The success of Condition Monitoring remains based on the knowledge of the machines, of the influence of defaults on physical values (temperature, vibration, motor current, ..), on the selection and definition of pertinent features sensitive to the failures to be monitored and predicted. In addition, the current progress of Artificial intelligence and data analytics can also improve the results of Condition monitoring. Well-chosen tools taken from the Artificial Intelligence world can contribute to setting and improving Predictive Maintenance. A short review of useful Machine-Learning tools will be presented, comparing supervised and non-supervised approaches. Then through an example applied to bearings and gears condition monitoring (in a wind-turbine drive train), we will show how, using unsupervised learning, it is possible to “learn” the normal operating state of the machines (taking into account multi-sensors approach) and automatically detect a deviation from this state, enabling the detection of bearings and gears defects.
1. How Machine-Learning can improve Predictive Maintenance?
2. What tools from Artificial Intelligence can really help Condition Monitoring Practitioners and what is unsupervised learning ?
3. How can we use Machine-Learning in real application to detect mechanical components failures ?
Sophie SIEG-ZIEBA graduated, more than 20 years ago, from University of Technololgy of Compiegne (France) as Mechanical Engineer, specialized in Acoustics and Vibrations. She then received a PhD in System Control, focusing on signal processing and data analysis for monitoring in industrial applications.