Identification of cavitation intensity for high-speed aviation hydraulic pumps using 2D convolutional neural networks with an input of RGB-based vibration data

Power density is an important attribute for aviation hydraulic pumps, which can greatly benefit from improving rotational speed. However, cavitation tends to occur in the pump at high rotational speeds, thus decreasing its volumetric efficiency and lifetime. Therefore, cavitation identification is essential and urgent for high-speed aviation hydraulic pumps. In this paper, we propose a real-time method for identifying the cavitation conditions based on the vibration signals measured at the pump housing. The collected three-channel vibration data are cut into frames to be transformed into RGB images and then these images are fed into a 2D convolutional neural network (CNN) to identify the levels of cavitation intensity. The experimental results show that the CNN model can achieve high accuracy rates when it accepts optimal RGB images. In addition, RGB images are found to be more robust against noise than their gray counterparts in the case of vibration-based cavitation identification.