Temporal Deep Learning-Based Prediction of Safety Failure Trends in Industrial Equipment

Abstract

Industrial equipment usually works in complex operating environments, where mechanical wear, temperature changes, load disturbances, and external environmental factors may jointly accelerate performance deterioration and induce safety risks. Therefore, forecasting the future degradation tendency of equipment is important for enhancing operational reliability and supporting condition-based maintenance. In this study, a time-series deep learning framework is developed for safety-oriented failure trend forecasting of industrial equipment. Multi-source monitoring data are first processed through missing-value treatment, abnormal-value correction, normalization, and sliding-window sampling. Then, a health indicator is established to describe the continuous deterioration process of equipment. On this basis, a TCN-Transformer network is constructed to extract both local abnormal variations and long-range temporal correlations from multivariate monitoring sequences. The forecasted health indicator is further integrated with a warning strategy to divide equipment conditions into four categories: normal operation, early degradation, moderate risk, and severe failure risk. Experimental results demonstrate that the developed model achieves better forecasting accuracy and warning effectiveness than LSTM, GRU, TCN, and Transformer benchmark models. The proposed framework offers a practical data-driven solution for degradation trend recognition, early risk warning, and maintenance decision support in industrial systems.