Artificial Intelligence and Machine Learning

Methods of survival analysis in the problem of predicting failure of equipment in industrial enterprises

A. N. Andronov, T. E. Badokina

Ogarev Mordovia State University, Saransk

Abstract: The article explores the application of survival analysis methods for predicting the time until failure of industrial equipment. Classical approaches such as the Kaplan-Meier method and the Cox model, as well as their modifications and machine learning techniques, including Random Survival Forests (RSF), are examined. Using real-world data from a meat processing plant, it is demonstrated that original parts have a lower risk of failure compared to non-original ones. The study also investigates the impact of various factors on the likelihood of industrial equipment failure using survival analysis methods. The Kaplan-Meier and Cox models demonstrated comparable accuracy, while weighted methods proved to be more adaptable to censored data. For quality assessment, metrics such as the Concordance Index, Brier Score, and Time-Dependent AUC were utilized.

Keywords: survival analysis, equipment failure, survival function, hazard function, Kaplan-Meier estimator, Cox proportional hazards model, Random Survival Forest.

UDC: 519.23, 004.85

Received: 16.03.2025
Revised: 25.05.2025

DOI: 10.26456/vtpmk737

Bibliographic databases:

•