Advanced Condition Monitoring of Complex Mechatronics Systems Based on Model-of-Signals and Machine Learning Techniques

Prognostics and Health Management (PHM) of machinery has become one of the pillars of Industry 4.0. The introduction of emerging technologies into the industrial world enables new models, new forms, and new methodologies to transform traditional manufacturing into intelligent manufacturing. In this context, diagnostics and prognostics of faults and their precursors has gained remarkable attention, mainly when performed autonomously by systems. The field is flourishing in academia, and researchers have published numerous PHM methodologies for machinery components. The typical course of actions adopted to execute servicing strategies on machinery components requires significant sensor measurements, suitable data processing algorithms, and appropriate servicing choices. Even though the industrial world is integrating more and more Information Technology solutions to keep up with Industry 4.0 new trends most of the proposed solutions do not consider standard industrial hardware and software. Modern controllers are built based on PCs and workstations hardware architectures, introducing more computational power and resources in production lines that we can take advantage of. This thesis focuses on bridging the gap in PHM between the industry and the research field, starting from Condition Monitoring and its application using modern industrial hardware. The cornerstones of this "bridge" are Model-of-Signals (MoS) and Machine Learning techniques. MoS relies on sensor measurements to estimate machine working condition models. Those models are the result of black-box system identification theory, which provides essential rules and guidelines to calculate them properly. MoS allows the integration of PHM modules into machine controllers, exploiting their edge-computing capabilities, because of the availability of recursive estimation algorithms. Besides, Machine Learning offers the tools to perform a further refinement of the extracted information, refining data for diagnostics, prognostics, and maintenance decision-making, and we show how its integration is possible within the modern automation pyramid