Domain Adaptation Remaining Useful Life Prediction Method Based on AdaBN-DCNN

• Published in: 2019 Prognostics and System Health Management Conference (PHM-Qingdao) pp. 1 - 6
• Main Authors: Li, Jialin, Li, Xueyi, He, David
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.10.2019
• Subjects: Adaptation models; adaptive batch normalization; Convolution; Data models; deep convolutional neural network; Engines; Predictive models; remaining useful life; time domain adaptation; Training
• DOI: 10.1109/PHM-Qingdao46334.2019.8942857

Prognostics and health management (PHM) has received much attention as an emerging discipline. And the prediction of remaining useful life (RUL) is the core of the PHM. The data-driven RUL prediction methods are more favored because they can be developed faster and cheaper. However, the existing data-driven prediction models usually can only be used under the same data domain (DD), and they require a lot of labeled data to retrain a new prediction model. So a domain adaptation prediction model is more desirable. In this paper, a domain adaptation RUL prediction model is proposed by integrating the adaptive batch normalization (AdaBN) into deep convolutional neural network (DCNN). The improved AdaBN-DCNN model can not only improve the accuracy of the prediction, but also adapt to the prognostic tasks under different DDs. The sliding time window (TW) and the improved piecewise linear RUL function are also used in this paper to improve the prediction capability of the model. The proposed RUL prediction model is validated using the C-MAPSS turbofan engine dataset provided by NASA. The prediction results show that the proposed model not only has a strong predictive power but also adapts to different DDs.