Abstract
New European regulations for reduction of car pollution set out the installation on-board of the EOBD (European On-Board Diagnostics) system, aiming to control the correct operation of engine components and functions having a direct effect on exhaust emissions.
During car operation, the EOBD system processes several diagnostic indexes expressing the efficiency level of engine components and functions monitored by the system. If a diagnostic index exceeds its fixed threshold, the EOBD system detects a fault.
The diagnostic indexes exhibit a random variability, which has to be studied by means of statistical methodolo-gies. This paper deals with the formulation of a statistical methodology that allows evaluating the optimum thresh-old for a diagnostic index, in terms of convenient balancing of false alarm and failure to detect risks. This method-ology has been applied to one of the lambda sensor diagnostic indexes, for a new car model prototype developed at Elasis research centre.
Fusing statistical and technological knowledge, an experimental plan has been designed and executed. For each experimental setting two runs are performed: an on road test and a standard cycle test (NEDC). The experi-mental results allowed formulating a probabilistic model, describing the evolution of the polluting emissions and the diagnostic index as function of the lambda sensor deterioration level. Subsequently, by means of a three-step procedure, it has been possible to attain the required evaluation of the threshold to be assigned to the diagnostic index.