Optimization of Fault Prediction by A.I. in Industrial Equipment: analysis of the operating parameters of a Bench Grinder

Predictive Maintenance (PM) plays a crucial role in maximizing efficiency and reducing costs associated with equipment and system maintenance in industrial companies. Recent advancements in Machine Learning (ML) have revolutionized PM by offering accurate and efficient fault prediction and maintenance planning capabilities. This research focuses on monitoring a bench grinder and observing sensors for temperature, current, angular velocity, and vibration under normal operating conditions. The objective is to predict failures based on specific variables related to the machine.

To develop the system, a prototype bench was designed to subject the machine to several working scenarios, collecting real-time sensor data. Data clusters were generated for each sensor, collecting 3000 samples over 7 consecutive days without faults and another 7 days with modified bench grinder behavior. Sampling was done at a rate of 1 second.
The performance of Decision Trees (DT), Support Vector Machines (SVM), Naive Bayes (NB), and K-Means + Neural Network (NN) algorithms was compared using the confusion matrix metrics. Each algorithm's performance was evaluated for RPM, current, temperature, and vibrations measures. The SVM algorithm showed the highest error for RPM with 43.5%. In contrast, all algorithms achieved minimal or zero errors for vibrations, indicating excellent performance.

These findings demonstrate the potential of ML algorithms in PM for the bench grinder. The results highlight the importance of selecting appropriate algorithms for specific measurements, with vibrations exhibiting the least error across all algorithms and contributes to optimize maintenance activities in industrial settings.