The problem
Industrial operators often depend on specialists to interpret alarms, error codes and technical documentation, resulting in delayed troubleshooting and unnecessary dependency on specialist support.
The discovery question was therefore not how can we use an LLM? but how can we reduce dependency on specialists when something goes wrong?
Discovery
- Interviewed system owners and testbed operators
- Studied incident and troubleshooting workflows
- Identified operational constraints and pain points
- Identified specialist dependency
- Identified that significant machine data existed but lacked contextual explanation
- Defined the AI-assistant concept around explainability and self-service
Product hypothesis: could AI translate industrial system context and specialist knowledge into understandable and actionable guidance?
Solution
Designed and developed a proof-of-concept AI assistant combining industrial machine data with domain knowledge and technical documentation to provide natural-language guidance.
Edge deployment kept industrial data on-site.
- Retrieval-Augmented Generation
- LangChain
- OPC UA
- Siemens S7 PLC
- Vector search over technical documentation
- Live machine data
- Edge deployment
Capabilities
- Root-cause explanations
- Step-by-step resolution
- Interactive maintenance guidance
- Natural-language alarm explanation
- Contextual technical information
Validation
Evaluated in the FESTO CP-Factory smart-factory testbed together with MITC partners.
User testing demonstrated that non-specialist users could successfully work through incidents using the AI assistant without waiting for specialist support.
Potential business value
The concept demonstrated potential to:
- Reduce troubleshooting time
- Reduce downtime exposure
- Improve first-line support
- Make specialist knowledge accessible at the point of operation
- Improve support scalability
Product perspective
The architecture was not limited to one troubleshooting scenario. The same combination of operational data, domain knowledge and AI assistance could support multiple reusable product capabilities.
- AI-powered technical support
- Maintenance decision support
- Field service
- Operational knowledge assistance
- Product selection
- Sales enablement
What this demonstrates
- Problem-first AI discovery
- Translating specialist knowledge for non-specialist users
- Rapid validation with real users
- Communication and usability as part of technical feasibility
- Judging AI by whether it improves the user's job rather than model sophistication alone
From problem to product opportunity
- Operational Problem
- Discovery
- AI Capability
- Validation
- Reusable Product Opportunities