Développement d’une approche de pronostic pour les équipements complexes permettant l’application de la maintenance prévisionnelle

Abstract

Predictive maintenance is a discipline that allows planning maintenance actions based on prognostic models. It aims to enhance the decision-making capabilities of organizations by allowing them to plan the right actions at the right time. Unlike reliability-based maintenance approaches, predictive maintenance approaches can consider the dynamic and individual aspects of each asset's data using prognostic algorithms. Based on equipment health data, its load conditions, and its operating environment, these algorithms can predict the occurrence of equipment failure modes as new data become available and estimate the uncertainties intrinsically associated with them. Despite the arrival of new sensor technologies and data acquisition devices, the lack of exploitable historical data remains a common problem for a large proportion of industrial applications in the field. In addition, a large majority of existing approaches are applied to specific equipment with a relatively limited number of critical components and failure mechanisms. In the industry, equipment is often complex and typically has multiple failure modes and failure mechanisms that can be interdependent and evolve across many components. Their diagnostic information comes from different sources (for example, measurements and inspections) at discreet intervals over time. As a result, component-level prognostic models have limited applicability to complex devices because they do not manage the inherent complexity of failure mechanisms or the interrelationships between components. It is therefore difficult to implement predictive maintenance approaches. To deal with these issues, the thesis proposes a holistic approach to the development of prediction models for complex equipment allowing the application of predictive maintenance. As a first step, a diagnostic and prognostic approach for complex equipment is proposed. The approach is essentially based on the understanding and the modelling of the physics of degradation (mechanisms of degradation) to identify and follow up the evolution of the failure mechanisms through a succession of physical states of degradation detectable by diagnostic tools. The diagnostic algorithm integrates information from various diagnostic tools and tracks the evolution of active failure mechanisms through graphical representations in the form of dynamic causal graphs. It is then possible to follow the evolution of the failure mechanism. Based on the diagnostic results, the prognostic algorithm propagates the failure mechanisms to failure modes of the equipment. It then allows the estimation of the remaining time to the occurrence of each failure mode for each prediction date. In a second step, a predictive maintenance approach is proposed based on the formalism of the prognostic algorithm. The approach differs from the other approaches proposed in the literature, as it enables the identification of the required maintenance actions according to the detected active failure mechanisms and the results of the prognostic algorithms. The results allow the organization to automatically identify any mitigation actions that are possible to avoid the occurrence of failure modes and suggest their potential effects on the system. To ensure the long-term viability of the approach a new method is proposed to evaluate the performance of the prognostic algorithms for repairable systems considering the organization's operational processes. Finally, the process of updating prognostic models is proposed considering the evolution of the business context and new information available. The proposed approach is successfully applied and validated in the case of hydroelectric generators considering their operational contexts. It shows great potential for applications in various industries for future research. The contributions of this thesis do not only focus on the development of prognostic algorithms, but also aim at ensuring the applicability of these algorithms in an operational context. This thesis proposes the development of a multi-failure mode prognosis approach for complex equipment considering the complexity of their failure mechanisms. For this, the definition of a mathematical formalism of diagnostic and prognostic algorithms based on graph theory is presented to ensure the genericity of the proposed algorithms and the interactive visualization of their results. In addition, it offers a predictive maintenance approach allowing to predict the intervals of application of the maintenance actions that will be required, considering the active detected failure mechanisms, prognosis results and operational context. To ensure the accuracy of the results, this work presents the first approach allowing the performance evaluation of prognosis algorithms for repairable systems and the definition of the acceptable prediction area based on their operational context. Finally, a process for updating and evolving prognostic algorithms is proposed to ensure their long-term viability.

Date	9 Mar 2020
Original language	French
Awarding Institution	École de technologie supérieure
Supervisor	Antoine Tahan (Supervisor) & Dragan Komljenovic (Co-supervisor)