6. Conclusions
This paper proposed a nonlinear integer program to determine an optimal plan of zero-defect, single-sampling by attributes for incoming inspections in assembly lines. In the presence of an inspection resource constraint, sample sizes of the parts waiting for incoming inspections must be determined simultaneously. The optimization model can take into account the parts’ heterogeneous quality features (e.g., NC risk, NC severity, lot size, and complexity of inspection) and determine the right sample size for each part. The paper proved the convexity feature of the expected total cost function for the zero-defect acceptance policy and, accordingly, recommended a three-step solution procedure. A real example of twenty parts was presented to illustrate the application of the optimization model.
This paper has built a foundation for practically meaningful extensions. The maximum allowable defective number in this paper is zero. The zero-defect acceptance policy may not be optimal in other circumstances. When it is greater than zero, the maximum allowable defective number becomes an additional decision variable besides the sample size. The sample size and the maximum allowable defective number adversely impact the expected NC cost, and the convexity of the objective function is no longer held on the entire decision space. These require a re-calibration of solution features. The rejection cost is low for incoming inspections and, thus, it was not considered in this paper. However, the rejection cost is an important cost component to be considered, for example, in outgoing inspections. By including the expected rejection cost in the objective function, the sample size decision seeks a trade-off among three cost components, which is more complex than the one presented in this paper. Another important extension of this paper is to model sample size decisions for multiple quality features at a part level and then integrate this lower level decision with the higher level decision analyzed in this paper. The lower level decision provides more accurate inputs to the higher level decision, while the latter poses a constraint for the former. The integration is a value-added, yet challenging, topic of research. All above-mentioned extensions are important subjects of future research.