Smart factory competition rules are being rewritten. Past assessments focused on equipment digital collection and cloud upload; excellence and flagship tiers now weigh AI scenario ratio, industrial agent count, and whether digital twins or embodied intelligence enter real lines. For SME manufacturers, this triggers two mistakes: assuming they will never qualify and buying only standalone machines, or stacking robots and large models from brochures while the shop still dispatches verbally.
A steadier view: ratings are a weather vane, not a work order. SME plants should harden each layer along collection → execution → quality → planning so agents have stations to stand on later.
Connectivity proves devices online, not production controlled
Many shops already see uptime, output, and alarms. Yet work orders stay on paper, material batches mismatch, rework never enters the system, night-shift handoffs rely on WeChat voice. This "digital equipment, analog management" structure makes any smart schedule idle: algorithms plan what foremen cannot execute, and execution never writes back.
Ask four questions first: Can each work order tie to equipment, people, and material batches? Can key process actual hours write back? Can defects split incoming, process, and equipment causes? Do rush orders and outsourcing enter one plan? These four "yes" answers matter more than another automation line for reaching "thinking."
Three MES modules are enough for phase one
Discrete manufacturers drown in MES feature lists. For most SME plants, phase one needs only three modules: work order execution, material error-proofing, quality records. Scheduling can start with finite daily capacity—not advanced APS day one. Vision inspection can start at the highest-complaint or highest-rework process, validate false positive and miss rates for three months—not plant-wide rollout.
- Work order execution: dispatch, reporting, mandatory downtime reason codes.
- Material error-proofing: issue, load, batch binding—reduce wrong material.
- Quality records: inspection items, defect codes, disposition statistics.
Industrial agents need stations, not just chat windows
Large models as shop Q&A look advanced but do not help scheduling. Valuable agents embed in existing actions: incoming defects auto-create supplier quality tickets; repeated equipment alarms suggest maintenance and open work orders; delivery risk flags capacity conflicts at order entry. They need tools (query inventory, open tickets, send messages), permissions, and human review for high-risk actions.
For manufacturing, the XYN digital intelligence system emphasizes running MES, QC, and warehouse execution first, then hanging reusable rules and assistants on processes. Plants need not become flagship samples overnight—but each year should name one process decision that moved from "the veteran remembers" to "system has records, rules iterate."
Cognitive autonomy is the direction. Plants with unstable execution layers only amplify chaos. Weave work orders, batches, and defects into three nets first—then smart manufacturing earns the next step.