Student project: LLM-Based Diagnosis and Maintenance for Electrochemical Sensors 

This project focuses on developing a preventive maintenance system, based on Large Language Models (LLM) for fault diagnosis with application to network electrochemical sensors. The goal, is to leverage the existing knowledge, mapped via knowledge graphs, and couple that with LLMs to interpret error logs, extract fault pathways, and generate root cause explanations with repair suggestions, therefore, helping support more flexible and faster, data-driven maintenance decisions. 

What you will do

The internship project is structured around a series of key tasks that guide the development of an intelligent fault diagnosis and preventive maintenance system for electrochemical gas sensors. These tasks cover understanding sensor failures, designing data models, processing real or simulated data, and integrating AI techniques for effective troubleshooting support.

• Literature Review: Study existing research on electrochemical sensor maintenance, fault diagnosis, knowledge graphs, and LLM-based retrieval-augmented generation (RAG).
• Project Planning: Develop a detailed timeline with milestones and deliverables spanning data collection, model development, and system integration.
• Data Collection: Gather real-world or simulated sensor error logs and operational data to create a realistic dataset for training and testing.
• Knowledge Graph Design and RAG Implementation: Design a schema for fault pathways and repair knowledge, then implement retrieval augmented generation to combine KG querying with LLM reasoning.
• Model Selection and Development: Choose and fine-tune LLMs and parsers to extract error codes and symptoms from natural language inputs.
• System Integration: Build the end-to-end pipeline combining data ingestion, KG querying, LLM inference, and output generation.
• Simulation/Demo: Design troubleshooting scenarios to validate system accuracy and usability; prepare demos showcasing fault diagnosis and maintenance recommendations .

What we do for you

• Work on a challenging, open ended problem with freedom to design innovative solutions.
• Receive guidance and support from a diverse team of experts in environmental science, sensor technology, and AI.
• Join the OnePlanet Data Science team, applying SOTA machine learning and big data frameworks to real environmental monitoring challenges.
• Collaborate with the broader Imec community, expanding your professional network and exchanging knowledge with leading scientists and engineers.
• Improve your coding and data science skills to meet industry standards.
• Access cloud computing resources to efficiently process large volumes of sensor and operational data.
• Gain hands on experience in AI driven preventive maintenance, knowledge graph design, and natural language processing within a meaningful environmental context.

Who you are

• Background on Computer Science, Electrical Engineering, or related areas.
• Proficiency in Python for data processing and model implementation.
• Familiarity with natural language processing (NLP) techniques and tools.
• Interest in knowledge graphs and graph databases (e.g., Neo4j, RDF) is advantageous.
• Willingness and ability to learn about Retrieval-Augmented Generation (RAG), which is essential for this project.
• Ability to work with structured and unstructured data, including sensor and error log data.
• Good analytical and problem-solving skills.
• Basic understanding of sensor technology or environmental monitoring is beneficial.
• Proficiency in using Git for version control and basic understanding of Agile/Scrum methodologies.
• A passion for applying AI and data science to solve environmental challenges.  

Interested

Does this position sound like an interesting next step in your career at imec? Don’t hesitate to submit your application by clicking on ‘APPLY NOW’.
Should you have more questions about the job, you can contact jobs@imec.nl.