<p>The design of closed-die forging processes requires determining process parameters such as the number of forming stages and die-surface geometry at each stage while satisfying evaluation indices such as forging load and shape accuracy. When single-stage forging is not feasible, many combinations of the number of stages and intermediate die-surface geometries must be explored, which are referred to as the “process layout” in this paper. This leads to a time-consuming and iterative trial-and-error process. In a previous study, an automatic design system for process layouts was proposed for disk-shaped products. The functional surface connection method was also introduced to ensure high flexibility in creating die-surface geometry. Subsequently, this method was integrated with the finite element method (FEM) and an optimization technique to establish a framework for process design. However, a major limitation was the high computational cost due to the large number of FEM simulations. In this study, a novel automatic design system was developed, the acceleration of which is achieved with machine learning (ML) models trained in advance instead of evaluations using FEM simulations. ML models were trained using datasets from process layouts generated for a variety of target shapes. By integrating the pre-trained ML models with an optimization algorithm, appropriate process layouts that satisfy the target requirements were generated. Results indicate that the ML-accelerated automatic design system can significantly reduce the computational cost of FEM simulations after the target shape is obtained. This reduction enables more efficient generation of process layouts.</p>