Real time fault diagnosis in industrial robotics using discrete and slantlet wavelet transformations

Abstract

Faults in industrial robotic systems can significantly impact operational performance and reliability, particularly in precision-driven environments. This study proposes a real-time, hardware-based fault diagnosis framework that integrates Discrete Wavelet Transform (DWT) and Slantlet Transform (SLT) for multi-joint fault detection in a LabVolt 5150 robotic arm. Acceleration data, captured via an ADXL345 sensor, were processed using DWT and SLT for feature extraction and subsequently classified using a Multilayer Perceptron Artificial Neural Network (MLP-ANN). The proposed method achieved 100% classification accuracy under both constant and variable fault conditions when using DWT, while SLT delivered faster processing times, reducing detection latency from 7.8 s (DWT) to 3.7 s. Notably, this work extends prior research by successfully diagnosing simultaneous faults in multiple robotic joints through real-world hardware experiments. Although emerging graph neural network (GNN) models, such as EGN-OOD, have demonstrated strong performance in mechanical system diagnostics, their application to real-time, multi-joint robotic fault detection remains limited. The results of this study provide valuable insights for selecting suitable algorithms in industrial applications, with future work aimed at integrating graph-based learning frameworks for enhanced adaptability and robustness.