Statistical issues in the development of an automotive on-board diagnostics

Abstract

The automotive on-board diagnostics (OBD) is a complex system whose aim is to monitor the state-of-health of another complex system, i.e. the vehicle engine. OBD aimed at the containment of polluting exhaust emissions is nowadays compulsory on every vehicle model introduced into the market. Increasingly stringent regulations and ever more demanding customers push from opposite sides towards a perfect working of OBD systems. Therefore manufacturers are globally interested in the high robustness and reliability of such critical systems. The aim of this presentation is to give an insight into the main statistical problems arisen during an extensive research project carried out in collaboration with an Italian manufacturer. In particular three aspects will be evidenced: a) the modeling and the analysis of variation aimed at understanding the causes of variation of diagnostic indexes and prioritizing improvement actions; b) the robust calibration of OBD systems by opportunely combining physical and computer experiments; c) the treatment of truly unequally spaced time series for the statistical process control in the burn-in phase. Some generalizations and directions for future research will be illustrated.