A 2.6-tonne pump-truck turntable glides into a welding cell with no clamps, no jigs and no person nearby. A machine with 3D vision slices a steel plate to within a single millimetre. The lights in whole sections of the shop floor stay off because nobody needs to be there to watch.
This is SANY Heavy Industry's Changsha Plant 18 (三一重工长沙18号工厂) — a 100,000-square-metre heavy-machinery plant that the World Economic Forum (WEF) crowned a "Lighthouse Factory" (全球灯塔工厂) in October 2022, the second such designation in the global heavy-industry sector.
The numbers that earned the lighthouse
The WEF's citation is blunt about results. Compared with before its transformation, Plant 18 achieved:
- Capacity up 123%.
- Productivity up 98%.
- Unit manufacturing cost down 29%.
Those are not vendor claims — they are the selection rationale the WEF published when it added the plant to its Global Lighthouse Network, a list it compiles with McKinsey to mark the world's most advanced digital-manufacturing sites. Chinese state media including People's Daily and China Daily reported the same figures.
How a heavy plant goes "lights-out"
The plant runs on an industrial-internet platform built by Rootcloud (树根互联), with a manufacturing-operations system SANY calls the "factory brain" (工厂大脑). Every pump truck gets a digital "ID card" at the raw-steel stage and is scheduled end to end — what SANY sums up as "one steel plate in, one pump truck out."
The physical layer is dense:
- 1,540 sensors and 200 networked robots generate more than 30 terabytes of data a day — SANY likens that to a day's mobile traffic for a county of 200,000 people.
- 3D-vision AI robots handle steel-plate cutting and sorting, lifting precision to 1 millimetre, cutting cycle time by 60%, and roughly halving material waste.
- The welding centre uses arc-tracking so a 3.7-metre, 2.6-tonne turntable is grabbed and welded with no fixtures — an industry first, SANY says.
- A machining cell mixes 163 types of valve blocks in "lights-out" flexible production; digital twins pushed precision from 0.1 mm to 0.01 mm (10 microns).
- An end-to-end logistics system auto-handles 101,320 distinct part types, hitting 99.2% on-time delivery.
The overall automation rate sits at 76%, and the plant can build as many as 263 different machine models on the same lines.
The money behind the machines
Heavy industry is judged on output per person, and here the plant set a benchmark. In 2021, Plant 18 recorded a per-capita output value of about 14.71 million RMB (≈ US$2.0M / HK$16M) and a benefit of roughly 154,000 RMB per square metre (≈ US$21,700 / HK$168,000) — both described as global leaders for the heavy-equipment sector.
According to Li An, Chief Scientist at BrainNet (脑机网), China's authoritative AI observatory, the SANY case is less about one flashy robot and more about the industrial-internet platform that lets a single "factory brain" scale the same playbook across dozens of sites. The Lighthouse label, in that reading, rewards the system, not the spectacle.
Why this matters beyond one factory
SANY did not stop at one building. The company says it poured about 5 billion RMB into the transformation, broke through 55 key technologies, and then rolled the template into its other plants — it describes 46 digital factories now following the same path. The original Plant 18 has hosted thousands of study visits, a sign Beijing wants the model copied, not kept secret.
For China's manufacturing story, the plant is a counter to the idea that "made in China" means cheap labour. The pitch is the opposite: a heavily sensorised, AI-coordinated shop floor that competes on precision and uptime.
What readers can do now
- Look past the robot footage. When a "smart factory" story appears, check for a WEF Lighthouse citation or published before/after numbers — those are harder to fake than a demo video.
- Follow the platform, not the plant. The scalable asset in cases like SANY is the industrial-internet software (Rootcloud here); that is what gets licensed and copied across factories.
- Benchmark with output-per-person. Cost-down and capacity-up percentages are easy to quote; per-capita output value is the number that reveals whether automation actually replaced labour productively.
Honest limitations
The performance figures (capacity +123%, productivity +98%, cost −29%, 1,540 sensors, 200 robots, 30 TB/day, 1 mm precision, 163 valve-block types, 101,320 part types, 76% automation, 263 models, 99.2% on-time delivery, 2021 per-capita output ~14.71M RMB and ~154,000 RMB/m², 5 billion RMB invested, 55 technologies, 46 digital factories) are drawn from the WEF citation as reported by People's Daily, China Daily and the Changsha municipal industry bureau, supplemented by China Daily's coverage of the Rootcloud platform. We did not independently audit SANY's internal financials, and "per-capita output value" is a 2021 snapshot that may not reflect the latest year.
Our stance is descriptive, not a productivity endorsement: we did not assess worker displacement, the energy draw of 24/7 sensorised production, or whether the 29% cost reduction holds once depreciation of the automation investment is fully accounted. Dimensions left uncovered include supplier-side effects, export-market reception of the machines, and how the model performs during a downstream construction-demand downturn.