Identification of qualitative regularities in the functioning of neural network models of a critical resource of lubricating oils

In the present work, it is proposed to compare two approaches to building a model of a critical resource of lubricating oils that describe the rate of change in the optical density of oils with time, depending on the duration and temperature of temperature control. The initial data for building models of a critical resource are the results of measurements of the optical density of oils. The data obtained as a result of experiments are processed using a neural network model with Bayesian regularization, which has high smoothness and works well in conditions of small training samples. In this case, emphasis is placed on the ability of the model to contribute to the mapping of the general laws governing the process of thermo-oxidative destruction for more detailed study. As a result, the approach in which the initial data for the model are calculated values of the differential estimates of the partial derivative obtained from the primary neural network model of optical density is more informative from the point of view of describing the qualitative patterns observed in lubricating oil under high temperatures.