Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Online-Adaptive Anomaly Detection for Defect Identification in Aircraft Assembly

Siddhant Shete, Dennis Mronga, Ankita Jadhav, Frank Kirchner

TL;DR

The paper tackles unsupervised anomaly detection for aircraft assembly by introducing an online-adaptive transfer-learning framework that selects visually similar training images and online-fits a normality model on EfficientNet B4 features. It jointly evaluates two similarity measures (SIFT-FLANN and Cosine) and two normality models (MVG and OCSVM), with adaptive thresholding to maintain robustness across environments. Empirical results on public benchmarks and lab datasets demonstrate high accuracy, notably up to 0.998 with Cosine+MVG, and show substantial training data reduction without sacrificing performance, outperforming the ET-NET ensemble. The approach offers practical potential for real-time quality control in aerospace manufacturing and can generalize to other industrial contexts.

Abstract

Anomaly detection deals with detecting deviations from established patterns within data. It has various applications like autonomous driving, predictive maintenance, and medical diagnosis. To improve anomaly detection accuracy, transfer learning can be applied to large, pre-trained models and adapt them to the specific application context. In this paper, we propose a novel framework for online-adaptive anomaly detection using transfer learning. The approach adapts to different environments by selecting visually similar training images and online fitting a normality model to EfficientNet features extracted from the training subset. Anomaly detection is then performed by computing the Mahalanobis distance between the normality model and the test image features. Different similarity measures (SIFT/FLANN, Cosine) and normality models (MVG, OCSVM) are employed and compared with each other. We evaluate the approach on different anomaly detection benchmarks and data collected in controlled laboratory settings. Experimental results showcase a detection accuracy exceeding 0.975, outperforming the state-of-the-art ET-NET approach.

Online-Adaptive Anomaly Detection for Defect Identification in Aircraft Assembly

TL;DR

The paper tackles unsupervised anomaly detection for aircraft assembly by introducing an online-adaptive transfer-learning framework that selects visually similar training images and online-fits a normality model on EfficientNet B4 features. It jointly evaluates two similarity measures (SIFT-FLANN and Cosine) and two normality models (MVG and OCSVM), with adaptive thresholding to maintain robustness across environments. Empirical results on public benchmarks and lab datasets demonstrate high accuracy, notably up to 0.998 with Cosine+MVG, and show substantial training data reduction without sacrificing performance, outperforming the ET-NET ensemble. The approach offers practical potential for real-time quality control in aerospace manufacturing and can generalize to other industrial contexts.

Abstract

Anomaly detection deals with detecting deviations from established patterns within data. It has various applications like autonomous driving, predictive maintenance, and medical diagnosis. To improve anomaly detection accuracy, transfer learning can be applied to large, pre-trained models and adapt them to the specific application context. In this paper, we propose a novel framework for online-adaptive anomaly detection using transfer learning. The approach adapts to different environments by selecting visually similar training images and online fitting a normality model to EfficientNet features extracted from the training subset. Anomaly detection is then performed by computing the Mahalanobis distance between the normality model and the test image features. Different similarity measures (SIFT/FLANN, Cosine) and normality models (MVG, OCSVM) are employed and compared with each other. We evaluate the approach on different anomaly detection benchmarks and data collected in controlled laboratory settings. Experimental results showcase a detection accuracy exceeding 0.975, outperforming the state-of-the-art ET-NET approach.
Paper Structure (15 sections, 2 equations, 5 figures, 1 table)

This paper contains 15 sections, 2 equations, 5 figures, 1 table.

Table of Contents

  1. INTRODUCTION
  2. RELATED WORK
  3. METHODOLOGY
  4. Training Data Selection
  5. SIFT-FLANN
  6. Cosine
  7. Feature Extraction
  8. Normality Model Fitting
  9. Multivariate Gaussian (MVG)
  10. One-class SVM (OCSVM)
  11. Adaptive Thresholding
  12. EXPERIMENTS
  13. Experimental Setup
  14. Results
  15. CONCLUSIONS

Figures (5)

  • Figure 1: Target application for anomaly detection: Section assembly in aircraft manufacturing nwzonline2023.
  • Figure 2: Proposed architecture design for adaptive online anomaly detection
  • Figure 3: Illustrative examples for the proposed anomaly detection method on different datasets.
  • Figure 4: Laboratory setup for anomaly detection.
  • Figure 5: AUROC Curves for Cosine-based (left) and SIFT/Flann-based approach.