Generative Neural Network-Based Online Domain Adaptation (GNN-ODA) Approach for Incomplete Target Domain Data

Recently, monitoring machine health with artificial intelligence (AI) models becomes more efficient using either vibration or audio signals. However, the vibration signals of machines with different tool materials or under different operating conditions are not consistent. Hence, the transfer learning algorithm used in this study is presented for domain adaptation to improve the inference accuracy. Herein, we introduce a domain adaptation technique to solve domain shift problems during machine health monitoring. Additionally, most existing articles are assumed to have obtained complete target data. However, these data are continuously acquired; that is, the target data are incomplete during the monitoring phase. Thus, a generative neural network-based online domain adaptation model (GNN-ODA) is proposed. This will improve the test accuracy by generating complete target data and further training the classifier using the complete source data. The experiments indicate that the proposed method outperforms other domain adaptation methods wherein target data were incomplete. The average accuracy on both the tool wear and bearing datasets exceeded 90% when the source and target domains were under similar operating conditions.