Lubricating oil system fault diagnosis algorithm based on multi-head attention-Markov transfer field-CNN fusion optimization

The lubrication system is vital for cooling, regulating and lubricating diverse mechanical components during the operation of an aeroengine. It also enhances the engine’s performance regulation under various operating conditions. Due to the increasing complexity of engine structures, the growing sophistication of intelligence, and the challenges posed by complex and variable flight conditions in an increasingly demanding external environment, the failure mode of the lubrication oil system has become increasingly intricate. This system’s current fault diagnosis process only considers the algorithmic model’s input mode for one-dimensional data. It fails to extract the strong mapping relationship between the feature parameters and the fault modes with high contribution parameters, resulting in poor final fault pattern recognition. To address these shortcomings, the proposed paper presents an optimization algorithm that utilizes a multi-head attention mechanism coupled with Markov transfer field sensing point parameter feature extraction and dimension transformation. The model employs the distinctive two-dimensional feature recognition capability of convolutional neural network networks to accurately diagnose lubricating oil system defects. The accuracy of the algorithm is verified by rigorous investigation. The study’s implementation process is complete, and the accuracy rate has increased by more than 95% compared to the unoptimized rate.