Chat Bot technology for industrial maintenance scenarios — A Practical semester @ TUM

I supervised a practical semester with 8 students at Technical University Munich (TUM). Main goal was the evaluation of Chat Bot technology for industrial maintenance/repair scenarios (hands & touch-free human-machine-interaction).

Example: A service technician needs support and receives repair instructions from a Chat Bot running on a Smart Watch.

• Watch the students kick off presentation to get an overview of the idea
  https://youtu.be/eIvhL9iIiJ4
• Watch the students final presentation for a live demo and results
  https://youtu.be/8jxiyftnTwk

Findings

• High quality speech recognition technology is required to ensure the industrial worker is understood correctly by the chat bot (think about background noise, accents, dialects, etc)
• Common chat bot request-response patterns are not suitable for (industrial) maintenance scenarios
• A chat bot-compatible documentation format is required for maintenance instructions, e.g. in form of a state machine. PDF, Word & Co won’t do it. Disadvantage: Documentation in such forms rarely exist.
• Alternatively an Artificial Intelligence / Learning approach is required to retrieve repair instructions from existing maintenance documentation (PDF, Word & Co).
• An autonomous and reliable Smart Watch without any paired Smart Phone is required. (The maintenance worker won’t deal with an Apple Watch and a paired iPhone in his pocket, both battery-intense, etc)
• An image recognition technology supports identifying the state and the context of the machine that shall be maintained. Note, this recognition service needs to be “trained” with corresponding pictures beforehand.

Used technologies: Ionic / Angular / Cordova, AIML, pandorabots.com, wit.ai, api.ai, custom state machine engine, image capturing/recognition service