Performance of hybrid modeling on a failure process in industrial systems

Abstract

A non-localised failure on a component can cause irreparable damage but it can also lead to the complete shutdown of the industrial system if it is not detected in time. Indeed, the first step in a failure process is the detection of the fault. Locating the fault is the second step in the process to know at which level of the system to intervene. Numerous methods for diagnosing industrial systems have already proved their worth. They are mainly based on physics-based behaviour laws. However, these behavioural models are generic and present difficulties of adaptation when applied to particular job profiles. Moreover, when dealing with complex systems, the implementation of behavioural laws the coupling of multiple components, interacting with each other with each other, is a laborious and time-consuming task. time-consuming task. The development of industrial systems instrumentation also encourages the use of the potential of real-time data collected on the systems. The problem in studying the data is the transparency of the models created solely from this data. The weight of the interactions present between the system's variables is not always identifiable. This means that the models developed from the data will not be easily transposable from one system to another, guaranteeing the same performance. To improve this adaptability, the idea is to draw on the knowledge of the system in question and to integrated into the modelling. For this purpose, models based on and those based on data learning will be coupled in order to data learning will be coupled in order to study the overall performance of this type of modelling. In the literature, this In the literature, this coupling is called hybrid modelling. To understand the construction process of such a model, the To understand the construction process of such a model, the study proposes to focus on the modelling of a DC electric motor. This application, which is widely studied in the literature, allows us to to exploit existing physical models of the system. The objective of this paper is therefore to study the The objective of this paper is therefore to study the performance of hybrid modelling to diagnose The objective of this paper is therefore to study the performance of hybrid modelling to diagnose the failures of a DC electric motor. To this end, the paper will describe the construction of the data-based model and the theoretical model. model and the theoretical model by discussing the capabilities and the capabilities and limitations of each model. The implementation of a The implementation of a hybrid approach will then be detailed. Finally, the performance of the implemented models will be presented.