Machine learning for predictive maintenance

Downtime due to sudden machine failure will cause much loss to the company. To overcome this, companies need to develop suitable maintenance strategy. Nowadays, machines in smart manufacturing provide large volume data to monitor machine conditions. Big data analytics becomes needed for processing large data, especially for predicting machine failure in Industry 4.0. Predictive maintenance works better than corrective or preventive maintenance. It can continuously monitor diagnostic and prognostic processes to predict future failures and the equipment's remaining useful life (RUL). Using real multi-sensor data and machine failure reports in industrial equipment, machine learning models can study data patterns and build failure prediction models based on real-time condition monitoring. The purpose of this study is to construct a diagnostic and prognostic model with tuning optimal machine learning parameters in support vector machine and random forest (RF) for classification of equipment conditions and RUL so that company can find out the prediction of future failure times. Also, the comparison of machine learning parameters and methods is carried out to determine which model has the highest accuracy. Based on each model's accuracy, RF performed better than SVM in diagnostic and prognostic models.