Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Tool Wear Prediction in CNC Turning Operations using Ultrasonic Microphone Arrays and CNNs

Jan Steckel, Arne Aerts, Erik Verreycken, Dennis Laurijssen, Walter Daems

TL;DR

Problem: Accurate tool wear prediction in CNC turning is challenging due to noisy environments and variable wear behavior. Approach: The authors integrate an ultrasonic microphone array with beamforming and a CNN to regress the Remaining Useful Life ($RUL$) of a carbide insert from spectrogram-like inputs $S(\omega,n)$. Data and architecture: The model is trained on a dataset of 350 workpieces on a Mazak QT10N lathe using Iscar GRIP 3015Y IC808 inserts, with data augmentation and an Adam optimizer. Findings and impact: The method achieves a maximum $RUL$ error of about 6% of total tool life and can also predict spindle current with around 1 A RMS error, demonstrating the viability of combining ultrasonic sensing with deep learning for predictive maintenance in automated turning.

Abstract

This paper introduces a novel method for predicting tool wear in CNC turning operations, combining ultrasonic microphone arrays and convolutional neural networks (CNNs). High-frequency acoustic emissions between 0 kHz and 60 kHz are enhanced using beamforming techniques to improve the signal- to-noise ratio. The processed acoustic data is then analyzed by a CNN, which predicts the Remaining Useful Life (RUL) of cutting tools. Trained on data from 350 workpieces machined with a single carbide insert, the model can accurately predict the RUL of the carbide insert. Our results demonstrate the potential gained by integrating advanced ultrasonic sensors with deep learning for accurate predictive maintenance tasks in CNC machining.

Tool Wear Prediction in CNC Turning Operations using Ultrasonic Microphone Arrays and CNNs

TL;DR

Problem: Accurate tool wear prediction in CNC turning is challenging due to noisy environments and variable wear behavior. Approach: The authors integrate an ultrasonic microphone array with beamforming and a CNN to regress the Remaining Useful Life () of a carbide insert from spectrogram-like inputs . Data and architecture: The model is trained on a dataset of 350 workpieces on a Mazak QT10N lathe using Iscar GRIP 3015Y IC808 inserts, with data augmentation and an Adam optimizer. Findings and impact: The method achieves a maximum error of about 6% of total tool life and can also predict spindle current with around 1 A RMS error, demonstrating the viability of combining ultrasonic sensing with deep learning for predictive maintenance in automated turning.

Abstract

This paper introduces a novel method for predicting tool wear in CNC turning operations, combining ultrasonic microphone arrays and convolutional neural networks (CNNs). High-frequency acoustic emissions between 0 kHz and 60 kHz are enhanced using beamforming techniques to improve the signal- to-noise ratio. The processed acoustic data is then analyzed by a CNN, which predicts the Remaining Useful Life (RUL) of cutting tools. Trained on data from 350 workpieces machined with a single carbide insert, the model can accurately predict the RUL of the carbide insert. Our results demonstrate the potential gained by integrating advanced ultrasonic sensors with deep learning for accurate predictive maintenance tasks in CNC machining.
Paper Structure (4 sections, 2 equations, 4 figures)

This paper contains 4 sections, 2 equations, 4 figures.

Table of Contents

  1. Introduction
  2. Hardware Setup
  3. Signal Processing and Experimental Results
  4. Discussion and Conclusion

Figures (4)

  • Figure 1: Overview of the hardware setup used in this paper. Panel a) shows the eRTIS sensor, an ultrasonic array sensor with 32 microphones arranged in a pseudorandom pattern. The sensor is waterproofed using a membrane material from the Acoustic Protective Material line manufactured by W. L. Gore and Associates. Panel b) shows the CNC lathe being used, a Mazak Quickturn 10N, using Iscar GRIP 3015Y grade IC808 inserts. Panel c) shows the inside eRTIS sensor mounted in the lathe. Panel d) shows the turned materials, and e) shows the model of the piece that has been turned during this experiment.
  • Figure 2: Spectrogram representation of the acoustic data recorded by the eRTIS sensors during a turning run. Panel a) shows the data of measured by the eRTIS sensor ouside of the machine, and panel b) shows the data recorded from inside the machine. In the low frequency parts, the signal is dominated by the motion sounds of the machine, and the high frequency parts of the spectrogram are generated by the cutting operations.
  • Figure 3: Overview of the CNN architecture, consisting of 4 convolutional layers, followed by a leaky ReLu activation, average pooling, batch normalization and dropout. After the convolutional layers, fully connected layers are used to form the regression outputs.
  • Figure 4: Results of the various metric predictions performed using the CNN networks based on the spectrogram inputs. Panel a) shows the run-number prediction based on a single spectrogram, and panel b) shows the prediction when using the predictions of five spectrograms centered around the desired run number. Panels c) and d) show the mean run number prediction error including the variance bands on the prediction (grey shadows). Panel e) shows the prediction of the spindle RMS current in Ampere, based on the acoustic data, and panel f) shows the error of the spindle current prediction.