Efficient symmetric comparison-based self-diagnosis using backpropagation artificial neural networks

One of the main challenging problems in distributed systems is the comparison-based self-diagnosis which aims at identifying the set faulty of nodes (or units) based on the matching and mismatching among the system´s nodes. The comparison-based self-diagnosis approach assigns tasks to the nodes that need to be diagnosed. The outcomes from each pair of units performing the same task are compared, and based on such comparison their fault status is identified. In this work, we consider symmetric t-comparison-based diagnosable systems in which at most t nodes can fail permanently at the same time. We introduce a novel backpropagation neural network-based (BPNN) approach to implement a new fault identification algorithm. The BPNN diagnosis algorithm uses the comparison approach to collect the agreements and disagreements among the nodes, and then uses these comparison outcomes to identify which nodes are faulty and which ones are fault-free. The BPNN-based diagnosis algorithm needs first to undergo an extensive training phase using various faulty situations. Results from a thorough simulation study demonstrate the effectiveness of the BPNN-based self diagnosis algorithm for randomly generated diagnosable systems, making it a viable addition to existing diagnosis algorithms. The novel approach is shown to scale very well for large systems.