Failure diagnosis and prognosis methods based on physical models

Equipment health is a major issue for manufacturers operating complex industrial systems. By monitoring the state of health throughout the life cycle, it is possible to detect the onset of deterioration, diagnose it and estimate the remaining life before failure (DEFAD, RUL). Over the last few decades, numerous methods have been developed in the field of CBM (condition-based monitoring-PHM [physical health management]), using specific models. This article presents the state-of-the-art in methods and tools based solely on physical models for diagnosing and prognosticating failures. The first section provides a brief reminder of the issues and objectives of CBM-PHM for residual life prediction (RUL), using the terminology defined by international standards NF EN 13306, ISO 13372 and ISO 13381. It then lists the phases of the CBM-PHM procedure adapted to physical models: observation, data pre-processing to identify model parameters and anomaly detection, diagnosis of degradation or failure based on analysis of model parameters (empirical or knowledge-based), prognosis in the presence of diagnosed degradation to estimate residual life, and decision-making for appropriate maintenance actions. It also provides a non-exhaustive list of the main metrics used to characterize RUL confidence and prediction. The second section is devoted to defining a topology of the various physical models in two categories. The first category is derived from the discipline of physics of failure (PoF), and is particularly well suited to modeling the reliability of electronic components and the failure mechanisms of materials used in mechanical engineering. For physical models for predicting the reliability of electronic components, the HDBK 217F, 217F plus™, FIDES, models are specified. For empirical models associated with material failure mechanisms, examples of empirical laws are presented: ISO 281 standard, Paris-Erdogan law, Whôler, Manson-Coffin... An example of a model used for the failure mechanisms of a wind turbine gearbox illustrates this issue. For the sake of completeness, the various concepts of statistical mechanics that take account of uncertainties in endurance testing are briefly reviewed. The second category of physical models is based on known laws of physics (electricity, electromechanics), whose mathematical structures and number of parameters are well known. These models are classified according to their characteristics (representation or knowledge model), properties and mathematical expressions (differential equations, equations of state, etc.). The third section is devoted to diagnostic and prognostic techniques, which consist in identifying and estimating parameters or state vectors in order to detect a characteristic evolution of degradation or failure. This section details the reference model method, the ordinary...