A Hybrid Machine Learning Based Data Fusion Strategy for Detection of Structural Damage

Abstract

One of the crucial aspects of ensuring safety and integrity in huge infrastructure is structural health monitoring. Recent advances of machine learning hybrid models have shown promising behaviour in enhancing both accuracy and resilience in structural health monitoring. The researchers have proposed various models for detection of the structural damages. However, the successful detection rate along with reliability is not up to the mark. The paper proposes a new hybrid model of CNN and LSTM networks that improve the detection and identification of damage in intricate structures. This model has made it possible to integrate the feature extraction capabilities of CNNs with the caparbilities of LSTMs in terms of temporal analysis. In this way, structural abnormalities could be detected accurately as time progresses. The proposed approach was tested on a multi-sensor dataset containing numerous damage scenarios which includes no damage, minor cracks, slight cracks, and excessive structural failures. With 1200 instances, the dataset became split into 70% for training, 20% for validation, and 10% for checking out. The version confirmed enormous improvements, attaining an accuracy of 85.6%, lack of 0.12, precision of 90%, bear in mind of 88%, F1 score of 89.5%, and an AUC of 0.94. Furthermore, the false bad price and false effective price were drastically reduced in comparison to conventional methods. Additionally, the hybrid model outperformed Probabilistic Neural Networks (PNN), which only finished an accuracy of 85% and an AUC of zero.87. The CNN-LSTM model's robustness in dealing with nonlinearities and its ability to perform successfully under noisy or incomplete facts make it enormously reliable for actual-global applications. Moreover, the automated characteristic extraction method eliminates the need for manual function engineering, simplifying implementation. Future paintings should consciousness on refining the model for real-time monitoring structures and extending its application to various different structural sorts.