Integrated Damage Location Diagnosis of Frame Structure Based on Convolutional Neural Network with Inception Module

Area of Science:

• Structural Engineering
• Artificial Intelligence
• Vibration Analysis

Background:

• Accurate damage location diagnosis is crucial for frame structure maintenance.
• Vibration data similarity and noise interference pose significant challenges.
• Existing methods struggle with complex structures and noisy environments.

Purpose of the Study:

• To develop a high-precision, noise-resistant neural network for frame structure damage location.
• To improve upon existing convolutional neural network models for fault diagnosis.
• To propose an integrated method to overcome single-sensor data limitations.

Main Methods:

• An improved convolutional neural network, named BICNN (convolutional neural network based on Inception), was developed by integrating the Inception module into TICNN.
• An integrated damage location diagnosis method was proposed to prevent misjudgments from single sensor data.
• The proposed method was tested on a four-story steel frame model from the University of British Columbia.

Main Results:

• The BICNN model achieved a diagnosis accuracy of 97.38%, surpassing other tested methods.
• The proposed method demonstrated significant advantages in noise resistance.
• The integrated approach effectively addressed the misjudgment issue associated with single sensor data.

Conclusions:

• The proposed BICNN method offers high accuracy and strong anti-noise ability for damage location diagnosis in frame structures.
• This approach is effective in solving accurate damage location diagnosis problems in complex frame structures, even under strong noise conditions.
• The study highlights the potential of advanced neural network architectures for structural health monitoring.