Wear State Score Prediction of Friction Testing Machine using Improved Ensemble Convolutional Neural Network

Oil monitoring is an effective means to ensure the safety of equipment operation. However, the existing online oil monitoring methods are difficult to measure and evaluate the health status of equipment in real time. Therefore, the wear state score prediction is quite significant. In this paper, we propose a wear state score prediction method named Improved Ensemble Convolutional Neural Network (IECNN). First, we use wear image sensors and metal particle sensors to obtain real-time dynamic monitoring data in the oil and screen out important wear features using Pearson correlation analysis. Second, we conduct similarity analysis to obtain the similarity between the initial state of wear and the current state features, which is used to reflect the wear state of equipment. Third, we designed an improved ensemble convolutional neural network composed of five blocks containing convolutional layers and maximal pooler layers to train the monitoring data. Finally, we adopt the integrated learning strategy to integrate the score prediction possibility obtained under different modes to obtain the wear state score value of the device. The proposed method is tested on the friction testing machine, and the whole life cycle of the oil under 4 working conditions. The experimental results show that the proposed method can accurately predict a score below the 0.3 alarm threshold as an early warning when the equipment is seriously worn, which is consistent with the offline detection report provided by the third-party testing institution. In contrast, it is significantly superior to existing feature similarity methods such as Phash and NCC.