Endi-QM: Energy Efficiency via Intelligent In-Process Quality Monitoring

Lab: Produktionsmanagement, Technologie und Werkzeugmaschinen Lab (PTW), TU Darmstadt Period: November 2022 – June 2024

Overview

Endi-QM applies deep learning and reinforcement learning methods to industrial sensor data and machine configuration parameters to predict product quality and optimize machine settings — reducing trial-and-error in manufacturing processes.

Contributions

• Developed an advanced program integrating a deep learning model (PyTorch) and actor-critic models (Stable-Baseline3) to optimize machine parameters and achieve desired product quality from initial machine and sensor data
• Customized a reinforcement learning environment via OpenAI Gym for this specific industrial task
• Conducted evaluations showing the trained agent iteratively adjusts parameters across various datasets, achieving desired product quality within 10 steps
• Explored multimodal fusion strategies to enhance classification performance across sensor modalities
• Refactored a GNN-based solution with GNNExplainer for interpretable model predictions
• Utilized Hydra + OmegaConf for efficient experiment configuration management
• Tracked experiments with MLFlow and TensorBoard for reproducibility and transparency
• Ensured compatibility with diverse industrial datasets including LEGO products and AUDI welding data