Utilizing machine learning for predictive maintenance of production machinery in small and medium enterprises

Predictive maintenance involves the early detection of potential machine failures and subsequent maintenance to prevent such failures. Machine learning is a pertinent statistical method for predictive maintenance, enabling the early detection of machine failures and the implementation of preventive measures through a model. The development of the machine learning model commences with data collection from the machine, encompassing vibration, acceleration, machine temperature, and machine sound, facilitated by a microcontroller equipped with sensors. Subsequently, the data undergoes cleaning, including removing outliers or missing values and standardization. Data is partitioned into 70% allocated for training and 30% for testing. After determining hyperparameters and their values through hyperparameter tuning, the training data is utilized to train machine learning models, such as K-nearest neighbor, decision tree, and random forest models. Post-training, the models are evaluated using the remaining test data, employing performance metrics such as accuracy, precision, recall, and F1-score. The random forest model excels due to its utilization of a substantial number of trees for predictions and the full exploitation of the variables which F1-score is 91.22%. The best-performing model is subsequently deployed into a monitoring system, providing real-time machine condition predictions. The deployment results validate the accurate prediction of machine failures.