Accelerated Artificial Neural Networks on FPGA for fault detection in automotive systems

Modern vehicles are complex distributed systems with critical real-time electronic controls that have progressively replaced their mechanical/hydraulic counterparts, for performance and cost benefits. The harsh and varying vehicular environment can induce multiple errors in the computational/ communication path, with temporary or permanent effects, thus demanding the use of fault-tolerant schemes. Constraints in location, weight, and cost prevent the use of physical redundancy for critical systems in many cases, such as within an internal combustion engine. Alternatively, algorithmic techniques like artificial neural networks (ANNs) can be used to detect errors and apply corrective measures in computation. Though adaptability of ANNs presents advantages for fault-detection and fault-tolerance measures for critical sensors, implementation on automotive grade processors may not serve required hard deadlines and accuracy simultaneously. In this work, we present an ANN-based fault-tolerance system based on hybrid FPGAs and evaluate it using a diesel engine case study. We show that the hybrid platform outperforms an optimised software implementation on an automotive grade ARM Cortex M4 processor in terms of latency and power consumption, also providing better consolidation.