Hybrid modelling in condition monitoring

Abstract: Assuring the reliability, availability, maintainability and safety of assets is key to business success. A logical first step is to consider the requirements of assets in the design process. However, these concepts must also be assured during the assets’ operation. Consequently, it is important to have knowledge of their actual condition.The condition monitoring of assets and their subsequent maintenance are changing with the rapid evolution of electronics and information and communication technologies. The contribution of such technologies to the monitoring of cyber-physical systems in the context of Industry 4.0 is important.In the era of big data, the ease of getting, storing and processing data is crucial. However, the trend towards big data is not as effective in the field of condition monitoring as in others. One of the challenges of today’s condition monitoring is the lack of data on those assets not allowed to operate beyond their pre-established maintenance limit. Datasets miss advanced degradation states of assets and fail to predict rarely occurring outliers, but both have a great impact on operation; in other words, data-driven methods are limited and cannot accurately tackle scenarios outside the training dataset.This thesis proposes augmenting such datasets with the addition of synthetic data generated by physics-based models describing the dynamic behaviour of assets. It argues a combination of physics-based and data-driven modelling, known as hybrid modelling, can overcome the aforementioned limitations. It proposes an architecture for hybrid modelling, based on data fusion and context awareness and oriented to diagnosis and prognosis.The thesis applies some of the key parts of this architecture to rotating machinery, developing a physics-based model for a rotating machine from an electromechanical point of view and following a multi-body approach. It verifies and validates the model following guidelines suggested in the literature and using experimental data acquired in predefined tests with a commercial test rig.The developed physics-based model is used to generate synthetic data in different degradation states, and these data are fused with condition monitoring data acquired from the test rig. A data-driven approach is used to train an algorithm with the resulting fused data, adapting the clusters obtained by an algorithm to the context in which the machine is operating. The hybrid model is applied specifically for fault detection, localisation and quantification. The use of context data is found to enhance the results and is the key to providing context-driven services in the future.In short, the model is ready to react to faults that have not occurred in reality, with a severity that has not been reached in a specific operating context but has been introduced in the physics-based modelling.