3D modelling of a frame assembly using deep learning and the Chu–Liu–Edmonds Algorithm. Issue 5 (15th July 2019)

• Record Type: Journal Article
• Authors: Cao, Hao; Mo, Rong; Wan, Neng
• Is Part Of: Assembly automation. Volume 39:Issue 5(2019)
• Page Start: 872
• Page End: 880
• Publication Date: 2019-07-15
• Subjects: Machine learning -- Frame assembly -- Intelligent design -- Minimum arborescence; Automation -- Periodicals; Automatic machinery -- Periodicals; Assembly-line methods -- Periodicals; Industrial engineering -- Periodicals; 670.42705
• DOI: 10.1108/AA-08-2018-0121
• Languages: English
• ISSNs: 0144-5154

Abstract

Abstract : Purpose: The proposed method is to generate the 3 D model of frame assemblies based on their topological model automatedly. It was a very demanding task and there was no appropriate automated method to facilitate this work. Design/methodology/approach: The proposed method includes two stages. The first stage is decisive. In this stage, a deep learning network and the Chu–Liu–Edmonds algorithm are used to recognize contact relations among parts. Based on this recognition, the authors perform a geometrical computation in the second stage to finalize the 3 D model. Findings: The authors verify the feasibility of the proposed method using a case study and find that the classification rate of the deep learning network for part contact relations is higher than 75 per cent. Furthermore, more accurate results could be achieved with modification by the Chu–Liu–Edmonds algorithm. The proposed method has lower computational complexity compared with traditional heuristic methods, and its results are more consistent with existing designs. Research limitations/implications: The paper introduces machine learning method into assembly modelling issue. The proposed method divides the assembly modelling into two steps and solves the assemble relation creatively. Practical implications: Frame assemblies are fundamental to many areas. The proposed method could automate frame assembly modelling in a viable way. It could benefit design and manufacture process significantly. Originality/value: The proposed method expands the application of machine learning into a new field. It would be more useful than simple machine learning in industry. The proposed method is better than general heuristic algorithms. It outputs identical results when the inputs are the same. Meanwhile, the algorithmic complexity in worst situation is better than general heuristic algorithms.