A useful summary of vibration-based SHM systems in aviation can be found in. The authors of applied the developed SHM system for monitoring an aircraft wing, while Mironov and Doronkin developed a system for monitoring composite helicopter blades based on their operational modal analysis. Examples of such systems can be found in. Special attention to vibration-based SHM systems is devoted to aircraft structures. Numerous applications of vibration-based SHM systems can be found in civil engineering, primarily in the monitoring of bridges and other civil structures. As examples of vibration-based SHM systems, several studies are available in the literature. An overview of such systems can be found in review papers. In the literature, numerous types and configurations of vibration-based SHM systems can be found, including purely experimental studies, numerical model-supported systems, or even systems based on a digital twin concept.
SHM systems based on this method use vibration measurements for structural evaluation, and the common ones include the following: natural frequencies, mode shapes or curvatures, or damping ratios. One of the oldest and most widespread SHM methods applied in numerous practical applications is the modal-based one. Examples of algorithms include the following: neural network classifiers, support vector machines, linear and quadratic discriminant analysis, kernel discriminant analysis, and nearest-neighbor classifiers. A possible output of such an assessment algorithm within an SHM system can be considered as a discrete class label, which represents Cartesian coordinates of location of damage sites and damage extent represented, e.g., by the loss of stiffness. In the same way, damage severity is assessed by assigning class labels to data, which correspond to various levels of damage extent. Thus, the damage localization is performed by an analysis of specific predefined substructures by assigning class labels for the data, which corresponds to damage in a given substructure. In supervised learning, the training data consist of both a set of feature vectors as well as their known class labels. In contrast to this, a supervised learning scheme provides the possibility of damage identification, i.e., it makes it possible to detect, localize, and assess the severity of damages.
These methods are focused on the detection of damage using a comparison of features that describe the condition of the intact and damaged structure. The unsupervised learning scheme leads to clustering analysis and some novelty detection methods, such as outlier analysis, kernel density methods, and auto associative neural networks. Their architecture and learning process are mainly dependent on the necessary level of damage identification. Pattern recognition methods can be classified into two learning (or training) schemes: supervised and unsupervised. In data-driven SHM, damage detection can be regarded as a problem of pattern recognition. Using the proposed approach, it was possible to precisely detect damage based on a limited number of strain sensors and mode shapes taken into consideration, which leads to efficient structural health monitoring with resource savings both in costs and computational time and complexity. The evaluation of selected sensor networks was performed and validated using machine learning techniques and visualized appropriately. The results of the calculations according to sensor placement methods were subsets of possible sensor network candidates, which were evaluated using the aggregation of different metrics. The advantage of the proposed approach is that information for sensor placement was used only from the structure’s healthy state. To solve this problem, different sensor placement methods with different constraint variants were applied. In this study, the problem of optimal sensor placement was investigated for a composite plate with simulated internal damage. Due to this, it is still an open problem to find a compromise between these two parameters.
Optimal sensor placement is one of the important issues in monitoring the condition of structures, which has a major influence on monitoring system performance and cost.
0 kommentar(er)
