Development of a technical condition assessment algorithm for complex systems based on probabilistic failure estimation

• Published in: Information Technology and Computer Engineering Vol. 22; no. 2; pp. 9 - 19
• Main Authors: Vychuzhanin, Vladimir, Vychuzhanin, Alexey
• Format: Journal Article
• Language: English
• Published: Vinnytsia National Technical University 05.09.2025
• Subjects: adaptive technical solutions; cognitive simulation; expert systems; integration of diagnostic methods; intelligent monitoring; predictive maintenance; probabilistic modelling
• ISSN: 1999-9941; 2078-6387; 2078-6387
• DOI: 10.31649/vitce/2.2025.09

This study proposed an integrated algorithm for assessing the technical condition of ship power plants, combining case-based reasoning (CBR), Bayesian networks, Markov processes, and cognitive simulation modelling. The algorithm was designed to enhance the accuracy and adaptability of diagnostics under conditions of uncertainty, limited data, and dynamic operational environments. The diagnostic process followed a multi-stage architecture that included the retrieval of historical failure cases, probabilistic correction based on interdependencies among components, modelling of component degradation over time, and adaptive scenario analysis. Each component of the algorithm plays a distinct role: CBR provides analogies to previously observed failures; Bayesian networks quantify probabilistic links between interrelated faults; Markov chains model the temporal degradation of equipment and estimate transition probabilities between operational states; and cognitive modelling allows the generation and testing of rare or cascading failure scenarios under variable conditions. The integration of these elements ensures that the algorithm dynamically updates failure probabilities and adapts to changing operational data. Simulation results demonstrated several improvements: the average prediction error for remaining useful life of components was reduced from 9% to 5.7%; the accuracy in identifying rare and cascading failures increased by 18% due to the use of cognitive modelling; and Bayesian correction reduced false positive diagnoses by 7.2% compared to baseline CBR systems. Overall, the predicted failure probability for 25,000 hours of operation was reduced from 83% (Bayesian-only model) to 68% with full model integration. The practical significance of the proposed algorithm lies in its ability to improve predictive maintenance planning, reduce equipment downtime, and increase the operational reliability of complex marine engineering systems. The modular architecture also enables the adaptation of the algorithm to various types of industrial technical systems