Using machine learning for service candidate sets retrieval in service composition of cloud-based manufacturing

Cloud manufacturing (CMfg) is a service-oriented manufacturing paradigm that is striving to produce highly customized products via sharing resources of multiple manufacturing providers. Distributed nature of this paradigm calls for addressing the service composition problem in order to achieve an optimal status in regard to such a collaboration. However, in the majority of research studies done on service composition in CMfg, the sets of candidate services on which the optimal composition is supposed to be conducted are assumed to be predefined. This is an extreme simplification and does not satisfy the actual requirements of cloud manufacturing. This study is aiming to propose a novel approach that first uses TF-IDF (term frequency-inverse document frequency) vectors extracted from the manufacturing capability data and machine learning algorithms to retrieve candidate sets for each corresponding subtask. Then, the optimal composite service is obtained for each scenario by using metaheuristic algorithms. The results prove the efficacy of this method in resulting in a more comprehensive approach for tackling the service composition problem in cloud manufacturing paradigm.