We discuss the optimal allocation based on a general Type-II progressive censored sample under a multi-level stress log-location-scale regression model. We propose to apply a meta-heuristic algorithm based on variable neighborhood search (VNS) approach to determine the optimal censoring schemes. It is found that our procedure with VNS algorithm can not only reduce the searching time significantly, but also give exactly the same optimal allocation and censoring schemes for small sample sizes as those obtained in an exhaustive search, and also reach the optimal or near-optimal objective values for large sample sizes within a reasonable computational time. The efficiency of the optimal allocation of units in a multi-level stress experiment is demonstrated through a real-life data set on the hardened steel specimens in a rolling contact fatigue test.
