Leveraging transfer learning and maintenance domain adaptation to improve useful life prediction accuracy of high pressure compressor of aeroengine

Predicting the Critical Useful Life (CUL) of aeroengine components, especially the High-Pressure Compressor (HPC), accurately serves both safety operational needs and maintenance optimisation. Machine learning shows promising results for CUL prediction; however, the missing point is having enough specific model and environment-related failure data to properly train the systems. The research investigates Transfer Learning (TL) and Maintenance Domain Approaches (MDA) to enhance CUL prediction models. The method draws information from aeroengine components that possess abundant failure records, intending to enhance performance in the target HPC despite restricted labelled data availability. The research analyses different MDA strategies that serve to link simulated or historical data to the actual operational conditions and maintenance practices of field operations. Real-world alongside simulated aeroengine data will serve this investigation, which evaluates different TL and MDA techniques with pre-trained model fine-tuning and feature design. Thus, through advanced HPC CUL prediction accuracy, the study predicts significant progress which will improve both predictive maintenance efficiency and aeroengine operational availability. Therefore, this approach employs knowledge in the domain while adopting it to field maintenance properties so that one can derive an accurate, practical CUL prediction for HPC. The proposed methodology provides the following enhanced outcomes:•The trained model has a fitness of 98.82 % and a minimal error index of 7.4926.•The proposed model worked against the test, where the outcome is highly linear with a minimal error index of 11.7181 and 97.02 % fitness.•HPC is found to be a more critical part of the engine.