CN-121980387-A - Abnormal data diagnosis method and system of structural health monitoring system

CN121980387ACN 121980387 ACN121980387 ACN 121980387ACN-121980387-A

Abstract

The invention discloses an abnormal data diagnosis method and system of a structural health monitoring system, and relates to the technical field of bridge data monitoring, comprising the following steps of collecting acceleration monitoring data to be diagnosed through the structural health monitoring system, and encoding the acceleration monitoring data to obtain a time-frequency diagram; and inputting the time-frequency diagram into a diagnosis model, and performing abnormal diagnosis on the acceleration monitoring data to obtain a diagnosis result. The self-attention residual error network model combining the convolutional neural network and the self-attention mechanism is introduced to the self-attention mechanism to enhance the modeling capability of the global dependency relationship on the basis of keeping the excellent local feature extraction capability of CNN, so that the richer global features of the data can be conveniently extracted, and the accuracy of data anomaly diagnosis is improved.

Inventors

CAO YANMEI
YANG YUANBO
ZHANG YUXIAN
TANG YIJING

Assignees

北京交通大学

Dates

Publication Date: 20260505
Application Date: 20260126

Claims (6)

1.An abnormal data diagnosis method of a structural health monitoring system is characterized by comprising the following steps: Collecting acceleration monitoring data to be diagnosed through a structural health monitoring system, and encoding the acceleration monitoring data to obtain a time-frequency diagram; The method comprises the steps of inputting a time-frequency diagram into a diagnosis model, and carrying out abnormal diagnosis on acceleration monitoring data to obtain a diagnosis result, wherein the diagnosis model comprises a feature extraction module, a self-attention residual error block group and a classification module, and the self-attention residual error block group comprises a plurality of first self-attention residual error blocks and second self-attention residual error blocks which are alternately stacked; The method comprises the steps of carrying out feature extraction, normalization and downsampling on a time-frequency diagram through a feature extraction module to obtain primary features of the time-frequency diagram, carrying out local feature extraction on the primary features through a first self-attention residual block, integrating a plurality of local features through a second self-attention residual block to extract global features, alternately extracting the local features and the global features to obtain a plurality of high-level features of the time-frequency diagram, mapping the high-level features through a classification module to obtain probability distribution, and carrying out anomaly diagnosis on the time-frequency diagram through the probability distribution.
2. The method for diagnosing abnormal data of a structural health monitoring system according to claim 1, wherein the first layer to the fifth layer of the first self-care residual block and the second self-care residual block are respectively a convolution layer, a batch normalization layer, a convolution layer, a batch normalization layer and a self-care module, the input and the output of the first self-care residual block are connected in a jump manner, the input and the output of the second self-care residual block are connected through the convolution layer, the plurality of first self-care residual blocks and the second self-care residual blocks perform feature extraction on the primary features to obtain a plurality of advanced features of the time-frequency diagram, and the method specifically comprises the following steps: The first self-attention residual block divides the primary feature into a plurality of local windows, and calculates self-attention in each local window to obtain local features; The second self-attention residual block uses a shift window self-attention mechanism to correlate the local features and calculate self-attention, and performs global information integration on the whole time-frequency diagram to obtain a cooperative change rule among different frequency components in the time-frequency diagram; The processing procedures of the first self-attention residual block and the second self-attention residual block are alternately and circularly iterated, and advanced characteristics of the time-frequency diagram are obtained gradually.
3. The method of claim 1, wherein the feature extraction module comprises a convolution layer, a batch normalization layer, and a maximum pooling layer, and the classification module comprises a global average pooling layer, a batch normalization layer, and an output layer.
4. A method of diagnosing abnormal data for a structural health monitoring system according to claim 1, and before the acceleration monitoring data are encoded, error data cleaning and data expansion are carried out on the acceleration monitoring data.
5. The method of claim 1, wherein the diagnostic result includes normal data, missing data, drift data, sub-minimum data, trend data, local gain data, outlier data, and overscan oscillation data.
6. An anomaly data diagnostic system for a structural health monitoring system, comprising: The acquisition module is used for acquiring acceleration monitoring data to be diagnosed through the structural health monitoring system and encoding the acceleration monitoring data to obtain a time-frequency diagram; the diagnosis module is used for inputting the time-frequency diagram into a diagnosis model and carrying out abnormal diagnosis on acceleration monitoring data to obtain a diagnosis result, wherein the diagnosis model comprises a feature extraction module, a self-attention residual block group and a classification module; the processing module is used for carrying out feature extraction, normalization and downsampling on the time-frequency diagram through the feature extraction module to obtain primary features of the time-frequency diagram, carrying out local feature extraction on the primary features through the first self-attention residual block, integrating a plurality of local features through the second self-attention residual block to extract global features, alternately extracting the local features and the global features to obtain a plurality of high-level features of the time-frequency diagram, and mapping the high-level features through the classification module to obtain probability distribution, and carrying out abnormality diagnosis on the time-frequency diagram through the probability distribution.

Description

Abnormal data diagnosis method and system of structural health monitoring system Technical Field The invention relates to the technical field of bridge data monitoring, in particular to an abnormal data diagnosis method and system of a structural health monitoring system. Background Over time, the bridge structure is subjected to the double effects of internal factors and external factors, the damage is gradually accumulated, more bridge disease problems occur, the structural performance is further reduced, and even safety accidents are caused. The safety and reliability of the bridge structure directly determine the smoothness of urban traffic and the safety of resident trip. Therefore, development and application of a structural health monitoring (Structural Health Monitoring, SHM) system are advanced, monitoring of long-term operation and safety states of the bridge is enhanced, and an informationized decision support system for bridge management and maintenance is established, so that the system becomes a necessary measure for bridge construction, management and maintenance at home and abroad. The SHM system is used for monitoring the structure, the state, the performance and the like of the bridge in real time through a plurality of sensor arrays by applying various sensing technologies and combining the knowledge of the related fields such as building structures, civil engineering, computer science, communication technology and the like, collecting and analyzing the related data of the bridge, analyzing and predicting the health condition, and providing data support and technical support for the maintenance and operation management of the bridge. More and more SHM systems are installed on various structural facilities such as dams, large bridges, high-rise buildings and the like, the dynamic response of the structure is measured permanently or periodically, the running condition of the structure is known in real time, and the evaluation of the structural state and the structural damage is carried out according to the abnormal state early warning of the SHM systems. Efficient and accurate state assessment of a structure generally requires obtaining high quality low noise monitoring data containing significant vibration information of the structure. When the SHM system is used for carrying out field implementation on the bridge structure, abnormal monitoring data are often generated due to the problems of sensor faults, transmission cable damage, electromagnetic interference and the like, and meanwhile, normal monitoring signals contain structure effective information and noise interference with different degrees due to the interference of monitoring objects and surrounding environments thereof and the influence of the sensor. Abnormal monitoring data and high noise data generated by the SHM system can influence the accuracy of structural state evaluation and structural modal identification, and the reliability of early warning and forecasting of the SHM system is reduced. Therefore, it is necessary to analyze and process massive monitoring data in the SHM system, and screen out normal data and abnormal data in the monitoring data through data abnormality diagnosis processing. Abnormality diagnosis of monitoring data of an existing SHM system is realized based on a deep learning model, which can autonomously learn highly abstract features from original data by using a neural network to complete classification tasks. However, when the deep learning model encounters a large amount of data, particularly when encountering a large amount of data monitored by long-term health of a structure, the data classification capability of the deep learning model is limited, and thus the accuracy of the identified abnormal data is limited. Disclosure of Invention Based on the defects in the prior art, the invention provides an abnormal data diagnosis method and system of a structural health monitoring system, which solve the problem that the accuracy of the identified abnormal data is limited due to the limited data classification capability of the existing deep learning model. The invention adopts the following technical scheme: In a first aspect, the present invention provides a method for diagnosing abnormal data of a structural health monitoring system, comprising the steps of: Collecting acceleration monitoring data to be diagnosed through a structural health monitoring system, and encoding the acceleration monitoring data to obtain a time-frequency diagram; The method comprises the steps of inputting a time-frequency diagram into a diagnosis model, and carrying out abnormal diagnosis on acceleration monitoring data to obtain a diagnosis result, wherein the diagnosis model comprises a feature extraction module, a self-attention residual error block group and a classification module, and the self-attention residual error block group comprises a plurality of first self-attention residual error blocks and second self-