Robots help make other robots by performing defined assembly and handling tasks inside a larger production system. ABB's Shanghai factory is a useful case study because its dated announcements describe the relationship between flexible work cells, mobile material delivery and digital production information.
The factory opened on December 2, 2022. This article follows the documented project history and explains the engineering ideas a reader can take from it. ABB's releases are company-reported evidence; the worked production example below is explicitly invented.
Separate the plan from the opening
| Date | Documented event | How to read it |
|---|---|---|
| October 27, 2018 | ABB announced a $150 million factory investment and an expected end-of-2020 start | A project plan and forecast |
| September 12, 2019 | ABB announced construction had begun and referred to opening during 2021 | A construction milestone with a revised forecast |
| December 2, 2022 | ABB reported the official opening in Kangqiao, Shanghai | An opening milestone, with company descriptions of the production system |
The 2018 announcement proposed interconnected automation islands and a digital model of the factory. The 2019 construction release described automated material delivery and a move between production cells. The 2022 opening report gave the production and research facility's area as 67,000 m² and the investment as $150 million.
Those dates explain why an old article's future tense needed replacement. They establish the reported sequence; they do not, by themselves, explain every schedule change. Keep an announced opening target separate from the date an organization actually reports opening.
Follow a unit of work through the system
ABB's opening report describes digitally connected modular cells served by autonomous mobile robots, with robotic screwdriving, assembly and material handling. The useful engineering idea is coordination: a station needs the correct parts, tools, instructions and acceptance result before the next step can proceed.
To understand a robot factory, imagine following one identified assembly through four questions. The sequence below is a conceptual reading framework, not ABB's disclosed work instruction.
- What arrives? Record which parts, kit and configuration belong to the assembly. A correctly placed part from the wrong variant still creates the wrong product.
- What operation happens? Identify the fixture, tool and required result. “Assembly” becomes understandable when it is broken into locating, fastening, connecting or checking a specific component.
- What evidence releases it? Ask which result permits the assembly to move on and how a failed or incomplete operation is identified.
- What moves next? Distinguish the product, its carrier, replenishment parts and reusable tooling. Each movement needs a destination that is ready to receive it.
Flexible cells can support different product routes or variants when the process is designed for them. That flexibility creates scheduling and coordination work: a station can wait for a missing kit even when its robot is healthy. A mobile platform can arrive on time but deliver to a blocked receiving position. Local automation and factory flow must be assessed together.
For a production reader, a useful exercise is to sketch the equivalent flow in an existing workshop. Mark where identity, location or readiness is currently communicated by a person. Those handoffs are candidates for clearer information and measurement before any equipment purchase.
Understand what the digital layer contributes
The 2018 plan called for a digital twin to help staff monitor and assess operations. A useful factory model connects its representations to actual configuration and observations. A layout model can test spatial arrangements; a timing model can explore queues; an operating-data view can show equipment state. Ask what the model contains, how it is updated and which decision it supports.
A digital representation becomes unreliable when its assumptions drift from the floor. A new fixture may change the reachable path, a software revision may alter timing, and product mix may change where queues form. Model ownership and change records therefore matter as much as the display.
ABB's 2024 industrial-AI article identifies AI and 2D vision for screw-hole localization on the Shanghai OmniCore controller production line. That is a concrete example of perception serving an assembly task. The article also makes performance claims; evaluating transfer to another line would require its test population, conditions and failure definitions.
Frame AI around an observable job: locate a feature, classify an inspection image or flag an unusual signal. Then identify the consequence of an incorrect output and the verification that follows. The presence of AI in one operation supplies no evidence that every factory decision is automated.
People remain responsible for defining the product, planning and improving processes, resolving exceptions, maintaining equipment and accepting changes. The reports describe automation of tasks, while a claim about total staffing or labor savings would require separate data. This case study makes no present-day staffing estimate.
Use a small model to see why coordination matters
Invented assembly-cell example
Suppose every unit needs 12 minutes of assembly and 6 minutes of testing. One assembly cell can supply 60 ÷ 12 = 5 units per hour; one test cell can handle 10. The ideal steady-state flow is limited to 5 units per hour by assembly.
With two independent assembly cells working in parallel, their combined nominal capacity becomes 10 units per hour. One test cell could match that flow if routing, loading and other conditions permit. Each unit still needs 18 minutes of processing across the two operations; parallelism changes output rate, not the sum of those processing times.
Add queues, travel, changeovers, downtime and rework to estimate actual lead time and good output. These invented numbers are unrelated to ABB's measured production.
The example suggests a disciplined question for a factory tour: which resource constrains the whole flow at the present product mix? A fast demonstration of one arm answers a narrower question. Ask for observed complete cycles and operating losses before inferring system throughput.
The manufacturing robot planning guide develops this distinction into a project brief. The autonomous navigation guide explains the positioning and route-following work behind mobile delivery.
Turn a factory story into an evidence record
Read each important claim with its date and scope. An announcement establishes what was proposed. An opening report establishes a reported milestone. A detailed operating study can establish performance under stated conditions. Give forecasts, company claims, measured results and your own inferences separate labels.
For “robots make robots,” write down one specific operation, one input, one output and one quality check you would want to see. Then ask how a failed result is handled and how that assembly retains its identity. This turns a slogan into a system you can explain.
Use the editable factory case-study record to map the evidence and work through the capacity example. It can also guide a student discussion or a factory visit without presenting a promotional claim as an independently measured result.