CN107092934A - A kind of large scale structure damnification recognition method based on three-level data fusion - Google Patents

Abstract

The invention discloses a large-scale structure damage identification method based on three-level data fusion, comprising the following steps: A, collecting displacement and acceleration information of a large-scale structure to be detected and performing data processing; B, adopting a three-level data fusion algorithm to perform step A Fuse the information collected in step B; C, identify the result obtained in step B, and obtain the detection result through numerical simulation, reduce the transmission of useless data and the transmission of redundant information). The invention designs a three-level fusion strategy to jointly complete the damage identification of the structure. First, the consistency fusion algorithm is used to perform data-level fusion on the information collected by a single sensor, so as to improve the accuracy of data collection and reduce the amount of data transmission of nodes; then use ACGA The ‑BP neural network uses the displacement of static measurement data and the natural frequency of dynamic measurement data as the input parameters of the network for preliminary damage identification; The results are fused again to make the final recognition result more accurate.

Description

A large-scale structural damage identification method based on three-level data fusion

technical field

The invention relates to a sensor technology, in particular to a large-scale structure damage identification method based on three-level data fusion.

Background technique

The health and safety of structures such as bridges in transportation hubs, super high-rise buildings that symbolize cities, large-scale sports centers for citizens' leisure and entertainment, and art centers are closely related to the people. However, due to the large volume, complex structure, long service life and large area of these structures, if they cannot be effectively monitored, protected and diagnosed, many uncertain factors will arise. In recent years, these civil structures have been affected by environmental or human factors during their service, and damage and collapse accidents have occurred frequently, causing serious social impact.

Although countries around the world began to pay attention to the damage of structures very early, it is difficult to fully identify the damage of structures due to the backwardness of traditional damage identification technology. At the beginning of the 21st century, engineering structures in service around the world entered a period of restoration. Because rebuilding bridges, dams and other large civil structures requires huge financial funds, countries all over the world regard old civil structures as precious wealth, and hope that through comprehensive diagnostic evaluation and targeted Repairing and strengthening to prolong its service life will save a lot of manpower and material resources. Wireless sensor network technology has very good prospects and important research significance. A large number of wireless sensor nodes are deployed in large structures, and data transmission is carried out between nodes through ZigBee communication protocol, which is neat and beautiful and saves the cost of cables. But at the same time we It should also be noted that the deployment of a large number of sensors also brings another problem. How to accurately identify structural damage from a large amount of redundant data requires another technology that is also the research focus of this paper—data fusion technology.

Contents of the invention

The purpose of the present invention is to provide a large-scale structural damage identification method based on three-level data fusion, so as to solve the problems raised in the above-mentioned background technology.

To achieve the above object, the present invention provides the following technical solutions:

A large-scale structural damage identification method based on three-level data fusion, comprising the following steps:

A. Collect the displacement and acceleration information of the large structure to be detected and perform data processing;

B. Using a three-level data fusion algorithm to fuse the information collected in step A;

C. Identify the result obtained in step B, and obtain the detection result through numerical simulation.

As a further technical solution of the present invention: the three-level data fusion algorithm includes data-level fusion, feature-level fusion and decision-level fusion.

As a further technical solution of the present invention: the data-level fusion adopts a consistency fusion algorithm, firstly, suspicious data is removed from several groups of data collected by a single sensor, and then the consistency fusion algorithm is used for fusion processing to obtain more accurate data.

As a further technical solution of the present invention: the feature-level fusion uses ACGA-BP neural network as a pattern recognizer, and uses frequency and displacement as input parameters respectively to realize preliminary recognition of structures.

As a further technical solution of the present invention: the decision-level fusion adopts the D-S evidence theory, and analyzes and discusses the comparison between using the D-S evidence theory to perform decision-level fusion on two preliminary recognition results compared to using frequency and displacement as neural network input parameters for damage The superiority of recognition.

As a further technical solution of the present invention: said step A includes the following steps: a, information acquisition, adopting various sensors according to the actual situation of the research object, and transferring the signal acquired by the sensor to the computer system after A/D conversion , b. Data preprocessing, using filtering or wild point elimination methods for data preprocessing, c, feature extraction, performing feature extraction on signals collected by sensors, and the extracted features are physical quantities with clear physical meaning or statistics without any physical meaning volume and its deformation.

Compared with the prior art, the beneficial effect of the present invention is: the present invention designs a three-level fusion strategy to jointly complete the damage identification of the structure, first uses the consistency fusion algorithm to perform data-level fusion on the information collected by a single sensor, and improves the data quality. Accuracy of collection can reduce the amount of data transmission of nodes; then use ACGA-BP neural network to use the displacement of static measurement data and the natural frequency of dynamic measurement data as the input parameters of the network for preliminary damage identification; finally use D-S evidence theory to The recognition results of the measurement data and the recognition results based on the dynamic measurement data are re-fused to make the final recognition results more accurate.

Description of drawings

Figure 1 is a structural diagram of the fusion strategy.

Figure 2 is a data fusion structure diagram.

detailed description

The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some, not all, embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by persons of ordinary skill in the art without making creative efforts belong to the protection scope of the present invention.

Please refer to Figure 1-2, a large-scale structural damage identification method based on three-level data fusion, including the following steps:

A. Collect the displacement and acceleration information of the large structure to be detected and perform data processing;

B. Using a three-level data fusion algorithm to fuse the information collected in step A;

C. Identify the result obtained in step B, and obtain the detection result through numerical simulation.

The three-level data fusion algorithm includes data-level fusion, feature-level fusion and decision-level fusion.

The data-level fusion adopts the consistency fusion algorithm. Firstly, several groups of data collected by a single sensor are removed from suspicious data, and then the consistency fusion algorithm is used for fusion processing to obtain more accurate data.

Feature-level fusion uses the ACGA-BP neural network as the pattern recognizer, and uses frequency and displacement as input parameters respectively to realize the preliminary recognition of the structure.

D-S evidence theory was adopted for decision-level fusion, and the superiority of using D-S evidence theory for decision-level fusion of two preliminary identification results compared with using frequency and displacement as neural network input parameters for damage identification was analyzed and discussed.

Step A includes the following steps: a, information acquisition, using various sensors according to the actual situation of the research object, and transferring the signals acquired by the sensors to the computer system after A/D conversion, b, data preprocessing, using filtering or The wild point elimination method performs data preprocessing, c. feature extraction, and performs feature extraction on the signal collected by the sensor. The extracted features are physical quantities with clear physical meaning or statistical quantities without any physical meaning and their deformations.

The working principle of the present invention is: the current large-scale structural damage identification is facing the following difficulties: (1) For a certain damage situation of the structure, it is difficult to accurately identify a single damage identification method, and the results obtained by using multiple different identification methods are often inconsistent, making it difficult to obtain accurate identification information from multiple results. (2) The monitoring system for large structures often obtains a variety of different characteristic information, and the results obtained by using different parameters for damage identification are also inconsistent. In order to solve the above problems, this paper designs a three-level fusion strategy based on the data fusion theory. At the data level, the method of consistent fusion is adopted to reduce the error caused by accidental factors; at the feature level, ACGA-BP neural network is used for fusion; at the decision level, D-S evidence theory is used for final fusion. In order to improve the accuracy of the results, this paper adopts the method of combining static measurement data and dynamic measurement data as the basis for judging structural damage. The structure diagram of the fusion strategy is shown in Figure 1.

Firstly, various data information is collected by the sensor. Before performing data fusion on a set of data collected by a single sensor, it is necessary to eliminate suspicious data. To determine whether the data meets the requirements, a threshold needs to be set. When the data is greater than this threshold The data can be regarded as suspicious data when , and this paper chooses the Grubbs criterion judgment method to distinguish suspicious data.

Since the data collected by the sensor is easily affected by environmental noise and accidental factors, the measurement data is inaccurate. Therefore, the similar information collected by a single sensor is first eliminated from suspicious data, and then the consistency fusion algorithm is used for data-level fusion. This data-level Fusion can effectively improve the accuracy of data collected by sensors, reduce the amount of data transmission of terminal nodes, and reduce the power consumption of nodes.

The ACGA-BP neural network is selected for feature-level fusion. The BP neural network has good recognition accuracy, but its convergence speed is slow and it is easy to fall into a local optimum. Therefore, the improved self-adaptive co-evolutionary genetic algorithm is used to search globally for the weights and thresholds of the neural network. This improves the accuracy and training time of the network. The natural frequency based on the dynamic measurement data and the displacement based on the static measurement data are respectively used as the input parameters of the network to obtain the preliminary damage identification results of the research object.

The preliminary damage identification results obtained by using the ACGA-BP network are still not highly accurate in terms of identification accuracy. Both the identification method based on static measurement data and the identification method based on dynamic measurement data have their own limitations. When the identification parameters exist When there is noise, there will be different degrees of misjudgment, and the D-S evidence theory adopted in this paper as a decision-level fusion scheme can effectively reduce the possibility of misjudgment, thereby greatly increasing the accuracy of the final recognition result.

The training of BP neural network is to modify the weights and thresholds by the method of gradient reduction, which is easy to fall into local optimum and affect the overall performance of the algorithm; with the widespread use of BP neural network, the traditional use of BP network to solve problems has In view of the shortcomings of BP network, many scholars have proposed various improvement strategies, such as adaptive learning rate method, conjugate gradient method, etc. These methods have improved the performance of BP neural network to a certain extent, but still cannot fundamentally Solve the problem of BP network weight and threshold training. The genetic algorithm has a strong global search ability, but its local search ability is insufficient. Therefore, the genetic algorithm is used to solve the weight and threshold of the BP network, and the combination of the two algorithms makes the performance of the entire network more superior.

The optimization of BP neural network using genetic algorithm is mainly divided into three parts: firstly, determine the structure of neural network, and then determine the individual information of genetic algorithm. Secondly, the adaptive co-evolutionary genetic algorithm is used to optimize the weight and threshold of the BP network. The fitness value of the individual is calculated by the fitness function, and the genetic algorithm finds the optimal individual through genetic operations. Finally, the weight and threshold information in the optimal individual is used as the initial value of the network.

Claims (6)

1. a large-scale structural damage identification method based on three-level data fusion, is characterized in that, comprises the following steps: A. Collect the displacement and acceleration information of the large structure to be detected and perform data processing; B. Using a three-level data fusion algorithm to fuse the information collected in step A; C. Identify the result obtained in step B, and obtain the detection result through numerical simulation. 2. A large-scale structural damage identification method based on three-level data fusion according to claim 1, wherein the three-level data fusion algorithm includes data-level fusion, feature-level fusion and decision-level fusion. 3. A large-scale structural damage identification method based on three-level data fusion shown in claim 2, characterized in that, said data-level fusion adopts a consistency fusion algorithm, and at first several groups of data collected by a single sensor are removed from suspicious The data is then fused with a consistent fusion algorithm to obtain more accurate data. 4. according to a kind of large-scale structural damage identification method based on three-level data fusion shown in claim 2, it is characterized in that, described feature level fusion adopts ACGA-BP neural network as pattern recognizer, takes frequency and displacement as input respectively Parameters to realize the preliminary identification of the structure. 5. A large-scale structural damage identification method based on three-level data fusion shown in claim 2, characterized in that, said decision-making level fusion adopts D-S evidence theory, and analyzes and discusses the use of D-S evidence theory for two preliminary identification results Advantages of performing decision-level fusion compared to mixing frequency and displacement as neural network input parameters for damage identification. 6. A kind of large-scale structural damage identification method based on three-level data fusion shown in claim 1, it is characterized in that, described step A comprises the following steps: a, information acquisition, adopt various different methods according to the actual situation of the research object The sensor, and the signal acquired by the sensor is transferred to the computer system after A/D conversion, b, data preprocessing, using filtering or wild point elimination method for data preprocessing, c, feature extraction, the signal collected by the sensor is characterized Extraction, the extracted feature is a physical quantity with clear physical meaning or a statistical quantity without any physical meaning and its deformation.