Smart Manufacturing
Industrial IoT AI After-Sales Agent
Put a field expert inside every machine
Consolidates scattered equipment manuals, fault SOPs, spare-parts lists, and operating procedures into a searchable production-line knowledge base. Field technicians ask questions in plain language or by alarm code and receive cited, actionable resolutions instantly — already live on Mengniu's Tiantrack conveyor line.
7×24
Round-the-clock self-service support
Cited
Every answer traceable and downloadable
On-premises
Production data never leaves the plant
Pain Points
Challenges
- Equipment manuals, alarm codes, and SOPs are spread across different people, roles, and systems, making fault diagnosis slow
- The volume of alarm codes is too large for field staff to memorize the meaning and resolution steps of each one
- Device-specific know-how and fault-handling expertise sit with senior staff and disappear when they transfer or leave
- When faults occur on night shifts or public holidays, there is no on-site expert to consult
- Paper manuals exist in multiple versions and are too complex to search quickly by hand
Capabilities
Core Capabilities
Natural-language Q&A with cited sources
Ask in everyday language and receive step-by-step resolution guidance with cited references and downloadable source documents.
Alarm-code-to-SOP lookup
Enter an alarm code (e.g., RADAR_FRONT_OBSTACLE) and instantly get the full troubleshooting checklist and resolution steps.
Compound-question decomposition
Multiple questions in a single query are automatically split into separate retrievals, merged into one response, and cited individually.
Voice input and audio playback
When hands are gloved, greasy, or occupied, complete the entire Q&A — asking and receiving answers — entirely by voice.
Photo and file queries
Photograph an alarm code on a device screen for instant OCR-based lookup; also supports direct Q&A over uploaded PDF, Word, and Excel files.
Structured knowledge management
One-click import of multi-format documents, automatic chunking into retrievable knowledge units, and search by group, category, or alarm-code tag.
Real-world Case
Mengniu Tiantrack Conveyor Line
The Mengniu Tiantrack conveyor line has integrated the production knowledge base module, giving on-site operators, maintenance staff, and managers a unified knowledge access point. When equipment faults, operating questions, or parts queries arise, staff can ask directly at a workstation terminal and receive cited, actionable resolutions — no more paper manuals or calls to the electrical team.
From "RGV Fault Handling Manual" — Rear obstacle detected （RADAR_BACK_OBSTACLE）
A standard safety alarm triggered by the rear radar detecting an obstacle. Resolution steps:
- Clear obstacles within 2 m of the rear — check the parts buffer area first
- Wipe the rear radar panel with a dry cloth
- Press the "Reset Radar" button on the vehicle panel
- If the alarm persists after 10 s, inspect the radar wiring
From "RGV Fault Handling Manual" — Vehicle position lost （POSITION_LOST）
The PLC cannot determine the vehicle's position — a critical fault that will halt dispatching. Resolution steps:
- Confirm the vehicle's actual current position on-site
- Write the corresponding position code to DB901.DBW148
- Trigger the position backfill and wait for PLC confirmation
- Cancel the vehicle's current task and re-issue the order
The conversations above use real knowledge entries from the RGV Fault Handling Manual, already live on the Mengniu Tiantrack line. Alarm codes map directly to resolution SOPs.
Ingested & Demo-ready Content
Covers the Knowledge Used Every Day On-site
Line Alarm Reference Table
Alarm codes and resolution descriptions for all line equipment
RGV Fault Handling Manual
SOPs for common vehicle faults — e-stop, radar obstacles, POSITION_LOST, and more
Lift & Slide Equipment Fault Manual
Troubleshooting for lifts, slide platforms, lift-gate photo-sensors, and other line devices
On-site Operating Procedures Manual
Shift startup checks, shutdown procedures, station call-for-parts, e-stop, and handover workflows
Spare Parts List
Part numbers, locations, replacement intervals, and suppliers for RFID readers, radar sensors, and more
PLC Configuration Manual
Network topology, PLC program structure, key parameters, and online monitoring methods
Business Impact
In Daily Use On the Floor
Fully on-premises deployment — production data and documentation never leave the plant. Reuses the existing line monitoring and WMS scheduling environment; no separate infrastructure needed.
-
Faster fault response
Alarm codes map directly to resolution SOPs, cutting the time equipment sits idle.
-
Institutional knowledge retained
Handling experience is captured in the knowledge base and stays put — regardless of staff turnover.
-
Faster onboarding for new staff
Self-service access to procedures and parts locations reduces dependence on senior colleagues.
-
Self-service on nights and holidays
When faults occur outside business hours, technicians can find resolution guidance on their own.
-
Single source of truth on-site
Centralized knowledge base — update once, effective everywhere, instantly.
Technical Architecture
RAG Retrieval Pipeline: Every Answer Is Traceable
Rather than sending questions directly to an LLM, we run a fully engineered retrieval pipeline — first retrieve relevant enterprise knowledge, then generate an answer, then verify every citation — eliminating hallucinations at the source.
Parse & OCR
PDF, Word, Excel, and image files are uniformly parsed into text; images go through OCR recognition.
Chunk into Knowledge Units
Long documents are split into retrievable knowledge entries, automatically tagged with alarm codes and category labels.
Hybrid Retrieval
Vector semantic search combined with keyword / alarm-code exact matching — dual-channel recall for maximum coverage.
Re-rank
A cross-encoder re-ranks candidate entries to improve Top-K precision.
Generation with Citations
The LLM generates answers grounded in the retrieved entries and is required to attach cited sources.
Citation Verification · Hallucination Guard
Every answer is checked line by line against the source text. When no evidence exists, the system declines to answer rather than fabricating.
Feedback Loop · Continuous Learning Resolution outcomes and human feedback flow back into the system, continuously improving retrieval ranking and the knowledge base — it gets more accurate over time.
Technical Specifications
Hard Specs. On-premises Ready.
Model-agnostic, hybrid retrieval, domestic-cloud compatible — designed for the reliability, security, and compliance demands of industrial environments.
- LLM
- Supports leading commercial and open-source / domestic models (Qwen, GLM, DeepSeek, etc.) — switchable and composable based on cost and compliance requirements
- Retrieval
- Vector semantic search + keyword / alarm-code exact matching (hybrid) · cross-encoder re-ranking · citation verification
- Deployment
- Public cloud / private cloud / on-premises; compatible with domestic IT environments (Xinchuang); models and data never leave the plant
- System Integration
- Standard API connectors for PLC monitoring, WMS scheduling, ticketing systems, IoT platforms, and enterprise messaging
- Document Formats
- PDF · Word · Excel · Markdown · Images (OCR)
- Multimodal Input
- Text · Voice (ASR + TTS) · Photo (OCR)
- Security & Compliance
- Data masking · granular permissions · operation audit logs · end-to-end traceability
Process
How It Works
From document import to field queries — a few steps turn scattered knowledge into actionable resolutions.
- 1
Import equipment manuals, fault SOPs, operating procedures, and parts lists into the system
- 2
The system automatically splits documents into retrievable knowledge units tagged with alarm codes
- 3
Field staff query by text, voice, or photo using plain language or alarm codes
- 4
Cited resolutions are returned and can be read aloud; handling experience is continuously fed back and retained
Foundation
Built on a Proven Platform
Every capability runs on the same stable, secure, and privately deployable foundation.
Want to see it running in your business?
Every business is different. Talk to us — we'll identify the right solution and get it live quickly.