127 / 2025-05-18 08:40:49

Quantitative relationship model of wear particles in different particle size classes and failure warning method

multivariate nonlinear regression, abrasive particles count, failure warning, fluid monitoring

The size distribution of abrasive particles produced by wear on rotating parts reflects the different wear stages of the equipment, and there is a correlation between the number of abrasive particles in different sizes. On-line oil abrasive sensors can collect the size and number of abrasive particles in the lubricating fluid, but there is a lack of methods to directly use this data for failure warning. For the failure warning of equipment rotating parts, the article constructs a model of the relationship between the number of abrasive particles in different size intervals based on the theory of multiple nonlinear regression, which can be used to estimate the number of abrasive particles in different size intervals and realize the failure warning of the equipment. Firstly, we need to carry out full-life experiments of target parts on the equipment and collect the abrasive grain data; secondly, we verify the correlation of the abrasive particles data between different sizes, and establish a model of the quantity relationship between abrasive particles through the multivariate nonlinear regression theory; finally, we use the model to predict the number of abrasive particles, and search for the time point when large abrasive particles, which are more threatening to the equipment, appear in large quantities, which is the theoretical failure warning point of the equipment. The experimental results show that the coefficient of determination R² of the multivariate nonlinear regression model established by this method can reach 91.55%, the mean absolute error (MAE) is 1.9944, and the root mean square error (RMSE) is 2.702. The failure warning is made 12.5h ahead of time, which is better than that of the simple univariate nonlinear regression model in judging the time when the number of abrasive particles rises sharply, and the model can realize the failure warning of large abrasive particles by using the number of medium-small abrasive particles. In addition, the accuracy and versatility of the method were further verified on other devices, and the warning of unexpected failures was made 5.5h in advance, which is of practical value.