Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

3D-CSAD: Untrained 3D Anomaly Detection for Complex Manufacturing Surfaces

Xuanming Cao, Chengyu Tao, Juan Du

TL;DR

This work tackles 3D anomaly detection on complex manufacturing surfaces without training data by transforming a single point cloud into a set of directional profiles, decomposing the surface into basic components, and applying Robust Principal Component Analysis on each component. The method combines profile generation, a novel component segmentation and cleaning (CSC) step, and component-based RPCA to model the reference surface as low-rank while isolating sparse anomalies. It demonstrates strong empirical performance against baselines on synthetic holes and scratches across multiple parts, with ablations confirming the critical role of CSC. The approach offers a practical, training-free solution for automatic surface quality inspection in manufacturing, with potential for real-time QA on complex geometries.

Abstract

The surface quality inspection of manufacturing parts based on 3D point cloud data has attracted increasing attention in recent years. The reason is that the 3D point cloud can capture the entire surface of manufacturing parts, unlike the previous practices that focus on some key product characteristics. However, achieving accurate 3D anomaly detection is challenging, due to the complex surfaces of manufacturing parts and the difficulty of collecting sufficient anomaly samples. To address these challenges, we propose a novel untrained anomaly detection method based on 3D point cloud data for complex manufacturing parts, which can achieve accurate anomaly detection in a single sample without training data. In the proposed framework, we transform an input sample into two sets of profiles along different directions. Based on one set of the profiles, a novel segmentation module is devised to segment the complex surface into multiple basic and simple components. In each component, another set of profiles, which have the nature of similar shapes, can be modeled as a low-rank matrix. Thus, accurate 3D anomaly detection can be achieved by using Robust Principal Component Analysis (RPCA) on these low-rank matrices. Extensive numerical experiments on different types of parts show that our method achieves promising results compared with the benchmark methods.

3D-CSAD: Untrained 3D Anomaly Detection for Complex Manufacturing Surfaces

TL;DR

This work tackles 3D anomaly detection on complex manufacturing surfaces without training data by transforming a single point cloud into a set of directional profiles, decomposing the surface into basic components, and applying Robust Principal Component Analysis on each component. The method combines profile generation, a novel component segmentation and cleaning (CSC) step, and component-based RPCA to model the reference surface as low-rank while isolating sparse anomalies. It demonstrates strong empirical performance against baselines on synthetic holes and scratches across multiple parts, with ablations confirming the critical role of CSC. The approach offers a practical, training-free solution for automatic surface quality inspection in manufacturing, with potential for real-time QA on complex geometries.

Abstract

The surface quality inspection of manufacturing parts based on 3D point cloud data has attracted increasing attention in recent years. The reason is that the 3D point cloud can capture the entire surface of manufacturing parts, unlike the previous practices that focus on some key product characteristics. However, achieving accurate 3D anomaly detection is challenging, due to the complex surfaces of manufacturing parts and the difficulty of collecting sufficient anomaly samples. To address these challenges, we propose a novel untrained anomaly detection method based on 3D point cloud data for complex manufacturing parts, which can achieve accurate anomaly detection in a single sample without training data. In the proposed framework, we transform an input sample into two sets of profiles along different directions. Based on one set of the profiles, a novel segmentation module is devised to segment the complex surface into multiple basic and simple components. In each component, another set of profiles, which have the nature of similar shapes, can be modeled as a low-rank matrix. Thus, accurate 3D anomaly detection can be achieved by using Robust Principal Component Analysis (RPCA) on these low-rank matrices. Extensive numerical experiments on different types of parts show that our method achieves promising results compared with the benchmark methods.
Paper Structure (25 sections, 7 equations, 12 figures, 3 tables, 1 algorithm)

This paper contains 25 sections, 7 equations, 12 figures, 3 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Works
  3. Training-based methods
  4. Supervised methods
  5. Unsupervised methods
  6. Untrained methods
  7. Nonmodel-based methods
  8. Model-based methods
  9. Methodology
  10. Overview
  11. Profile Generation
  12. Segmenting into profiles
  13. Functional curve fitting
  14. Domain alignment
  15. Component Segmentation & Cleaning
  16. ...and 10 more sections

Figures (12)

  • Figure 1: Motivation of proposed 3D anomaly detection for complex manufacturing surfaces with a single sample. (a) is the input parts. (b) is the surface generation process from the perspective of design. (c) is the low-rank matrix stacked by multiple profiles, and (d) is anomaly detection results based on one single point cloud sample.
  • Figure 2: Illustration of the causes of error points during profile generation. (a) The points distribution point cloud in the surface pointed to by the arrow, where blue points are the surface points and red points are edge points. (b) The generated deviated profile (in yellow) caused by unstructured edge points. (c) the real generated profiles. (d) The profile matrix with error points.
  • Figure 3: The framework of the proposed method, which consists of three major steps, i.e., profile generation, component segmentation and cleaning module (CSC), and component-based RPCA (C-RPCA).
  • Figure 4: The procedure of profile generation step.
  • Figure 5: The illustration of profile generation step. (a) the illustration of the slicing plane on input point cloud data. (b) the sliced 3D profiles. (c) 2D profiles fitted by B-spline curve. (d) the matrix stacked by the aligned 2D profiles.
  • ...and 7 more figures