Modular design and optimization of complex equipment via integrated machine learning and MCDM: A case study

Complex equipment modular design must integrate uncertain expert judgments and multiple component relationships. This study proposes a framework combining a five-dimensional design structure matrix, spectral clustering improved with dung beetle optimization, and game-theoretic multi-criteria ranking. The matrix captures physical, functional, structural, morphological, and sustainability relationships in a gangue-sorting machine. Relative to standard spectral clustering, the method increased intra-module cohesion by 5.44% and reduced inter-module coupling by 18.95%, while giving the best overall balance across the tested alternatives. Scheme rankings were stable across the tested grey relational coefficients. The framework supports transparent module partitioning and scheme selection, but requires validation on larger equipment and additional mechanical products.