By Kai-yuan Teng, Chen Kang Kang
From CommonWealth Magazine (vol. 807)
Source：Pei-Yin Hsieh
From Shenzhen, to Taiwan, and Mexico, future factories built by AI robots have already become a reality in hundreds of processes at Foxconn, closing the gap to fully autonomous facilities. How was this accomplished?
Nvidia chief Jensen Huang’s whirlwind visit to Taiwan this past June set imaginations racing regarding AI manufacturing. In the future, “the factories will orchestrate robots. And those robots will be building products that are robotic. Robots interacting with robots, building products that are robotic,” asserted Huang.
And that future is now.
Worldwide manufacturing giant Hon Hai Precision (known as Foxconn Technology Group internationally), valued at US$91 billion, opened the conglomerate’s first “lighthouse factory” in 2023.
The “lighthouse” factory designation is jointly conferred by the World Economic Forum (WEF) and the McKinsey consulting company. In the Industry 4.0 revolution, these are deemed model factories best equipped to utilize such technologies as big data, AI, and the industrial internet of things (IIoT) to guide the manufacturing industry’s transformation and enhancement.
Hon Hai's Longhua factory in Shenzhen, which started construction in 1996, is so large that it looks like a town. It became HonHai's first lighthouse factory in 2019. (Source: Hon Hai)
At the Longhua factory, the starting point of Foxconn founder Terry Gou’s expansion into China in 1996, the facility once spanned an area equivalent to nearly 250 football fields and employed nearly 400,000 workers. Now, the workforce has been reduced to 250,000, yet the production value remains unaffected.
Foxconn is renowned for its superior molding technology. At work among the molds at the Longhua factory, engineers display a system on screens that tracks all data related to the molding process. Employees can quickly locate and modify files according to customer needs and swiftly create molds.
Precise molds are the foundation of Hon Hai's fortune. (Photo: Pei-Yin Hsieh)
Foxconn boasts the largest fleet of machine tools worldwide, with nearly 70,000 units performing thousands of operations such as metal cutting, grinding, and polishing to produce finished products.
In the past, machine tooling relied on skilled craftsmen to listen to the sound of the equipment, observe the state of the materials and factory temperature, and adjust the parameters and operate the machinery based on experience. However, people are bound to make mistakes, and the complex process of adjusting machines could take hours.
Foxconn long ago sought to automate its operations. More than a decade ago, Terry Gou brought in Jay Lee to serve as vice chairman to spearhead the group’s smart manufacturing efforts.
Foxconn then set about integrating Industry 4.0 into its production lines. Although one executive later noted that this could at best be described as “automation.”
The challenge lies in the rapid iteration of products in contract manufacturing. When a customer changes the design, such as switching from a square to a round metal component, the production line - developed at significant cost - must be completely reconfigured.
In Longhua, they still preserve old but easy-to-use mold intelligent management system. (Photo: Pei-Yin Hsieh)
“We pretty much used all the fancy technology available at the time,” relates Chen-Ting Wu, director of Foxconn’s smart manufacturing division. “Because it was advanced but impractical, like an angel, we called it the ‘angel factory,’” he adds.
Chen-Ting Wu, director of Foxconn’s smart manufacturing division. (Photo: Pei-Yin Hsieh)
However, from big data and the IIoT, to cloud computing and AI, "the past five years have seen exponential advancements in smart manufacturing," said Wu, 42, who holds a Ph.D. in industrial engineering and algorithms from National Tsing Hua University. Over the past seven years, Wu has witnessed the evolution of AI manufacturing from its infancy.
In 2019, Foxconn combined big data, sensors, IoT, and automation tool arrays to introduce new equipment in its factories so as to keep pace with clients’ rapid product iterations.
Chen-Ting Wu notes that two main technologies were critical in getting the “angel factories” down to earth.
First, industrial IoT communication protocols have increasingly aligned with internet protocols, lowering the barriers to entry. The second key factor is breakthrough advancements in neural network training for AI models in recent years.
“'Angel Factory' technology has now come down to earth,” Wu described.
The Shenzhen Guanlan factory has taken the implementation the farthest.
The entire factory operates without the need for lighting. Sensors on the smart machines and logistics transport vehicles display green, red, and yellow signals, indicating whether the machines are operating normally, waiting for materials, or experiencing malfunctions.
Taking the day’s work orders, the cloud-based AI factory’s brain directs autonomous robotic arms (AMR) to move around throughout the facility.
Approaching a "vending machine" holding 21 tools, the AMR selects the correct tool with the aid of AI-directed vision and positioning technology.
Robotic arm finds the right cutting tools. (Photo: Pei-Yin Hsieh)
Next, the automated transport robot heads to a specific machine tool, as instructed by AI, and using its robotic arm for visual positioning, installs the tool on the machine. The entire process of framing products is completed automatically.
Underpinning this manufacturing production line that has clients so excited is the perfect combination of cloud-based training and edge AI responses.
Eight years ago, Foxconn launched a large-scale transformation plan, eliminating precision machines that were incapable of thinking.
As AI smart factory model training requires fast computational power, it is handled by the cloud. Foxconn has uploaded all relevant engineering management documents, production systems, daily factory records, and on-site maintenance manuals to the cloud to facilitate the training of AI models for different manufacturing processes.
Chen-Ting Wu demonstrates how the factory’s AI-powered Situation Room works, where six screens display temperature fluctuations, production capacity utilization, and the quantity of completed products. Managers are able to monitor machine health at all times, and rely on AI to assign tasks.
In the past, factory production scheduling was a major source of stress for plant managers. However, these days AI systems in factories can automatically handle scheduling, accounting for various contributing factors. Tasks are assigned to each machine through computer commands, making it clear how much capacity each workstation can handle, thereby reducing the pressure on plant managers.
Chen-Ting Wu even described a scenario seemingly fitting for a sci-fi film, saying “Foxconn has over 60,000 CNC machines, which I can control from my office.” However, this kind of remote control has not been achieved in practice at this time, so plant managers continue to exercise on-site control.
This system can assist all the human machine managers throughout the plant. Equipped with augmented reality (AR) glasses, when issues with the machines arise, they can conduct repairs as instructed by the glasses.
With a dashboard, the headquarter can control everything in the factory. (Photo: Pei-Yin Hsieh)
And the results are astonishing.
In its evaluation of the Guanlan plant, the World Economic Forum wrote: “in order to respond to the rapid release of new communications products and the demand for strict quality standards, the scaled deployment of 37 Industry 4.0 cases has accelerated new product introductions by 29 percent, accelerated mass production by 50 percent, reduced defect rates by 56 percent, and saved 30 percent on manufacturing costs.”
At Guanlan, Foxconn operates Asia's first AI-powered anodizing plant.
Anodizing is a critical process that gives metals used in cars, phone cases, and other products their shiny surface, allowing them to be dyed in such attractive colors as green, pink, and blue. This process plays a crucial role in determining a product's appearance.
In the past, operators had to hang 50-kilogram plates like laundry on racks before robotic arms would dip them into dyeing vats.
Due to the high-risk nature of this manufacturing, not only are workers at risk of getting injured, but a scratched metal plate can spoil the project, resulting in major losses. Consequently, anodization is the most labor-intensive area of the plant.
Now, however, robots handle the task.
Employees at the Longhua factory are operating the machine according to the instructions of AR glasses. (Photo: Pei-Yin Hsieh)
Robotic arms load the metal plates, which then go through chemical polishing, oxidation, sealing, and dyeing. These four steps involve being lifted by overhead cranes and dipped into more than 100 liquid tanks.
Previously, a skilled technician would manually control the process, observing and relying on experience to determine whether the metal components should be immersed for 100 or 130 seconds, depending on their condition that day.
Today, Foxconn collects data such as the concentration, conductivity, and temperature of 100 liquid tanks, and uses AI models to analyze how these data arrays affect the dyeing outcome. “The AI model can automatically recommend the proper dyeing time,” shared a supervisor at the site.
The entire anodizing process can be completed at the touch of a button.
The number of employees at Hon Hai has declined from its peak, yet smart manufacturing keeps up the manufacturing volume. (Photo: Pei-Yin Hsieh)
The final stage - quality control - has also been taken over by AI models.
In the dark AI visual inspection room, metal components rotate rapidly inside a chamber, continuously illuminated by flashes of light. At an average interval of one image every 0.6 seconds, a total of 154 pictures are taken for optical inspection to determine the presence of any defects.
Using big data to train AI models to rapidly adjust procedures has allowed Foxconn to become the enterprise group with the most facilities certified as lighthouse factories by the World Economic Forum.
From Shenzhen to Chengdu, Zhengzhou, Wuhan, Taiwan, the United States, and Mexico, Foxconn’s AI-driven smart manufacturing is being rapidly replicated, greatly reducing the pressure of building new factories, labor shortages, and management challenges.
On the flip side, then, why is Foxconn recruiting workers for its Zhengzhou factory?
The differences between the Zhengzhou facility and the Longhua and Guanlan “angel” factories that CommonWealth visited are stark. On the one hand are lights-off factories reliant on AI, big data and robots, while Zhengzhou’s assembly line stands on the other.
The difference is that the production lines that CommonWealth visited handle front-end manufacturing for various Foxconn products, while
Zhengzhou is an assembly line.
Robotics experts acknowledge that the complex electronics product assembly is the most difficult area to automate at present. The breakthrough could come from the humanoid robots that Jensen Huang’s Nvidia and Tesla are rushing to invest in.
In January, Goldman Sachs released the report “Humanoid Robot: The AI Accelerant,” boldly predicting that humanoid robots could appear in factories within the next few years.
The final chapter of Foxconn’s angel factories could go like this: the customer places an order, and the AI factory automatically produces parts. Next, humanoid robots, equipped with neuron-filled hands as nimble as human fingers, would then assemble the products and deliver them to consumers.
Hon Hai Foxconn's factory in Longhua, Shenzhen. The Future Factory has come to the present. (Photo: Pei-Yin Hsieh)
Foxconn is already thinking about the transformation of its Zhengzhou plant, announcing an additional investment of NT$4.5 billion in late July for a new division headquarters and seven centers focusing on strategic sectors such as electric vehicles, energy storage batteries, digital health, and robots.
Once the employer of over one million workers, just 900,000 remain at Foxconn. And how many will remain in the future is difficult to predict. What is certain is that working alongside robots is not the future, but the present.
