Abstract:This study proposed a scheme based on Machine Learning (ML) models to detect geometric defects of additively manufactured objects. The ML models are trained with synthetic 3D point clouds with defects and then applied to detect defects in actual production. Using synthetic 3D point clouds rather than experimental data could save a huge amount of training time and costs associated with many prints for each design. Besides distance differences of individual points between source and target point clouds, this scheme uses a new concept called "patch" to capture macro-level information about nearby points for ML training and implementation. Numerical comparisons of prediction results on experimental data with different shapes showed that the proposed scheme outperformed the existing Z-difference method in the literature. Five ML methods (Bagging of Trees, Gradient Boosting, Random Forest, K-nearest Neighbors and Linear Supported Vector Machine) were compared under various conditions, such as different point cloud densities and defect sizes. Bagging and Random Forest were found the two best models regarding predictability; and the right patch size was found to be at 20. The proposed ML-based scheme is applicable to in-situ defect detection during additive manufacturing with the aid of a proper 3D data acquisition system.

What problem does this paper attempt to address?

### What problem does this paper attempt to solve? This paper aims to propose a machine learning (ML) model-based approach to detect geometric defects arising during the additive manufacturing (AM) process. Specifically: 1. **Background and Challenges**: - While additive manufacturing has advantages in complex geometric design, it has limitations in production speed, cost, and precision. - It is estimated that surface defects lead to about 10% failure rate in additive manufacturing parts. These defects include porosity, layer shifts, inclusions, delamination, and lack of fusion, all of which can cause geometric inaccuracies and affect mechanical performance. - Traditional 2D image processing methods are not precise enough in detecting small-sized defects, especially in high-precision AM processes. - Real-time defect detection during production, rather than measurement in the post-processing stage, helps save time and costs. 2. **Proposed Method**: - The paper proposes a new scheme that uses synthetic 3D point cloud data to train machine learning models, which are then applied to defect detection in actual production. - This scheme introduces a new concept—"patch"—to capture macroscopic information of nearby points, enhancing the effectiveness of model training and implementation. - Using synthetic data instead of experimental data can significantly reduce training time and costs. 3. **Experimental Validation**: - By comparing the prediction results of experimental data with different shapes, it was found that the proposed scheme outperforms the existing Z-difference method. - Five machine learning methods (Bagging of Trees, Gradient Boosting, Random Forest, K-nearest Neighbors, and Linear Supported Vector Machine) were compared and tested under different point cloud densities and defect sizes. - Experimental results show that Bagging and Random Forest are the two models with the best prediction performance. In summary, the main goal of this paper is to develop an efficient and accurate machine learning method for real-time defect detection in the additive manufacturing process, thereby improving production efficiency and product quality.

Geometrical defect detection for additive manufacturing with machine learning models

Geometrical defect detection on additive manufacturing parts with curvature feature and machine learning

A Novel Defect Detection System for Complex Freeform Surface Structures

Online Quality Monitoring in Material Extrusion Additive Manufacturing Processes Based on Laser Scanning Technology

Defect Classification for Additive Manufacturing with Machine Learning

Visual Detection of Surface Defects Based on Self-Feature Comparison in Robot 3-D Printing

Detecting and classifying hidden defects in additively manufactured parts using deep learning and X-ray computed tomography

Rapid surface defect identification for additive manufacturing with in-situ point cloud processing and machine learning

An In-Situ Deep Learning-Based Defect Detection Technology for Additive Manufacturing Process

Comparative Analysis of Deep Learning-Based Defect Monitoring in Metal Additive Manufacturing

Real-time Defect Detection of Laser Additive Manufacturing Based on Support Vector Machine

Multi-view Online Vision Detection Based on Robot Fused Deposit Modeling 3D Printing Technology

Defect detection of 3D printing surface based on geometric local domain features

A Real-Time Defect Detection Strategy for Additive Manufacturing Processes Based on Deep Learning and Machine Vision Technologies

Defect Analysis of 3D Printed Cylinder Object Using Transfer Learning Approaches

On the application of machine learning for defect detection in L-PBF additive manufacturing

Applied AI in Defect Detection for Additive Manufacturing: Current Literature, Metrics, Datasets, and Open Challenges

Machine learning algorithms for defect detection in metal laser-based additive manufacturing: A review

Acoustic feature based geometric defect identification in wire arc additive manufacturing

Explainable deep neural network for in-plain defect detection during additive manufacturing

Uni-3DAD: GAN-Inversion Aided Universal 3D Anomaly Detection on Model-free Products