PointSGRADE: Sparse Learning with Graph Representation for Anomaly Detection by Using Unstructured 3D Point Cloud Data

Abstract

Surface anomaly detection using unstructured 3D point cloud data poses challenges due to the irregular data structure and varying anomaly patterns. This paper introduces PointSGRADE, a sparse learning framework with graph representation designed to detect anomalies on smooth free-form surfaces. The method formulates the problem as a penalized optimization problem and solves it using a majorization-minimization framework, demonstrating high accuracy and robustness in case studies.

Assumption:

• (1) Smooth free-form surface: limited overall curvature; neighborhood approximated well by local plane
• (2) Sparse anomaly
• (3) Gaussian measurement noise

Goal

Propose a computational efficient method for sparse anomaly detection of smooth free-form surface using one single point cloud sample.

Overall framework of PointSGRADE.

Formulation and graph representation for smooth free-form surface.

Juan Du

Assistant Professor

My research interests include knowledge-infused data science for quality improvement, industrial data analytics and machine learning, and system informatics and control for manufacturing applications.