Smart Factory Architecture Done Right: Headroom for Edge AI, NAS, HA, Firewall
Piecemeal factory IT is the single biggest reason Edge AI, MES, and digital-twin rollouts become expensive. QAAD designs a layered architecture with 5–7 years of headroom, led by engineers who truly understand FDI/EPE manufacturing operations.
A common reality in Vietnamese factories: IT infrastructure is built piecemeal — buy a switch when ports run out, add a server when an app lags, plug a PC in when a new machine arrives. Three to five years later, when the business wants to roll out Edge AI, MES, or a digital twin, the flat L2 network cannot be segmented, core bandwidth is saturated, the racks are full, and swapping a single switch takes a whole day of downtime. Retrofit costs grow exponentially.
Designing it right on day one doesn't cost much more — but in five years it saves hundreds of thousands of dollars and hundreds of production hours.
A layered architecture for the smart factory
A multi-line plant is a compact data centre. You need five clear layers — from the shop floor to the cloud:
The four pillars you cannot skip
- NAS + DR + Snapshot: Production data (machine logs, camera feeds, AI models) must not be lost. Large-capacity NAS + hourly snapshots + off-site DR.
- HA (High Availability): Dual firewalls (active/passive), dual core switches (MLAG/VSS). A single point of failure halts the whole plant — unacceptable.
- Segmented Firewall: Separate OT network from IT office; VLANs for cameras, Edge AI, guest Wi-Fi. A single ransomware-infected PC doesn't take the plant down.
- Endpoint Security (EDR): Trend Micro Apex One or equivalent on every endpoint, feeding a SIEM. Attacks are caught early.
Edge AI — why you must leave headroom
When a line needs AI Inspection or Sound Inspection, the model must run right at the edge — never through the cloud:
- Latency: < 100 ms to reject a defective unit at the right position. A cloud round-trip is seconds.
- Privacy: Line camera/audio is trade-secret data — it must not leave the plant.
- Reliability: Losing Internet cannot stop production.
- Cost: Millions of inferences/day × cloud API = not viable.
Design headroom so Edge AI fits in later:
- Core bandwidth provisioned at 2× current demand.
- 20% spare switch ports on every rack.
- PoE++ on access switches so AI cameras don't need separate power.
- Reserved VLANs / IP ranges for Edge AI, IoT, SCADA separation.
- Racks with dual power + UPS and room for 2–3U Edge AI boxes.
A central data system — the prerequisite for plant-wide AI
Without a central data lake, every line is a silo — you can't train statistically meaningful models, can't do predictive maintenance, can't produce plant-wide reports. QAAD insists on a data architecture from day one: raw (machine logs) → processed (MES) → analytics (BI / ML).
Process digitisation — not "putting paper forms in an app"
- Video SOPs in multiple languages (see this post).
- MES directly wired to PLC / SCADA — no manual entry.
- Audit trails auto-generated for ISO 9001 / IATF 16949 / GMP.
- Integration: ERP ↔ MES ↔ WMS ↔ QMS — not islands.
The cost of getting it wrong — hard numbers
Real numbers we've witnessed:
- Rewiring a 5,000 m² factory from flat L2 to segmented L3 ≈ 150–300K USD + 1–2 weeks downtime.
- Low-end NAS bought year one, replaced year three = ~40K USD + data-migration risk.
- No HA firewall: a single 4-hour outage on a charger line ≈ 80K USD lost output + FDI-customer penalties.
- No VLAN segmentation: one ransomware event → 3 days of full plant downtime; recovery bill in six figures.
QAAD — not a hardware reseller, but a complete architecture partner
Our difference: a team that understands how a factory actually operates, not just the technology. QAAD engineers have walked the floors of FDI/EPE plants in Bac Ninh, Phu Tho, Binh Duong, Dong Nai — understanding the process from "first shift sample" to "end-of-line acceptance".
More importantly, QAAD delivers a single end-to-end stack, not just a pile of equipment:
- Infrastructure hardware: Cisco networking (switches/firewalls), NAS, servers, UPS, Edge AI boxes, GPU clusters, endpoint security appliances.
- Business software: MES, WMS, QMS, SOP training system, BI dashboards, ERP integration — built in-house or delivered through our partners.
- Specialised AI models: AI Inspection (machine vision), Sound Inspection (acoustic), Predictive Maintenance — built for each customer problem, not shrink-wrapped products.
- Operation & handover: SLAs, internal IT training, operations runbooks.
One partner, accountable from the first metre of cable to the final AI model — no finger-pointing between five or six different vendors when something breaks.
QAAD's smart-factory consulting covers:
- As-is assessment: infrastructure, processes, people, 5-year plan.
- Reference architecture: network, data, AI, security, software — clearly layered.
- Detailed BOM: Cisco networking, Trend Micro security, NAS, AI servers, GPU boxes.
- Software stack: MES / SOP / BI / AI inference — build vs. buy decisions.
- Phased rollout plan so production never stops.
- In-house IT training — clean handover.
Our mission
Quality Assurance · Active Development — our customers don't pay for a pretty deck. They pay for an architecture that lives 5–7 years, ready to accept Edge AI, MES, IoT, or a digital twin whenever they decide.
If you're building a new plant, expanding a line, or preparing to deploy AI — talk to QAAD before signing hardware POs. A single free consultation can save you a six-figure mistake.