رگرسیون بردار پشتیبانی برای نظارت بر بهداشت بر روی خط ساختارها در مقیاس بزرگ
|کد مقاله||سال انتشار||مقاله انگلیسی||ترجمه فارسی||تعداد کلمات|
|24804||2006||15 صفحه PDF||سفارش دهید||محاسبه نشده|
Publisher : Elsevier - Science Direct (الزویر - ساینس دایرکت)
Journal : Structural Safety, Volume 28, Issue 4, September 2006, Pages 392–406
Large-scale, structural health monitoring remains a challenge especially when I/O measurement data are contaminated by high-level noise. A novel approach that uses incremental support vector regression (SVR), a promising statistics technology, is proposed for large-scale, structural health monitoring. Due to the potential properties of this novel SVR, the SVR-based approach makes structural health monitoring accurately and robustly. A sub-structure strategy is utilized to reduce the number of unknown parameters in the health monitoring formula, thereby making large-scale structural health monitoring possible. Lastly, an incremental SVR training algorithm adopted for the SVR-based approach not only markedly reduces computation time, but identifies structural parameters on-line. Numerical examples show that results of this SVR-based approach for large-scale structural health monitoring are accurate and robust, even when observed data are contaminated with different kinds and intensity levels of noise.
Civil, mechanical, or aerospace structures may be damaged during strong natural disasters or rapidly deteriorate because of age, underlining the need for a robust methodology to monitor structural health. Although much attention has been paid to damage detection problems using such methods as neural networks and statistic filters ,  and , structural health monitoring remains a challenge, particularly when input and output (I/O) measurements are contaminated by high-level noise. Related function estimation methods therefore need further investigation to ensure robust structural health monitoring. For linear regression problems, robustness statistics aim at describing the structure best fitting the bulk of the empirical data using the function where (xi, yi) is a set of measurements (i = 1, …, N), w a regression coefficient vector, and b the model offset. The classical regression approach is the least squares (LS) method, which provides easy computation but is sensitive to outliers, i.e., a data point that is located far from the rest of the data. Support vector regression (SVR) is a novel statistical technology with robustness. A least squares version (LS-SVR) extends the SVR theory. It offers a concise formulation and rapid computation, but is not robust to non-Gaussian noise ,  and . A comparison of these regression methods (LS, SVR, and LS-SVR) is provided to clarify their basic concepts and mathematical backgrounds. Next, an SVR-based approach for large-scale structural health monitoring by a sub-structural strategy is suggested. Lastly, a 30-DOF shear-building and a 52-DOF truss are used as structural health monitoring examples to show the efficiency of the proposed approach.
نتیجه گیری انگلیسی
An incremental SVR-based approach is proposed for structural health monitoring. A 30-DOF shear building structure and a 52-DOF truss were investigated, which demonstrated that the proposed approach identifies structural parameters rapidly, accurately, and robustly, even when observed data are polluted by different kinds and intensities of noise. This novel SVR-based approach has the following features: 1. It is efficacious for structural health monitoring, and has robust properties. The ‘Max-Margin’ idea and the novel ε-insensitive loss function utilized in the SVR formulations gives the SVR excellent robust capabilities, providing accurate and robust identified structural parameters. 2. An incremental training algorithm is incorporated to train the SVR formulation sequentially. This not only significantly reduces computation time, but confer structural health monitoring on-line. A local identification strategy used in the proposed approach makes large-scale structural health monitoring possible. By dividing the entire structure into smaller sub-structures, the SVR works in a reduced dimension, guaranteeing that the proposed approach works rapidly for large-scale structural identification. In addition, it is an ‘output-only’ identification technique when the excitation force is outside the sub-structure studied, because no force measurement is required. In brief, this proposed technique gives special robust performances for large-scale structural health monitoring, even when I/O data are corrupted with high-level, non-Gaussian noise. Certain aspects require further study, such as extending the SVR-based approach to time-varying structural identifications.