Cavitation Prediction Method of Lubricating Oil Pump Based on Bayesian Optimized Convolutional Neural Network

The lubricating oil pump is the power for the circulating flow of oil, which is one of the critical components to ensure the proper operation of the lubricating system. Its cavitation caused by the high-altitude and high-speed working environment is an issue that cannot be ignored. And the pressures are the most direct indicators characterizing the state of the oil. This paper focuses on the internal meshing gear pump as the research subject, performs a simulated high-altitude test. According to the decrease in the oil flow rate, the pump cavitation is classified into 4 levels. Meanwhile, cavitation prediction models were developed using a Convolutional Neural Network (CNN). The raw pressure signals of the inlet and outlet were utilized as inputs to models, while the cavitation degree of the pump served as the output. A comparison of the prediction results from different signal sources was conducted to investigate the sensitivity of two pressure signals to the cavitation of the pump. To enhance the prediction performance, the Bayesian Optimization algorithm was integrated into models for hyperparameter optimization. Subsequently, BO-CNN models tailored to different signal sources were developed, which significantly enhanced model accuracy. Finally, the prediction results were evaluated and compared across multiple dimensions. The results show that the CNN model can effectively extract the cavitation degree signal. Both the prediction accuracy for inlet pressure and outlet pressure exceed 90%. Notably, the outlet pressure exhibits superior performance. The BO-CNN model, constructed based on the outlet pressure, achieves the highest prediction accuracy of 98.99%.