Maren Dangut

Deep learning approaches are continuously achieving state-of-the-art performance in aerospace predictive maintenance modelling. However, the data imbalance distribution issue is still a challenge. It causes performance degradation in predictive models, resulting in unreliable prognostics which prevents predictive models from being widely deployed in realtime aircraft systems. The imbalanced classification problem arises when the distribution of the classes present in the datasets is not uniform, such that the total number of instances in a class is significantly lower than those belonging to the other classes. It becomes more challenging when the imbalance ratio is extreme. This paper proposes a deep learning approach using rescaled Long Short Term Memory (LSTM) modelling for predicting aircraft component replacement under imbalanced dataset constraint. The new approach modifies prediction of each class using rescale weighted cross-entropy loss, which controls the weight of majority classes to have a less contribution to the total loss. The method effectively discounts the effect of misclassification in the imbalanced dataset. It also trains the neural networks faster, reduces over-fitting and makes a better prediction. The results show that the proposed approach is feasible and efficient, achieving high performance and robustness via skewed aircraft central maintenance datasets.

The use of aircraft operation logs to develop a data-driven model to predict probable failures that could cause interruption poses many challenges and has yet to be fully explored. Given that aircraft is high-integrity assets, failures are exceedingly rare. Hence, the distribution of relevant log data containing prior signs will be heavily skewed towards the typical (healthy) scenario. Thus, this study presents a novel deep learning technique based on the auto-encoder and bidirectional gated recurrent unit networks to handle extremely rare failure predictions in aircraft predictive maintenance modelling. The auto-encoder is modified and trained to detect rare failures, and the result from the auto-encoder is fed into the convolutional bidirectional gated recurrent unit network to predict the next occurrence of failure. The proposed network architecture with the rescaled focal loss addresses the imbalance problem during model training. The effectiveness of the proposed method is eval...

Aircraft fault detection and prediction is a critical element of preventing failures, reducing maintenance costs, and increasing fleet availability. This paper considers a problem of rare failure prediction in the context of aircraft predictive maintenance. It presents a novel approach of predicting extremely rare failures, based on combining two deep learning techniques, auto-encoder (AE) and Bidirectional Gated Recurrent Unit (BGRU) network. AE is modified and trained to detect rare failure, and the result from AE is fed into the BGRU to predict the next occurrence of failure. The applicability of the proposed approach is evaluated using real-world test cases of log-based warning and failure messages obtained from the aircraft central maintenance system fleet database and the records of maintenance history. The proposed AE-BGRU model is compared with other similar deep learning methods, the proposed approach is 25% better in precision, 14% in the recall, and 3% in G-mean. The resu...