Rework production system is a typical structure in practical manufacturing, where the defective parts detected by the inspection machine are sent to rework rather than being directly scrapped. Due to the special structure, more complex coupling phenomena between machines and buffers are introduced, especially when machines are unreliable and buffers are finite. In addition, under the intelligent manufacturing environment, customized demands trigger small-lot orders become a common occurrence, which lead to traditional steady-state analysis inapplicable. Therefore in this letter, we study dynamic process modeling and real-time performance evaluation problems of the rework system. Practical issues, such as unreliable machines, finite buffers, and small-lot orders are taken into account. Specifically, a novel dynamic decomposition and aggregation algorithm is proposed. The proposed method is validated to have a precise performance measures prediction effect, as well as an efficient computational performance.