Remaining useful life prediction of aero-engine based on an enhanced probSparse self-attention

Predicting the remaining useful life (RUL) of aircraft engines allows for advanced estimation of its health status, serving as a crucial element in the overall health management system. Traditional RUL prediction models utilizing long short-term memory and convolutional neural network technologies suffer from high complexity and limited generalization capabilities. Thus, an enhanced probSparse self-attention mechanism inspired by the Informer is introduced and applied to the RUL prediction of the commercial modular aero-propulsion system simulation (CMAPSS) datasets. Firstly, probSparse self-attention is employed to diminish the computational complexity of the traditional mechanisms. On this basis, a gated convolution neural network is incorporated to extract the correlation characteristics between sensor signals and flight conditions, and a fully connected neural network is utilized to fit the mapping relationship between the high-dimensional features outputted by the self-attention network and the engine RUL. Ultimately, the CMAPSS datasets are leveraged for training and testing the model, and the simulation results demonstrate that the proposed methodology achieves a reduction in RUL prediction errors by 41.0% for the individual degradation and 41.3% for the simultaneous degradation respectively under high-altitude conditions.