Development of anomaly detection system by cutting sounds using deep learning

Abstract

With the development of Internet of Things (IoT) technology in the production field, it has become possible to collect various time-series signals. This has led to increasing interest in data-based deep learning methods for anomaly detection, replacing rule-based methods that rely on theoretical or experimental laws. This study was performed to develop an AI model to detect cutting process anomalies, such as cutting tool wear, based on time-series signals, such as vibration, sound, and acoustic emissions. The proposed method consists of two functional elements: a Hilbert-Huang Transform (HHT) that extracts instantaneous amplitude and frequency data as features from time-series signals, and a Convolutional Neural Network (CNN) that determines the degree of abnormality based on the extracted features. Here, we report the performance of the proposed method, which was verified using pseudo-signals and then applied to cutting sounds.