We will investigate the multi-step-stress test on repairable systems with the basic life distribution being a non-homogeneous Poisson process. Two different types of repairable systems are considered. The first, simple non-homogeneous Poisson model (SNHPM), where each experimental unit can be fixed to good-as-new for testing again after failure. The second type is the cumulative exposure non-homogeneous Poisson model (CENHPM). Each experimental unit is fixed to test again after failure but failure rate is affected by previous exposure. Inferential procedures and optimum designs are discussed for multi-step accelerated life test plans. Maximum likelihood estimators of unknown parameters are derived. We obtain Fisher’s information matrix and use that to find optimum time to change the stress. A real world example of accelerated life testing is presented, and numerical estimates based on our results are obtained and discussed.