Optimum multistage deep drawing process design using artificial neural network-based forming quality evaluation function

Effectively implementing the multistage deep drawing process is challenging due to the accumulation of material, geometric, and contact nonlinearities, as well as the complexity introduced by discrete forming stages and intermediate failures. Optimizing this process requires the design and integration of discrete forming stages to establish an optimized manufacturing plan. This paper presents a multistage deep drawing recommendation system that combines a novel forming quality evaluation function, called the Integrated Multistage Deep Drawing Evaluation Function (IMSDDEF), with the Non-dominated Sorting Genetic Algorithm III (NSGA-III). The system aims to suggest near-optimal forming plans with a minimal number of forming stages, while accounting for blank material properties, target-net shape, product quality, and the intermediate failures. Artificial Neural Network surrogate models for classification and regression have been developed, trained on finite element simulation data with plastic anisotropy to predict product formability, minimum height, and thickness based on design variables at each forming stage. Flowchart-based decision-making logic in IMSDDEF integrates these surrogate models to determine key performance metrics, including the appropriate number of forming stages, as well as the product's shape loss and minimum thickness. IMSDDEF coordinates with NSGA-III to derive Pareto-optimal solutions within a minute, enabling trade-offs between conflicting objectives. From the resulting Pareto front, the most suitable forming plan is selected through detailed comparative analysis. Experimental validation is conducted to confirm the reliability and effectiveness of the proposed system. The novel function IMSDDEF could be further generalized to other sequential material processing optimization scenarios to optimize manufacturing plans beyond multistage deep drawing.