Robot Welding Digital Cell: An Edge-Computing System That Connects MES to the Welding Torch

The Scenario: Digitalization Challenges of Robot Welding Lines

Rail-transit equipment (high-speed trains/locomotives/urban-rail vehicles) sets the industry ceiling for welding quality — body welds carry structural safety, and each welding cell (vision/back-gouging/welding/slag-removal) has its own control system, but they are isolated:

• Four subsystems, four data silos: vision inspection, back-gouging, welding and slag-removal run separately; process parameters (current/voltage/speed/gas pressure) are scattered and cannot be aggregated
• MES cannot reach the device layer: MES dispatches work orders, but terminal execution, process parameters and quality results are unknown to MES — planning and execution disconnect
• Manual quality traceability: weld quality recorded per weld/layer/pass by hand; on failure, nobody can say which weld, which parameter, which batch
• No unified alarming: equipment faults, parameter limits and quality rejects have no unified sound-light alarm or push notification; response depends on people watching
• Vision has a blind spot: Phase-1 vision checks weld appearance (porosity/undercut/forming), but internal weld quality (cold weld, lack of fusion, abnormal heat input) is invisible to vision — process-signal-level monitoring is needed

Solution: The Edge-Computing System as the Welding Cell's Digital Hub

This solution builds a turnkey system around the edge controller (TSingley TiMC-EI844):

Customer MES/ERP (RESTful API + JWT auth)
        ↑ HTTPS / MQTT
Edge controller TiMC-EI844 (TSingley IIoT adapter V1.0)
  ├─ Production / BI / Messaging / Edge compute / IIoT — five modules
        ↑ Southbound: OPC UA / Modbus TCP / Siemens S7
  ├─ Vision inspection system — weld quality + anomaly photos
  ├─ Back-gouging control system — gouging pressure/speed
  ├─ Welding control system — current/voltage/speed/pressure (per weld/layer/pass)
  └─ Slag-removal control system — slag status

Core Capability 1: Cloud-Network-Edge-Device Integration

• Northbound to MES/ERP: RESTful API (HTTPS + JWT auth) receives work orders and reports completion; MQTT lightweight IoT protocol for data platforms
• Southbound to four subsystems: vision via OPC UA/Modbus TCP; back-gouging/welding/slag-removal via Siemens S7 (native PLC DB block bytes) — heterogeneous devices unified
• Network: star topology with the edge controller as hub, private IP range + VLAN isolation, no IP conflicts

Core Capability 2: Five Software Modules, Full Coverage

Core Capability 3: Real-Time Parameter Acquisition, Weld-Level Traceability

The key to weld quality traceability is recording process parameters per product-weld-layer-pass:

• Real-time acquisition: welding current/voltage/speed/pressure auto-sampled at ≥1Hz; event-triggered on quality anomalies/device alarms; manual trigger for debugging
• Level-by-level traceability: every weld (weld NO/layer NO/pass NO) records current, voltage, speed, pressure, slag status, weld image features and quality verdict (pass/rework/scrap) — failures pinpoint to an exact weld
• Quality loop: vision system judges pass/fail, auto-photographs on anomaly; results correlate with work order, equipment and operator (shift), meeting ISO 9001 quality-record requirements
• Storage: 1TB holds ≥90 days at 1Hz sampling, integrity ≥99.5%, incremental backup + NAS backup machine

Core Capability 4: Online Welding-Quality Monitoring (Phase 2) — Dual Mistake-Proofing

Phase-1 vision solves weld appearance, but internal quality (cold weld, lack of fusion, abnormal heat input) can only be seen in process signals. The customer's Phase-2 plan introduces online welding-quality monitoring, reusing TSingley's proven CNN+VAE+rule-engine fusion from ultrasonic welding, forming dual mistake-proofing with vision:

• Two-channel complement: vision checks appearance (porosity/undercut/forming/spatter), process-signal monitoring checks internals (current/voltage/speed/pressure curve features) — appearance-passing does not equal internally-passing; both must pass for OK
• CNN curve classification: a lightweight 1D-CNN infers in real time, learning the current-voltage-speed curve shapes of normal/cold-weld/lack-of-fusion/abnormal-heat-input, judged instantly at the edge
• VAE anomaly detection: a variational auto-encoder trains on normal data only and catches novel welding anomalies the CNN has never seen — weld conditions vary too much to pre-label every defect
• Explainable rule engine: arc-voltage offset, welding-speed offset, pressure fluctuation — physical features translated into concrete metrics a process engineer understands, solving the human-trust problem
• Seamless with Phase-1 data flow: monitoring results merge into the edge system's production details (weld/layer/pass + quality verdict); vision and process-signal results shown side by side — traceability upgrades from 'appearance record' to 'internal+external evidence'
• Alarm linkage: welding-process anomalies instantly trigger the tri-color sound-light alarm + message push, same chain as Phase-1 messaging

Phase 2 fully reuses the Phase-1 edge platform: add models, rules and acquisition channels — no architecture change, no production stop. That is the depth value of 'reuse + light customization'.

Core Capability 5: Message Linkage and On-Site Alarming

• Tri-color sound-light alarm linked to the messaging module: normal = green steady, warning = yellow flash (near threshold/needs maintenance), fault = red flash + buzzer (≥85dB)
• Audit logs for key operations; three-level permissions (operator/admin/engineer) prevent parameter misconfiguration

Core Capability 6: Turnkey Delivery, Reliable Operation

• Console workbench (piano-style) integrates all hardware: edge controller/industrial power/switch/display/alarm/UPS, DIN-rail standard mounting
• UPS protection: 2-6ms switch to battery on mains loss, ≥30min runtime, low-battery auto safe-shutdown — no data loss
• Industrial-grade hardware: Intel 12th-gen i5 + 16GB DDR5 + 1TB SSD, wide voltage DC9-36V, MTBF ≥8000h, adapted to dusty/EMI welding shops

Value: The Welding Cell from Silo to Closed Loop

Deployment & Reuse

Deployed in rail-transit equipment manufacturing (CRRC ecosystem): Phase 1 delivers the edge-computing system + vision monitoring; Phase 2 plans online welding-quality monitoring (CNN+VAE+rule engine). Built on the standard edge-computing product with light per-line customization (protocol integration, work-order fields, report config, monitoring models). Reusable for: construction machinery, shipbuilding, pressure vessels, steel structures — robot-welding lines, especially factories with MES but unconnected device layers.

From Silo to Closed Loop: A Digital Hub for the Welding Cell

Four subsystems, four data sets, one digital hub — the edge-computing system turns a robot-welding cell from 'every system for itself' into a fully linked closed loop: Phase-1 vision mistake-proofing, Phase-2 process-signal mistake-proofing, traceability from weld/layer/pass level to internal-plus-external evidence.

Which data on your welding lines is still unconnected? Talk to us for a peer-industry reference.