Robot fault prediction based on improved GM(1,1) model and RBF

Introduction As the core equipment of intelligent manufacturing, the operational stability of industrial robots directly affects production efficiency and safety. However, long-term operation under complex working conditions can easily result in mechanical wear, electrical failures, and other issues, resulting in an average fault repair time of 4-8 hours. Methods A new hybrid prediction method combining the grey model and radial basis function is designed. The sensitivity problem of the grey model initial value is optimized through initial value correction, and the non-linear fitting ability of the neural network is combined. At the same time, the extreme value method is used to dynamically adjust weights to ensure real-time adaptability. Results The experiment is based on an industrial dataset: improving the grey model to increase accuracy by 40%. The combined model reduces the prediction error threshold to 0.07 meters per second, with a correlation coefficient of 0.95, enhancing accuracy, stability, and robustness, providing a reliable solution for complex engineering environments. Discussion This study provides a reliable solution for predictive maintenance of industrial robots, which can further optimize the predictive performance under ultra-low speed conditions and multi-fault coupling scenarios in the future.