Pattern Classification And Clustering Recognition Research On Bridge Structural Damage Identification

Because bridge`s high engineering capacity, cost and important in the transporteconomy, and ensure compliance with life safety standards during its operation, it isnecessary to do early structural damage identification and assessment for it. Bridgedamage and extent of identification is the core of the study. When the modal feature isnot known, pattern recognition is extensive research, and is a typical method fordamage detection of the used to help. Pattern recognition is implicit identifyingrelevant and irrelevant variable nonlinear relationship skills, with self-learning andfault-tolerant capabilities. Advantages that enable it to properly minimize responsemeasurement, and the negative effects of structural finite element model. For largestructures like bridges, we are obviously not going to get to know each individual,which means that each of the measured part of state, but through pattern recognitionand effective realization of bridge damage through fast, accurate and intelligentrecognition, so as to guarantee the safety of major engineering structures like bridges,completeness, suitability and durability. Yet there has been little system uses a rangeof pattern recognition algorithm of static and dynamic study on damage detection ofbridge. The full text in this area of research as follows:1. Using pattern recognition of bridge damage identification based on theassumption that the authenticities of the data, but those data are massive, itseffectiveness is difficult to test through conventional means. Modeling method andintelligent algorithm for optimal sensor placement is key to solving the problem ofchoosing, therefore the optimal sensor placement problem in this paper from thistwo-pronged fix. Was the establishment of the modal type of optimal sensorplacement for the random variable expected single-objective and multi-objectiveinteger programming model, and second, advantages of using DNA genetic algorithmto solve this problem, and design an algorithm, and finally by XuGe bridge forinstance validation algorithm is feasible and effective.2. Using the kind of pattern recognition method SVM (support vector machine)for bridge static damage identification, the key lies in how to construct the similarity between the simulated undamaged and damaged as training set through ANSYS andthe engineering practice to show its anti-interference ability; and how to use a noisetest set to determine the accuracy of the location and the degree of the damage inorder to manifest its ability to distinguish damage. According to this problem, firstly,a high precision pattern recognition result is presented by the static load deflectionresponse of the mode under different noise conditions and different loadingconditions in detail; secondly, the effectiveness of SVM is proved by the use ofcomparative analysis with the professional data mining software WEKA. After that,there is another problem that how to recognize static load pattern。Probability skylinesmainly extend the problem on uncertain data set, and according to the instance ofdistribution, the dominance relations of the class of objects are evaluated on thewhole, that is to say the advantages and disadvantages relations of the data areestimated. The bridge damage static load data exactly right have this kind ofadvantages and disadvantages relations, so probability skyline is applied to the bridgestatic load pattern recognition in this paper, and a good recognition effectiveness isshowed by the results, and also a certain actual applied value is given. Using the kindof pattern recognition method SVM (support vector machine) for bridge frequencydomain damage identification, the key lies in how to choose the node and its damageunit, and how to compute its identification precision and anti-interference ability.According to these two problems, firstly damage identification object is chosen whichis the point of Xuge Bridge optimal placement of the sensors in this paper, and thenthe accuracy of the location, the degree of the damage and noise test of Xuge Bridgeare presented by SVM pattern recognition method. At last, the comparative analysis isgiven by use of the professional data mining software WEKA to prove that thismethod has certain rationality and advantage. Vehicle moving over bridge of thetime-domain damage pattern recognition is mainly discussed in this article, and thetrouble is how to obtain the corresponding damage criterion by use of ANSYSsoftware to simulate calculation. The speed response of the vehicle moving over thebridge can be obtained according to the sampling interval of the measuring pointbased on energy ratio time-domain damage criterion, and then energy ratio before andafter damage can be use of damage criterion for the measuring point. SVM pattern recognition method has been proved to be a good damage recognition method. So theSVM damage identification method is presented in this paper by use of energy ratiocriterion.3. The step by step pattern recognition methods of bridge damage identificationare divided into two steps: damage location recognition and damage degreerecognition, and the key is the classification and regression problems respectively.The SOM neural network is proposed to do deal with location identification problemby the use of cluster analysis, and RBF neural network is presented to solve damagerecognition problem by the use of regression analysis. Based on damage identificationresults of the main girder and arch tower of Xuge Bridge in the frequency domain, theeffective of this method is shown, and the characteristic analysis of the identificationresults is given. This paper focuses on frequency domain damage locationidentification by use of pattern recognition methods without prior knowledge.Damage location identification is a key step to damage identification, and it can onlybe clustering recognition to solve in the case of no prior knowledge. Though there aremany clustering methods, there is no a common universal clustering method whichcan be applied to all the clustering problems, so clustering ensemble algorithm isproposed and proved to solve more problems. In this paper, the damage locationidentification of a truss structure and Xuge Bridge are recognized based on theCo-occurrence similarity (CSCE) and matrix transformation clustering ensemble, andit achieved full recognition. Finally, comparison analysis is also presented to provethe validity of this method in the paper by use of the professional data miningsoftware WEKA. The clustering method based on rough set can integrate collectionmethod and probability method to calculate similarity, and have very good clusteringeffectiveness. Bridge damage locations are recognized by rough clustering method inthis paper, and it achieved full recognition. It is compared with the fuzzy clustering.According to the method it also can get the sample results of the attribute reductionand reduction rules which can provide reference data in order to further study forsample characteristics.