Auto-parts Manufacturer Multi-plant Smart Cloud and AI Inspection

Challenge

Inter-plant MPLS was expensive and slow to scale, and ERP / MES moved to cloud with degraded experience; nightly data sync with German HQ took over six hours; manual visual inspection on the line missed defects and left no traceable quality data.

GSS scope of responsibility

GSS owned WAN and cross-border circuit design, carrier coordination, CPE and field delivery, infrastructure deployment for the hybrid cloud platform and edge AI inspection clusters, and 24/7 managed operations throughout. ERP / MES / PLM applications and inspection-model business rules were owned by the customer IT and quality teams.

Solution

GSS rebuilt the WAN as SD-WAN + dedicated-circuit hybrid with a Shanghai ↔ Frankfurt IEPL, delivered a hybrid cloud management platform for ERP / MES / PLM, and deployed AI visual inspection clusters at the edge in all five plants — all operated by the GSS NOC.

Background

The company operates five plants in East and Central China, two R&D centres and a Shanghai head office, with IT architecture set by its German parent: ERP runs in the HQ data centre while MES and PLM are moving to cloud. The all-MPLS network carried a high monthly bill and new sites took 8–12 weeks; nightly engineering-data sync to Germany routinely exceeded six hours, so German colleagues started their day on stale drawings.

On the shop floor, critical parts (gears, precision castings) relied on manual visual inspection with a miss rate near 2%, and quality results could not be tied back to batch, machine or process parameters.

Solution

Network and cloud (traditional network business)

• Hybrid WAN: MPLS retained for core plant traffic, overlaid with dual-internet SD-WAN and application-aware routing — ERP / MES on circuits, video and office SaaS on internet — tripling total bandwidth while cutting the monthly bill.
• Cross-border circuit: a 200M Shanghai ↔ Frankfurt IEPL, contracted and invoiced by GSS in China with the overseas leg delivered by upstream partners, under 190 ms latency, carrying PLM sync and cross-border SAP access.
• Hybrid cloud management platform spanning Alibaba Cloud, on-prem VMware and HQ Azure, providing per-plant chargeback, automated deployment and security baseline checks for the MES / PLM migration.
• All sites operated 7×24 by the GSS NOC with bilingual support; German HQ has direct access to live dashboards for China’s network and cloud.

Edge AI inspection (AI business)

• Edge GPU clusters: a two-node inference cluster in each of the five plants; industrial-camera images are inferred locally in under 200 ms per part, independent of the WAN.
• Defect models covering gear burrs, casting porosity, surface scratches and dimensional deviation, initially trained on three months of historical images and continuously retrained from manual re-check results.
• Unified AI call control: LLM calls for process Q&A and inspection-report generation all go through the GSS Unified AI API Gateway, with quota per plant / department, input and output tokens metered separately, and monthly AI-spend attribution per plant visible to German HQ.
• Quality data loop: verdicts are written to MES with batch, machine ID and process parameters, creating a fully traceable quality record; model updates are pushed to all five plants through the cloud platform’s pipeline.
• China–Germany collaboration: HQ quality teams view China inspection dashboards and sample images in real time over the IEPL.

Results

Metric	Before	After
Monthly WAN cost	Baseline	−34%
Site bandwidth	20–50M MPLS	100–300M hybrid
New site lead time	8–12 weeks	2–3 weeks
Nightly cross-border sync	> 6 h	~50 min
Inspection miss rate	~2%	~0.3% (−85%)
Inspection time per part	8–12 s manual	< 0.2 s
Quality traceability	None	100% batch-linked

Delivered in two phases, completing in November 2024; GSS continues to operate the network, cloud platform and edge AI clusters.