Quaternion Signal Analysis for Detection of Broken Rotor Fault Degrees in Induction Motors

Abstract

Fault detection in induction motors is essential for maintaining the reliability of industrial operations. In practical applications, induction motors experience gradual wear on critical components, such as rotor bars, affecting their performance. This paper introduces a new methodology for modeling predictive wear functions related to rotor faults in induction motors, providing accurate forecasts and optimal performance through Quaternion Signal Analysis in the time domain. Our approach accurately detects wear in broken rotor bars and anticipates their degradation over time. The methodology involves coupling four vibration signals from the motor, representing them as quaternion coefficients, and calculating their rotational attributes to derive a statistical mean. We employ polynomial and Fourier regression techniques to construct a predictive wear function. We assess its accuracy through root mean square error (RMSE) analysis, which improves with increased sample size and regression complexity. Our findings indicate that polynomial regression, particularly at the second degree, achieves superior RMSE results compared to Fourier regression, even within limited sample windows. This approach offers a robust framework for early fault detection and wear prediction in induction motors, supporting enhanced maintenance strategies.