Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

An Unsupervised Time Series Anomaly Detection Approach for Efficient Online Process Monitoring of Additive Manufacturing

Frida Cantu, Salomon Ibarra, Arturo Gonzales, Jesus Barreda, Chenang Liu, Li Zhang

TL;DR

The paper tackles unsupervised anomaly detection in online additive manufacturing by introducing CSSAD, a framework that combines a GPU-accelerated convolution-based matrix profile (Conv-MP) with semantic segmentation to locate the onset of defects in high-resolution sensor time series. By leveraging Nearest Neighbor Arc Density and a bias-corrected density measure (CAD), the method identifies anomalous regions with a controllable threshold, enabling efficient online monitoring. On seven real-world AM datasets, CSSAD achieves a mean F1-score of 0.812 and low false positives, outperforming several baselines and benefiting from significant GPU acceleration. The approach offers practical value for real-time process monitoring and cyber-physical anomaly detection in additive manufacturing, with potential extensions to online batch processing for evolving printing regimes.

Abstract

Online sensing plays an important role in advancing modern manufacturing. The real-time sensor signals, which can be stored as high-resolution time series data, contain rich information about the operation status. One of its popular usages is online process monitoring, which can be achieved by effective anomaly detection from the sensor signals. However, most existing approaches either heavily rely on labeled data for training supervised models, or are designed to detect only extreme outliers, thus are ineffective at identifying subtle semantic off-track anomalies to capture where new regimes or unexpected routines start. To address this challenge, we propose an matrix profile-based unsupervised anomaly detection algorithm that captures fabrication cycle similarity and performs semantic segmentation to precisely identify the onset of defect anomalies in additive manufacturing. The effectiveness of the proposed method is demonstrated by the experiments on real-world sensor data.

An Unsupervised Time Series Anomaly Detection Approach for Efficient Online Process Monitoring of Additive Manufacturing

TL;DR

The paper tackles unsupervised anomaly detection in online additive manufacturing by introducing CSSAD, a framework that combines a GPU-accelerated convolution-based matrix profile (Conv-MP) with semantic segmentation to locate the onset of defects in high-resolution sensor time series. By leveraging Nearest Neighbor Arc Density and a bias-corrected density measure (CAD), the method identifies anomalous regions with a controllable threshold, enabling efficient online monitoring. On seven real-world AM datasets, CSSAD achieves a mean F1-score of 0.812 and low false positives, outperforming several baselines and benefiting from significant GPU acceleration. The approach offers practical value for real-time process monitoring and cyber-physical anomaly detection in additive manufacturing, with potential extensions to online batch processing for evolving printing regimes.

Abstract

Online sensing plays an important role in advancing modern manufacturing. The real-time sensor signals, which can be stored as high-resolution time series data, contain rich information about the operation status. One of its popular usages is online process monitoring, which can be achieved by effective anomaly detection from the sensor signals. However, most existing approaches either heavily rely on labeled data for training supervised models, or are designed to detect only extreme outliers, thus are ineffective at identifying subtle semantic off-track anomalies to capture where new regimes or unexpected routines start. To address this challenge, we propose an matrix profile-based unsupervised anomaly detection algorithm that captures fabrication cycle similarity and performs semantic segmentation to precisely identify the onset of defect anomalies in additive manufacturing. The effectiveness of the proposed method is demonstrated by the experiments on real-world sensor data.

Paper Structure

This paper contains 17 sections, 8 equations, 6 figures, 2 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Problem Setting
  3. Related Work
  4. Background
  5. Proposed Method
  6. Background on STOMP for Matrix Profile
  7. Convolution-based Matrix Profile
  8. Semantic Segmentation-based Anomaly Detection
  9. Experiments
  10. Datasets
  11. Experiment setup
  12. Baseline Methods
  13. Evaluation Metrics
  14. Experiment 1: Performance on Anomaly Detection in Real-world data.
  15. Experiment 2: Ablation Study on Threshold Parameter
  16. ...and 2 more sections

Figures (6)

  • Figure 1: Problem Setting
  • Figure 2: Overall Framework for Convolutional Semantic Segmentation Anomaly Detection (CSSAD)
  • Figure 3: top: nearest neighbor arc formed by nearest neighbor subsequence pairs $T_{100, 50}$ and $T_{450, 50}$ (in red). Bottom: arc count over time series $T$.
  • Figure 4: (a) Corrected density plots (CAD) of a sensor data containing anomalous defects (i.e. anomaly data) with ground truth defects (anomaly region) highlighted. (b) CAD of on the same data removing the defects part (i.e., clean data).
  • Figure 5: The x-axis on one of the 3D printer sensor datasets with red highlighting the area affected by the attack and blue highlighting the burn-in part of the data that was excluded.
  • ...and 1 more figures

Theorems & Definitions (5)

  • Definition 1
  • Definition 2
  • Definition 3
  • Definition 4
  • Definition 5