Redundant Dissimilar Sensor Fusion with Dynamic Driver Input Classification and Graceful Degradation for Drive-by-Wire Applications

Abstract

Dissimilar sensor redundancy with force and displacement measurements can provide extended fault coverage for drive-by-wire applications. However, large variances occur when correlating these inputs in dynamic measurements. To address this issue, this paper introduces a novel fault tolerant sensor fusion concept to fuse redundant kinetic and kinematic sensors dynamically and accurately. It consists of a driver input condition classifier, and a graceful degradation scheme voter. The classifier determines the states of several driver input conditions in real-time, which are then utilized as a pointers to allocate the best fitting driver input kinetic-kinematic response profile during dynamic operation for sensor data conversion. The graceful degradation voter module facilitates voting and fusing of the converted redundant sensor data. Functional safety is further maintained through a graceful degradation scheme to tolerate the presence of different fault scenarios in the system. The proposed sensor fusion concept is implemented on a brake-by-wire pedal test bench. Test results show significant performance gains in driver command accuracy for fault-tolerant dissimilar redundant sensor fusion.