Intelligent System and Method for Deformation Monitoring and Early Warning of Steel Space Frame Structure

Abstract

The present invention relates to the field of structural safety monitoring, in particular to an intelligent system and method for deformation monitoring and early warning of steel grid structures. The system adopts distributed monitoring nodes, which are flexibly arranged at key parts of the steel grid structure to collect deformation data in real time; combined with advanced data acquisition and transmission equipment to ensure the efficient and accurate transmission of data; the core data processing and analysis system uses deep learning algorithms such as convolutional neural networks to deeply analyze the deformation data and accurately identify complex deformation patterns and potential risks; the system realizes all-round and multi-dimensional real-time monitoring of the steel grid structure, significantly improving the comprehensiveness and accuracy of data; at the same time, the application of intelligent algorithms greatly improves the accuracy and timeliness of early warning, providing a strong guarantee for the safety of the steel grid structure.

Description

Technical Field

[0001] The present invention relates to the field of structural safety monitoring, and particularly to an intelligent system and method for deformation monitoring and early warning of steel grid structures. By using distributed monitoring nodes, intelligent sensing technology, and deep learning algorithms, it realizes real-time monitoring, data analysis, and risk early warning of the deformation state of steel grid structures. Background Art

[0002] Due to its good mechanical properties and high space utilization rate, steel grid structures are widely used in large public buildings, stadiums, exhibition centers and other places. However, due to the complexity of its structural characteristics and use environment, steel grid structures are prone to problems such as deformation and stress concentration during long-term use. If not discovered and processed in time, it may lead to serious safety accidents.

[0003] Traditional steel grid structure deformation monitoring technologies mainly rely on manual regular inspections or simple sensor detections, which have problems such as discontinuous monitoring, insufficient data analysis, and untimely early warning. There are also some automated monitoring systems in the prior art, but they generally have the following deficiencies: First, the number of monitoring nodes is limited, making it difficult to construct a complete structural deformation scenario; second, the data analysis method is single, making it difficult to extract effective information from a large amount of data; third, the early warning mechanism is simple, only based on fixed thresholds for judgment, and cannot adapt to complex and changing structural states; fourth, there is a lack of in-depth analysis of deformation data, making it difficult to discover potential risks.

[0004] With the development of artificial intelligence and deep learning technologies, it has become possible to apply these advanced technologies to the field of steel grid structure monitoring. Convolutional neural networks are particularly suitable for processing structurally deformed data with spatial correlation due to their strong feature extraction and pattern recognition capabilities. However, how to construct a deformation monitoring and early warning system based on convolutional neural networks and how to ensure the accuracy and reliability of the system are still technical problems to be solved urgently.

[0005] Therefore, there is an urgent need for a system and method that can monitor the deformation of steel grid structures in real time and perform data analysis and risk early warning through intelligent algorithms to improve the safety and service life of steel grid structures. Summary of the Invention

[0006] The object of the present invention is to provide an intelligent system and method for deformation monitoring and early warning of steel grid structures. By means of a distributed monitoring node network, the deformation data of steel grid structures is collected in real time, and combined with intelligent algorithms such as convolutional neural networks, the deformation data is deeply analyzed and risk predicted, so as to achieve accurate evaluation and timely early warning of the safety state of steel grid structures, and improve the safety and service life of steel grid structures.

[0007] The present invention proposes an intelligent system for deformation monitoring and early warning of steel grid structures, including:

[0008] Steel grid structure;

[0009] Monitoring nodes, which are distributed in the steel grid structure for real-time collection of deformation data;

[0010] Data acquisition and transmission equipment for acquiring and transmitting the deformation data;

[0011] Data processing and analysis system for receiving the deformation data and performing deformation early warning analysis based on a convolutional neural network.

[0012] Preferably, the monitoring nodes are used to monitor the deformation of the steel grid structure, including displacement, deformation, stress, strain or stress-strain.

[0013] Preferably, the data acquisition and transmission equipment includes sensors, analog-to-digital converters, signal transmitters and signal receivers to realize the functions of real-time acquisition, automatic analysis and early warning of the deformation data of the steel grid structure, including transmitting the deformation data to the data processing and analysis system for further processing in a wireless or wired manner.

[0014] Preferably, the data processing and analysis system is also provided with a data storage unit for storing the deformation data and recording the historical deformation data of the steel grid structure.

[0015] Preferably, the data processing and analysis system is also provided with a data comparison unit for comparing the deformation data with the historical deformation data to analyze the deformation trend of the steel grid structure.

[0016] Preferably, the data processing and analysis system is also provided with a data correction unit for correcting abnormal or non-compliant deformation data.

[0017] Preferably, the data processing and analysis system draws two-dimensional or three-dimensional curves and / or surfaces of the deformation data characteristics through the deformation data.

[0018] Preferably, the data processing and analysis system performs deformation warning analysis using a convolutional neural network, including the following steps: S110, extracting sample images, where the number of sample images is greater than 20,000; S120, preprocessing the sample images, including image enhancement and normalization; S130, using the convolutional neural network as a basic model to train the sample images and obtaining a classification model; S140, establishing an output layer of the convolutional neural network to obtain a relationship prediction diagram between each classification of deformation data and the warning level; S150, inputting each classification of deformation data into the classification model to obtain the prediction probability of each classification of deformation data; S160, calculating the sum of the prediction probabilities of each classification of deformation data and performing normalization processing; S170, multiplying the prediction probability by the risk value for each classification of deformation data to obtain the risk prediction for each classification of deformation data; S180, judging whether there is a deformation risk for each classification of deformation data according to the risk threshold.

[0019] Preferably, the data processing and analysis system is further provided with a data analysis unit to analyze the deep reasons for the deformation data of the steel grid structure, including performance measurement, error rate statistics, and fault mode identification.

[0020] The monitoring and warning method based on the steel grid structure deformation monitoring and warning intelligent system includes the following steps:

[0021] S10, installing monitoring nodes distributedly in the steel grid structure for real-time collecting deformation data at the target positions of the steel grid structure;

[0022] S20, collecting and transmitting the deformation data through data collection and transmission equipment to realize the functions of real-time acquisition, automatic analysis, and warning of the deformation data of the steel grid structure;

[0023] S30, receiving and processing the deformation data through the data processing and analysis system, and performing deformation warning analysis on the deformation data based on a convolutional neural network;

[0024] S40, analyzing the deep reasons for the deformation data of the steel grid structure, and continuously optimizing the data processing and analysis system by quantifying the problem impacts such as performance measurement and error rate statistics.

[0025] The present invention has the following beneficial effects:

[0026] 1. Realize all-round and multi-dimensional real-time monitoring of the steel grid structure through a distributed monitoring node network, improving the comprehensiveness and accuracy of monitoring data;

[0027] 2. Use intelligent algorithms such as convolutional neural networks to analyze deformation data, which can identify complex deformation patterns and potential risks, greatly improving the accuracy and timeliness of warning;

[0028] 3. A complete data processing and analysis chain is constructed, including units such as data storage, data comparison, data correction, and data analysis, ensuring the quality of monitoring data and the reliability of analysis results;

[0029] 4. Support two-dimensional or three-dimensional visualization display of deformed data, enabling engineers to intuitively understand the structural deformation state, facilitating decision-making and processing;

[0030] 5. Through in-depth cause analysis and system continuous optimization mechanism, the system can continuously improve itself to adapt to the monitoring requirements of different steel grid structures. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 is a schematic diagram of the overall architecture of the intelligent system for steel grid structure deformation monitoring and warning of the present invention;

[0032] Figure 2 is a schematic diagram of the distribution of monitoring nodes of the present invention;

[0033] Figure 3 is a schematic diagram of the structure of the data acquisition and transmission device of the present invention;

[0034] Figure 4 is a schematic diagram of the module composition of the data processing and analysis system of the present invention;

[0035] Figure 5 is a flow chart of deformation warning analysis based on convolutional neural network of the present invention;

[0036] Figure 6 is a flow chart of the steps of the monitoring and warning method of the present invention;

[0037] Figure 7 is a schematic diagram of the architecture of the convolutional neural network model in the present invention;

[0038] Figure 8 is an example diagram of deformed data visualization in the present invention;

[0039] Figure 9 is a schematic diagram of the process of in-depth cause analysis in the present invention;

[0040] Figure 10 is a flow chart of the implementation of the adaptive threshold adjustment mechanism in the present invention;

[0041] Figure 11 is a schematic diagram of the architecture of the multi-modal data fusion algorithm in the present invention;

[0042] Figure 12 is a schematic diagram of the spatio-temporal correlation analysis and anomaly detection method in the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0043] Please refer to the attached Figures 1-12, The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. These drawings and embodiments are only used to illustrate the present invention, but should not be construed as limiting the present invention.

[0044] Embodiment 1: Overall System Architecture

[0045] As Figure 1 shown, the intelligent system for deformation monitoring and early warning of the steel grid structure of the present invention mainly includes the following parts: steel grid structure 1, monitoring nodes 2, data acquisition and transmission equipment 3, and data processing and analysis system 4.

[0046] The steel grid structure 1 is the monitoring object of this system, which can be the steel grid structure in various large buildings, such as the roof of a stadium, an exhibition center, etc. The monitoring nodes 2 are distributed at key positions of the steel grid structure 1. Preferably, these key positions include structural support points, the location with the largest span, stress concentration areas, etc., and are used to collect deformation data of the steel grid structure 1 at these positions in real time.

[0047] The data acquisition and transmission equipment 3 is connected to the monitoring nodes 2 and is used to collect and transmit the deformation data obtained by the monitoring nodes 2. The data processing and analysis system 4 is connected to the data acquisition and transmission equipment 3, receives the deformation data, and stores, processes, analyzes, and gives early warnings to these data.

[0048] In a preferred embodiment of the present invention, the number of monitoring nodes 2 is configured according to the complexity and size of the steel grid structure 1, and the number can vary from 10 to 200. When the area of the steel grid structure 1 does not exceed 5000 square meters, the number of monitoring nodes 2 is preferably 10 to 50; when the area is between 5000 and 20000 square meters, the number of monitoring nodes 2 is preferably 50 to 100; when the area exceeds 20000 square meters, the number of monitoring nodes 2 is preferably 100 to 200. This configuration method ensures the comprehensiveness and representativeness of the monitoring data, and at the same time avoids the cost increase caused by redundant configuration.

[0049] Embodiment 2: Implementation Method of Monitoring Nodes

[0050] As Figure 2 shown, the monitoring node 2 of the present invention is designed to monitor various deformation parameters of the steel grid structure 1, including displacement 21, deformation 22, stress 23, strain 24, or stress-strain 25.

[0051] In an embodiment of the present invention, the displacement 21 is measured by a high-precision displacement sensor. The measurement range is preferably 0 - 100 mm, and the accuracy is preferably 0.01 mm. The deformation 22 is measured by a deformation sensor. The measurement range is preferably 0 - 50 mm, and the accuracy is preferably 0.005 mm. The stress 23 is measured by a stress sensor. The measurement range is preferably 0 - 500 MPa, and the accuracy is preferably 0.1 MPa. The strain 24 is measured by a strain gauge. The measurement range is preferably 0 - 3000 με, and the accuracy is preferably 1 με. The stress-strain 25 is measured by a comprehensive stress-strain sensor.

[0052] These different types of sensors enable the monitoring node 2 to comprehensively capture the deformation state of the steel grid structure 1, providing multi-dimensional data support for subsequent analysis. In practical applications, one or more of the above sensors can be selectively configured according to the characteristics of the steel grid structure 1 and the monitoring requirements.

[0053] Embodiment 3: Implementation method of data acquisition and transmission equipment

[0054] As Figure 3 shown, the data acquisition and transmission equipment 3 of the present invention includes a sensor 31, an analog-to-digital converter 32, a signal transmitter 33, and a signal receiver 34, realizing the functions of real-time acquisition, automatic analysis, and early warning of the deformation data of the steel grid structure 1.

[0055] The sensor 31 is connected to the monitoring node 2 and is used to convert the physical deformation amount into an electrical signal. The analog-to-digital converter 32 is connected to the sensor 31 to convert the analog electrical signal into a digital signal. The signal transmitter 33 is connected to the analog-to-digital converter 32 and is responsible for sending the digital signal wirelessly or wired. The signal receiver 34 is arranged on one side of the data processing and analysis system 4, receives the signal from the signal transmitter 33, and transmits it to the data processing and analysis system 4 for further processing.

[0056] In a preferred embodiment of the present invention, the sampling frequency of the sensor 31 is set to 10 Hz, that is, 10 data are collected per second. This frequency can not only capture the dynamic deformation process of the steel grid structure 1 but also not generate too much redundant data. The resolution of the analog-to-digital converter 32 is preferably 16 bits, which can meet the requirements of high-precision deformation measurement. The signal transmitter 33 can use a 4G / 5G wireless communication module or an industrial Ethernet wired connection, and a suitable method is selected according to the on-site environment. The signal receiver 34 is equipped with a data caching function, which can temporarily store data during network fluctuations to avoid data loss.

[0057] Embodiment 4: Data storage unit of the data processing and analysis system

[0058] As Figure 4As shown in the figure, the data processing and analysis system 4 of the present invention includes modules such as a data storage unit 41, a data comparison unit 42, a data correction unit 43, and a data analysis unit 44.

[0059] The data storage unit 41 is used to store the deformation data from the data acquisition and transmission device 3 and record the historical deformation data of the steel grid structure 1. In an embodiment of the present invention, the data storage unit 41 adopts a distributed storage architecture, including a real-time database and a historical database. The real-time database stores high-frequency deformation data for the most recent 7 days, using in-memory database technology to ensure the speed of data access and storage; the historical database stores complete historical deformation data, using relational database technology to support complex data queries and analysis.

[0060] To improve data management efficiency, the data storage unit 41 also implements data hierarchical storage and automatic archiving functions. Specifically, for data that is more than 30 days old, the system will change its storage mode from high-frequency sampling to low-frequency sampling, for example, reducing the sampling frequency from the original 10 Hz to 0.1 Hz, that is, storing data every 10 seconds. For data that is more than 1 year old, the system will perform data compression processing, only retaining data at key change points to save storage space. This hierarchical storage strategy ensures data integrity while significantly reducing storage costs.

[0061] Example 5: The data comparison unit of the data processing and analysis system

[0062] As Figure 4 shown in the figure, the data comparison unit 42 of the present invention compares the real-time collected deformation data with the historical deformation data in the data storage unit 41 to analyze the deformation trend of the steel grid structure 1.

[0063] In an embodiment of the present invention, the data comparison unit 42 adopts a multi-level comparison strategy. First is the benchmark comparison, comparing the current deformation data with the structural design benchmark value to evaluate whether the structural deformation is within the allowable range. Second is the historical same-period comparison, comparing the current deformation data with the data at the same time period and under the same temperature conditions in history to identify abnormal changes. Finally is the trend comparison, through time series analysis technology, comparing the time change trend of the deformation data to predict the future development direction of deformation.

[0064] Specifically, the data comparison unit 42 uses the following mathematical model for deformation trend analysis:

[0065] Assume that the deformation data of the steel grid structure 1 at time point is , and the historical same-period deformation data is , then the relative change rate can be expressed as:

[0066] ,

[0067] When exceeds the preset threshold (preferably, set to 5%), the system will mark this point as a potential anomaly point.

[0068] For the prediction of deformation trend, the data comparison unit 42 uses the exponential smoothing method:

[0069] ,

[0070] wherein, is the predicted deformation value at the next time point, is the smoothing coefficient (preferably, takes a value of 0.3), is the predicted deformation value at the current time point. This method can effectively smooth short-term fluctuations and highlight long-term deformation trends.

[0071] Example 6: The data correction unit of the data processing and analysis system

[0072] As Figure 4 shown, the data correction unit 43 of the present invention is used to correct abnormal or non-compliant deformation data to ensure the accuracy of subsequent analysis.

[0073] In an embodiment of the present invention, the data correction unit 43 adopts a three-level data correction strategy: outlier correction, missing value filling, and noise filtering.

[0074] Outlier correction uses statistical methods based on the mean value and standard deviation of the deformation data, and regards the data outside the range as outliers. For the detected outliers, the system will not directly delete them, but correct them in the following way:

[0075] ,

[0076] For data missing situations, the data correction unit 43 uses linear interpolation to fill:

[0077] ,

[0078] When the number of consecutive missing data points does not exceed 5, the above method is used; when the continuous missing exceeds 5 points, the system will generate a missing data warning, indicating that there may be sensor failures or communication problems.

[0079] For data noise, the data correction unit 43 uses moving average filtering:

[0080] ,

[0081] Among them, is the filter window size (preferably, takes the value of 5). This method can effectively filter high-frequency noise and retain the effective information of the deformed data.

[0082] Example 7: Visualization Function of Data Processing and Analysis System

[0083] As Figure 8 shown, the data processing and analysis system 4 of the present invention can draw two-dimensional or three-dimensional curves and / or surfaces of the deformed data characteristics through the deformed data, facilitating engineers to intuitively observe the deformation state of the steel grid structure 1.

[0084] In an embodiment of the present invention, the two-dimensional curve is mainly used to display the change trend of single-point deformation over time. The system supports multiple curve chart types, including line charts, scatter plots, and area charts. Engineers can select any monitoring node 2 to view the deformation data change curve within a specified time period. At the same time, the system also supports the overlay display of multi-point data, facilitating the comparison of the deformation conditions at different points.

[0085] The three-dimensional surface is used to display the deformation distribution of the entire steel grid structure 1. The system constructs a complete structural deformation surface model according to the spatial positions and deformation data of each monitoring node 2 through triangulation and interpolation algorithms. This visualization method can intuitively display the spatial distribution of the deformation, helping engineers identify the deformation concentration areas and potential risk points.

[0086] Specifically, when constructing the three-dimensional deformation surface, the system adopts the radial basis function (RBF) interpolation algorithm:

[0087] ,

[0088] Among them, represents the deformation value at the spatial point , represents the coordinates of the spatial point, represents the th coordinate of the monitoring node 2, is the radial basis function (preferably, the Gaussian function is adopted, where is the shape parameter and takes the value of 0.1), is the weight coefficient, which is determined by solving a system of linear equations. This interpolation method can smoothly reconstruct the deformation field of the entire structure and obtain reasonable estimated values even in areas where the monitoring nodes 2 are sparse.

[0089] Example 8: Deformation Early Warning Analysis Based on Convolutional Neural Network

[0090] As Figure 5 shown, the data processing and analysis system 4 of the present invention uses a convolutional neural network for deformation early warning analysis, including the following steps:

[0091] S110, extract sample images, where the number of sample images is greater than 20000. In an embodiment of the present invention, the sample images are derived from two parts: one is the structural deformation images generated from historical monitoring data, and the other is the structural deformation scenario images simulated by finite element analysis. The sample images include four categories: normal state, slight deformation, moderate deformation, and severe deformation, and the number of samples in each category is kept basically balanced to avoid bias in model training.

[0092] S120, preprocess the sample images, including image enhancement and normalization. Image enhancement includes operations such as contrast adjustment, histogram equalization, and edge enhancement to improve the clarity and feature recognizability of the images. Normalization scales the image pixel values to the range [0,1], which helps to accelerate model training and improve training stability. Specifically, the normalization formula is:

[0093] ,

[0094] where is the original image pixel value, and are the minimum and maximum pixel values of the image, respectively.

[0095] S130, using the convolutional neural network as the basic model, train the sample images and obtain a classification model. In an embodiment of the present invention, an improved VGG16 architecture is used as the basic model, which includes 13 convolutional layers and 3 fully connected layers. The convolutional layers use 3x3 small convolutional kernels to achieve the effect of a large receptive field by stacking small convolutional kernels, while reducing the number of parameters. Each group of convolutional layers is followed by a max pooling layer to reduce the feature map size and extract significant features. The training process uses a batch size of 64, an initial learning rate of 0.001, and a learning rate decay strategy, where the learning rate is halved every 10 epochs. The training termination condition is that the accuracy of the validation set does not increase for 5 consecutive epochs.

[0096] S140, establish the output layer of the convolutional neural network to obtain a relationship prediction map between each classification of deformation data and the warning level. The output layer contains 4 neurons, corresponding to the four categories of normal state, slight deformation, moderate deformation, and severe deformation respectively. The Softmax activation function is used to convert the output into a probability distribution:

[0097] ,

[0098] where is the The original output scores for each category, is the predicted probability for the

[0099] S150. Input the classified deformation data into the classification model to obtain the predicted probabilities for each classification of the deformation data. In practical applications, the system converts the real-time collected deformation data into images in the same format as the training samples, and then inputs them into the trained classification model to obtain the predicted probabilities.

[0100] S160. Calculate the sum of the predicted probabilities for each classification of the deformation data and perform normalization processing. Normalization processing ensures that the sum of all predicted probabilities is 1, which is convenient for subsequent risk assessment:

[0101] ,

[0102] S170. For each classification of the deformation data, multiply the predicted probability by the risk value to obtain the risk prediction for each classification of the deformation data. In an embodiment of the present invention, the risk values corresponding to the four categories are: normal state is 0.1, slight deformation is 0.4, moderate deformation is 0.7, and severe deformation is 1.0. The risk prediction value is calculated as follows:

[0103] ,

[0104] S180. Determine whether there is a deformation risk for each classification of the deformation data according to the risk threshold. Preferably, the risk threshold is set to 0.6. When the risk prediction value of any category exceeds this threshold, the system determines that there is a deformation risk and triggers a warning at the corresponding level. The setting of the risk threshold is based on a large amount of experimental data analysis and expert experience, which can ensure the warning sensitivity while minimizing the false alarm rate.

[0105] Embodiment 9: The data analysis unit of the data processing and analysis system

[0106] As Figure 9 shown, the data analysis unit 44 of the present invention is used to analyze the deep reasons for the deformation data of the steel grid structure 1, including performance measurement 45, error rate statistics 46, and fault mode recognition 47.

[0107] Performance measurement 45 mainly focuses on the key performance indicators of the system, including data acquisition accuracy rate, transmission delay, warning timeliness, etc. In an embodiment of the present invention, the data acquisition accuracy rate is required to reach more than 99.9%, the transmission delay is controlled within 100 milliseconds, and the warning response time does not exceed 1 second. These indicators are evaluated through system log analysis and regular tests to ensure that the system continuously meets the monitoring requirements.

[0108] The error rate statistics 46 track various errors during the operation of the system, including sensor failures, communication interruptions, data processing anomalies, etc. The system classifies errors into four levels: minor (does not affect normal monitoring), moderate (affects some functions), severe (affects core functions), and critical (system paralysis). By analyzing the frequency, distribution, and trend of error occurrences, identify the weak links in the system and provide a basis for system optimization.

[0109] Fault mode recognition 47 constructs a feature pattern library for common faults of the steel grid structure 1 based on historical fault data and expert knowledge. By matching and analyzing the real-time deformation data with these patterns, the system can identify potential structural problems, such as settlement of support points, instability of members, and loosening of node connections. In an embodiment of the present invention, the fault mode recognition adopts a hybrid reasoning method based on rules and cases, which not only utilizes the professional knowledge in the field of structural engineering but also can continuously improve the pattern library through case learning.

[0110] Embodiment 10: Method for Deformation Monitoring and Early Warning of Steel Grid Structure

[0111] As Figure 6 shown, the method for deformation monitoring and early warning of the steel grid structure of the present invention is based on the aforementioned intelligent system and includes the following steps:

[0112] S10, distributively install monitoring nodes 2 in the steel grid structure 1 for real-time collection of deformation data at the target positions of the steel grid structure 1. The installation positions of the monitoring nodes 2 are determined by finite element analysis, and the deformation-sensitive areas and key support points are preferably selected. In an embodiment of the present invention, the installation process follows the principle of zoning division - combination of coarse and dense - key strengthening, that is, first divide the steel grid structure 1 into several regions according to its functions and structural characteristics, then arrange the bases of the monitoring nodes 2 in each region, and then increase the point density in the key regions.

[0113] S20, collect and transmit the deformation data through the data collection and transmission device 3 to achieve the functions of real-time acquisition, automatic analysis, and early warning of the deformation data of the steel grid structure 1. The data collection adopts a combination of timed triggering and event triggering. The collection frequency of timed triggering is 10Hz. When a sudden deformation event is detected, the system will automatically increase the collection frequency to 50Hz to capture the dynamic process of deformation. The data transmission adopts a data compression and hierarchical transmission strategy, transmitting the compressed data under normal conditions and the original data under emergency conditions to ensure the timely acquisition of effective data in various situations.

[0114] S30, receiving and processing deformation data through the data processing and analysis system 4, and performing deformation warning analysis on the deformation data based on the convolutional neural network. The system first performs quality inspection and preprocessing on the received data, and then stores it in the database, while performing real-time analysis and early warning judgment. The analysis process of the convolutional neural network is as described in Example 8, and the accurate prediction of deformation risk is achieved through eight steps of extracting sample images, preprocessing, model training, establishing output layers, calculating prediction probabilities, normalization processing, risk prediction and risk judgment.

[0115] S40, analyzes the deep causes of the deformation data of the steel grid structure 1, and continuously optimizes the data processing and analysis system 4 by quantifying the impact of the problem such as performance measurement 45 and error rate statistics 46. The deep cause analysis adopts the process of feature extraction-pattern matching-cause inference-verification feedback to extract key features from the deformation data, match them with known failure modes, infer possible causes, and verify them through further monitoring or on-site inspection. The system also implements the knowledge base self-learning function, and continuously enriches and improves the failure mode library by recording the results of each analysis and processing, thereby improving the accuracy of future analysis.

[0116] Preferably, the method of the present invention also includes visual presentation of the analysis results and multi-channel early warning. Visual presentation intuitively displays the deformation state of the structure by drawing two-dimensional or three-dimensional curves and / or surfaces; multi-channel early warning ensures that the early warning information can be conveyed to the relevant responsible persons in a timely manner through SMS, email, mobile application push, etc. The system also supports an automatic upgrade mechanism for the early warning level. When the system detects risks for multiple consecutive times and the risk value shows an upward trend, the early warning level will automatically increase, triggering a higher level of emergency response.

[0117] Embodiment 11: Adaptive threshold adjustment mechanism

[0118] like Figure 10 As shown, the present invention further provides an adaptive threshold adjustment mechanism as an important supplement and innovation to the fixed threshold judgment method. The mechanism can automatically adjust the risk judgment threshold according to the characteristics, environmental conditions and historical data of the steel grid structure 1, significantly improving the adaptability and early warning accuracy of the system.

[0119] In traditional structural monitoring systems, risk judgment usually relies on pre-set fixed thresholds, which is difficult to cope with complex and changeable actual situations. For example, under different temperature conditions, the normal deformation range of the steel grid structure 1 will be different; for another example, for structures with different service lives, the appropriate warning thresholds should also be different. The adaptive threshold adjustment mechanism of the present invention provides a dynamic and intelligent solution to these problems.

[0120] This mechanism mainly includes the following core components: an environmental factor compensation module 51, a structural characteristic analysis module 52, a historical data learning module 53, and a threshold optimization module 54. These modules work together to achieve intelligent adjustment of the threshold.

[0121] The environmental factor compensation module 51 is responsible for analyzing the influence of environmental conditions (such as temperature, humidity, wind force, etc.) on the deformation of the steel grid structure 1. In an embodiment of the present invention, this module constructs a relationship model between environmental factors and structural deformation based on structural mechanics and the theory of thermal expansion of materials. Taking temperature compensation as an example, the module uses the following formula to calculate the deformation amount caused by temperature change:

[0122] ,

[0123] where, is the deformation amount caused by temperature change, is the linear expansion coefficient of steel (preferably, the value is ), is the structural characteristic length, is the temperature change value. Through this formula, the system can calculate the normal deformation amount caused by temperature change and eliminate it from the total deformation amount to obtain a more accurate evaluation of the abnormal deformation amount.

[0124] The structural characteristic analysis module 52 is responsible for analyzing the inherent characteristics of the steel grid structure 1, such as stiffness, strength, service life, etc., and adjusting the threshold parameters according to these characteristics. In an embodiment of the present invention, this module uses a structural response evaluation method based on finite element analysis to construct a mapping relationship between structural characteristics and safe deformation limits. Specifically, the module first establishes a finite element model of the steel grid structure 1, and then through multi-condition analysis, obtains the safe deformation ranges of different regions under different conditions. These range values will be used as the basic reference values for threshold adjustment.

[0125] The historical data learning module 53, based on historical monitoring data, extracts the deformation patterns and rules of the steel grid structure 1 through statistical analysis and machine learning methods. In an embodiment of the present invention, this module uses a clustering analysis algorithm based on self-organizing mapping (SOM) to identify the typical deformation patterns of the structure from a large amount of historical data. The core steps of the algorithm are as follows:

[0126] 1. Data preprocessing: Standardize the historical deformation data to eliminate the influence of dimension and scale;

[0127] 2. SOM network initialization: Create a two-dimensional grid, where each node represents a potential deformation pattern prototype, and the number of nodes is preferably to ;

[0128] 3. Competitive learning: For each input sample , calculate its Euclidean distance from all node weight vectors , and find the winning node corresponding to the minimum distance :

[0129] ,

[0130] 4. Weight update: Update the weight vectors of the winning node and the nodes within its neighborhood:

[0131] ,

[0132] where is the learning rate (preferably with an initial value of 0.1, decreasing over time), is the neighborhood function (preferably using a Gaussian function, with an initial neighborhood radius of 1 / 3 of the grid size, decreasing over time);

[0133] 5. Repeat steps 3 and 4 until the network converges;

[0134] 6. Cluster analysis: Analyze the formed SOM network to identify typical deformation patterns and their characteristics.

[0135] Through the above algorithm, the system can learn the normal deformation patterns and abnormal deformation patterns of the steel grid structure 1 from historical data, providing data support for threshold adjustment.

[0136] The threshold optimization module 54 comprehensively considers the outputs of the environmental factor compensation module 51, the structural characteristic analysis module 52, and the historical data learning module 53, and dynamically adjusts the risk judgment threshold. In an embodiment of the present invention, this module adopts a multi-factor comprehensive evaluation method based on fuzzy logic to achieve the optimized adjustment of the threshold. Specifically, the module first converts each input factor into a fuzzy set, then calculates the comprehensive evaluation value through fuzzy inference rules, and finally obtains the finally adjusted threshold through the defuzzification process. The core threshold adjustment formula is as follows:

[0137] ,

[0138] where is the reference threshold (preferably 0.6), , and are the adjustment coefficients based on environmental factors, structural characteristics, and historical data respectively, and the value range of each coefficient is usually [-0.3, 0.3].

[0139] This adaptive threshold adjustment mechanism can dynamically adjust the warning threshold according to the actual situation of the steel grid structure 1, greatly improving the intelligence and warning accuracy of the system. For example, in hot summer weather, the system will appropriately increase the deformation threshold caused by thermal expansion; for old structures, the system will lower the deformation threshold to increase the safety margin; for areas that have experienced anomalies many times in history, the system will set a more sensitive threshold to strengthen monitoring.

[0140] Embodiment 12: Multimodal Data Fusion Algorithm

[0141] As Figure 11 shown, the present invention further provides a multimodal data fusion algorithm for integrating data from different types of sensors (displacement, deformation, stress, strain, etc.) to achieve a more comprehensive and accurate structural state assessment. This algorithm makes full use of the complementarity and synergy between multi-dimensional monitoring data, significantly improving the monitoring ability and warning accuracy of the system.

[0142] Traditional structural monitoring systems usually process different types of monitoring data independently and lack the ability to comprehensively analyze multi-source heterogeneous data. The multimodal data fusion algorithm of the present invention solves this problem and realizes all-round fusion from the data layer, feature layer to the decision layer.

[0143] This algorithm mainly includes the following core modules: data preprocessing module 61, feature extraction module 62, feature fusion module 63 and decision fusion module 64. These modules constitute a complete multimodal data fusion system.

[0144] The data preprocessing module 61 is responsible for cleaning, standardizing and synchronizing different types of original sensor data. Since the sampling frequencies, value ranges and noise characteristics of different types of sensors are different, data preprocessing is a necessary step for fusion analysis. In an embodiment of the present invention, this module first calibrates the timestamps of various data to ensure the time consistency of the data; then uses the Z-score normalization method to unify different types of data to the same scale:

[0145] ,

[0146] where is the original data value, is the mean value of this type of data, is the standard deviation, is the normalized data value. For noise processing, the module selects an appropriate filtering algorithm according to the characteristics of different sensors, such as low-pass filtering for displacement data and median filtering for strain data.

[0147] The feature extraction module 62 is responsible for extracting effective features from various preprocessed data. In one embodiment of the present invention, this module adopts different feature extraction strategies for different types of data: for displacement data, time-domain features (such as mean, standard deviation, peak value, etc.) and frequency-domain features (such as main frequency, energy distribution, etc.) are extracted; for stress-strain data, statistical features (such as extreme value, change rate, etc.) and pattern features (such as typical stress-strain curve features) are extracted; for deformation data, spatial distribution features (such as deformation area size, shape, etc.) are extracted. Specifically, taking the time-domain feature extraction as an example, the module calculates the following feature indicators:

[0148] 1. Mean: ,

[0149] 2. Standard deviation: ,

[0150] 3. Peak-to-peak value: ,

[0151] 4. Kurtosis coefficient: ,

[0152] 5. Skewness: ,

[0153] For frequency-domain features, the module first converts the time-domain data to the frequency domain through the fast Fourier transform (FFT):

[0154] ,

[0155] Then the following frequency-domain features are extracted:

[0156] 1. Main frequency: The frequency point with the largest amplitude in the spectrum

[0157] 2. Band energy: The sum of the spectrum energy within a specific frequency band

[0158] 3. Spectrum entropy: , where is the normalized spectrum amplitude

[0159] The feature fusion module 63 fuses the features extracted from different types of data to generate a comprehensive feature vector. In one embodiment of the present invention, this module adopts a feature fusion algorithm based on a deep autoencoder. The autoencoder consists of an encoder and a decoder. The encoder maps the input features to the latent space, and the decoder then attempts to reconstruct the original features from the latent space. By minimizing the reconstruction error, the autoencoder can learn an effective representation of the data. Its network structure is as follows:

[0160] 1. Input layer: The dimension is the sum of the dimensions of various features, such as ;

[0161] 2. The encoder layer contains multiple hidden layers that gradually reduce the dimension, such as ;

[0162] 3. Latent space layer: Represents the fused feature space, usually with a dimension of 32;

[0163] 4. Decoder layer: Symmetric to the encoder layer, gradually restoring the dimension, such as ;

[0164] 5. Output layer: Has the same dimension as the input layer and is used to reconstruct the original features;

[0165] The mean squared error is used as the loss function during the training process:

[0166] ,

[0167] where is the original feature of the th sample, and is the reconstructed feature.

[0168] After training, the output of the latent space layer is the fused feature representation, which retains both the key information of various types of original data and captures the correlations between different types of data.

[0169] The decision fusion module 64 generates the final risk assessment result based on the fused features and combines the outputs of multiple classifiers. In one embodiment of the present invention, this module uses an ensemble learning method to combine the decision results of multiple base classifiers. Specifically, the module uses three different classifiers: Support Vector Machine (SVM), Random Forest (RF), and Deep Neural Network (DNN). Each classifier independently classifies the fused features, and then the final decision is obtained through weighted voting:

[0170] ,

[0171] where represents the predicted probability of the th classifier for class , and is the weight of the th classifier (determined by the performance on the validation set. Preferably, the weight of SVM is 0.3, the weight of RF is 0.3, and the weight of DNN is 0.4).

[0172] To further improve the classification accuracy, the decision fusion module 64 also uses the Dempster-Shafer evidence theory for uncertainty processing. By treating the outputs of each classifier as evidence and combining these evidences based on specific rules, this theory can better handle the uncertainty and conflict situations in classification.

[0173] This multi-modal data fusion algorithm significantly improves the system's perception ability and analysis accuracy of the state of the steel grid structure 1. For example, when the displacement data shows that there is deformation at a certain part of the structure, while the stress data shows that the stress at that part is within the normal range, traditional single-modal analysis may be difficult to judge the actual situation; through multi-modal fusion analysis, the system can combine multi-dimensional information to make a more accurate judgment, such as identifying that this is normal seasonal thermal expansion and contraction rather than structural damage.

[0174] Example 13: Spatiotemporal Correlation Analysis and Anomaly Detection

[0175] As Figure 12 shown, the present invention also provides an anomaly detection method based on spatiotemporal correlation analysis for mining hidden patterns and potential risks in the deformation data of the steel grid structure 1. By analyzing the correlation of deformation data in the time and space dimensions, this method can identify abnormal phenomena that are difficult to discover by traditional methods and improve the early warning ability of the system.

[0176] In the complex steel grid structure 1, there is usually a certain spatiotemporal correlation in the deformation data of each monitoring point. For example, the deformation trends of adjacent regions usually have similarity; the deformation data of the same region at different time points usually has temporal continuity. The spatiotemporal correlation analysis method of the present invention is precisely based on these characteristics, and discovers potential structural anomalies by identifying data points that violate the normal correlation pattern.

[0177] This method mainly includes the following core modules: a spatial correlation analysis module 71, a temporal correlation analysis module 72, a spatiotemporal joint analysis module 73, and an anomaly detection module 74. These modules together constitute a complete spatiotemporal correlation analysis framework.

[0178] The spatial correlation analysis module 71 is responsible for analyzing the correlation of deformation data between different monitoring points of the steel grid structure 1. In an embodiment of the present invention, this module constructs a spatial correlation model of structural deformation based on graph theory and spatial statistics principles. The specific steps are as follows:

[0179] 1. Construct a spatial relationship graph of monitoring points: Treat each monitoring point as a node in the graph, and establish edge connections according to the physical distance and structural connection relationship between monitoring points;

[0180] 2. Calculate the spatial autocorrelation index: Use Moran's I to evaluate the global spatial autocorrelation:

[0181] ,

[0182] wherein, is the number of monitoring points, and are the deformation values of point and point respectively, is the average deformation value, is the spatial weight (usually defined based on inverse distance or topological relationship);

[0183] 3. Local spatial autocorrelation analysis: The Local Indicators of Spatial Association (LISA) statistic is used to evaluate the local spatial autocorrelation of each monitoring point:

[0184] ,

[0185] 4. Spatial hot spot analysis: The Getis-Ord Gi* statistic is used to identify spatial clustering patterns (hot spots and cold spots):

[0186] ,

[0187] wherein, .

[0188] Through the above analysis, the system can identify which areas have significant spatial clustering of deformation and which monitoring points have inconsistent deformation with the surrounding environment, thereby discovering potential local anomalies.

[0189] The time correlation analysis module 72 is responsible for analyzing the correlation patterns of deformation data in the time dimension. In an embodiment of the present invention, this module uses time series analysis and change point detection methods to identify time anomalies in the deformation data. The specific steps are as follows:

[0190] 1. Time series decomposition: Decompose the deformation time series into a trend term, a seasonal term, and a residual term:

[0191] ,

[0192] wherein, is the trend term, is the seasonal term, is the residual term;

[0193] 2. Autocorrelation analysis: Calculate the autocorrelation function (ACF) and partial autocorrelation function (PACF) of the time series to identify time-dependent patterns:

[0194] ,

[0195] Among them, is the autocorrelation coefficient of lag ;

[0196] 3. ARIMA model fitting: Based on the identified time-dependent pattern, construct an ARIMA(p,d,q) model:

[0197] ,

[0198] Among them, is the lag operator, and are the autoregressive and moving average polynomials respectively, is the order of differencing, is white noise;

[0199] 4. Change point detection: Use the CUSUM (Cumulative Sum) or PELT (Pruned Exact Linear Time) algorithm to detect change points in the time series:

[0200] ,

[0201] When exceeds a specific threshold, a change point is considered to exist.

[0202] Through the above analysis, the system can identify abnormal changes in the deformed data over time, such as mutations, trend changes, or disruptions in seasonal patterns, etc., which may all be early signals of structural state changes.

[0203] The spatio-temporal joint analysis module 73 fuses the results of spatial and temporal correlation analyses to achieve a more comprehensive understanding of correlation. In an embodiment of the present invention, this module uses the spatio-temporal autoregressive model (STAR) and spatio-temporal kernel density estimation (ST-KDE) methods to construct a spatio-temporal joint distribution model of the deformed data. The general form of the STAR model is:

[0204] ,

[0205] Among them, represents the deformation value at position at time , is the spatial lag parameter, is the time lag parameter, is the spatio-temporal interaction parameter, is the random error term.

[0206] The ST-KDE method estimates the spatio-temporal joint density function through the following formula:

[0207] ,

[0208] Among them, represents the spatial position and time point, represents the sample point, and are the spatial and temporal kernel functions respectively and are the spatial and temporal bandwidth parameters respectively.

[0209] Through these models, the system can comprehensively understand the correlation patterns of deformation data in the spatio-temporal dimension, providing a richer information basis for anomaly detection.

[0210] Based on the spatio-temporal correlation analysis results, the anomaly detection module 74 identifies data points that violate the normal correlation pattern and gives an anomaly warning. In one embodiment of the present invention, this module adopts a multi-dimensional anomaly detection method based on Mahalanobis distance and Local Outlier Factor (LOF). Specifically, for each data point, calculate its Mahalanobis distance in the spatio-temporal joint feature space:

[0211] ,

[0212] Among them, is the spatio-temporal feature vector of the data point, is the feature mean vector, is the feature covariance matrix. When exceeds a specific threshold (usually 3.0), mark this point as a potential anomaly.

[0213] For the calculation of the Local Outlier Factor, first determine the of the data point -nearest neighbors (preferably ), then calculate its local reachability density and the Local Outlier Factor relative to the neighbor points:

[0214] ,

[0215] Among them is the of -nearest neighbor set, is the local reachability density of . When is significantly greater than 1 (usually taking the threshold 1.5), it is considered that is a local outlier.

[0216] In addition, the module also combines expert rule knowledge to classify and assess the risks of detected anomalies. For example, points that are abnormal only in the spatial dimension may be judged as local deformation; points that are abnormal only in the temporal dimension may be judged as instantaneous disturbances; and points that are abnormal in both the spatial and temporal dimensions may be judged as serious risks and require immediate action.

[0217] This anomaly detection method based on spatiotemporal correlation analysis can capture abnormal patterns that are difficult to detect with traditional single-point monitoring, especially those that are not obvious at a single monitoring point but deviate significantly from normal behavior in the overall spatiotemporal pattern. For example, the deformation value of a certain monitoring point may be within the allowable range, but its change pattern is inconsistent with the surrounding points or inconsistent with the historical pattern, which may be an early signal of local damage to the structure. By detecting these abnormal patterns early, the system can provide effective early warning before structural problems expand, greatly improving the safety level of the steel grid structure1.

[0218] In summary, the steel grid structure deformation monitoring and early warning intelligent system and method provided by the present invention realizes comprehensive monitoring and accurate early warning of the deformation state of the steel grid structure through distributed monitoring nodes, intelligent data analysis and convolutional neural network early warning. The system also innovatively introduces an adaptive threshold adjustment mechanism, a multimodal data fusion algorithm, and a spatiotemporal correlation analysis and anomaly detection method, which greatly improves the intelligence, adaptability and early warning accuracy of the system. These innovations enable the system to perceive the safety status of the steel grid structure more comprehensively and accurately, effectively improve the safety and service life of the structure, and have broad application prospects.

[0219] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, various improvements and modifications to the present invention without departing from the principles and spirit of the present invention should be considered as the protection scope of the present invention.

[0220] The above description is only a preferred embodiment of the present invention and does not limit the present invention in any way. Any simple modification, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention shall fall within the protection scope of the technical solution of the present invention.

Claims

1. Intelligent system for monitoring and early warning of deformation of steel grid structure, characterized by: include: Steel grid structure; Monitoring nodes are distributedly arranged in the steel grid structure and are used to collect deformation data in real time; Data acquisition and transmission equipment, used for acquiring and transmitting the deformation data; A data processing and analysis system is used to receive the deformation data and perform deformation warning analysis based on a convolutional neural network; the monitoring node is used to monitor the deformation of the steel grid structure, including displacement, deformation, stress, strain or stress-strain, and the data processing and analysis system performs deformation warning analysis using a convolutional neural network, including the following steps: S110, extracting sample images, wherein the number of sample images is greater than 20,000; S120, preprocessing the sample images, including image enhancement and normalization; S130, training the sample images based on the convolutional neural network as a basic model, and obtaining a classification model; S140, establishing an output layer of the convolutional neural network, and obtaining a relationship prediction diagram between each classification of deformation data and a warning level; S150, inputting each classification of deformation data into the classification model to obtain a prediction probability of each classification of deformation data; S160, calculating the sum of the prediction probabilities of each classification of each deformation data, and performing normalization processing; S170, for each classification of deformation data, multiplying the prediction probability by the risk value to obtain a risk prediction for each classification of deformation data; S180, judging whether there is a deformation risk for each classification of deformation data according to the risk threshold; The convolutional neural network adopts the improved VGG16 architecture as the basic model, which includes 13 convolutional layers and 3 fully connected layers; the convolutional layer adopts a small 3x3 convolution kernel, and achieves the effect of a large receptive field by stacking small convolution kernels, while reducing the number of parameters; each group of convolutional layers is followed by a maximum pooling layer to reduce the size of the feature map and extract significant features; The output layer contains 4 neurons, corresponding to the four categories of normal state, slight deformation, moderate deformation and severe deformation; The risk threshold is a threshold determined by an adaptive threshold adjustment mechanism, which includes an environmental factor compensation module, a structural characteristic analysis module, a historical data learning module and a threshold optimization module. The environmental factor compensation module, the structural characteristic analysis module, the historical data learning module and the threshold optimization module work together to achieve intelligent adjustment of the threshold.

2. The steel grid structure deformation monitoring and early warning intelligent system according to claim 1 is characterized in that: The data acquisition and transmission equipment includes sensors, analog-to-digital converters, signal transmitters and signal receivers, which realize the functions of real-time acquisition, automatic analysis and early warning of deformation data of steel grid structures, including transmitting the deformation data to the data processing and analysis system in a wireless or wired manner for further processing.

3. The steel grid structure deformation monitoring and early warning intelligent system according to claim 1 is characterized in that: The data processing and analysis system is also provided with a data storage unit for storing the deformation data and recording the historical deformation data of the steel grid structure.

4. The steel grid structure deformation monitoring and early warning intelligent system according to claim 3 is characterized in that: The data processing and analysis system is also provided with a data comparison unit for comparing the deformation data with the historical deformation data to analyze the deformation trend of the steel grid structure.

5. The steel grid structure deformation monitoring and early warning intelligent system according to claim 1 is characterized in that: The data processing and analysis system is also provided with a data correction unit to correct abnormal or non-compliant deformation data.

6. The steel grid structure deformation monitoring and early warning intelligent system according to claim 1 is characterized in that: The data processing and analysis system draws a two-dimensional or three-dimensional curve and / or surface of deformation data characteristics through the deformation data.

7. The steel grid structure deformation monitoring and early warning intelligent system according to claim 1 is characterized in that: The data processing and analysis system is also provided with a data analysis unit for analyzing the underlying causes of the deformation data of the steel grid structure, including performance measurement, error rate statistics and fault mode identification.

8. A monitoring and early warning method based on the steel grid structure deformation monitoring and early warning intelligent system according to any one of claims 1 to 7, comprising the following steps: S10, distributively installing monitoring nodes in the steel grid structure to collect deformation data at target positions of the steel grid structure in real time; S20, collecting and transmitting the deformation data through data collection and transmission equipment to realize the functions of real-time acquisition, automatic analysis and early warning of deformation data of steel grid structure; S30, receiving and processing the deformation data through the data processing and analysis system, and performing deformation early warning analysis on the deformation data based on convolutional neural network; S40, analyzing the deep-seated causes of the deformation data of the steel grid structure, and continuously optimizing the data processing and analysis system by quantifying the impact of the problem such as performance measurement and error rate statistics.

Images