Framework for Offline Data-Driven Aircraft Fault Diagnosis

This paper proposes a framework for aircraft fault diagnosis based on the offline analysis of flight data. It overcomes the limitations of current data-driven approaches by combining steps based on both real data, obtained from aircraft flight data records, and simulated data, generated from aircraft models. The framework explores unsupervised and supervised methods, resulting in a proactive approach to flight safety and speeding the learning of fault cases. The influence of both temporal data representation and sensor selection on fault diagnosis performance is analyzed. The framework is organized into four phases (initial, training, operation, and improvement) that cover the aircraft system lifecycle. We used the hierarchical clustering algorithm in the unsupervised part and an ensemble of three algorithms (k-nearest neighbors, decision trees, and neural networks) in the supervised one. The framework is evaluated using an aircraft electrohydraulic actuating system as the case study, for which we obtained a balanced accuracy of 96% in the operation phase and of 90.4% in the improvement phase. The contribution of the framework is also accessed through a comparison with results obtained using only supervised methods. It confirms that the combination of supervised and unsupervised methods improves the performance of the fault diagnosis system.