Forecasting Imminent Failures in Electrical Industrial Centrifuge using Machine Learning

Predictive maintenance detects early signs of functional loss in electrical equipment and is a widely used strategy to minimize industrial costs. Machine learning models, the predictive maintenance state-of-the-art, forecast early failures in a myriad of industrial equipment but their application in electrical industrial centrifuge, an equipment whose failure can be critical, was rarely investigated. We hypothesize that machine vibration, temperature readings, lubrication levels, and motor current values help to forecast mechanical failures in an electrical industrial centrifuge. We validate our hypothesis by collecting 23,000 monitoring samples of an electrical industrial centrifuge and its respective history of 757 breakdowns over a 3-yr period, training a Random Forest machine learning model with the first two years, and testing in the third year. Our results show that the trained model can forecast failures with a F1-score of. 34, establishing a baseline for forecasting imminent failures in industrial centrifuge. Our proposed approach aids technician inspections, reduces factory downtime, and potentially saves tens of thousands of dollars every year.