Master Thesis

**Your Job**:
Join AMI, the Artificial Materials Intelligence division at the IET-3 at Forschungszentrum Jülich to develop an Autonomous Multi-Agent Performance Prediction Framework for PEM water electrolysis. As an MSc student, you will design and implement a suite of AI “agents” (autoencoders, statistical models, LSTMs and LLM-based rule engines) that process historical and live sensor data (voltage, current, temperature, pressure, flow) to forecast electrolyzer efficiency, degradation and stability. You will build data pipelines, refine predictive rules through a review-and-correction loop, and integrate an Aggregator Agent to fuse all outputs into one robust performance prediction-advancing explainable, self-improving AI for sustainable hydrogen production.

Your tasks in detail:
1. Data Preparation & Preprocessing
- Collect and clean historical and live PEM electrolyzer sensor streams (voltage, current, temperature, pressure, flow) from our partners.
Segment data into fixed‐length windows, normalize values, and engineer features (e.g., rolling means, deltas).

2. Agent Implementation
- Statistical Agent: Develop regression/control‐chart methods to capture trends and threshold breaches.
Forecasting Agent: Implement LSTM/Transformer networks for time‐series performance forecasting.

3. Rule-Based Agent: Encode expert‐driven and LLM-generated prediction rules into a modular, executable library.
- Rule Management & Refinement
- Rule Generation Agent: Use templates or LLM prompts to propose new performance‐prediction rules when agents disagree or fall below accuracy targets.
- Review Agent: Monitor inter-agent conflicts, false positives/negatives, and flag poor predictions for rule updates.
Correction Agent: Validate, debug, and sanitize new or modified rules (syntax, logic, coverage) before deployment.

4. Collaboration & Documentation
- Work closely with project supervisors and partner labs to share progress, datasets, and interim results.
- Maintain clear, up-to-date documentation of data pipelines, model code, rules library, and evaluation results.
- Present findings in regular meetings and contribute to thesis chapters or conference papers as needed.

**Your Profile**:

- Enrollment in a Master’s Program; Fields such as Data Science, Computer Science, Materials Science, Engineering, Physics, or a related domain.
- Proficient in Python; experience or eagerness to learn ML libraries (PyTorch, TensorFlow, scikit-learn),
- Familiarity with time-series analysis and forecasting methods (LSTM/Transformer).
- Experience with rule-based and LLM-driven systems is a plus.
- Passion for sustainable energy solutions, with an emphasis on hydrogen materials.
- Creative problem solver who enjoys interdisciplinary research and collaborative teamwork.
- Proficient in English (written and spoken).
Comfortable engaging with international and cross-functional teams.

**Our Offer**:
Hands-On multi-Agent AI Project

Play a central role in developing a novel, high-impact performance prediction system for PEM water electrolyzers.

Cutting-Edge Research Environment

Access to modern HPC infrastructure, AI toolkits, and an internationally recognized research community.

Professional Development

Build critical skills in time-series data workflows, multi-agent architectures, and rule-based/LLM-driven model development—highly sought after in both academia and industry.

Flexible MSc Thesis Structure
- Typical duration of 6-12 months aligned with Master’s project requirements.
Potential to co-author research papers or present findings at conferences, depending on project progress.

Support & Guidance
- Supervision from experienced researchers in a highly collaborative setting.
Opportunity to network with experts at FZ Jülich, and other Helmholtz centers