Individual Confidence Intervals in JMP(R): A Unified Cost-Saving Approach to Control Limits and Capability Evaluation in Validation and Batch Release ( 2019-EU-45MP-087 )
Feb 11, 2019 4:33 PM
| Last Modified: Mar 13, 2019 10:46 AM
Level: Intermediate Job Function: Analyst / Scientist / Engineer Per Vase, Managing Partner, NNE Torben Bygvraa Rasmussen, Principal Consultant, NNE
Health authorities demand statistical assurance that batches after validation have sufficient quality. The Mixed Models platform in JMP can, from validation data, predict future batch performance using individual confidence intervals by setting batch as a random factor. If an individual confidence interval with alpha = acceptable quality level (AQL) falls inside the specification interval, there is less than AQL risk that a given future observation will be outside specification. In addition, individual confidence limits can be used as control limits after validation in continued process verification. Mixed models can be built for mean, moving range and range within and thereby control limits can be established for all charts in the JMP three-way chart. By using Box-Cox transformation, asymmetric limits will be obtained for the range within. Using mixed models, systematic variation in subgroups from cavities or positions, for example, can also be handled. By setting batch a systematic factor in batch release, pooling of within-batch variation is obtained, leading to reduced need for sample size. Real-world examples will be shown where a classical control limit and capability index approach will fail, but where a unified individual confidence limit approach succeeds, both concerning control limits and capability evaluation.