Multi-stage Group Testing with (r,s)-regular design Algorithms
Michael Balzer
[stat.ME]
In industrial engineering and manufacturing, quality control is an essential part of the production process of a product. To ensure proper functionality of a manufactured good, rigorous testing has to be performed to identify defective products before shipment to the customer. However, testing products individually in a sequential manner is often tedious, cumbersome and not widely applicable given that time, resources and personnel are limited. Thus, statistical methods have been employed to investigate random samples of products from batches. For instance, group testing has emerged as an alternative to reliably test manufactured goods by evaluating joint test results. Despite the clear advantages, existing group testing methods often struggle with efficiency and practicality in real-world industry settings, where minimizing the average number of tests and overall testing duration is critical. In this paper, novel multistage (r,s)-regular design algorithms in the framework of group testing for the identification of defective products are investigated. Motivated by the application in quality control in manufacturing, unifying expressions for the expected number of tests and expected duration are derived. The results show that the novel group testing algorithms outperform established algorithms for low probabilities of defectiveness and get close to the optimal counting bound while maintaining a low level of complexity. Mathematical proofs are supported by rigorous simulation studies and an evaluation of the performance.