The 7th session of 2024 ICIM lecture series focusing on machine learning methods for design-for-manufacturing and manufacturing process
On August 18th, 2024, the International Coalition of Intelligent Manufacturing (ICIM) successfully held the 7th session of the "Win-win Cooperation: Intelligent Manufacturing in a Global Vision" online lecture series. Professor Emeritus David W. Rosen from Georgia Institute of Technology is specially invited to share "Machine learning methods for design-for-manufacturing and manufacturing process". His wonderful lecture has gained wide attention and attracted over 2,000 views.
Figure 1: Professor Emeritus David W. Rosen from Georgia Institute of Technology
With the rapid development of manufacturing industry, the complexity of design requirements and the diversity of manufacturing processes have become increasingly prominent. In order to cope with these challenges and achieve efficiency and accuracy of manufacturing process, machine learning has been leveraged as a core engine to promote transformation and upgrading of manufacturing industry. In this lecture, Professor Rosen explained the innovative application of 3D CNN (Convolutional Neural Network) in additive manufacturing, and analyzed in detail how to accurately classify and optimize the manufacturing process through CNN. In addition, he explored the fusion of conditional Generative Adversarial Networks (cGAN) and topology optimization, and showed how this combination can be used to achieve generative design that both meets functional requirements and efficient optimization. Professor also explained how to use CNN and cGAN to realize Design for Manufacturing, and the combination of generative design and DFM, which drew a promising blueprint for future manufacturing.
Professor Rosen pointed out that the combination of CNN with topology optimization for simulation optimization of additive manufacturing can make simulation results closer to the actual production, reduce errors, so as to greatly shorten the time and improve the overall efficiency.
In modern manufacturing, in order to improve production efficiency and product quality, it is particularly important to select manufacturing processes based on the shape characteristics of parts. Professor Rosen said that accurately capturing the 3D information of parts by point cloud is highly adaptable and can also retain the geometric details, which provide rich data sources for subsequent analysis and processing.
Figure 2: Shape descriptor using a CNN method
He pointed out that traditional topology optimization often requires repeated verification, which will consume a lot of computing resources and time, while generative design combining topology optimization and cGAN can generate a wide variety of designs that look good enough and can meet specific engineering and functional requirements at the same time.
Figure 3: Comparison of traditional topology optimization method and new method combining cGAN
In the product life cycle from design to manufacturing, it is crucial to satisfy both the functionality and manufacturability. Professor Rosen said that by combining DFF with DFM, both the functionality and manufacturability can be realized through iterations by DFF and DFM modules.
At the end of the Q&A session, he answered the questions around the limitations of AI in intelligent manufacturing, solving complex engineering problems using CNN, and materials selection. His unique insights and rich practical experience won wide praise from the audience.
The lecture was hosted by Dr. Pei Huang, Deputy Secretary-General of ICIM and CEO of e-works. It is worth mentioning that Professor Yong Chen from the University of Southern California, also participated in the discussion.
The live broadcast concluded in a warm atmosphere, receiving highly praise from the audience. As the 7th lecture of ICIM's " Win-win Cooperation: Intelligent Manufacturing in a Global Vision" series, it has broadened the audience's vision and deepened the understanding of applying machine learning in manufacturing design and process selection. It has also provided new directions for future development of manufacturing industry.
Figure 1: Professor Emeritus David W. Rosen from Georgia Institute of Technology
With the rapid development of manufacturing industry, the complexity of design requirements and the diversity of manufacturing processes have become increasingly prominent. In order to cope with these challenges and achieve efficiency and accuracy of manufacturing process, machine learning has been leveraged as a core engine to promote transformation and upgrading of manufacturing industry. In this lecture, Professor Rosen explained the innovative application of 3D CNN (Convolutional Neural Network) in additive manufacturing, and analyzed in detail how to accurately classify and optimize the manufacturing process through CNN. In addition, he explored the fusion of conditional Generative Adversarial Networks (cGAN) and topology optimization, and showed how this combination can be used to achieve generative design that both meets functional requirements and efficient optimization. Professor also explained how to use CNN and cGAN to realize Design for Manufacturing, and the combination of generative design and DFM, which drew a promising blueprint for future manufacturing.
Professor Rosen pointed out that the combination of CNN with topology optimization for simulation optimization of additive manufacturing can make simulation results closer to the actual production, reduce errors, so as to greatly shorten the time and improve the overall efficiency.
In modern manufacturing, in order to improve production efficiency and product quality, it is particularly important to select manufacturing processes based on the shape characteristics of parts. Professor Rosen said that accurately capturing the 3D information of parts by point cloud is highly adaptable and can also retain the geometric details, which provide rich data sources for subsequent analysis and processing.
Figure 2: Shape descriptor using a CNN method
He pointed out that traditional topology optimization often requires repeated verification, which will consume a lot of computing resources and time, while generative design combining topology optimization and cGAN can generate a wide variety of designs that look good enough and can meet specific engineering and functional requirements at the same time.
Figure 3: Comparison of traditional topology optimization method and new method combining cGAN
In the product life cycle from design to manufacturing, it is crucial to satisfy both the functionality and manufacturability. Professor Rosen said that by combining DFF with DFM, both the functionality and manufacturability can be realized through iterations by DFF and DFM modules.
At the end of the Q&A session, he answered the questions around the limitations of AI in intelligent manufacturing, solving complex engineering problems using CNN, and materials selection. His unique insights and rich practical experience won wide praise from the audience.
The lecture was hosted by Dr. Pei Huang, Deputy Secretary-General of ICIM and CEO of e-works. It is worth mentioning that Professor Yong Chen from the University of Southern California, also participated in the discussion.
The live broadcast concluded in a warm atmosphere, receiving highly praise from the audience. As the 7th lecture of ICIM's " Win-win Cooperation: Intelligent Manufacturing in a Global Vision" series, it has broadened the audience's vision and deepened the understanding of applying machine learning in manufacturing design and process selection. It has also provided new directions for future development of manufacturing industry.
