Development of a methodology for predicting the residual life of hydraulic cylinders of earthmoving machines

Abstract

The article is devoted to the development of a comprehensive methodology for predicting the residual life of hydraulic cylinders of earthmoving machines based on the integration of physical and mathematical models of degradation with machine learning methods. The study is based on the analysis of multiparametric data on acoustic emission, pressure in the hydraulic system, vibration characteristics and operating temperature. An innovative architecture of the forecasting system is proposed, including a multilevel model of seal degradation, an adaptive signal filtering algorithm, and a deep learning neural network LSTM with physical limitations. The experimental verification was carried out on a sample of 847 hydraulic cylinders of various types of earthmoving machines with cyclic loads from $2.3\times10^{5}$ to $4.7\times10^{6}$ operating cycles. The developed methodology ensures 87.3% accuracy in predicting the remaining resource with a planning horizon of up to 2,000 operating hours, which is 24,6% higher than existing approaches. The correlation coefficient between the predicted and actual values is $r=0.912$. The average relative error of forecasting does not exceed 12,4%. The practical implementation of the technique in maintenance systems reduces the likelihood of sudden failures by 34,8% and optimizes preventive maintenance intervals. The results obtained open up prospects for the creation of intelligent lifecycle management systems for hydraulic drives of heavy machinery.