Based on regression statistical fitting of performance degradation response data, a reliability assessment model of variable speed complex degradation was established for multi-stage degraded products. First, piecewise regression fitting was used to model performance degradation data, and the maximum likelihood function of change point distribution was derived. Secondly, due to the complexity of the model, the mayimum likelihood estimation of change point estimation cannot be displayed. Therefore, layered Bayes model is used for modeling, and Gibbs sampling technique in Markov chain Monte Carlo algorithm is used for parameter diagnosis of the model. The change point of the model is estimated and verified by using Schwarz information criterion criterion and empirical likelihood ratio. Finally, according to the definition of product failure degradation, the reliability function of the product under the failure mode is derived. Based on Mate30 game performance data modeling and analysis, the multi-stage degradation model is proved to be highly adaptable and feasible. Compared with traditional two-stage modeling, multi-stage degradation modeling reduces unnecessary data waste and improves reliability of product reliability evaluation.