Machine Learning–Based Estimation of Risk Priority Number for Dumper Failure Analysis in Opencast Metal Mines

Heavy Earth-Moving Machines (HEMM), particularly dumpers, play a critical role in opencast metal mining, where harsh operating conditions frequently lead to equipment failures affecting safety, productivity, and maintenance costs. Traditional Failure Mode and Effects Analysis (FMEA) prioritize failures using the Risk Priority Number (RPN), calculated from Severity (S), Occurrence (O), and Detection (D) scores. However, this approach relies heavily on expert judgment and static evaluation, limiting its ability to incorporate historical operational data. This study proposes a data-driven framework integrating Machine Learning (ML) techniques with FMEA to improve failure risk prioritization for mining dumpers. The analysis is based on a dataset of 208 breakdown events collected over 2.5 years from 20 Komatsu HD785 dumpers, covering 15 major failure categories. Although limited in size, the dataset provides meaningful insights into failure behaviour. Severity (S) was assessed based on safety, operational, and economic impact indicators, while Occurrence (O) and Detection (D) were derived from historical failure frequency and monitoring effectiveness. Key input features included equipment age, failure type, subsystem category, maintenance history, and downtime characteristics. Gradient Boosting Regression models were developed to estimate S, O, and D scores, which were subsequently used to compute RPN values. The models demonstrated strong predictive performance, with R² values of 0.963 for Severity, 0.979 for Occurrence, and 0.981 for Detection. However, these results should be interpreted cautiously due to potential overfitting arising from the limited dataset. Engine-related failures were identified as the highest-risk category (RPN ≈ 240), while brake and suspension systems exhibited moderate risk levels.The proposed ML-assisted FMEA framework enables data-supported risk prioritization and supports maintenance decision-making in opencast mining. Further validation u

View article page

Download PDF