Rotating machinery is central to transportation and power generation, and maximizing its uptime while minimizing unplanned maintenance is important from safety and economic perspectives. Roller bearings are ubiquitous
components in such machinery and are very often the cause of failures; thus, condition monitoring of roller bearings is a topic of key interest to many industries. Historically accomplished using human expertise and experience to
estimate the condition of a machine from a limited set of signals and curated statistical features, machine learning has become an effective tool in bearing fault identification with modern computing and algorithmic advances. In this regard, two major challenges exist: first, the availability of inservice data from bearings containing faults is often rare or difficult to obtain, and moreover, for machines being deployed to the field for the first time, no such data exists. Secondly, the extracted statistical measures (features) must be chosen to maximize the classifier accuracy (or another metric), and their selection of these to maximize accuracy can be a challenging task. We directly address these challenges with two separate approaches. With validation against four experimental datasets, we show that in the absence of recorded in-service data, machine learning algorithms trained using simulated bearing vibration signals (i.e. simulation-driven machine learning) can classify bearing race faults with greater accuracy than those trained using data collected from other machines. Next, we propose convolutional neural networks and nearest-neighbor dynamic time warping (NNDTW) as statistical feature-free methods to detect bearing race faults using a signal processing pipeline based on angle synchronous averaging. We show that these methods offer superior accuracy from the simulation-driven perspective and can predict an inservice wind turbine fault over one month before failure.
Camci, F., Medjaher, K., Zerhouni, N., & Nectoux, P. (2013). Feature evaluation for effective bearing prognostics. Quality and reliability engineering international, 477-486.
Janssens, O., Slavkovikj, V., Vervisch, B., Stockman, K., Loccufier, M., Verstockt, S., . . . Van Hoecke, S. (2016). Convolutional neural network based fault detection for rotating machinery. Journal of Sound and Vibration, 331-345.
Keogh, E., & Ratanamahatana, C. A. (2005). Exact indexing of dynamic time warping. Knowledge and information systems, 358-386.
LeCun, Y., & Bengio, Y. (1995). Convolutional networks for images, speech, and time series. The handbook of brain theory and neural networks.
McFadden, P., & Toozhy, M. (2000). Application of synchronous averaging to vibration monitoring of rolling element bearings. Mechanical Systems and Signal Processing, 891-906.
Rauber, T. W., de Assis Boldt, F., & Varejao, F. M. (2015). Heterogeneous feature models and feature selection applied to bearing fault diagnosis. IEEE Transactions on Industrial Electronics, 637-646.
Sassi, S., Badri, B., & Thomas, M. (2007). A numerical model to predict damaged bearing vibrations. Journal of Vibration and Control, 1603-1628.
Siemens Industry Software NV. (2016). Siemens LMS Imagine.Lab Amesim (Version 15).
Sobie, C. a. (Submitted 01/2017). Simulation-Driven Machine Learning: Bearing Fault Classification. Submitted to Mechanical Systems and Signal Processing.