He, Miao Development of Deep Learning Based Methodology on Rotating Machine Fault Diagnosis Rotating machines are widely used in various industrial applications. It is necessary to implement the condition based maintenance for rotating machines to prevent failures, increase reliability and decrease maintenance cost. Traditionally, the most critical issue in developing rotating machine fault detection and diagnosis methods is to extract and quantify the complicated signal processing-based fault features. With the combination of data mining techniques, faults can be diagnosed accurately using previously extracted features. However, nowadays there are challenges in using existing methods for rotating machinery fault diagnosis. In the age of Internet of Things and Industrial 4.0, massive real-time data were collected from health monitoring systems for fault diagnosis. The traditional methods to extract features from monitoring data manually with expertise in signal processing and prior knowledge in fault diagnosis is rarely accomplishable on a machinery big data platform. Therefore, a novel methodology that can automatically extract the adaptive fault features from monitoring data and, diagnose the fault pattern intelligently, is expected to realize rotating machinery fault detection and diagnosis on machinery big data platform. With its deep architecture, deep learning can automatically extract features from the data and hence eliminate the process of handcrafting features from the data. Though there is a growing interest in using deep learning for machinery fault detection and diagnosis, some challenges still exist. However, the raw monitoring data were processed with complicated signal processing algorithms such as wavelet-package transform (WPT) , or pre-processed to obtain features . The complicated signal processing is still required in many reported deep learning based fault diagnosis applications in literature. The current selection of deep learning architecture is trial and error based. The selection of deep learning architecture has not been well investigated yet. Until now, only the vibration condition monitoring data was studied with the application of deep learning based approaches. Other monitoring data such as acoustic emission (AE) data and piezoelectric (PE) data have yet to be processed with deep learning based approaches. In this research, novel deep learning based methodologies that can automatically extract the adaptive fault features from monitoring data and intelligently diagnose the faults with machinery big data is developed to address the issues stated above. Specifically, the following new effective and efficient rotating machine fault diagnosis are presented: a deep learning based approach for bearing fault diagnosis using AE signals, a deep learning based approach for simultaneous bearing fault diagnosis and fault severity detection using vibration signals, a deep learning based approach for planetary gear box (PGB) fault diagnosis, a signal processing integrated deep learning approach for bearing fault diagnosis using vibration signals. The realization of adaptive feature extraction and learning can reduce the ratio of training samples to testing samples. Furthermore, a novel signal processing integrated deep learning method is proposed to capture the hidden time and frequency features in the monitoring data. The introduction of signal processing into deep learning method provides a view of effective deep learning method on time series monitoring signals. To validate the proposed methodology on rotating machinery fault diagnosis, data collected from a bearing test rig and a planetary gear box (PGB) test rig were used. The data was collected from the runs on bearings and gears with seeded typical faults. Vibration and AE data were collected at the bearing test rig, while vibration, AE, and PE data were collected on the PGB test rig. Fault Diagnosis, Rotating Machine, Condition Based Maintenance, Deep Learning 2018-11-27
    https://indigo.uic.edu/articles/thesis/Development_of_Deep_Learning_Based_Methodology_on_Rotating_Machine_Fault_Diagnosis/10831151