Sensor Compensation in Motor Drives using Kernel Regression

Sensors are essential in feedback control systems, because the performance is dependent on the measurements. Fault in sensors may lead to intolerable degradation of performance and even to instability. Therefore, the high performance expected with vector control may not be achieved with fault in sensors. Several approaches related to fault tolerant motor control have already been proposed. However, most of them consider the sensors fault-free and work about faults in motors and actuators. Furthermore, the purpose of this work is not only sensor fault tolerance but also sensor fault compensation. In a standard fault tolerant approach, the fault would be detected and the sensor would be isolated. The faulted sensor may have an off-set or scaling error and could still be used if its error is compensated. In this paper, this is done with a mathematical solution based on kernel regression that can compensate the measurement error generating more accurate and reliable estimates. This technique is described and applied in motor drives. Simulated and experimental results are presented and discussed.