet de-noising and HHT was adopted to generate the chatter features when analyzing the cutting vibration from the internal turning operation [32]. Besides, selecting proper features is significantly important to enhance the performance of chatter detection techniques before applying the classification algorithms, but it is still hard to identify which features can represent more chatter information. Feature selection of the most significant features correlated to the cutting process is among the most challenging when using multi-sensor signals. Chatter detection consists of processing large amounts of vibration data, highlighting certain features in them that correspond to a certain mode of operation of the machine, and choosing the appropriate classifier model. In literature, various techniques are utilized for the feature selection process [33], [34]. Among these approaches, the recursive feature elimination (RFE) method provides effective performance for feature selection compared with metaheuristic algorithms [27], [35]. After selecting the important features, effective chatter detection methods require an effective classification technique to provide good performance with high accuracy. However, traditional approaches consume a long period of time and make it impossible to monitor the cutting vibration in real-time. To automate chatter detection, artificial machine learning techniques are used to do an excellent job of classification. This approach usually includes signal analysis and artificial intelligence. Recently, there are many machine learning algorithms such as random forest (RF), gradient boosting, support vector machines (SVM), k-nearest neighbors (kNN), and artificial neural networks (ANNs) that have been adopted to identify chatter [36], [37], [38]. Among these techniques, time–frequency image features obtained from continuous wavelet transform (CWT) and the SVM approach provide effective performance to detect milling chatter [39], [40]. However, a number of parameters of CWT are required to be optimized making it less useful for practical applications [14].

Based on the above literature review, this paper introduces a new smart multi-sensor data fusion for chatter detection in the milling machines. It is carried out depending on an effective multi-sensor data fusion to overcome the high cost and complicated implementation of the traditionally-used dynamometers. Furthermore, the paper introduces intelligent techniques to classify the cutting stabilities based on the selected features. The proposed system can provide a reliable and effective infrastructure for practical applications. The following points conclude the contributions of this paper,

  • This paper introduces a cost-effective multi-sensor data fusion scheme for the milling chatter detection with easy implementation in practical applications. The WPD parameters including the mother wavelet and the decomposition level are optimized by using kurtosis and crest factors to achieve a better performance.

  • The frequency bandwidths related to resonant frequencies from both sound and vibration signals are considered as effective features for chatter recognition. Furthermore, the irrelevant statistical features of the WPD coefficients are determined and removed utilizing the recursive feature elimination method to enhance the performance of the classification models.

  • The proposed multi-sensor data fusion scheme confirms an excellent chattering detection with an excellent accuracy of 97.66% that is significantly greater than the accuracy based on the traditional schemes.

The rest parts of this paper are sorted as follows. Section 2 illustrates the experimental setup and data acquisition of vibration and sound signals; the signal processing and feature selection methodology are illustrated in Section 3; Section 4 gives an overview of machine learning classifiers for milling chatter detection; the results and discussions are presented in Section 5; finally, Section 6 recorded the conclusion and future works of this research.

Section snippets

Experiment setup and data acquisition

A high accuracy 5-Axis trunnion table machining center (Tongtai CT-350) was used to obtain the experimental data for the training model. The machine has a direct-drive type oil–air lubrication spindle with a maximum spindle speed of 15 000 rpm. The workpiece of Al6061-T6 which is widely utilized in the automobile and aerospace areas was used for the chatter experiments. The mill cutter is an uncoated carbide that has two flutes with a 12 mm diameter, and the tool has a helix angle of 26° and a

Signal processing methodology

Cutting chatter vibration occurs frequently during metal cutting processes, particularly when cutting conditions with high cutting speeds and large engagement angles are used. The tool is therefore vibrated with a chatter frequency. The purpose of using sensors is to capture the chatter vibration during the cutting process. In this study, microphone and accelerometer sensors, which are more affordable and proper for the workshop, are effectively used to measure the occurrence of chatter

Machine learning classifiers

Chatter detection consists of processing large amounts of cutting data, highlighting certain features corresponding to a certain mode of operation of the machine, and choosing the appropriate classifier model. The traditional approach to detect chatter is post-processing, but it is time-consuming and makes it impossible to monitor the cutting vibration in real-time. To automate the chatter detection, artificial machine learning techniques are used to do an excellent job as classifiers. A

Results and discussions

In this work, a total of sixteen features was extracted from the WPD coefficients related to the resonant bandwidth frequencies. Eight features described in Table 2 were from the sound and the other eight features were from the vibration signals. Then the proposed BPNN was applied to those features for the chatter detection. All data were normalized and randomly separated into a 75% training dataset and a 25% testing dataset. In this work, the signal analysis and chatter detection were carried

Conclusions

Chatter vibration in milling represents a big issue against the improvement of metal cutting processes. This paper developed a new chatter identification scheme using multi-sensor data fusion and advanced intelligent techniques. The new scheme can be implemented easily at a low cost comparing with the traditional chatter detection schemes. Both cutting vibration and sound signals are measured and analyzed by an effective signal analysis technique using wavelet packet decomposition. Besides, the

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Acknowledgments

This study was funded by the Ministry of Science and Technology (MOST) in Taiwan under Grant no. MOST 110-2222-E-011-002-, and the center for cyber–physical system by the Ministry of Education (MOE) in Taiwan .

References (55)

  • PengC. et al.

    A new method for the prediction of chatter stability lobes based on dynamic cutting force simulation model and support vector machine

    J Sound Vib

    (2015)
  • CaoH. et al.

    Chatter identification in end milling process using wavelet packets and Hilbert–Huang transform

    Int J Mach Tools Manuf

    (2013)
  • BondaA.G.Y. et al.

    Vibration signature based stability studies in internal turning with a wavelet denoising preprocessor

    Measurement

    (2020)
  • KouG. et al.

    Evaluation of feature selection methods for text classification with small datasets using multiple criteria decision-making methods

    Appl Soft Comput

    (2020)
  • ChandrashekarG. et al.

    A survey on feature selection methods

    Comput Electr Eng

    (2014)
  • RostamiM. et al.

    Review of swarm intelligence-based feature selection methods

    Eng Appl Artif Intell

    (2021)
  • YesilliM.C. et al.

    On transfer learning for chatter detection in turning using wavelet packet transform and ensemble empirical mode decomposition

    CIRP J Manuf Sci Technol

    (2020)
  • ChenY. et al.

    Feature extraction using dominant frequency bands and time-frequency image analysis for chatter detection in milling

    Precis Eng

    (2019)
  • KuljanicE. et al.

    Multisensor approaches for chatter detection in milling

    J Sound Vib

    (2008)
  • ChenY. et al.

    An intelligent chatter detection method based on EEMD and feature selection with multi-channel vibration signals

    Measurement

    (2018)
  • MafarjaM. et al.

    Whale optimization approaches for wrapper feature selection

    Appl Soft Comput

    (2018)
  • AltintasY.A. et al.

    Manufacturing automation: Metal cutting mechanics, machine tool vibrations, and CNC design

    Appl Mech Rev

    (2001)
  • SmithK.S. et al.

    Machining dynamics: Frequency response to improved productivity

    (2008)
  • FaassenR.

    Chatter Prediction and Control for High-Speed Milling

    (2007)
  • ZhaoY. et al.

    Maxent feature-based reliability model method for real-time detection of early chatter in high-speed milling

    ISA Trans

    (2020)
  • LinR. et al.

    Instantaneous frequency estimation for wheelset bearings weak fault signals using second-order synchrosqueezing S-transform with optimally weighted sliding window

    ISA Trans

    (2021)
  • VashishtR.K. et al.

    Online chatter detection for milling operations using LSTM neural networks assisted by motor current signals of ball screw drives

    J Manuf Sci Eng

    (2020)
  • Cited by (65)

    • Improved STFT analysis using time-frequency masking for chatter detection in the milling process

      2024, Measurement: Journal of the International Measurement Confederation
    • Milling chatter detection based on information entropy of interval frequency

      2023, Measurement: Journal of the International Measurement Confederation
    • Comprehensive approach toward IIoT based condition monitoring of machining processes

      2023, Measurement: Journal of the International Measurement Confederation
    View all citing articles on Scopus
    View full text