Thermal and Depth Fusion for Predictive Maintenance: A Gated Cross-Modal Attention Framework
Abstract—Predictive maintenance (PdM) in industrial environments demands accurate, early fault detection across heterogeneous equipment types. Single-modality sensing approaches—whether thermographic or depth-based—fail to capture the full spectrum of fault precursors inherent to rotating machinery, electrical systems, and pressurised infrastructure. This paper proposes TDFNet, a Thermal-Depth Fusion Network featuring a novel Gated Cross-Modal Attention (GCMA) mechanism that dynamically weights the contribution of co-registered infrared thermal and structured-light depth streams based on operational context. A temporal convolutional network (TCN) module with dilated causal convolutions models degradation trends across eight-frame inspection windows. We introduce ThermDepth-Industrial, a benchmark dataset comprising 47,200 annotated thermal-depth frame pairs spanning seven fault categories from three industrial facilities. TDFNet achieves 94.7% test accuracy and an AUC of 0.989, surpassing all unimodal and fusion baselines by a margin of 2.3 to 13.0 percentage points. The proposed system meets real-time deployment constraints, running at 28.3 fps on an NVIDIA Jetson AGX Xavier with TensorRT INT8 quantisation. Ablation experiments confirm the independent contribution of GCMA, TCN, and heterogeneous pre-training to overall system performance. Dataset and code are made publicly available to facilitate reproducibility.
Registration ID: IJVRA_702423 Published ID: IJVRA2603770
"Thermal and Depth Fusion for Predictive Maintenance: A Gated Cross-Modal Attention Framework", IJVRA - International Journal of Versatile Research and Analysis (www.IJVRA.org), ISSN:2984-8903, Vol.4, Issue 3, page no.937-947, March-2026, Available :https://ijpub.org/ijvra/papers/IJVRA2603770.pdf
Paper Reg. ID: IJVRA_702423
Published Paper Id: IJVRA2603770
Research Area: Other area not in list
Country: dhule, Maharashtra, India
ISSN: 2984-8903 | IMPACT FACTOR: 9.12 Calculated By Google Scholar | ESTD YEAR: 2023
An International UGC CARE JOURNAL PUBLICATION Low Cost (₹599), Scholarly Open Access, Peer-Reviewed, Refereed Journal Impact Factor 9.12 Calculate by Google Scholar and Semantic Scholar | AI-Powered Research Tool, Multidisciplinary, Monthly, Multilanguage, Crossref DOI Member Journal Indexing in All Major Database & Metadata, Citation Generator
Publisher: IJVRA (IJ Publication) Janvi Wave
© 2026 - Authors hold the copyright of this article. This work is licensed under a Creative Commons Attribution 4.0 International License and The Open Definition.
This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (CC BY 4.0). 🛡️ Disclaimer:
The content, data, and findings in this article are based on the authors’ research and have been peer-reviewed for academic purposes only. Readers are advised to verify all information before practical or commercial use. The journal and its editorial board are not liable for any errors, losses, or consequences arising from its use.
: Click Here to Get
About Hard Copy and Transparent Peer Review Report
