Data Driven Prediction of Motor Status for Optimizing the Control Using Machine Learning

Abstract

The AC electric motor has been invented for over 200 years ago, and during this long period of development, it has become one of the most fundamental components of modern industry. Nowadays, it is used almost everywhere. From small pump components to large cruise ship steering systems, industrial AC motors are especially indispensable in a wide range of applications. Consequently, the less energy motors consume, the less environmental impact, which directly contributes to sustainable industrial practices. Particularly there are strict regulations that promote energy efficiency in Europe, which demand continuous improvement in motor design and operation. In addition to improving motor efficiency, another optimized energy consumption strategy is reducing AC motor idle time. That leads to wasteful power usage.

This thesis aims to find a method for identifying idle time and other operational states of industrial AC motors. The first step is that a tri-axial accelerometer sensor needs to be installed to collect analog-to-digital converted (ADC) vibration data from the motor, ensuring that the dynamic mechanical behavior of the system is captured in sufficient detail. The second step is data preprocessing and feature extraction that need to be performed for machine learning. The third step is that three different machine learning models are developed and trained using labeled data to learn the characteristics between motor states. The final step is that, based on the best prediction result, the selected machine learning model is used to determine the operational state of the motor. Model performance is evaluated through parameters such as accuracy, precision, recall, and F1 score, which provide a comprehensive assessment of classification quality.

The result shows that the One Class SVM model with the best thresholds achieved the highest accuracy of 0.98 and F1 score of 0.97 in identifying motor states. This confirms the feasibility of vibration machine learning for reliable motor state identification in practical industrial scenarios. The method enables automatic detection of idle periods and it can be integrated into real-time monitoring to support energy-saving strategies and to reduce unnecessary power consumption. Beyond energy efficiency, such approaches also contribute to predictive maintenance and sustainability in smart manufacturing environments.