A look at the Simulia User Day conference hosted by Dassault Systèmes Korea on June 11th.
An example of the use of AI in manufacturing technology, where there is a perception that the results of AI are difficult to see, attracted attention. Officials from major domestic manufacturers such as Samsung Electronics and LG Electronics appeared directly and shared examples of combining AI and engineering.
Shin Young-sung (신용선), an expert from the Samsung Electronics Industry Cooperation Team, introduced an example of antenna performance prediction technology using machine learning algorithms. He emphasized that AI can also be used in engineering fields such as antenna design.
Singh, who has been involved exclusively in antenna development for more than 20 years, said that even with advances in AI, he didn’t think applying machine learning to antenna design would have much benefit from an engineering perspective. He said there was a reason for that.
First, the performance of an antenna is greatly affected not only by the structure of the antenna itself, but also by the surrounding structure of the vehicle or equipment to which the antenna is attached, as well as by complex changes in the media. There is also no big data, which is essential for training AI models.
Even if simulation data is obtained using a finite element method (FEM) solver, it is time-consuming and costly. Because the correlation between input design variables and output results exhibits high-dimensional nonlinear behavior, it is not easy to model predictive models using mathematical modeling.
Under these circumstances, Mr. Shin changed his mind in March of last year, when he was working on a project to develop an antenna AI model, and began to think that AI could also be used in antenna development.
“I didn’t know any coding languages or how to build datasets,” he said. “With support from Dassault Systèmes, we launched a proof-of-concept (PoC) project to build an analytical automation process in a virtual space and directly demonstrate the feasibility of applying AI.”
According to Shin, the results of the project can be summarized as follows. In terms of work efficiency, the CST analysis process, which previously had to be repeated hundreds of times, has been significantly reduced. Additionally, by selecting only design candidates that are highly likely to meet specifications at the initial design stage, we have shortened development lead time by several dozen times. In terms of technology applications, it has been proven that reliable predictive metamodels can be built even with limited datasets.
Shin said the company plans to expand its reach beyond a single antenna module to composite multiband antennas for LTE and Wi-Fi, and to build a company-wide standard design validation infrastructure that includes geometry variables such as body-chassis structure, glass, and bracket locations. “When engineers control AI pipelines with clear design goals and domain knowledge, they can eliminate iterative tuning and analysis time and use them as strategic engineering tools to accelerate critical design decisions,” he said.
Sang-hyuk Choi (최상혁), a senior researcher at LG Electronics Industrial Technology Research Institute, introduced how to build an SPDM CAE database using an AI agent.
LG Electronics began piloting an SPDM (simulation process and data management) platform based on Dassault Systèmes’ solution at the end of 2019, and completed deployment at its headquarters in April 2022. For the first few years, the problem was that engineers at the working level were reluctant to use the system because the CAE organizations in each business division were dispersed. So LG Electronics took on the task of evolving a report file that remained a simple data lake into a quantitative numerical database.
The direction was not to change the way engineers work. As before, we focused on having users upload only PowerPoint reports and the system automatically extracting the figures and converting them into a database.
A problem we encountered during the project was that typical large language model levels were unable to accurately read table and object data within complex PowerPoint files. To overcome this, we introduced the vision language model Qwen3-VL, which simultaneously analyzes visual information, and built a four-step data extraction process. The constructed database will be used in the field for outlier analysis, design pattern analysis, trend analysis, etc.
Choi cited the main accomplishment as building a system that converts unstructured report data into structured assets without changing the way working-level engineers work. “We have overcome the limitations of existing methods that required direct input of numbers and maximized database accessibility and convenience with an AI agent pipeline that guarantees 100% autonomy in the field,” he said, adding, “We have completed the foundation for a data pipeline that can provide high-quality engineering nutrients to AI models.”
