Explainable Anomaly Detection for Industrial IoT Data Streams
Ana Rita Paupério, Diogo Risca, Afonso Lourenço, Goreti Marreiros, Ricardo Martins
TL;DR
The paper addresses real-time anomaly detection in industrial IoT under limited labels by proposing a collaborative DSM framework that pairs an online, adaptive Onl-iForest with interactive learning via incremental PDPs and streaming ICE-based explanations. It introduces streaming feature-effect analysis and a FI score to render unsupervised anomalies actionable for operators. A Jacquard loom case study demonstrates how FI and PDPs illuminate fault-related patterns beyond raw anomaly scores, while acknowledging data sparsity and the need for continued work on bearing fault prediction. The approach aims to enhance interpretability, trust, and timely maintenance decisions in resource-constrained industrial settings.
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.