نوع مقاله : مقاله پژوهشی
نویسندگان
1 گروه مهندسی عمران، دانشکده فنی، دانشگاه خوارزمی
2 گروه عمران، دانشکده فنی و مهندسی دانشگاه خوارزمی تهران ایران
3 استادیار / دانشگاه خوارزمی
چکیده
کلیدواژهها
موضوعات
عنوان مقاله [English]
نویسندگان [English]
The challenge of identifying damage-indicative features that are robust to noise in time-series data has long hindered the effectiveness of traditional machine learning-based structural damage detection methods. The time-consuming preprocessing procedures required by these methods also contributed to reduced accuracy and performance. However, the advent of deep learning has led to an increased exploration of the use of deep architectures in structural health monitoring. Despite these advancements, many deep learning approaches are resource-intensive and require significant data for training, making them less feasible for real-time applications. As a solution, we propose using 2-D convolutional neural networks (2-D CNNs) that integrate feature extraction and classification into a single entity. Our method employs a network of lighted CNNs instead of deep ones and utilizes raw acceleration signals as input, overcoming the limitations of previous approaches. Using lighted CNNs, in which everyone is optimized for a specific element, increases the accuracy and makes the network faster to perform. Also, a new framework is proposed for decreasing the data required in the training phase. We verified our method on Qatar University Grandstand Simulator (QUGS) benchmark data provided by Structural Dynamics Team. The results showed an improvement in accuracy over other methods, and the running time was adequate for real-time applications.
کلیدواژهها [English]
)