A Fault Diagnosis and Prediction Method for Wireless Sensor Networks Based on Attention Mechanism Fused Neural Networks

ABSTRACT The industrial internet of things (IIoT) confronts dual challenges in wireless sensing networks (WSN): achieving interoperability among heterogeneous devices while ensuring mission‐critical security and ultrareliable low latency communication (URLLC). Current solutions exhibit two key limitations: inability to capture multiscale temporal dependencies in nonstationary sensor streams and lack of physics aware deep learning architectures for security sensitive predictive maintenance. To address these gaps, this article proposes IIoT‐FSO, an intelligent framework integrating deep learning with free‐space optical (FSO) channel dynamics. Leveraging FSO strengths (GHz level modulation bandwidth, millimeter precision beam confinement, and electromagnetic resilience), the framework establishes new benchmarks for atmospheric channel modeling in adaptive photonic systems. A core innovation lies in a hybrid neural architecture that combines gated recurrent units (GRUs), long short‐term memory (LSTM) networks and attention mechanisms. This architecture integrates spatial pattern recognition capabilities of CNNs, temporal dynamics modeling by GRU and LSTM (for capturing spatiotemporal dependencies), and adaptive weighting of critical turbulence features across scales via the attention mechanism. Experimental results demonstrate significant performance improvements in FSO communication link prediction. Multivariate analysis (attenuation, additional losses, and beam divergence based on sliding window algorithm) shows that the attention‐LSTM model achieves 76.83% accuracy, while the CNN‐GRU‐LSTM architecture reaches 80.13% accuracy, both outperforming conventional methods. This article advances the technical readiness of IIoT for secure and low latency industrial applications, highlighting its potential to reshape smart manufacturing paradigms enabled by WSN. A neural network model is based on attention mechanisms for the industrial internet of things free‐space optical (IIoT‐FSO) framework. The model effectively captures the dynamic variations of the atmospheric channel and enhances the reliability of wireless sensor networks through multiscale spatiotemporal feature extraction.