Flexural fracture strength, fracture locations, and Monte Carlo predictions for a silicon nitride by ten US laboratories

The work reported was conducted to provide a basis for a number of structur al ceramic mechanical property standardization activities in the United Sta tes, Germany, Japan, and Sweden, A comparison of key property values of a c ommercial silicon nitride determined in a number of laboratories was a majo r objective, The work reported was conducted by 10 U.S. laboratories on GN- 10 silicon nitride, and represented the U.S. work within an International E nergy Agency program including the United States, Germany, Japan, and Swede n, Fracture location analyses showed that fracture location within the inne r span often was not a linear function of location within the span. Some of this behavior was explained by random sampling effects based upon simulati on predictions, but some was apparently dependent upon friction within the fixtures in spite of efforts to minimize it. Flexural strengths were measur ed at 25 degrees and 1250 degrees C in air and were analyzed using the two- parameter Weibull model in terms of in and a, using both linear regression (LR) and maximum likelihood (ML) methods. Under the measurement conditions for the 10 room-temperature strength sets, the value of the ML estimator fo r in varied by as much as 36%, while the value for the sigma(Theta) paramet er estimator varied only 3.3%, The LR estimator for tn varied by about 54%. For the high-temperature specimens, the ML estimator for in varied by 48% while the LR estimator varied by 38%, Ranked fracture location analysis sho wed that the high-temperature fracture locations were more random than thos e in the room-temperature specimens, and was probably due to friction in th e high-temperature fixtures. There was little pin rolling ability in many o f the high-temperature fixtures used. Monte Carlo and one-way analysis-of-v ariance (ANOVA) methods provided insight into the consistency of the streng th values. Monte Carlo predictions showed that for room-temperature strengt h, the maximum likelihood estimator in for all 10 Laboratories fit within t he 10% and 90% confidence bounds for 30 specimen sets. The dispersion of th e high-temperature data was such that the nl estimator satisfied the model only at the 1% acid 99% confidence levels for the 15 specimen sets. ANOVA r esults showed that for the room-temperature flexural strength, data from al l 10 laboratories were not distinguishable for this evaluator at the 95% co nfidence level and that scatter within individual data sets was a larger ef fect than was the variation between the data sets. For the high-temperature data, the results from one laboratory were clearly outside the allowable r ange at this confidence level.