Transfer Learning Based Equipment Status Estimation System

The status of production equipment in industries is prone to be changed by various factors, which might affect the machining quality and even result in defects. Thus, anomaly detection for equipment is an indispensable procedure. Typically, using supervised learning-based techniques is a common solution for monitoring the equipment state. However, these approaches need a large amount of manpower and time to obtain enough labeled data. In recent, Autoencoder (AE) becomes a popular network structure for unsupervised deep anomaly detection. This paper proposes a so-called Tool Performance Estimation (TPE) scheme based on AE. TPE defines three generic scenarios: (I) equipment abnormality, (II) parameter abnormality, and (III) other abnormality, for any type of machinery equipment in industries to perform systematic troubleshooting by analyzing the severity of deviation between normal and abnormal data. TPE also monitors and ensures its performance by automatically updating models inside it. In the practical example, it is verified that TPE can timely warn and help users to analyze and troubleshoot when abnormalities occur in production.