Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

CoopASD: Cooperative Machine Anomalous Sound Detection with Privacy Concerns

Anbai Jiang, Yuchen Shi, Pingyi Fan, Wei-Qiang Zhang, Jia Liu

TL;DR

A pre-trained model is employed as the backbone of the ASD model to improve its robustness and develop specialized techniques to stabilize the model under a completely non-iid and domain shift setting.

Abstract

Machine anomalous sound detection (ASD) has emerged as one of the most promising applications in the Industrial Internet of Things (IIoT) due to its unprecedented efficacy in mitigating risks of malfunctions and promoting production efficiency. Previous works mainly investigated the machine ASD task under centralized settings. However, developing the ASD system under decentralized settings is crucial in practice, since the machine data are dispersed in various factories and the data should not be explicitly shared due to privacy concerns. To enable these factories to cooperatively develop a scalable ASD model while preserving their privacy, we propose a novel framework named CoopASD, where each factory trains an ASD model on its local dataset, and a central server aggregates these local models periodically. We employ a pre-trained model as the backbone of the ASD model to improve its robustness and develop specialized techniques to stabilize the model under a completely non-iid and domain shift setting. Compared with previous state-of-the-art (SOTA) models trained in centralized settings, CoopASD showcases competitive results with negligible degradation of 0.08%. We also conduct extensive ablation studies to demonstrate the effectiveness of CoopASD.

CoopASD: Cooperative Machine Anomalous Sound Detection with Privacy Concerns

TL;DR

A pre-trained model is employed as the backbone of the ASD model to improve its robustness and develop specialized techniques to stabilize the model under a completely non-iid and domain shift setting.

Abstract

Machine anomalous sound detection (ASD) has emerged as one of the most promising applications in the Industrial Internet of Things (IIoT) due to its unprecedented efficacy in mitigating risks of malfunctions and promoting production efficiency. Previous works mainly investigated the machine ASD task under centralized settings. However, developing the ASD system under decentralized settings is crucial in practice, since the machine data are dispersed in various factories and the data should not be explicitly shared due to privacy concerns. To enable these factories to cooperatively develop a scalable ASD model while preserving their privacy, we propose a novel framework named CoopASD, where each factory trains an ASD model on its local dataset, and a central server aggregates these local models periodically. We employ a pre-trained model as the backbone of the ASD model to improve its robustness and develop specialized techniques to stabilize the model under a completely non-iid and domain shift setting. Compared with previous state-of-the-art (SOTA) models trained in centralized settings, CoopASD showcases competitive results with negligible degradation of 0.08%. We also conduct extensive ablation studies to demonstrate the effectiveness of CoopASD.
Paper Structure (18 sections, 6 equations, 4 figures, 4 tables, 2 algorithms)

This paper contains 18 sections, 6 equations, 4 figures, 4 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. Related Work
  3. Machine ASD in Centralized Settings
  4. Anomaly Detection in Decentralized Settings
  5. Proposed Method
  6. ASD Model
  7. Local Training Process
  8. Global Aggregation Process
  9. Experiment
  10. Dataset
  11. Hyperparameter Settings
  12. Experiment Results
  13. Ablation Study
  14. Centralized versus Decentralized
  15. Regularization Methods
  16. ...and 3 more sections

Figures (4)

  • Figure 1: Architecture of the ASD Model in CoopASD. The feature extractor $f(\cdot)$ is updated globally and shared among factories, while the linear classifier $c_i(\cdot)$ and KNN detector $g_i(\cdot)$ are uniquely constructed and preserved locally.
  • Figure 2: Training process viewed in parameter space. The red and blue ovals denote the sweet spots for anomaly detection and machine attribute classification respectively. The ASD model is trained by classifying machine attributes since labeled anomalies are not provided for training. Therefore, one can not tell when to stop training, and the model is likely to be overfitted or underfitted.
  • Figure 3: Training and detection procedures of CoopASD.
  • Figure 4: Comparison of different sampling probability