Elsevier

Mechanical Systems and Signal Processing

Volume 85, 15 February 2017, Pages 809-826
Mechanical Systems and Signal Processing

Audio signal analysis for tool wear monitoring in sheet metal stamping

https://doi.org/10.1016/j.ymssp.2016.09.014Get rights and content

Highlights

  • This is the first study using audio signals for progressive die wear monitoring.

  • Signal extraction recovers audio generated by stamping from corrupted recordings.

  • Audio spectra and band power variations show good correlation with wear level.

  • Signal strength increment can be detected significantly prior to process failure.

  • Extracted signal shows strong correlation with increasing wear at 4.2–5.4 kHz range.

Abstract

Stamping tool wear can significantly degrade product quality, and hence, online tool condition monitoring is a timely need in many manufacturing industries. Even though a large amount of research has been conducted employing different sensor signals, there is still an unmet demand for a low-cost easy to set up condition monitoring system. Audio signal analysis is a simple method that has the potential to meet this demand, but has not been previously used for stamping process monitoring. Hence, this paper studies the existence and the significance of the correlation between emitted sound signals and the wear state of sheet metal stamping tools. The corrupting sources generated by the tooling of the stamping press and surrounding machinery have higher amplitudes compared to that of the sound emitted by the stamping operation itself. Therefore, a newly developed semi-blind signal extraction technique was employed as a pre-processing technique to mitigate the contribution of these corrupting sources. The spectral analysis results of the raw and extracted signals demonstrate a significant qualitative relationship between wear progression and the emitted sound signature. This study lays the basis for employing low-cost audio signal analysis in the development of a real-time industrial tool condition monitoring system.

Introduction

Sheet metal stamping is the most frequently used primary manufacturing method for mass-production of automotive body and structural components. Since, global automotive production exceeds 60 million vehicles per year, and there are over one hundred sheet metal components on average in each car, this potentially has a very significant number of applications even for just the automotive industry alone. Hence, even a small improvement could be highly influential to the product quality and cost efficiency. The increasing trends towards advanced- and ultra-high strength steels (AHSS and UHSS) in the automotive industry for continued vehicle light-weighting has increased forming forces, die wear and galling and, therefore, has resulted in premature die failures. These faults are difficult to predict at the design stage and may lead to expensive tool replacements and additional costs due to poor part quality, machine downtime and unscheduled maintenance [1]. Hence, stamping tool condition monitoring is a timely need. The aim of implementing a monitoring system is to achieve uninterrupted production and prevent the adverse consequences described. However, current techniques for assessing tool wear and part quality rely on manual visual inspection of formed part surface quality or the tool surface condition. These methods only provide information about the state of the tooling after the wear has become severe and, therefore, can be too late to prevent production downfall. Hence, it is a non-trivial requirement to have a system for online monitoring of tool condition that can provide the early detection of tool wear and identification of the current tool state.

Du et al. [2] presented a comprehensive review on the sensing and online monitoring techniques for sheet metal stamping processes. Other earlier reviews and research articles [3], [4], [5], [6] provide evidence of progress in this area, but still demonstrate the current need for a high-performance condition monitoring system. Tonnage or force monitoring has been the most popular [7], [8] while other commonly used sensors include acceleration [9], proximity, optical [5], [6], and acoustic emission [10], [11]. Du. et al. [2] reviewed strain, acceleration, force and proximity sensors and argued that force and proximity sensors could be effective, but the sensor cost is too high. Acceleration sensors are cost effective, but very high computational power is necessary to convert the signal into useful information. Strain and acoustic emission sensor signals are rich in wear information, but can be difficult to install, since they must be mounted on the tool or sheet surfaces [12]. Despite this research, there is still an unmet demand from manufacturing industries for a low-cost easy to set up condition monitoring system which is capable of detecting the onset of wear.

Audible acoustic sensors (microphones) have a higher potential to meet this demand compared to other acoustic sensor types - such as vibration (acceleration) and acoustic emission sensors - due to their low-cost, non-disruption to the ongoing operation and simplicity in sensor placement and setup. However, the high sensitivity to background noise has hindered the employment of microphones in real applications. Audio has been employed in many other manufacturing applications, except stamping, such as diesel engine injection monitoring [13], rolling element bearing defect detection [14] and monitoring of machining/cutting operations such as turning [15], [16], [17], micro milling [18], [19], and drilling [20], [21]. Downey et al. [22] conducted audio analysis to study a single point machining operation in an environment where no other machine operations occurred at the same time. Unfortunately, this will not be the case in a real production environment. Additionally, this study and other research works [12], [23] were founded on the characteristic that experienced operators can differentiate the good quality and degraded process conditions by hearing changes in the emitted sound. Depending on the manufacturing process and application examined, the emitted audio signal characteristics in both time and frequency domains can significantly vary as evidenced by the wide range of literature findings. For example, Eneyew et al. [20] observed a decreasing trend of sound amplitude with the number of holes drilled, while Downey et al. [22] detected an increasing trend in amplitude with the number of machining cycles. A shift in signal energy from a lower frequency range to a higher frequency range is observed with wear formation in micro-milling operations [12] and in turning operations [15]. However, in metal cutting it has been shown that higher amplitude signal peaks appear in the upper frequency range for sharp tools and in the lower frequency range for worn tools [24].

However, these existing techniques focus on identifying few distinct wear levels or faults such as slug, misfeed, or tool breakage. Furthermore, to the authors’ best knowledge, no work in the literature has examined the use of audio sensors for progressive wear monitoring in sheet metal stamping. One of the reasons for this may be that even an experienced stamping process operator cannot distinguish the wear state of the tools by relying on his/her hearing. Hence, it is still unknown if emitted audio carries any information about wear progression of the stamping tools. To ensure this, it is necessary to obtain the audio emitted by stamping. However, it is not straightforward to isolate the stamping signal since most of the industrial production lines use mechanical presses with progressive die sets or transfer press systems, which integrate several operations – such as piercing, trimming, feeding, clamping, restrike, etc. –within a single stamping press stroke. Our recent work [25] has shown that these additional operations may emit higher intensity sound signals compared to stamping, making it difficult for the human ear to recover the stamping signal.

Therefore, it is particularly useful to determine firstly if the emitted sound from the stamping operation contains any useful information related to wear progression. If this is the case, sound-based condition monitoring systems would be highly preferred by manufacturing industries due to the advantages discussed earlier. In the studies discussed above, several techniques have been developed/employed for data processing based on fast fourier transform (FFT) analysis, cepstrum, wavelet analysis [26], singular spectrum analysis (SSA) [16], [17], and Hilbert Huang transform (HHT) [27], [28]. Speech recognition systems [23], pattern recognition systems [24], neural network systems [29], and support vector machines [16] have been used in the decision making phase. Nevertheless, there is a lack of robust methods which consider the disturbances by other processes/components of the machine and background signals. This point is a key impediment for application to the real production environment, and therefore, there is a need for techniques to recover information from corrupted sensor signals.

Blind signal separation (BSS) is an effective technique capable of recovering unobserved source signals from observed mixtures without any information of the mixing system [30], [31], [32], [33]. BSS first emerged as a biological problem in 1982 and then evolved to successful usage in several applications such as communication, audio processing (speech and music) and image processing [34]. BSS also has been employed in mechanical applications, including rotating machinery condition monitoring [35], [36], fault detection [37] and condition monitoring of manufacturing processes like milling [38], [39], [40]. Among the BSS techniques for manufacturing process monitoring, Shao et al. [38] and Shi et al. [39] focused on single channel source separation, using power signals and audio signals respectively. Both of these studies face computational complexity due to the selection of a single sensor to deal with higher sensor cost and installation inconveniences. As initial studies, they have conducted experimental analysis on the extreme conditions (i.e. sharp tool condition against severe tool breakage) and have noted the need for future investigations into minor tool breakages or tool wear. The direct usage of above techniques for our application is hindered because the BSS algorithm development depends on original source characteristics, which differ from application to application.

Hence, a blind signal separation technique, appropriate for the application, should be developed. Since the knowledge of original signal characteristics and the mixture model is unavailable, system parameterization was necessary and was previously conducted by Ubhayaratne et al. [25]. In this analysis, the audio signals generated by the stamping operation were modeled as a BSS system and a simple semi-blind signal extraction technique was developed. This is not a final signal separation solution. However, it served the purpose of recovering the stamping signal so that it could be employed in studying the correlation between wear progression and audio emissions. However, in [25] this signal extraction technique was tested only on four distinct tool/operation conditions: (i) unworn dies with lubrication; (ii) unworn dies without lubrication; (iii) severely worn dies with lubrication; and (iv) and severely worn dies without lubrication. This technique has not been tested on continuous wear progression. However, the ability to identify changes in the signals during continuous wear progression is a critical step in developing a condition monitoring method. The current study investigates the correlation between audio emissions and continuous wear progression, which is bridging the gap between existing research. Also, in [25] qualitative analysis of the time domain waveform and the quantitative analysis of time domain signal features, such as maximum amplitude and the RMS value, were employed. Even though time domain features were capable of distinguishing distinct wear levels, they fail to demonstrate a significant pattern with continuous wear progression. Thus, frequency domain analysis was introduced in the current study.

This paper provides the first steps towards the aim to develop a real-time industrial stamping tool condition monitoring system. A mechanical press with a channel forming tool set was used for the experiments. Stamping was started with unused die corner radii inserts and run continuously until the die radii surfaces showed severe wear. At this point, the surface quality of the stamped parts was unacceptable, and the process would typically be aborted in a real production environment. Conventional microphones were used for recording the emitted sound. These microphone recordings were a mixture of the signal of interest and other corrupting signals due to the nature of the process, and so the stamping related audio signal was recovered using the technique in [25]. Spectral analysis was conducted on raw and extracted signals, and the observations are discussed in correlation with the wear conditions that are qualitatively observed on the formed channel part surfaces, both visually and via profilometry measurements.

The overall contribution of this study compared to existing work can be summarized as follows:

  • i.

    Study of signal behaviour with continuous wear progression, instead of specific tool defects or few distinct wear levels, which has not been studied before for sheet metal stamping processes.

  • ii.

    The new design of the experimental setup to generate the data specific to the condition explained in i. The experiment was carefully designed to generate progressive wear with near-production industrial stamping conditions, preventing other faulty conditions such as misfeed, slug or intermittent wear conditions.

  • iii.

    Application of frequency domain analysis instead of time domain waveform analysis because the time domain analysis employed in [25] does not show a significant correlation with continuous wear progression.

The rest of the paper is organized as follows. 2.1 Stamping equipment, 2.2 Materials, 2.3 Data acquisition setup explains the stamping equipment, materials and the data acquisition setup used for the experiments, respectively. Section 2.4 describes the conducted experiments in detail, and Section 2.5 details the surface quality inspection using profilometry. Section 3.1 presents the signal extraction technique used as a pre-processing step and Section 3.2 gives the details on the spectral analysis conducted on raw and processed data to compare the signal behaviour with wear progression. The results obtained by the application of the proposed technique and four other existing techniques are presented in Section 4. Section 5 provides a discussion of the results and future work, and the concluding remarks are in Section 6.

Section snippets

Experimental outline

This experiment aims to determine if the simple audio analysis could be employed to detect the die wear and galling, and the primary objectives of the experimental design were to:

  • a)

    closely replicate the industrial stamping operation and near-production process conditions;

  • b)

    generate progressive wear, preventing other possible faults such as slug, misfeed, and intermittent wear conditions.

Signal processing

Stamping operations, in general, would generate noisy audio because this operation is conducted following, or at the same time as, several other manufacturing operations. Hence, a semi-blind signal extraction algorithm, which is detailed in Section 3.1, was employed to extract the audio associated with the stamping operation only. Then spectral analysis was conducted on raw and extracted data in order to study the correlation between the wear formation of stamping dies and the audio emissions.

Visual inspection

Beginning with the unused die inserts, a total of 1500 parts were stamped until failure of the die insert was judged. The progression of the scratches on the sidewalls of the parts, due to wear of the die insert, is summarized by the photographs shown in Fig. 4. The analysis in this study focussed on the sidewalls on one side of the stamped channels, which contacted the left side die insert. When the test was stopped at 1500 parts, severe scratches were not evident on the other side of the

Discussion

This study has identified that there is a significant qualitative correlation between the progressive wear conditions of stamping dies and the emitted audio signature during the stamping operation. To the authors’ knowledge, this is the first study of its kind to use audio sensor signals for progressive tool wear analysis of the stamping process. Although other signal types – including tonnage, acceleration, and acoustic emission – have been examined previously, audio may be a preferred

Conclusion

In this paper, experiments were conducted to study the emitted sound behaviour during the wear progression of stamping tools. Spectral analysis and the aggregate power analysis were conducted on raw and extracted audio signals using semi-blind signal extraction techniques.

The key findings of this research and their significance are summarized below:

  • 1.

    A semi-blind signal separation algorithm has been employed to allow the audio signal of the stamping operation to be recovered from the mixed

Acknowledgments

This research was supported by an Australian Research Council (Australia) Linkage Project (LP120100239) and the Ford Motor Company University Research Program (URP Number: 2010-4003R). The authors would like to thank Dr. Hamid Abdi and Dr. Harisankar Parippaayillam for their contribution in the data acquisition system development.

References (51)

  • Y. Xiang

    Blind source separation based on constant modulus criterion and signal mutual information

    Comput Electr Eng

    (2008)
  • D. Peng et al.

    Underdetermined blind separation of non-sparse sources using spatial time-frequency distributions

    Digit. Signal Process.: Rev. J.

    (2010)
  • R. Boustany et al.

    Blind extraction of a cyclostationary signal using reduced-rank cyclic regression—a unifying approach

    Mech. Syst. Signal Process.

    (2008)
  • R. Boustany et al.

    A subspace method for the blind extraction of a cyclostationary source: Application to rolling element bearing diagnostics

    Mech. Syst. Signal Process.

    (2005)
  • N. Bouguerriou et al.

    Novel cyclostationarity-based blind source separation algorithm using second order statistical properties: theory and application to the bearing defect diagnosis

    Mech. Syst. Signal Process.

    (2005)
  • H. Shao et al.

    Power signal separation in milling process based on wavelet transform and independent component analysis

    Int. J. Mach. Tools Manuf.

    (2011)
  • M.P. Pereira et al.

    Sliding distance, contact pressure and wear in sheet metal stamping

    Wear

    (2010)
  • M.P. Pereira et al.

    The effect of the die radius profile accuracy on wear in sheet metal stamping

    Int. J. Mach. Tools Manuf.

    (2013)
  • M.P. Pereira et al.

    Wear at the die radius in sheet metal stamping

    Wear

    (2012)
  • R. Du, Y.S. Xu, A review on the sensing and on-line monitoring of sheet metal stamping operations, Proceedings of the...
  • R. Aronson

    Intelligent presses

    Manuf. Eng.

    (1998)
  • T. Beard

    New force in stamping process control

    Mod. Mach. Shop

    (1989)
  • G. Brueninghaus

    Process monitoring in the sheet metal processing industry

    VDI-Z.

    (1988)
  • G. Brueninghaus

    Process supervision of automatic stamping presses

    Wire

    (1994)
  • M. Doolan et al.

    Use of image recognition techniques in the analysis of sheet metal forming force signature curves

    J. Manuf. Sci. Eng.

    (2003)
  • Cited by (54)

    • Metal forming defect detection method based on recurrence quantification analysis of time-series load signal measured by real-time monitoring system with bolt-type piezoelectric sensor

      2022, Mechanical Systems and Signal Processing
      Citation Excerpt :

      Arun et al. [21] measured acoustic emission signals during the grinding process to monitor the surface roughness obtained throughout the grinding cycle and used various machine learning algorithms to predict the condition of the grinding tool. Ubhayaratne et al. [22] monitored the tool wear conditions of the sheet metal stamping process through audio signal analysis. A bolt-type piezoelectric sensor (Fig. 1) based on the “sensoric fastener” technology was proposed by Groche and Brenneis [23]; another piezoelectric load sensor is the “PiezoBolt,” developed by ConSenses GmbH [24].

    • Development and testing of a wireless rotating triaxial vibration measuring tool holder system for milling process

      2020, Measurement: Journal of the International Measurement Confederation
      Citation Excerpt :

      The cutting process is a complex dynamic physical process. A variety of sensors have been employed to acquire the dynamic information indirectly, such as cutting forces [6–10], spindle currents [11,12], vibrations [13,14], sound [15,16] and acoustic emission (AE) [17,18], etc. The generated vibration signals contain a wealth of useful information, which is often used for chatter detection [19], surface roughness prediction [20], tool condition monitoring [21], etc.

    • An industrially viable wavelet long-short term memory-deep multilayer perceptron-based approach to tool condition monitoring considering operational variability

      2023, Proceedings of the Institution of Mechanical Engineers, Part E: Journal of Process Mechanical Engineering
    • Predictive maintenance methodology in sheet metal progressive tooling: a case study

      2023, International Journal of System Assurance Engineering and Management
    View all citing articles on Scopus
    View full text