IOP Conference Series: Materials Science and Engineering

Tracking the wear states of tools on punching machines is necessary to reduce scrap rates. In this paper, we propose a method to estimate wear state of punches using Temporal Convolutional Network Autoencoder (TCN-Autoencoder), one of the deep learning techniques for learning time-series information with convolutional architecture. Approach involves inputting raw time-series information, such as sensor, vibration and audio data, into TCN-Autoencoder, and calculating the reconstruction error between the output and the input data. The reconstruction error is used as “anomaly score” and indicates the distance from the normal state. By training TCN-Autoencoder only with data annotated as “normal” state, the reconstruction error becomes larger when inputting abnormal state data, which corresponds the wear state of the punch. Performance is evaluated on experimental measurement data that spans various wear states of the punch. The results showed our model can estimate anomalies faster than the conventional machine-learning-based anomaly estimation method, while maintaining the high estimation accuracy. This is due to TCN-Autoencoder being able to learn from both frequency and time domain.

The amount of information contained in process signals such as acoustic emission and force signals has proven vital for the detection of changes in physical conditions or quality feature prediction in sheet metal forming applications. Both signal types have also been researched in the context of wear detection, yet systems that reliably identify the wear state at a given time in sheet metal forming processes based on these signals do not exist. This paper proposes an architecture to assess the wear increase within a given time frame in an experiment based on an autoencoder. The ability of autoencoders to encode and decode signals has been widely studied and this approach leverages the fact that autoencoders are more likely to learn representative encodings on stable and homogeneous signals than on heterogeneous signals with high fluctuations. This approach utilizes the circumstance that high tool wear leads to changes in the signal and signal fluctuation. In consequence, autoencoders can be utilized to track tool wear progression without the need for labelled data. The findings show a strong similarity to physical models for the wear progression of tool components, indicating the validity of this approach. Additionally, an analysis of the signals yields characteristic effects of the considered force signals that could specifically represent wear resistance.