主讲人概况：夏敏，1987年4月生，2017年获得加拿大英属哥伦比亚大学机械工程专业博士学位，2017年至2019年在英属哥伦比亚大学从事博士后研究，现在英国兰卡斯特大学从事讲师工作。主要从事机电系统健康监测，模式识别，深度学习等领域研究，目前以第一编辑或通讯编辑在IEEE Transactions on Industrial Informatics，IEEE Transactions on Mechatronics等期刊上发表SCI论文10余篇（ESI高被引一篇），他引300余次；同时是多个SCI期刊的客座编委。
报告摘要：The growing demand for high quality, low cost and highly customized products requires increasing reliability of machinery in the production systems. At present, online diagnosis and prognosis procedures for machine conditions are implemented to carry out condition-based maintenance where an accurate and appropriate maintenance strategy is needed. Due to the wide use in various industries and costly and catastrophic outcome their malfunction can bring, condition monitoring and fault diagnosis of rotating machinery; e.g., roller bearings and gear transmissions, is an active research area. With the advances in sensor technologies and signal processing, various signals such as vibration, acoustic emission, temperature, pressure, and current have been investigated and implemented in monitoring the conditions of rotating machinery. Waveform signals are widely studied in fault diagnosis since they can be accurate indicators of the health status of most rotating machinery. Features of the signals in the time domain, frequency domain and time-frequency domain have been studied and used for fault diagnosis of rotating machinery, assisted by methods of artificial intelligence.
Most of the traditional approaches rely on manual feature extraction, which requires significant prior knowledge of signal processing and diagnostic expertise. Also, the existing algorisms are for specific issues and therefore are case sensitive and not meant for general applications. In recent years, deep neural networks (DNNs) have been investigated and have shown promising capability of capturing representative features from raw data through multiple non-linear transformations across their deep structures. DNNs have been implemented in many applications such as computer vision, natural language processing, speech recognition, and bioinformatics with outstanding performances compared with the approaches that use traditional manually designed features. Due to the superior capability of DNNs in feature learning and classification, they have recently attracted the attention of researchers in machine fault diagnosis.
This talk will first present a stack denoising autoencoder-based intelligent fault diagnosis method that can make use of the unlabeled data. By using small amount of labeled data, the DNN model can be further fine tuned to achieve satisfactory diagnosis result.
A regular fully connected DNN has limited application in more complex problems due to the exponential growth of parameters when more layers were added to the model. Compared to the standard deep neural networks with all fully connected layers, a convolutional neural network (CNN) has much fewer connections and parameters so that it is easier to train with less computational resources and available training data. With the linear and nonlinear layers in the model, CNN has shown strong capability in learning sensitive and robust features. Recent research has shown that an estimator employing multiple sensors and sensor fusion techniques can provide enhanced and robust estimates.
Second, this talk will discuss a novel fault diagnosis approach for rotating machinery based on CNN. Raw vibration signals are directly used as the input to the model to detect different failures. Signals from multiple sensors are fused at the data level in this model to increase the accuracy and reliability of the diagnosis. Representative features are extracted automatically through feature learning of the CNN-based model.
Remaining useful life prediction is the core task in prognosis which is the other main aspect of machine health monitoring. Traditional data driven approaches of RUL prediction rely heavily on manual feature extraction and selection using human expertise. This talk will present an innovative two-stage automated approach to estimate the RUL of bearings using deep neural networks (DNNs). A denoising autoencoder-based DNN is used to classify the acquired signals of the monitored bearings into different degradation stages. Representative features are extracted directly from the raw signal by training the DNN. Then, regression models based on shallow neural networks are constructed for each health stage. The final RUL result is obtained by smoothing the regression results from different models. The proposed approach has achieved satisfactory prediction performance for a real bearing degradation dataset with different working conditions.