Manufacturing
Three layers, one team, one contract
A factory intelligence system is three layers of different engineering disciplines: field, basic control, and optimisation. Clients used to engage an instrument supplier, a control contractor and a software house separately — three vocabularies, brittle interfaces, and a lot of pointing at each other when something failed. Those three layers now close inside one team: one contract, one point of contact, and interface problems that are ours to solve.
Safety statement: what we build is a basic process control and production optimisation system. Interlocks, limit protection, emergency shutdown and manual takeover are provided by field hardware and an independently installed safety instrumented system. Failure, downtime, misjudgement or deactivation of our system does not affect those functions — per GB/T 50770-2013 and GB/T 21109.1-2007.
Who owns which layer
| Layer | What it covers | Who owns it |
|---|---|---|
| Layer 3 · Optimisation | Process modelling, setpoint decisions, multi-loop coordination, and the web management and optimisation system | Xiwen Liu and the modelling team |
| Layers 1 and 2 · Field and basic control | Control scheme design, instrument and actuator selection, control system configuration and loop commissioning, and field data acquisition | Our partner and his control engineering team |
| Safety instrumented system | Trip on over-temperature, isolate on leak, stop feed on cooling loss | Outside our scope; implemented separately by a licensed contractor |
The lower layers are owned by someone who has built them
Partner · Industrial control and data acquisition
More than a decade in process industry, having led the build and commissioning of control systems and data acquisition across multiple production lines. He owns the field and basic control layers: control scheme design, instrument and actuator selection, system configuration and loop commissioning, and getting field data up reliably and completely — a good model at layer three is worthless if what feeds it is wrong.
What this means for the client
-
No supplier coordination to runControl, software and AI vendors speak different languages and argue at the interfaces. With one prime contractor those arguments happen inside our team.
-
No hiring for the projectThis kind of role pays off during construction; once the plant is running, demand drops sharply and keeping the headcount is not economic. We assemble what the project needs.
-
One contract, one acceptance standardNo more \u201cevery supplier says they are finished and the system still does not run.\u201d
-
Someone to hold responsibleOne party to call. Prime contractor responsibility — scope, schedule, interfaces, function and warranty — is written into the contract.
This is not a new pairing. The two worked together for two years and jointly delivered an AI expert system for a large pyrometallurgical production line: he owned control and data acquisition, founder Xiwen Liu owned the expert system, agent orchestration and process modelling. The upper and lower layers each have an equal owner, and the interface sits between two people rather than between two companies.
One clarification about "web-based"
Operating the plant and understanding the plant are two different screens. The control-room station stays on the industrial network, supplied with the control system, and carries real-time operation and alarm acknowledgement. What we build carries trends, reports, management dashboards, anomaly alerts, AI setpoint recommendations and mobile access. We advise against replacing the control-room station with a browser. That is an architectural principle, not a statement about capability.
R&D and commercial capability, from group-level delivery
Electron microscopy recognition
Quantifies grain size, density, porosity and fines from micrographs, generates particle-size distributions automatically and accepts manual correction; 91.27% mean accuracy on sphericity.
Virtual laboratory
Process-data modelling that predicts precursor properties from raw-material inputs and analyses the driving factors, cutting trial-batch cost.
Industry intelligence agent
An LLM + RAG + agent core — search, industry radar, voice of customer, internal intelligence — closing the loop from collection to analysis to decision support.
Contact
Start with one conversation about the case
If you have a specific case — one production line, one faculty, one department, one clinical pathway — we'll do a feasibility conversation at no charge and tell you whether it's worth building.