Investigating the Impact of Signal Resolution on Machine Learning Based Multi-Class Fault Detection

Signal resolution is crucial for diverse applications, notably impacting accuracy, signal-to-noise ratio, and system performance. Today, microcontrollers (MCUs) designed for industrial purposes provide analog-to-digital converter (ADC) resolutions ranging from 10 to 16 bits. Therefore, evaluating the ADC resolution of machine learning algorithms is crucial to enhance the system's performance and cost-effectiveness. This study explores the impact of signal resolutions of 10, 11, 12, and 16-bit sensor data, emphasizing their significance in machine learning classification tasks specifically aimed at motor fault detection. By conducting controlled experiments, this study utilizes a comprehensive sensor dataset obtained from two industrial motors to assess the effects on signal quality and information retrieval. Since AC drives account for over 50% of global electricity consumption and find widespread use in various industrial applications, the findings shed light on the trade-offs associated with different resolution levels in real-world applications, particularly industrial settings. The research contributes to signal processing understanding, aiding the selection of resolutions for specific contexts, and facilitating informed decisions in motor-driven systems and machine learning classification.