Application of optimized convolutional neural network to fixture layout in automotive parts

Abstract Fixture layout is a complex task that significantly impacts manufacturing costs and requires the expertise of well-trained engineers. While most research approaches to automating fixture process use optimization or rule-based frameworks, this paper presents novel approach using supervised learning. The proposed framework replicates 3-2-1 locating principle fixtures for sheet metal designs. This ensures correct fixing an object by restricting its degrees freedom. One main novelty topographic maps generated from design data as input convolutional neural network (CNN). These are created projecting geometry onto plane converting Z coordinate into gray-scale pixel values. also in ability reuse knowledge about fixturing lay out new workpieces integration with CAD environment add-in. results hyperparameter-tuned CNN regression show high accuracy fast convergence, demonstrating usability model industrial applications. was first tested automotive b-pillar designs found have (≈ 100 % ) classifying these offers promising design.