Feature Evaluation based on Correlation Coefficient Filter Method for Electrical Motor Condition Monitoring

In the realm of electrical motor condition monitoring, identifying optimal features is crucial for enhancing predictive maintenance strategies. This paper focuses on analyzing features for low speed electrical motor condition monitoring, aiming to determine the optimal number of features that can be monitored with fewer sensors without significantly compromising the accuracy and precision of the predictive models. This study utilizes conventional statistical feature extraction methods, initially considering a set of 30 features derived from three sensors. Feature selection is performed using Pearson and Spearman correlation coefficient filter methods to identify the most relevant features. Subsequently, a systematic reduction of feature from thirty to three is carried out. The filtered features are then evaluated using three different machine learning models. The resultant accuracy and precision of these model are compared to assess the impact of feature reduction on model performance. The findings demonstrate that it is possible to maintain high accuracy and precision with a reduced number of features, thereby optimizing sensor usage and improving the efficiency of condition monitoring systems for electrical motors.