چكيده به لاتين
Given the important role of economic issues in today's world and the need to prevent sudden structural damages, which leads to a great loss of lives and properties, the science of structural health monitoring has reached a major turning point in its evolution. One of the branches of this science is using signal processing based methods. Moreover, the methods based on artificial intelligence are very popular today. Common methods of structural health monitoring in this area often use the signal processing and pre-processing stage, which leads to an increase in the volume of calculations and a decrease in the prediction speed. In addition, in most of these methods, the number of sensors in the structural health monitoring system is considered constantly. In this dissertation, an attempt has been made to design an efficient algorithm using machine learning methods to detect the failure in structures, so that by reducing the number of sensors required for health monitoring, the volume of computational operations is reduced. In this way, this method will be more cost-effective. Two important innovations in this research are the elimination of the signal processing stage and the consideration of sensors in the feature selection stage. In this study, by defining the term feature-sensor, despite other common methods, in addition to selecting the superior features, the best sensors have been determined as the second innovation and is accomplished for the first time on this bridge to take a big step in the direction of low cost and more efficient structural health monitoring. Finally, after implementing this method on the data of the benchmark structure, among the 16 sensors located on the deck of this bridge, 2 sensors were introduced as the best sensors and the feature of Shannon entropy was introduced as the best feature to be used as the most proper input for machine learning. In the end, the proposed method depicted very high accuracy and was more successful in comparison with other methods.
Keywords: structural health monitoring (SHM), damage detection, machine learning, classification
