Process Capability Evaluation Research Via Continuous Ranked Probability Score

Process capability analysis is an important content of quality management.When the evaluation of the process capability is needed in the enterprise,the normality based process capability indices are commonly used.But in practice,not all the process distributions are normal.In this case,using traditional methods to evaluate the process capability may lead to mistake.To evaluate the capability of non-normal process distribution,scholars have proposed many methods.The two commonly used process capability indices for non-normal distributions are percentile based indices and yield based indices.Percentile based indices use 0.135,50 and 99.865 percentiles of the process distribution to reflect the process fluctuation degree,and then give an evaluation result;Yield based indices use proportion of non-conforming to reflect the process capability.While when the process distribution is non-normally distributed,these partial features may not reflect the uniformity of the process.To solve this problem,CRPS(continuous ranked probability score)is applied to process capability analysis and CRPS based process capability evaluation indices are proposed.An example and simulations show that CRPS based indices are more stable and accurate indicators than the existing indices in reflecting the degree of process fluctuation.In the evaluation of the process capability,in order to ensure the validity of the data,it is necessary to evaluate the capability of the measurement system.The traditional measurement system capability evaluation methods are R&R% and P/T%.When the measurement system error and the process fluctuation are normal,this method can effectively evaluate the measurement system capability.However,in non-normal cases,this method may lead to mistake.For non-normal measurement system capability analysis,CRPS is also used and CRPS based measurement system capability evaluation indices are also proposed.Actual case and simulation are used to illustrate that CRPS based indices can evaluate the measurement system capability when the measurement system error and process are non-normally distributed.