Identification and Training of AI Models for In-Situ Defect Detection in Electron Beam Melting Additive Manufacturing

Institute

Lehrstuhl für Werkstofftechnik der Additiven Fertigung

Type

Master's Thesis /

Content

 

Description

Motivation

Electron Beam Melting (EBM) is a key additive manufacturing (AM) process used for producing high-performance metal components, particularly in the aerospace and medical sectors. Despite its advantages in material utilization and design freedom, EBM is susceptible to several defect types such as porosity, lack of fusion, and surface irregularities. These defects can significantly influence the mechanical performance and reliability of printed parts.

In-situ monitoring technologies, such as Backscattered Electron (BSE) imaging and Computed Tomography (CT) scans, provide valuable process data that can be leveraged to identify and understand these defects. Artificial Intelligence (AI) offers powerful tools for automated defect recognition and classification, but the selection and training of suitable AI models require careful consideration of data characteristics, model complexity, and computational efficiency.

 

Objective

 

The goal of this thesis is to identify, train, and evaluate suitable AI models for detecting and classifying different types of defects in EBM-manufactured components using in-situ process data. The student will investigate and compare different learning approaches to determine the most promising architecture for robust and accurate defect detection.

The final outcome should include a trained and validated AI model, together with a systematic comparison of different model types and a discussion of their applicability for integration into an in-situ process monitoring framework.

 

Proposed Work Packages

1. Literature Review and Data Analysis
   • Review state-of-the-art research on AI-driven defect detection in additive manufacturing.
   • Analyze available EBM in-situ datasets (e.g., BSE images, CT data) and identify relevant defect types.
   • Investigate Computer Vision models for image-based feature extraction and defect identification in EBM data.
2. Model Selection and Data Preparation
   • Identify suitable AI architectures (e.g., Convolutional Neural Networks and other relevant).
   • Preprocess and label data.
3. Model Training and Performance Evaluation
   • Implement and train selected models using available datasets.
   • Compare model performance using relevant metrics (e.g., accuracy, precision, recall).
4. Validation and Discussion
   • Validate the models.
   • Discuss advantages, limitations, and integration potential into real-time monitoring systems.

Requirements

• Background in Mechanical Engineering, Data Science, Computer Science, or related field
• Basic understanding of additive manufacturing processes (preferably PBF-Processes)
• Experience with Python and common machine learning frameworks (e.g., TensorFlow, PyTorch, scikit-learn)
• Interest in applied AI and manufacturing process analysis

Tags

MAT Geitner, MAT Studi

Possible start

sofort

Contact

Claudia Geitner
Phone: +49 (0) 89 289 - 55326
claudia.geitnertum.de