The construction of covering arrays with a small number of rows is a difficult optimization problem. Covering arrays generated by greedy methods are often far from optimal, while many metaheuristics and search techniques become inefficient once larger instances are concerned. In this work, we propose to incorporate improvement heuristics directly into the constructing process of widely used in-parameter-order (IPO) algorithms for covering array generation. We discuss how this approach can significantly reduce the search space of the heuristics and implement some of the discussed concepts in the SIPO algorithm, which enhances greedy FIPOG algorithms with Simulated Annealing. Using SIPO, we improved the best known upper bound on the number of rows of binary CAs of strength $6$ for $43$ different instances.

https://link.springer.com/chapter/10.1007/978-3-030-79987-8_40