When does a bridge become an aeroplane?

Tina A. Dardeno,Lawrence A. Bull,Nikolaos Dervilis,Keith Worden

2024-11-27

Abstract:Despite recent advances in population-based structural health monitoring (PBSHM), knowledge transfer between highly-disparate structures (i.e., heterogeneous populations) remains a challenge. It has been proposed that heterogeneous transfer may be accomplished via intermediate structures that bridge the gap in information between the structures of interest. A key aspect of the technique is the idea that by varying parameters such as material properties and geometry, one structure can be continuously morphed into another. The current work demonstrates the development of these interpolating structures, via case studies involving the parameterisation of (and transfer between) a simple, simulated 'bridge' and 'aeroplane'. The facetious question 'When is a bridge not an aeroplane?' has been previously asked in the context of predicting positive transfer based on structural similarity. While the obvious answer to this question is 'Always,' the current work demonstrates that in some cases positive transfer can be achieved between highly-disparate systems.

Machine Learning

What problem does this paper attempt to address?

The problem that this paper attempts to solve is to achieve effective knowledge transfer between highly heterogeneous structures (i.e., structures with great differences). Specifically, the author explores how to bridge the information gap between two completely different structures, such as bridges and airplanes, by introducing intermediate structures, thereby achieving the effective transfer of data and features from one structure to another.

### Problem Background

In the field of Structural Health Monitoring (SHM), especially Population - Based Structural Health Monitoring (PBSHM), knowledge transfer between different structures is a challenge. Traditional linear machine - learning methods have difficulty dealing with non - Euclidean data, while geometric methods are more suitable for handling these complex, curved manifold structures. When the difference between two structures is very large, direct knowledge transfer often has poor results.

### Solution

The author proposes a method to achieve progressive transfer from one structure to another by generating a series of parameterized intermediate structures. These intermediate structures can gradually transition from one structure to another by adjusting material properties, geometric shapes, and other parameters. The core ideas of this method are:

1. **Parameterized Modeling**: Generate a series of continuously changing intermediate structures by changing parameters such as material properties and geometric shapes.
2. **Progressive Transfer**: Use Support Vector Machine (SVM) classifiers and normal - condition alignment techniques to gradually transfer knowledge from the source structure to the target structure.
3. **Kernel Function Selection**: Use linear kernels and Geodesic Flow Kernels (GFK) for classification to improve the transfer effect.

### Experimental Design

To verify the effectiveness of this method, the author conducted the following experiments:

- **Model Development**: Generated a series of intermediate structure models from simple bridges to simplified airplanes.
- **Data Generation**: Generated data sets for normal and damaged conditions for each structure and added noise to simulate the actual environment.
- **Transfer Strategy**: Used normal - condition alignment and SVM classifiers (using linear kernels and geodesic flow kernels respectively) for knowledge transfer.

### Experimental Results

- **Linear Kernel**: As the number of intermediate structures increases, the transfer accuracy improves significantly. When using 78 intermediate structures, the prediction accuracy rate reached 99.5%.
- **Geodesic Flow Kernel**: Regardless of the number of intermediate structures used, the performance of the geodesic flow kernel is better than that of the linear kernel. In particular, when using 13 or more intermediate structures, the average prediction accuracy rate reached 100%.

### Conclusion

This research shows that by introducing intermediate structures, effective knowledge transfer can be achieved between highly heterogeneous structures, especially between structures with great differences such as bridges and airplanes. This provides theoretical and technical support for the application of similar transfer - learning methods in more complex structural health monitoring tasks in the future.

### Formula Summary

- **Geodesic Flow Kernel Formula**:
\[
\langle z^\infty_i, z^\infty_j \rangle = \int_0^1 (\Phi(t)^T x_i)^T (\Phi(t)^T x_j) \, dt = x_i^T G x_j
\]
where \( G \) is a positive semi - definite matrix, defined as:
\[
G = Q \begin{bmatrix} V_1 & 0 \\ 0 & -\tilde{V}_2 \end{bmatrix} \begin{bmatrix} \Lambda_1 & \Lambda_2 \\ \Lambda_2 & \Lambda_3 \end{bmatrix} \begin{bmatrix} V_1^T & 0 \\ 0 & -\tilde{V}_2^T \end{bmatrix} Q^T
\]
where the elements of \( \Lambda_1, \Lambda_2, \Lambda_3 \) are respectively:
\[
\lambda_{1i} = \frac{1 + \sin(2\theta_i)}{2\theta_i}, \quad \lambda_{2i} = \fra