A Digital Transformation Framework for Real-Time Vibration Monitoring Using Virtual Sensors and Machine Learning

This study introduces a real-time vibration monitoring and fault diagnosis framework for industrial machinery, focusing on the Y2-132M-4 three-phase asynchronous motor. The system overcomes the limitations of traditional manual inspections by employing virtual sensors that simulate three-axis vibration signals (AccX, AccY, AccZ) using open-source data, eliminating the need for physical hardware during development. A machine learning-based classification model, specifically Random Forest, is integrated to detect mechanical and electrical faults with high accuracy. To enhance system performance, the TRIZ methodology is applied to resolve engineering trade-offs, such as processing delays and data loss. Key features such as mean, standard deviation, peak-to-peak values, and frequency components are extracted from the signal for model training. The resulting Random Forest model achieved a classification accuracy of 92%, outperforming other models like K-Nearest Neighbors, Logistic Regression, and Decision Tree classifiers. The system is deployed on a web-based platform that supports real-time monitoring through dynamic visualization and a sliding window mechanism for continuous alerting. This work demonstrates the feasibility of using virtual sensors combined with machine learning to support predictive maintenance strategies. The framework also lays the foundation for future work involving physical sensor integration, deep learning models, and broader industrial deployment.