Online Learning Based Underwater Robotic Thruster Fault Detection

This paper presents a novel online learning-based fault detection designed for underwater robotic thruster health monitoring. In the algorithm, we build mathematical model between control variable and propeller speed by fitting collected work status data to model. To improve accuracy of modeling, multi-center PSO algorithm with memory ability is utilized optimize modeling parameters. Additionally, update mechanism accommodate change sea environment. During operation, robot predicted through model, residuals are used avoid false alarm, an adaptive strategy established based on mechanism. The proposed method has been extensively evaluated using different robotics, trial simulation, pool test experiment experiment. Compared fixed model-based method, prediction MAE learning at least 37.9% lower than that show no misdiagnosis in experiments, while misdiagnosed. Experimental results competitive terms accuracy, adaptability, robustness.