Quality Assessment in the Presence of Additional Data in Photovoltaics

Acceptance sampling represents an important tool for quality control. The practical methods of choice for non-normal variables are attribute sampling and variables sampling assuming normality applied to averages instead of single observations. Both methods usually lead to very large sample sizes and are therefore infeasible in practice if observations are expensive. We discuss and extend recent results developed for the photovoltaic industry and actively used there. Here – and presumably in other industries as well – additional data are available which can be used to construct valid and asymptotically optimal sampling plans for non-normal measurements. Consistency and asymptotic optimality of the sampling plans, which are random in our setup, as well as asymptotic normality of the required sample size are established under weak assumptions. We also provide sensitivity studies dealing with the effects of a systematic bias (shift) between the additional data and the lot (shipment), which may matter in practice. The new plans are investigated by simulations to some extent.