Explainable Anomaly Detection for Industrial IoT Data Streams

2025-12-09

AI Summary

To address the challenges of constrained computational resources and delayed/missing labels in industrial IoT data streams—rendering supervised methods ineffective—this paper proposes an online unsupervised anomaly detection and interpretable decision-making framework for real-time operational maintenance. Methodologically, it integrates a lightweight online Isolation Forest with a novel incremental Partial Dependence Plot (iPDP) and a dynamic feature importance assessment mechanism based on ICE curve shifts, enabling adaptive threshold tuning and root-cause attribution. An edge-optimized real-time inference framework ensures low-latency deployment. Evaluated on a Jacquard loom unit, the approach achieves millisecond-level detection and interpretable early warning of bearing failures, significantly enhancing detection robustness and operational trustworthiness under weakly supervised conditions.

Technology Category

• Data Mining & Knowledge Management: Anomaly/Outlier Detection
• Machine Learning: Calibration & Uncertainty Quantification
• Knowledge Representation and Reasoning: Diagnosis and Abductive Reasoning

Application Category

• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• Economics, Online Markets and Human Computation: Data quality aspects of human-annotated datasets
• Responsible Web: Machine-in-the-loop, human agency and autonomy

Abstract

Industrial maintenance is being transformed by the Internet of Things and edge computing, generating continuous data streams that demand real-time, adaptive decision-making under limited computational resources. While data stream mining (DSM) addresses this challenge, most methods assume fully supervised settings, yet in practice, ground-truth labels are often delayed or unavailable. This paper presents a collaborative DSM framework that integrates unsupervised anomaly detection with interactive, human-in-the-loop learning to support maintenance decisions. We employ an online Isolation Forest and enhance interpretability using incremental Partial Dependence Plots and a feature importance score, derived from deviations of Individual Conditional Expectation curves from a fading average, enabling users to dynamically reassess feature relevance and adjust anomaly thresholds. We describe the real-time implementation and provide initial results for fault detection in a Jacquard loom unit. Ongoing work targets continuous monitoring to predict and explain imminent bearing failures.

Problem

• Develops an unsupervised anomaly detection framework for IoT data streams
• Integrates human-in-the-loop learning to handle delayed or missing labels
• Enhances interpretability with incremental feature importance and threshold adjustment

Innovation

• Unsupervised anomaly detection with human-in-the-loop learning
• Online Isolation Forest with incremental interpretability features
• Real-time feature relevance assessment via deviation-based scoring