Structural damage identification using phase space matrix with convolutional neural networks

Abstract

Structural health monitoring is an important field for ensuring the safety and reliability of structures. This paper proposes a structural damage identification approach using phase space matrix with Convolutional Neural Networks (CNNs). The proposed approach has two key advantages. First, it is highly sensitive to structural damage, while remaining robust to measurement noise and modelling uncertainties. This is achieved by transforming the raw sensor data into a multi-dimensional phase space matrix via delay-coordinate embedding, which captures the underlying dynamics of the structure. The developed CNN model, incorporating convolutional layers, batch normalization and the Rectified Linear Unit (ReLU) activation, is then trained to identify both the location and severity of structural damage. Second, the proposed approach only requires a few sensor measurements to identify structural damage, making it cost-effective and practical for real-world applications. The effectiveness and performance of the proposed approach are investigated through numerical simulations of an eight-story shear-type steel frame model and experimental verifications on a steel frame structure under hammer impact excitations. The results demonstrate that the proposed approach achieves a high accuracy in detecting structural damage locations and quantifying damage severity, even in the presence of high levels of measurement noise and finite element modelling uncertainty. The proposed approach shows a great potential for enhancing the efficiency and accuracy of structural health monitoring and damage detection for structures with only limited sensor measurement data.