In the wood manufacturing industry, specifically within the edgebanding pro-
cess, maintaining high quality requires a deep understanding of complex ma-
chine interactions. When a defect occurs, it currently takes a highly experi-
enced operator to diagnose the root cause and adjust the machine parameters
correctly. However, such qualified personnel are becoming increasingly scarce,
and the ”trial-and-error”method used to find optimal parameters is costly in
terms of both time and material. While high-end automated sensor systems
can solve this, they are often too expensive for manufacturers. A promising
alternative is a ”Human-in-the-Loop” approach: using an AI agent to act as a
digital expert, leading the operator through a diagnostic conversation to accu-
rately identify the defect. This creates a scalable way to achieve high-quality
production without the need for prohibitive hardware investments

GOALS:

This thesis involves designing a system where an AI agent guides operators through structured interactions to
diagnose production defects. You will leverage a knowledge base of defects, attributes, and visual aids to enable
high-certainty classification of defect type, position, and dimensions. The work requires analyzing industrial quality
assurance needs and developing conversational logic—comparing AI architectures to identify the most reliable and
efficient method for data extraction. The outcome is a prototype converting diagnostic conversations into structured
technical reports.
Key tasks include:

• Analysis of the domain-specific ”defect library”(attributes, types, and visual examples).
• Development of a conversational strategy/flow to guide non-expert users.
• Implementation of an AI agent prototype (e.g., utilizing LLMs, RAG, or decision-logic).
• Development of a method to calculate the certainty/confidence of the final classification.

HELPFUL PRIOR KNOWLEDGE:

• Experience with modern AI frameworks, Knowledge Graphs and information modeling
• Comfortable working independently to design a software prototype