MLP neural network with an optimal architecture for modeling the ECAP-C process

The Equal Channel Angular Pressing-Conform (ECAP-C) process is one of the Severe Plastic Deformation (SPD) methods used to create the ultrafine-grained structure. The current study investigates an optimized artificial neural network for modeling the ECAP-C process based on experimental tests and finite element methods. The ECAP-C process of AA 7075 was performed for validation of the finite element model. After validating, the design of experiments was carried out using the response surface method. The process parameters include the rotary wheel radius, the rod contact angle, the die channel angle, outer corner angle of die, the friction coefficient (between rod and roller, and between rod and die), and the aspect ratio of the die channel. Moreover, the responses are the required torque, yield strength, output rod curvature, and strain non-uniformity at the rod cross section. For modeling with the neural network, a Multi-Layer Perceptron (MLP) network is considered, while its structure is optimized using two metaheuristic methods, i.e., Particle Swarm Optimization (PSO) and Whale Optimization Algorithm (WOA). It was found that the MLP network with two hidden layers can efficiently predict the process outputs, with 40 and 7 neurons in its first and second hidden layers, respectively. Furthermore, the comparison of experimental and numerical results has an acceptable agreement.