A Method of Computing Bivariate Normal Probabilities: With an Application to Handling Errors in Testing and Measuring

Many manufactured products are 100 per cent tested with the idea of insuring that each unit of product meets the performance specifications. The general procedure is to set the test specification limits, which are the limits used for product unit acceptance, at or arbitrarily near the performance specification limits. These are established by engineering requirements and require a test-set accuracy and precision of a specified amount with respect to the performance specification limits or some nominal value. Eagle 1 and many others have studied the problem of locating test specification limits with respect to performance specification limits under various conditions of test-set precision when testing itself is subject to random error. He pointed out that, when random errors of testing exist, two types of errors or mistakes can occur which should be taken into account in setting test specification limits. The first error, called consumer's loss (CL), is defined as the probability that nonconforming product units will be accepted. The second error, called pro* Sandia Corporation, Albuquercpie, N. M. 553