Unidentified and undetected faults in heavy earthmoving machinery (HEMM) are linked with high failure frequency and downtime across the industries. Draglines are capital-intensive HEMM deployed in large surface coal mines for stripping overburden, and drag system failures contribute to substantial downtimes; thus impacting its availability, reliability and productivity. Application of an effective fault detection and analysis method can reduce its failure frequency as well as downtime. Therefore, the present paper demonstrates a data-driven approach for fault analysis of drag system using inference-based Bayesian network (BN) through generated sensor data, and logbook records of 28 months. The test dataset is used for building a 16-node three-layer BN. Historical fault records, experts' opinion, and characteristics of faults helped defining the threshold limits for fault type identification. Thereafter identified faults, based on their dependency of occurrence on cause(s)-symptom(s) in the causal model, are categorized as catastrophic fault, degraded fault, or intermittent fault. Finally, the fault analysis results are validated through three-axiom based sensitivity analysis, and their prediction accuracy revealed the capability of the model for successful identification of faults. The parameters sensitive to faults are expected to act as a guiding tool for condition-based maintenance of dragline to improve its reliability and availability.
展开▼
