Cross working conditions manufacturing process monitoring using deep convolutional adversarial discriminative domain adaptation network

In recent years, with the increase in the variety of customer demands, the different quality demands have improved the diversity of manufacturing working conditions. The previous process monitoring approaches assume that the training and testing process signals are collected from the same working conditions. Unfortunately, in the real manufacturing process, it is expensive and even impossible for people to collect enough abnormal process data in some working conditions to build a monitoring model. A two-stage network–deep convolutional adversarial discriminative domain adaptation network (DCDAN) is proposed in this work. By introducing deep learning theory into domain adaption, the proposed DCDAN exploits the generator and discriminator with abnormal process signals in the source working condition to adapt normal signals in the target working condition, and thus improve the accuracy of the monitoring model. With continuously generating abnormal signals for cross-working condition monitoring tasks, the proposed DCDAN can provide reliable cross-working conditions manufacturing process monitoring results when abnormal data in the target domain are not available. Extensive empirical evaluations on eight monitoring tasks of the milling process validate the applicability and practicability of DCDAN.