From Data to Physics: Signal Processing for Measurement Head Degradation

Abstract

Data-driven Prognostic and Health Management is gaining more and more popularity, to the point where it is often seen as a potential solution for any maintenance problem. To the opposite, it is known that, when the physic of a system is known and tractable, it is usually the most efficient way to infer relevant indicators for detecting system behaviour changes. And sometimes, when the physic of a system is only partially known, data-driven approaches can highlight critical information that would help to make the right physical assumptions, which in turn, can be used for elaborating the right health indicators.   This paper presents a real-life example of such process. With focus on the estimation of the degradation of a measurement system head, a data-driven approach, with its limitations, helped to identify the right signal processing tools with which to analyse the data. With the results obtained from Spectral Analysis with Windowed Fourier Transform, the physics of the system could be hypothesized, and then used for deriving the most relevant indicators for monitoring the degradation. These indicators have been implemented online and so far used successfully on several machines encountering degradation of their measurement head.   This successful ``reversed'' case study of degradation monitoring is a strong reminder of the benefits of field symbiosis: data-driven approach made possible the understanding of the physics, which in turn could be used for data monitoring.