Fault tolerance requirements in postal automation: A case study

This paper presents the fault tolerance requirements of the image processing module of an automatic mail piece sorting system. This embedded system requires high computing performance that can only be delivered by parallel processing modules. However, the region-of-interest (ROI) analysis application is prone to data-induced software errors. After describing the ROI analysis application, its fault and failure model are investigated. Considering the particularities of the application, the requirements for fault tolerance for this type of embedded parallel applications are identified. Detection concentrates on the data-dependent processing modules; isolation confines the faulty areas locally and the recovery investigates the re-integration in the application. The ROI analysis application allows that no output is associated with some mail pieces, such that the recovering easily gets in pace with the rest of the mail piece sorter. From these requirements, a fault tolerance framework has been derived to improve the dependability of embedded applications without hard real-time constraints. This is done at low cost by flexible and yet standardised techniques.