Data-Driven Diagnostics and Prognostics for Complex Systems: Methods and Applications

Recent advances in artificial intelligence or machine learning have the potential to significantly improve the effectiveness and efficiency of diagnostic and prognostic techniques. The objective of this research is to develop novel data-driven predictive models with machine learning and deep learning algorithms that allow one to model the degradation, detect the faults, as well as predict the remaining useful life (RUL) of complex systems, including bearings, gearboxes, and Lithium-ion (Li-ion) batteries. First, an enhanced ensemble learning algorithm is developed to improve the accuracy of RUL prediction by selecting diverse base learners and features at different degradation stages. The proposed method with increased diversity in base learners and features was demonstrated to be more accurate than other reported algorithms. Second, a convolutional long short-term memory (Conv-LSTM) approach is introduced to accurately classify the type, position, and direction of gear faults under different operating conditions by extracting spatiotemporal features from multiple sensors. The proposed method achieved 95% classification accuracy of fault type and 80% classification accuracy of fault location. Third, a deep learning method that combines convolutional neural networks (CNN) and bi-directional long short-term memory (BiLSTM) is developed to predict the discharge capacity and the end-of-discharge (EOD) of Li-ion batteries. The results show that by considering the discharge capacity estimated by CNN, the MAPE of EOD prediction using BiLSTM decreased from 8.52% to 3.21%. Fourth, a physics-informed machine learning method that combines the calendar and cycle aging (CCA) model and a LSTM model is developed to predict battery degradation behavior and RUL under different working conditions. The results show that the proposed method can predict the RUL of batteries accurately (10% in term of MAPE)