Executive SummaryThe inert strength and fatigue performance of a diesel engine exhaust valve made from silicon nitride (Si~Nq) ceramic were assessed. The Si~Na characterized in this study was manufactured by Saint Gobain / Norton Industrial Ceramics and was designated as NT551. The evaluation was performed utilizing a probabilistic life prediction algorithm that combined censored test specimen strength data with a Weibull distribution function and the stress field of the ceramic valve obtained from finite element analysis. The major assumptions of the life prediction algorithm are that the bulk ceramic material is isotropic and homogeneous and that the strength-limiting flaws are uniformly distributed.The results from mechanical testing indicated that NT551 was not a homogeneous ceramic and that its strength was a function of temperature, loading rate, and machining orientation.Fractographic analysis identified four different failure modes; 2 were identified as inhomogeneities that were located throughout the bulk ofNT551 and were due to processing operations. The fractographic analysis concluded that the strength degradation of NT551observed from the temperature and loading rate test parameters was due to a change of state that occurred in its secondary phase.. Pristine and engine-tested valves made from NT551 were loaded to failure and the inert strengths were obtained. Fractographic analysis of the valves identified the same four failure mechanisms as found with the test specimens.The fatigue performance and the inert strength of the Si~NAvalves were assessed from censored and uncensored test specimen strength dat~respectively. The inert strength failure probability predictions were compared to the inert strength of the Si~Nqvalves.. . . The inert strength failure probability predictions were more conservative than the strength of the valves. The lack of correlation between predicted and actual valve strength was due to the . nonuniform distribution of inhomogeneities present in NT551. For the same reasons, the predicted and actual fatigue performance did not correlate well.