ارائه روش‌ هوشمند ارتعاش-پایه جهت شناسایی آسیبها در تیرهای فولادی بر اساس توزیع شبه ویگنر-ویل هموار شده و شبکه عصبی مصنوعی

Structural health monitoring (SHM) is a health assessment process that uses an automatic monitoring system to use cost-effective strategies for condition-based maintenance of structures. The number of structures is increasing, and the majority of structures have deteriorated and worn out due to the lack of use of health measurement during operation. Due to the importance of the subject, during the previous decades, detailed research in the field of vibration-based methods for health measurement of structures are increasing. In this research, a new methodology for detecting damage using signal processing in steel beams will be investigated.
In order to control the presented method, experiments were conducted in the laboratory on a steel beam which was an IPE 14 steel beam. To conduct the experiments, accelerometer sensors were installed under the steel beam. Then, an exciting load was applied to the steel beam and its vibrations were recorded with the sensors. 7 different damage scenarios were considered and damages were created in the steel beam. Similar to the healthy state, vibration-base tests were conducted on the steel beam with different damage scenarios. Considering the need for multiple data to train the artificial neural network, the steel beam was modeled in the Abaqus finite element software. In fact, the model was created completely based on the laboratory. In order to ensure the model of the model components, the results of the model analysis were evaluated and compared with the results obtained in the laboratory and the accuracy of the performance was ensured.
After ensuring the correct performance of the finite element model, 80 damage scenarios were defined. The scenarios were created in the finite element model and after analysis, the vibrations were recorded. The vibrations were processed with a Smoothed Pseudo Wigner-Ville Distribution (SPWVD). SPWVD is a time-frequency analysis tool designed to improve the readability of the classical Wigner-Ville Distribution by reducing its inherent cross-term interference while maintaining good time-frequency resolution. In fact, The SPWVD is a powerful tool for high-resolution time-frequency analysis when cross-term reduction is needed. Signal processing was performed in Matlab software. Thus, time-frequency plans were calculated. The obtained characteristics were used as input for training, evaluation and testing of the neural network.
To evaluate the error of the network used, the regression correlation diagram and the RMSE and MSE criteria were used. The results obtained from the test data and evaluation of the neural network indicate a very low error rate in damage identification and it can be concluded that the results of the calculations are very accurate and close to the expected numbers.
To identify damage in the specified locations, prediction scenarios were used and the network performance in identifying damage was evaluated under 4 scenarios. The results of this section also show that there is less than 1 percent error in identifying each damage, which is very high accuracy. In this study, the ability of the proposed method to identify multiple damages simultaneously was also evaluated. The proposed method is able to identify multiple damages simultaneously with very good accuracy. However, considering the capabilities of the proposed method, its use for detecting damage in steel beams is recommended.