Artificial Neural Network Approach for Estimating Operating Parameters for Predictive Maintenance of Hydraulic Circuit

This paper deals with the problem of model interpretability of neural network black box models in the terms of predictive maintenance. A testing device for the evaluation of predictive AI models has been presented. The experiment consisted in testing a feedforward neural network model, which was designed for the approximation of a complex multiparametric function and prediction of the monitored parameter (temperature of the working fluid in the tank). Favorable results were obtained for predicting the value of temperature in the working fluid tank based on the other simulation parameters and the simulation run time. Verification of the reliability of the prediction was carried out by additional neural network testing on new data. Correlation surface plots were also plotted and analyzed for the extracted dependencies between the parameters used for the prediction of the parameter of interest using the neural network. The achieved results have perspectives for processing the results obtained by neural network prediction in the field of predictive analytics and maintenance.