Delayed reconfigurable manufacturing systems (D-RMS), a specialized subset of reconfigurable manufacturing systems (RMS), have been introduced to address the convertibility challenges inherent in traditional RMS. An exclusive part family formation method of D-RMS based on machine learning is proposed in this chapter. Firstly, a similarity coefficient that considers the characteristics of delayed reconfiguration is constructed. The positions of common operations within their respective sequences are analyzed, with earlier common operations increasing the likelihood of parts being grouped into the same family. To further refine this analysis, the concept of the longest relative position common operation subsequence (LPCS) is introduced, which accounts for the relative positions of common operations. Additionally, the position differences and discontinuities of LPCSs within the corresponding operation sequences are examined. Secondly, K-medoids is adopted to group parts into families based on the similarity among parts. Finally, a case study is conducted to demonstrate the effect of the proposed method.