Trustworthy Equipment Monitoring via Cascaded Anomaly Detection and Saliency-Guided Inspection

Predictive maintenance demands accurate anomaly detection and actionable, interpretable explanations. While fusing sensor time-series with thermal imagery is promising, naive fusion can actually degrade performance. This paper provides evidence-based guidelines for interpretability-driven model selection, demonstrating that cascaded architectures outperform end-to-end fusion in industrial monitoring. We propose a two-stage approach: Stage 1 uses a Random Forest on statistical sensor features for robust anomaly detection (94.70% macro F1), while Stage 2 employs a convolutional thermal encoder with spatial attention for post-detection inspection. Rigorous analysis reveals traditional machine learning significantly outperforms deep learning baselines (Cohen's d = 3.04-8.81) on low-noise, filtered sensor data. Additionally, we introduce a diagnostic protocol using gate weight analysis to quantify modality bias, preventing over-reliance on visually rich but less informative data. Our explainability pipeline integrates Shapley-based sensor ranking with spatial attention. Finally, perturbation audits confirm thermal attention acts as a spatial regularizer, with the thermal encoder achieving a 78.49% fault-detection F1, demonstrating learned sensitivity to fault presence despite poor severity grading.

• Publication year

  2025

• Venue

  Unknown venue

• Publication date

  2025-12-31

• Fields of study

  Computer Science, Engineering

• Identifiers
  arXiv 2512.24755
• External record

  Open on Semantic Scholar

• Source metadata

  Semantic Scholar

• No claims are published for this paper.

• No concepts are published for this paper.

• Fast Anomaly Detection for Vision-Based Industrial Inspection Using Cascades of Null Subspace PCA Detectors

  2025cited by this paper
• Cross-modal gated feature enhancement for multimodal emotion recognition in conversations

  2025cited by this paper
• Mitigating modality imbalance in multimodal sentiment analysis via emotion-enriched visual encoding and pyramid gated fusion

  2025cited by this paper
• UAV Anomaly Detection Model Based on Integrated Multi-Modal Neural Network and Neural Architecture Search

  2024cited by this paper
• Regulating Modality Utilization within Multimodal Fusion Networks

  2024cited by this paper
• Balancing Multimodal Learning via Online Logit Modulation

  2024cited by this paper
• Pre-gating and contextual attention gate - A new fusion method for multi-modal data tasks

  2024cited by this paper
• Classifier-guided Gradient Modulation for Enhanced Multimodal Learning

  2024cited by this paper
• Twin Graph-Based Anomaly Detection via Attentive Multi-Modal Learning for Microservice System

  2023cited by this paper
• Multimodal Representation Learning by Alternating Unimodal Adaptation

  2023cited by this paper
• Balanced Multimodal Learning via On-the-fly Gradient Modulation

  2022cited by this paper
• Evaluating Weakly Supervised Object Localization Methods Right

  2020cited by this paper
• A Survey on Deep Learning for Multimodal Data Fusion

  2020cited by this paper
• Applications of machine learning to machine fault diagnosis: A review and roadmap

  2020cited by this paper
• From local explanations to global understanding with explainable AI for trees

  2020cited by this paper
• A systematic literature review of machine learning methods applied to predictive maintenance

  2019cited by this paper
• Explainable Artificial Intelligence (XAI): Concepts, Taxonomies, Opportunities and Challenges toward Responsible AI

  2019cited by this paper
• Deep learning and its applications to machine health monitoring

  2019cited by this paper
• Multimodal Transformer for Unaligned Multimodal Language Sequences

  2019cited by this paper
• RISE: Randomized Input Sampling for Explanation of Black-box Models

  2018cited by this paper
• Stop explaining black box machine learning models for high stakes decisions and use interpretable models instead

  2018cited by this paper
• Deep learning for time series classification: a review

  2018cited by this paper
• Deep learning for smart manufacturing: Methods and applications

  2018cited by this paper
• Cascade R-CNN: Delving Into High Quality Object Detection

  2017cited by this paper
• Multimodal Machine Learning: A Survey and Taxonomy

  2017cited by this paper
• Attention is All you Need

  2017cited by this paper
• A New Deep Learning Model for Fault Diagnosis with Good Anti-Noise and Domain Adaptation Ability on Raw Vibration Signals

  2017cited by this paper
• A Unified Approach to Interpreting Model Predictions

  2017cited by this paper
• Diagnosis of the three-phase induction motor using thermal imaging

  2017cited by this paper
• Decoupled Weight Decay Regularization

  2017cited by this paper
• Condition Monitoring of Bearing Damage in Electromechanical Drive Systems by Using Motor Current Signals of Electric Motors: A Benchmark Data Set for Data-Driven Classification

  2016cited by this paper
• Grad-CAM: Visual Explanations from Deep Networks via Gradient-Based Localization

  2016cited by this paper
• Deep Residual Learning for Image Recognition

  2015cited by this paper
• Learning Deep Features for Discriminative Localization

  2015cited by this paper
• Neural Machine Translation by Jointly Learning to Align and Translate

  2014cited by this paper
• Infrared thermography for condition monitoring – A review

  2013cited by this paper
• PRONOSTIA : An experimental platform for bearings accelerated degradation tests.

  2012cited by this paper
• Multimodal Deep Learning

  2011cited by this paper
• Scikit-learn: Machine Learning in Python

  2011cited by this paper
• Random Forests

  2001cited by this paper
• Long Short-Term Memory

  1997cited by this paper
• An introduction to predictive maintenance

  1989cited by this paper

• No citing papers are available for this paper.