Hybrid As-Operated Digital Twin of an Aircraft Brake

Le résumé fourni par la source

This paper proposes a Digital Twin-based approach to predict friction wear of an aircraft braking system between required brake overhauls, based on individual aircraft operating conditions. The classification of Condition-Based Maintenance (CBM) as a Type III approach in SAE ARP6887 introduces significant challenges for purely statistical or data-driven algorithms. Models must be component-specific and sensitive to varying operational conditions, leading to domain shift, data sparsity, and limited generalization across fleets. Ensuring robustness, interpretability, and certifiability under these constraints are open research challenge. Conventional data analytics approaches are of limited use in scenarios characterized by a lack of run-to-failure data. On the other hand, the usage of model-based approaches can be limited due to complex physics modeling. Hybrid technologies are a more recent research area in the field of CBM, integrating prior physical knowledge with Machine Learning (ML) solutions. In this context, the As-Operated Digital Twin paradigm has emerged as a promising framework, representing the virtual counterpart of physical assets to reflect actual in-service condition and usage history. It can provide several benefits in the field of CBM, as it provides real-time insights into the health and degradation status of the monitored component and reduces the common challenges of a lack of run-to-failure data and a lack of collocated sensors with respect to the source of degradation. This article proposes a Hybrid As-Operated Digital Twin architecture, relying on the Archard equation to model the degradation phenomenon. The unknown degradation parameters of Archard’s law are usually modeled with empirical equations based on laboratory data, limiting the validity of the model to the laboratory domain. To improve model generalization and reduce the number of experiments to fit the degradation parameters, a physics-informed Recurrent Neural Network (RNN) model is proposed. The proposed Hybrid As-Operated Digital Twin includes the physics-informed RNN model and the heat-sink thermal model of the brake system. The robustness of the model is studied, employing Monte Carlo methods for uncertainty quantification. The discussed methodology is demonstrated on a sample aircraft brake model with laboratory-domain data.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.