Attention-Fused Staged DWT-LSTM for Fault Diagnosis of Embedded Sensors in Asphalt Pavement

Jiarui Zhang ^1,2, Haihui Duan ^1,2, Songtao Lv ^1,2

⁰National Engineering Research Center of Highway Maintenance Technology, Changsha University of Science & Technology, Changsha 410114, China.

Materials (basel, Switzerland) +

|

August 28, 2025

View abstract on PubMed

Summary

This study introduces an intelligent framework using Discrete Wavelet Transform and a staged, attention-based Long Short-Term Memory network for accurate asphalt pavement sensor fault diagnosis. The method achieves 98.89% accuracy, outperforming others and identifying complex faults.

Area Of Science

Civil Engineering
Sensor Technology
Artificial Intelligence

Background

Embedded sensors in asphalt pavement are crucial for monitoring, but suffer from data scarcity and difficulty in identifying compound faults.
Existing fault diagnosis methods struggle with complex sensor failures, compromising pavement monitoring reliability.

Purpose Of The Study

To develop an intelligent diagnostic framework for embedded asphalt pavement sensors.
To address challenges in identifying short-term, long-term, and compound sensor faults.
To improve the reliability and accuracy of pavement monitoring data.

Main Methods

A "Decomposition-Focus-Fusion" framework integrating Discrete Wavelet Transform (DWT) and a staged, attention-based Long Short-Term Memory (LSTM) network.
DWT for multi-scale feature extraction; Bidirectional LSTM (Bi-LSTM) and stacked LSTM for short-term and long-term fault characteristics, respectively.
An attention network for intelligent weighting and fusion of sub-model outputs to classify eight sensor operational states.

Main Results

The proposed framework achieved a mean diagnostic accuracy of 98.89% (±0.0040) via 5-fold cross-validation.
Significantly outperformed baseline models (SVM, KNN, unified LSTM) in accuracy and compound fault identification.
Demonstrated rapid response to dynamic faults and computational efficiency for real-time monitoring.

Conclusions

The integrated DWT and attention-based LSTM framework provides a precise and robust solution for asphalt pavement sensor fault diagnosis.
Each component of the "Decomposition-Focus-Fusion" architecture is critical for the model's superior performance.
The framework effectively distinguishes challenging fault types and complex compound faults, enhancing pavement monitoring reliability.

Keywords:

DWT-LSTM asphalt pavement attention mechanism fault diagnosis sensor state monitoring