Integrated Approach to the Diagnostics of Technical Systems Using Machine Learning and Fuzzy Logic Methods

Timely diagnostics of deviations in the operation of technical systems is crucial to prevent emergencies and minimize the time to eliminate them. This task requires the development of effective methods for predicting the state of technical objects. The performance of prognostic models faces two key challenges: the need to optimize the hyperparameters of the models and a pronounced imbalance in the source data when information on contingencies is insufficient. In this paper a new integrated approach is proposed to solve this problem using machine learning methods and fuzzy logic. The approach includes automated selection of the most informative parameters of the object’s operation, application of hybrid sampling algorithms to eliminate class imbalance, optimization of hyperparameters of machine learning models, and use of fuzzy logic to assess the state of the entire system. The approach is implemented as a software complex in Python, which provides automation of the process of building models and visualization of analysis results. The verification of the software was carried out on hydraulic system data. It was found that the application of used approach makes it possible to increase the accuracy of predicting the state of the system components (the F-score metric has increased by 54-67% and the AUC-ROC - by 22-28%) compared to models without balancing and selecting hyperparameters based on the grid search method. For the diagnostics of the entire system, the Mamdani fuzzy model was constructed. The results obtained confirm the effectiveness of the proposed approach for monitoring complex technical systems.