Overall grouting compactness detection of bridge prestressed bellows based on RF feature selection and the GA-SVM model

Since the grouting quality of a bridge bellows directly affects the safety and durability of an entire bridge structure, it is crucial to study efficient and accurate detection methods. With most methods focusing on horizontal detection, there has been very little research on longitudinal detection that can identify overall grouting compactness quickly. Therefore, this study proposes a longitudinal transmission detection method based on random forest (RF) feature selection and the genetic algorithm optimization support vector machine (GA-SVM) model. In this method, first, the waveform data of the channels with different grouting density in the model test were collected and processed. Second, the time domain and frequency domain features of the waveform were extracted. Third, the random forest was used to screen out the important features. Finally, the SVM model whose parameters were optimized by GA was trained and used for prediction. In this study, eight features out of forty features were screened out by RF, and their feature importance reached 97.45%. Moreover, the model prediction effect after feature selection was significantly better than that before selection. Compared with other optimization algorithms and prediction models, the GA-SVM, whose mean square error (MSE) for the test set was only 0.0013 and whose goodness of fit reached 0.9825, had the best prediction effect. This method could significantly improve the detection efficiency and prepare for defect location and quantification.

Automated summary

The study proposes a longitudinal transmission detection method based on RF feature selection and GA-SVM model, which quickly detects the grouting compactness of bridge bellows by screening important features and optimizing parameters. Eight important features were selected out of 40, and the prediction effect was better compared to other optimization algorithms and models.