Verification of the technical equipment degradation method using a hybrid reinforcement learning trees–artificial neural network system

Abstract This article reports on a study aimed to assess the extent to which engine operation time can be prolonged without oil change in systems working in a discontinuous mode. This applies particularly to elements of critical infrastructure or vehicles performing occasional work. The analysis of longitudinal oil sample data has enabled us to study long-term changes in the levels of particles and substances in the oil. This new database will support the validation and verification of the team's former research and determine the practical implications emerging from the revised results. The data were investigated by means of a synergy-based method combining the computational powers of reinforced decision trees and artificial neural networks. The complex selection of signals from laboratory tests of oil samples obtained from over a decade of vehicle testing is the factor that facilitated establishing optimal system condition parameters. The proposed reinforced learning tree (RLT)-based model is an implementation of stochastic gradient boosted decision trees – a method that is particularly useful for predictive data mining. By combining RLT with artificial neural network modelling, we have produced and validated an effective tool for predicting the investigated technical system's operation time. A distinct advantage of using neural networks on a new set of variables derived from an authentic tribological system is that it enables a genuine evaluation of the model's operation. The advantages of the proposed method are: i) high-accuracy technical system condition and reliability assessment; ii) improved equipment usage planning; iii) optimising the safe use of the equipment and its lifetime cost.