Chuncheon biotech complex tests AI-driven 'virtual factory' to cut trial and error for small firms

"Our goal is to standardize biotech manufacturing data and build a standard AI model. If startups use this model, they can reduce trial and error and enter the field much faster."

Lee Beom-ho, head of the Korea Industrial Complex Corp.'s Chuncheon branch, laid out the AI transformation, or AX, ambitions for Hupyeong Industrial Complex during a visit Monday. The plan is to accumulate data from biotech manufacturing processes that workers have long managed through experience alone, have AI analyze it, and then roll out a validated model to small and medium-sized enterprises and startups.

At the Chuncheon Biotech Industry Promotion Institute inside the complex, an AX pilot program was under way, connecting data generated across biotech manufacturing processes — fermentation, extraction, concentration and drying — to AI systems. The core aim goes beyond simple automation: AI learns from accumulated data to predict quality anomalies and support workers' decision-making.

"Where conventional smart factories focused on collecting and monitoring equipment data through IoT sensors, the WISE Factory connects production, quality and equipment data to AI," said Lee Tae-seop, head of the production support division at the Chuncheon Biotech Industry Promotion Institute. "We are building a system in which AI analyzes and interprets data to support decision-making on the production floor."

The project is part of the "AX Demonstration Complex Construction Project" led by the Ministry of Trade, Industry and Energy and the Korea Industrial Complex Corp. A total of 19.98 billion won ($14.4 million) will be invested through 2028 to build a 1,713-square-meter WISE Factory at the Chuncheon Biotech Industry Promotion Institute. The facility will house an AX flagship factory, a manufacturing AI open lab and a virtual factory demonstration platform, serving as a hub for spreading the Manufacturing AI Transformation, or M.AX, initiative.

Fermentation is one of the clearest examples. When cultivating microorganisms, variables such as temperature, pH and pressure all affect product quality. Workers previously checked and recorded process conditions by hand; going forward, sensors will collect data in real time and AI will analyze it alongside historical production and quality records.

"AI will analyze real-time process data together with past quality data to identify conditions that affect quality and flag potential anomalies," Lee Tae-seop said. Currently, products are inspected after production is complete to determine whether they meet standards. Once AX is applied, AI will analyze factors affecting quality during the production process itself and predict potential problems in advance. Vision AI will also be introduced in tableting and capsule processes to detect products that deviate from normal shape or form. Final process decisions and accountability, however, remain with human workers; AI supports their judgment rather than replacing it.

RMS Platform, the AI supplier for the project, is building a system that recreates about 30 types of equipment used in biotech manufacturing processes in a virtual space, allowing companies to test production conditions before committing to a real run. "Temperature, humidity and pressure were things workers recorded by hand and judged by feel — now that data will be captured from equipment and learned by AI," said Lee Yun-ho, chief executive of RMS Platform. "The biggest difference from conventional automation is that AI uses data as the basis for supporting the decisions needed to run a process."

The approach is particularly valuable for small and medium-sized enterprises and startups that cannot afford their own production facilities. The Chuncheon Biotech Industry Promotion Institute already has equipment for extraction, concentration, fermentation, drying and tableting, and about 100 companies use its machines roughly 2,000 times a year, according to Lee Tae-seop. Because companies bring their own raw materials and use shared equipment, and because biotech manufacturing processes share similar equipment, sensor placements and data types such as temperature and pressure readings, standardizing process data and building AI models is comparatively straightforward.

Digital twin technology will allow companies to adjust variables such as temperature and pressure in a virtual factory and predict outcomes before any real equipment is switched on. "Once the integrated AX platform is built, companies will be able to run simulations from their own sites without visiting in person and predict what results different condition values will produce," Lee Tae-seop said.

Lee Yun-ho projected that a standardized AX model, once established, could cut by more than 30 percent the costs that biotech startups incur in preparation, operations and trial and error.

The biotech sector is inherently more cautious about adopting new equipment or processes than other manufacturing industries. Product safety and quality are directly at stake, and changing production equipment or processes can trigger costly re-validation requirements under Good Manufacturing Practice, or GMP, standards. High upfront costs and a shortage of specialist staff add to the burden, putting AX adoption out of reach for many small and medium-sized enterprises.

"Companies recognize the need for AI, but most still lack a concrete strategy or clear understanding of which parts of their processes to apply it to," said Jo Mi-jeong, head of the AI innovation division at the Gangwon Information and Culture Industry Promotion Agency. "Drawing out the knowledge and know-how that skilled floor workers have kept in handwritten records and in their heads — turning that into data — is also part of what this project is here to do."

The Korea Industrial Complex Corp. plans to demonstrate AX's effectiveness at the public pilot factory first, lowering the barrier for private companies to follow. "If the public sector proves it first — if data actually accumulates and the benefits of adoption are clearly shown — companies will start saying they want to try it too," Lee Beom-ho said. "The goal is to build a standard AI model, reduce trial and error for startups, and spread the approach to other biotech companies and industrial complexes."

Produced in partnership with the Korea Industrial Complex Corp.