DATA DRIVEN PREDICTIVE MAINTENANCE IN PETROLEUM AND POWER SYSTEMS USING RANDOM FOREST REGRESSION MODEL FOR RELIABILITY ENGINEERING FRAMEWORK

Abstract

This study addresses the persistent problem that predictive maintenance in petroleum and power systems is frequently implemented with fragmented data architectures, limited analytics capability, and weak linkage to reliability engineering outcomes—factors that collectively constrain its influence on equipment availability, maintainability and risk reduction. The purpose of the research is to develop and empirically evaluate a data-driven predictive maintenance framework that embeds a Random Forest regression model within a structured reliability engineering context. A quantitative, cross-sectional, case-based design is adopted, combining survey responses from 210 professionals in petroleum and power enterprises (70 percent response rate from 300 questionnaires) with 12,480 equipment time-series records from critical assets such as transformers, gas turbines and centrifugal compressors. Key variables include data quality, system integration, staff competency, organizational support, perceived predictive maintenance effectiveness, adoption intensity, and a quantitative degradation or risk score derived from operational parameters. The analysis plan integrates descriptive statistics, reliability and validity testing, Pearson correlations, multiple regression, and Random Forest regression compared directly with linear regression. The measurement model demonstrates strong internal consistency (Cronbach’s alpha 0.86 to 0.91), and the TOE-based predictors explain 62 percent of the variance in perceived effectiveness (adjusted R² = 0.62), with organizational support (β = 0.31, p < .001) and data quality (β = 0.28, p < .001) emerging as dominant drivers of successful adoption. The Random Forest model achieves R² = 0.87 and RMSE = 0.94 on a 0–10 risk scale, outperforming linear regression (R² = 0.71, RMSE = 1.45) and highlighting load factor, temperature, vibration intensity and oil quality as the most influential degradation predictors. These findings imply that petroleum and power enterprises can materially strengthen reliability-centered and risk-based maintenance decisions by jointly investing in data governance, system integration, analytics skills development, and ensemble machine learning approaches. The study provides both methodological and practical contributions by demonstrating how predictive models, when aligned with reliability engineering principles and organizational readiness, can deliver measurable improvements in asset health forecasting and operational risk management.