In the quest for operational excellence within the energy and manufacturing sectors, predictive maintenance techniques have become increasingly sophisticated. This paper introduces the SVM-TESA model, a novel predictive maintenance framework developed through the analysis of a uniquely compiled dataset from an electrical power plant and a mechanical base company, comprising 2147 samples. The SVM-TESA model innovatively combines the classification power of Support Vector Machines with Time-Enhanced Semantic Analysis to mine insights from complex maintenance logs. It leverages temporal data and natural language processing to forecast equipment failures with an accuracy rate of 90.1%. Our research delineates the model’s construction, highlighting the integration of time series analysis, text mining enhancements, and ensemble learning methods to optimize predictive accuracy. The efficacy of SVM-TESA is evidenced by its performance on the dataset, which showcases its potential to transform maintenance protocols by significantly reducing downtime and maintenance costs. The implications of this model extend to a broad range of industrial applications, promising a new benchmark in predictive maintenance strategies.