Impact Load Localization Based on Multi-Scale Feature Fusion Convolutional Neural Network.

Area of Science:

• Engineering
• Machine Learning
• Structural Health Monitoring

Background:

• Impact load localization is crucial for complex structures like ships.
• Traditional methods often require manual feature extraction, which can be complex and time-consuming.

Purpose of the Study:

• To propose a multi-scale feature fusion convolutional neural network (MSFF-CNN) for automated impact load localization.
• To improve the accuracy and efficiency of impact load identification and location in ship structures.

Main Methods:

• An end-to-end machine learning model directly processes raw vibration signals.
• Utilizes four independent convolutional layers with varying kernel sizes for automatic feature learning and concatenation.
• Employs data normalization and L2 regularization to enhance data and prevent overfitting.
• A softmax classification layer is used for classification and localization.

Main Results:

• The MSFF-CNN achieved a classification and localization accuracy of 94.29% on a ship's stern compartment model.
• Demonstrated improved feature extraction capabilities, combining local perception and global vision.
• Outperformed traditional Convolutional Neural Network (CNN) approaches.

Conclusions:

• The MSFF-CNN method effectively enhances the classification ability for impact loads.
• The approach shows significant potential for practical engineering applications in structural health monitoring.