Product configuration design is instrumental in combining available configurable components to rapidly generate new products that align with customised requirements, aiding manufacturing enterprises in enhancing production efficiency and reducing development costs. Nevertheless, traditional methods remain inadequate for effectively leveraging historical configuration knowledge and rapidly generating new configuration design schemes devoid of established references. Inspired by the combined effects of the associative learning of the cerebellum and the reward modulation learning of the basal ganglia on goal-oriented behaviours, this paper proposes a cognitive computing method for product configuration design based on associative rule mining and deep reinforcement learning. For one thing, the benchmark learning-based association rule mining method, which simulates the cerebellum's function, automatically extracts configuration design knowledge that connects customer requirements with module instances, thereby promoting the end-to-end reusing of historical configuration design solutions. For another, the improved deep Q-network is proposed to mimic the function of the basal ganglia, which is employed to generate new configuration design schemes in the absence of historical precedents. A case study on the configuration design of the essential service water system of a nuclear power plant is implemented to demonstrate the reliability and reasonability of the proposed approach.
Stored only when both papers are in the corpus — 12 citations exist upstream.
Imported from OpenAlex (CC0) — nothing here is generated.