GPS Signal Fault Diagnosis for Unmanned Rollers Based on Total Disturbance Observation and Support Vector Machine

The roller is a typical articulated multi-body vehicle with multi-degree of freedom in motion. Accurate and reliable position and heading angle measurements are important foundations for the accurate path-following of unmanned rollers. Due to the poor operation environment of the roller, the positioning signal often drifts or jumps, which affects the reliable operation of the system. To achieve reliable fault diagnostic in the positioning system, in this paper, a novel solution that combines total disturbance observation and support vector machine (SVM) classification, is proposed. A multi-body kinematic model is established with steering wheel angle and vehicle speed as inputs, and with the longitude, latitude and heading angle as outputs. The discrepancy of model estimates from the measured value is treated as total disturbance, to be estimated by the extended state observer. Then the estimated total disturbance, together with the measured position and heading angle are input into the support vector machine for faults classification. Experimental results show that the fault diagnosis accuracy is 95%, the improvement in accuracy and computational time is 9% and 12% respectively, compared with the conventional solution that only based on SVM.