Deep learning-based concrete filled steel tube node safety state evaluation method

Abstract

The invention discloses a deep learning-based safety state evaluation method for a concrete-filled steel tube node. The method comprises the following steps of: 1, constructing a comprehensive safety evaluation index of the concrete-filled steel tube node; 2, collecting dynamic performance data of the nodes, and performing denoising, normalization and feature extraction processing to form multi-dimensional normalized time sequence data; 3, calculating the weight of each index, and optimizing weight distribution; 4, inputting the multi-dimensional normalized time sequence data into a deep learning model, carrying out model training, verification and optimization in combination with optimized weight distribution, and outputting a preliminary comprehensive safety score and a long-term performance prediction result of the node; and 5, further comprehensively evaluating the safety state of the node, determining a final comprehensive safety score, and grading to formulate corresponding grading maintenance suggestions. And 6, dynamically predicting the future performance change trend of the node, and quantifying the fatigue life, the crack growth rate and the material degradation condition of the node. The method is suitable for dynamic safety evaluation and long-term performance prediction of the concrete-filled steel tube node.

Description

Technical Field

[0001] The present invention belongs to the technical field of civil engineering, and specifically relates to a safety state evaluation method for concrete-filled steel tubular joints based on deep learning. Background Art

[0002] As a key connection form in the field of civil engineering, concrete-filled steel tubular joints are widely used in important structures such as high-rise buildings, bridges, and industrial plants, and their safety directly affects the stability of the overall structure. However, there are certain limitations in the current technical means for evaluating the safety of concrete-filled steel tubular joints. The existing technologies mainly include the following:

[0003] (1) Safety evaluation method based on the traditional Analytic Hierarchy Process (AHP)

[0004] Technical principle: Construct a safety evaluation index system for the joint based on expert experience, calculate the weights of each index using the analytic hierarchy process, and then comprehensively evaluate the safety of the joint through a mathematical model.

[0005] Disadvantages or problems:

[0006] 1) Strong subjectivity: The evaluation system and weight assignment rely on expert experience, lack objective data support, and are easily interfered by human factors.

[0007] 2) Limited computing power: The traditional analytic hierarchy process cannot efficiently process the multi-dimensional data of the joint under complex load conditions and is difficult to reflect the dynamic changes of the joint performance.

[0008] 3) Insufficient index coverage: It mainly focuses on a single bearing capacity index and ignores the comprehensive performance such as the fatigue life, deformation performance, and long-term durability of the joint.

[0009] (2) Safety evaluation method based on the fuzzy comprehensive evaluation method

[0010] Technical principle: By constructing a multi-layer fuzzy evaluation model, fuzzify various performance indexes of the concrete-filled steel tubular joint and calculate the comprehensive safety score of the joint.

[0011] Disadvantages or problems:

[0012] 1) Low data dependence: The fuzzy comprehensive evaluation method fails to fully utilize the dynamic monitoring data of the joint and only evaluates based on laboratory test data or limited historical data, making it difficult to truly reflect the performance of the joint under actual working conditions.

[0013] 2) Weak dynamic evaluation ability: The existing methods are mainly used for static safety evaluation and lack the ability to predict the dynamic performance such as fatigue damage, crack propagation, and long-term aging of the joint under complex working conditions.

[0014] (3) Insufficient processing of index correlation: There may be coupling relationships between different performance indicators (such as bearing capacity and deformation performance), and existing methods fail to scientifically quantify these relationships, resulting in possible biases in evaluation results.

[0015] (3) Safety evaluation method for concrete-filled steel tubular joints based on static experiments

[0016] Technical principle: By applying static loads to the joints under laboratory conditions, measuring the deformation, stress, bearing capacity and other performances of the joints, and evaluating their safety.

[0017] Disadvantages or problems:

[0018] (1) Single experimental conditions: Static experiments are mostly carried out under idealized conditions, ignoring the influence of complex loads and environmental factors (such as temperature and humidity, corrosion) that the joints are subjected to in the actual engineering environment, and the evaluation results may not conform to the actual situation.

[0019] (2) Lack of dynamic response ability: Static experiments cannot evaluate the dynamic behaviors such as fatigue performance, crack propagation rate and long-term durability of the joints, and it is difficult to provide a scientific basis for the whole life cycle design and maintenance of the structure.

[0020] (3) Low economy and efficiency: Static experiments usually consume a large amount of time and cost, and are not suitable for popularization and application in large-scale engineering projects.

[0021] (4) Safety evaluation method for joints based on finite element numerical simulation

[0022] Technical principle: With the help of finite element analysis software (such as ABAQUS), establish a numerical model of the concrete-filled steel tubular joint, and calculate the mechanical properties and failure behaviors of the joint by simulating complex loads and boundary conditions.

[0023] Disadvantages or problems:

[0024] (1) Model simplification affects accuracy: The finite element model needs to simplify the actual joint, which may ignore complex behaviors such as material nonlinearity and crack propagation, resulting in a large deviation between the evaluation result and the actual situation.

[0025] (2) Limited data input: The simulation analysis relies on static or laboratory data and does not combine real-time monitoring data, so it cannot comprehensively reflect the performance state of the joint under actual working conditions.

[0026] (3) Large amount of calculation: Finite element analysis under complex working conditions requires high computing resources and time, which limits its real-time performance and practicality.

[0027] To sum up, the existing technologies have the following prominent disadvantages and problems in the safety evaluation of concrete-filled steel tubular joints:

[0028] (1) Limitations of evaluation methods: Existing methods mostly adopt traditional analytical methods and experimental data, unable to handle the dynamic performance under complex loads and environmental conditions, and lacking the ability to predict long-term safety.

[0029] (2) Incompleteness of the index system: Existing technologies overly focus on a single bearing capacity index for the safety evaluation of joints, while ignoring comprehensive performances such as fatigue life, deformation performance, and long-term durability. The evaluation results may not comprehensively reflect the true safety state of the joints.

[0030] (3) Lack of dynamic monitoring and data-driven support: Existing methods mainly rely on static experimental data or limited historical data, and do not fully utilize the dynamic monitoring data obtained by modern sensing technologies, resulting in insufficient real-time reflection of the joint performance state.

[0031] (4) Absence of dynamic safety evaluation: Current technologies are difficult to achieve the dynamic safety assessment of joints under complex loads, long-term use, or harsh environmental conditions, especially lacking effective prediction means in aspects such as fatigue damage, crack propagation, and material aging.

[0032] (5) Limitations in result application: Existing technologies mostly stay at the level of simple safety scoring or grade division, and fail to provide specific maintenance decision-making suggestions based on the evaluation results, restricting their engineering application value. Summary of the Invention

[0033] In view of the problems existing in the above-mentioned prior art, the present invention provides a safety state evaluation method for concrete-filled steel tube joints based on deep learning. The purpose is to construct a comprehensive safety evaluation system for concrete-filled steel tube joints including bearing capacity, deformation performance, fatigue life, durability, and joint stability. This system includes data collection and processing, index weight calculation, deep learning model prediction, fuzzy comprehensive evaluation, and hierarchical maintenance suggestions. Combining dynamic monitoring data and using deep learning models to accurately evaluate the safety state of joints under complex loads and environmental conditions, outputting the joint state through comprehensive scoring and safety grades, and providing hierarchical maintenance suggestions to overcome the subjectivity and limitations of traditional evaluation methods.

[0034] To achieve the above purpose, the specific solutions of the present invention are as follows:

[0035] The safety state evaluation method for concrete-filled steel tube joints based on deep learning includes the following steps:

[0036] Step 1, construct a comprehensive safety evaluation index system for concrete-filled steel tube joints, and the comprehensive safety evaluation index system includes a bearing capacity index, a deformation performance index, a fatigue life index, a durability index, and a joint stability index;

[0037] Step 2: According to the comprehensive safety evaluation index system constructed in Step 1, collect the dynamic performance data of the displacement, strain, crack propagation, material aging and environmental parameters of the concrete-filled steel tubular joints, and perform denoising, normalization and feature extraction on the collected dynamic performance data to form multi-dimensional normalized time series data;

[0038] Step 3: Calculate the weights of the comprehensive safety evaluation index system described in Step 1 and optimize the weight distribution;

[0039] Step 4: Input the multi-dimensional normalized time series data described in Step 2 into the deep learning model, and combine with the optimized weight distribution in Step 3 to perform model training, verification and optimization, and output the preliminary comprehensive safety score and long-term performance prediction results of the concrete-filled steel tubular joints. The long-term performance prediction results include fatigue life, crack propagation rate and material deterioration rate;

[0040] Step 5: Based on the preliminary comprehensive safety score described in Step 4, use the fuzzy comprehensive evaluation method to further comprehensively evaluate the safety status of the concrete-filled steel tubular joints, determine the final comprehensive safety score, divide the comprehensive safety into grades according to the final comprehensive safety score, and formulate corresponding hierarchical maintenance suggestions according to the divided grades;

[0041] Step 6: Based on the long-term performance prediction results described in Step 4, further dynamically predict the future performance change trend of the concrete-filled steel tubular joints, and quantify the fatigue life, crack propagation rate and material deterioration of the joints.

[0042] Furthermore, the bearing capacity indexes described in Step 1 include ultimate bearing capacity and yield bearing capacity, the deformation performance indexes include ultimate deformation and stiffness degradation rate, the fatigue life indexes include fatigue limit and fatigue crack propagation rate, the durability indexes include environmental durability and material deterioration rate, and the joint stability indexes include stability coefficient.

[0043] Furthermore, the calculation formula of the ultimate bearing capacity is as follows:

[0044]

[0045] Where: F max is the maximum load in the test of the concrete-filled steel tubular joint; A is the cross-sectional area of the concrete-filled steel tubular joint; the calculation formula of the yield bearing capacity is as follows:

[0046]

[0047] Where: Fyield is the yield load; A is the cross-sectional area of the concrete-filled steel tubular joint;

[0048] The calculation formula of the ultimate deformation is as follows:

[0049]

[0050] Where: δ max is the ultimate deformation; L is the length of the concrete-filled steel tubular joint;

[0051] The calculation formula of the stiffness degradation rate is as follows:

[0052]

[0053] Where: K i is the initial stiffness; K f is the final stiffness;

[0054] The calculation formula of the fatigue limit is as follows:

[0055]

[0056] Where: ΔF fatigue is the fatigue limit load, and A is the cross-sectional area;

[0057] The calculation formula of the fatigue crack growth rate is as follows:

[0058]

[0059] Where: a f is the final crack length; a i is the initial crack length; N is the number of cycles;

[0060] The calculation formula of the material deterioration rate is as follows:

[0061]

[0062] Where: P 0 is the initial performance value; P t is the performance value after t years; t is the time;

[0063] The calculation formula of the stability coefficient is as follows:

[0064]

[0065] Where: F cr is the critical load, and F applied is the actual acting load.

[0066] Furthermore, the steps of denoising, normalizing, and feature extraction for the dynamic performance data in step 2 are as follows:

[0067] Step 21, Denoising: Use the wavelet denoising algorithm to denoise the dynamic performance data described in Step 2. The dynamic performance data is multi-dimensional data. Remove the random noise in the signal, eliminate the outliers in the multi-dimensional data, and use the linear interpolation method or the mean filling method to process the missing values in the multi-dimensional data;

[0068] Step 22, Normalization: Normalize the multi-dimensional data described in Step 21 to form multi-dimensional normalized time series data. The calculation formula for normalization is as follows:

[0069]

[0070] In the formula, X normalized represents the normalized dynamic performance data; X represents the dynamic performance data; X min represents the minimum value of the dynamic performance data; X max represents the maximum value of the dynamic performance data;

[0071] Step 23, Feature Extraction: Use principal component analysis to extract the main features of the multi-dimensional normalized time series data described in Step 22. Set the cumulative contribution rate to be retained at 99%, determine the number of principal components, and extract the key feature values from the multi-dimensional normalized time series data. The key feature values include the maximum value, the minimum value, and the change rate. Classify the processed dynamic performance data according to time and working conditions.

[0072] Further, the steps for optimizing the weight allocation in Step 3 are as follows:

[0073] Step 31, Construct an index comparison matrix through the analytic hierarchy process and invite experts in the field of civil engineering to score;

[0074] Step 32, Calculate the weights and verify the consistency ratio to ensure that CR < 0.1. Otherwise, adjust the scores;

[0075] Step 33, Adjust the preliminary weights in multiple rounds to form the weight allocation results for the middle layer and the bottom layer indicators, and verify the weight allocation results with the experimental data.

[0076] Further, the structure of the deep learning model in Step 4 includes an input layer, a hidden layer, and an output layer. The input layer is used to receive multi-dimensional normalized time series data. The format of the input time series data is (T, F), where T represents the time step and F represents the number of features;

[0077] The hidden layer contains a three-layer network structure. The first layer is a long short-term memory network layer for extracting the long-term dependence features of multi-dimensional normalized time series data. The second layer is a convolutional neural network layer for extracting the local features of time series data. The third layer is a fully connected layer for integrating the extracted features and mapping them to the output result;

[0078] The output layer is used to output the comprehensive safety score and the long-term performance prediction value of the concrete-filled steel tubular joint. The scoring range of the comprehensive safety score of the concrete-filled steel tubular joint is 0 to 1.

[0079] Furthermore, the steps of the long-term performance prediction in step 4 are as follows:

[0080] Step 41, use a deep learning model to obtain the prediction results of the fatigue life, crack propagation rate, and material deterioration rate of the concrete-filled steel tubular joint within a specific future time period, and output the time series prediction curve of the performance change of the concrete-filled steel tubular joint;

[0081] Step 42, formulate a maintenance and reinforcement plan for the concrete-filled steel tubular joint according to the prediction results of step 41.

[0082] Furthermore, the steps of determining the final comprehensive safety score in step 5 are as follows:

[0083] Step 51, construct a membership function: according to the evaluation criteria of each underlying index of the comprehensive safety evaluation index, define the membership function respectively; the underlying indexes include ultimate bearing capacity, yield bearing capacity, ultimate deformation, stiffness degradation rate, fatigue limit, fatigue crack propagation rate, environmental durability, material deterioration rate, and stability coefficient;

[0084] Step 52, calculate the membership values: according to the membership function defined in step 51, calculate the membership values of each underlying index, and form a fuzzy evaluation matrix with the membership values of all underlying indexes;

[0085] Step 53, comprehensively calculate the safety score: combine the fuzzy evaluation matrix obtained in step 52 with the weights determined in step 3, and determine the final comprehensive safety score of the concrete-filled steel tubular joint through the fuzzy comprehensive evaluation method.

[0086] Furthermore, the scoring range of the final comprehensive safety score in step 5 is 0 to 1, and the grade is divided into grade A, grade B, grade C, grade D, and grade E;

[0087] The comprehensive safety score range of grade A is 0.9 to 1.0, indicating that the comprehensive safety performance reaches an excellent level, and the maintenance suggestion is that the joint performance is good, and only a routine inspection is required every six months;

[0088] The comprehensive safety score range of grade B is 0.7 to 0.9, indicating that the comprehensive safety performance reaches a good level, and the maintenance suggestion is to check every quarter and perform local maintenance in a timely manner;

[0089] The comprehensive safety score range for Class C is 0.5 - 0.7, indicating that the comprehensive safety performance is at a medium level. The maintenance suggestion is that there are certain potential hazards and comprehensive reinforcement is required during the next major overhaul.

[0090] The comprehensive safety score range for Class D is 0.3 - 0.5, indicating that the comprehensive safety performance is at a poor level. The maintenance suggestion is that the safety is poor and reinforcement or partial replacement is required as soon as possible.

[0091] The comprehensive safety score range for Class E is 0 - 0.3, indicating that the comprehensive safety performance is at an extremely poor level. The maintenance suggestion is that there are serious potential hazards and it is necessary to immediately stop using and conduct large-scale replacement.

[0092] Advantages of the present invention

[0093] 1. The safety state evaluation method for concrete-filled steel tubular joints based on deep learning of the present invention collects multi-dimensional data of concrete-filled steel tubular joints under complex working conditions, including deformation, strain, crack propagation rate and environmental durability of the joints, by introducing dynamic monitoring devices such as laser displacement sensors, strain monitors, and non-destructive testing equipment. Wavelet denoising and normalization techniques are used for data processing to avoid the influence of human interference and subjective factors, and comprehensively improve the objectivity and scientificity of the evaluation.

[0094] 2. The present invention constructs a complete safety evaluation index system for concrete-filled steel tubular joints, including five categories of indicators: bearing capacity, deformation performance, fatigue life, durability and joint stability, covering key parameters such as ultimate bearing capacity, stiffness degradation rate, fatigue crack propagation rate, material deterioration rate and stability coefficient. The index system comprehensively covers the performance of concrete-filled steel tubular joints in aspects such as short-term bearing capacity, long-term fatigue life and durability; and classifies the joints according to the comprehensive safety score, and provides detailed maintenance suggestions from "excellent" that requires no maintenance to "extremely poor" that requires stopping use and comprehensive replacement. It fully considers the multi-dimensional performance requirements of the joints in practical engineering applications. Compared with the limitation of traditional methods that overly rely on a single bearing capacity index, the multi-dimensional index system of the present invention can comprehensively reflect the comprehensive performance of the joints, has engineering practicability and decision-making guiding significance, and can effectively reduce the maintenance cost and safety risk during the use of the structure. Especially for various failure modes under complex loads and environmental conditions. The weights of each index are determined by the analytic hierarchy process (AHP) and the Delphi method to ensure the scientificity and rationality of the weight distribution, and the correlation and coupling relationship between each index is quantified by the fuzzy comprehensive evaluation method. And it is applicable to engineering application scenarios of various concrete-filled steel tubular joints, including the safety assessment of joints in structures such as high-rise buildings, bridges, large stadiums and industrial factories. Whether it is laboratory research or on-site real-time monitoring, it can be flexibly adapted, and has significant engineering application value.

[0095] 3. The present invention combines dynamic monitoring data with deep learning technology to achieve the dynamic safety assessment of concrete-filled steel tubular joints under complex loads and environmental conditions: Dynamic monitoring can obtain multi-source data such as joint deformation, strain, and crack propagation in real time and capture the performance changes of joints under complex working conditions such as fatigue damage, crack propagation, and long-term aging. The dynamic monitoring ability enables the present invention to be applicable not only to the safety assessment of joints under laboratory conditions but also to provide real-time feedback on the state changes of joints in engineering practice, offering real-time early warning and prediction capabilities. It solves the problem that traditional evaluation methods are mostly based on static experimental data and cannot effectively evaluate the performance changes of joints under complex loads and long-term working conditions. Through the deep neural network (DNN) model, the performance of concrete-filled steel tubular joints under various dynamic working conditions is predicted and evaluated, which can accurately capture dynamic behaviors such as joint fatigue damage, crack propagation, and material aging, realize accurate modeling of high-dimensional and non-linear data, and enhance the ability to capture the dynamic performance changes of concrete-filled steel tubular joints. The deep learning model is based on multiple rounds of iterative training, combined with multi-dimensional data input and adaptive parameter optimization, and can predict the long-term safety state of concrete-filled steel tubular joints, providing a scientific basis for the whole life cycle management of structures. By combining dynamic performance data monitoring with deep learning technology, the present invention effectively makes up for the reliability deficiency problems caused by low data dimension and weak analysis ability in traditional methods; the evaluation results are based on a scientific index system and data analysis methods, with higher accuracy and credibility. The dynamic prediction method overcomes the limitations that existing static experiments and finite element simulations cannot reflect the performance changes of joints in real time. The deep learning model can automatically learn key features from large-scale monitoring data, replacing traditional evaluation methods based on expert experience, and significantly reducing the errors caused by subjectivity. Through the fuzzy comprehensive evaluation method, weighted comprehensive analysis is carried out on each evaluation index, combined with the prediction results of the deep learning model, ensuring the accuracy and rationality of the final score. At the same time, it can output the safety level of the concrete-filled steel tubular joint and the corresponding detailed score, solving the problem that the evaluation results in the existing technology are mostly simple classifications and do not reflect the weights and correlations of various indicators in the evaluation.

[0096] 4. On the basis of outputting the comprehensive safety score, the present invention divides the safety state of concrete-filled steel tubular joints into five grades and provides classified maintenance suggestions for different safety grades. The maintenance suggestions combine dynamic monitoring data and long-term performance prediction, providing scientific support for joint maintenance decisions in engineering practice and significantly improving the safety and operation efficiency of engineering structures.

[0097] 5. The deep learning technology of the present invention is used for dynamic performance modeling and prediction, and the fuzzy comprehensive evaluation method is used for the comprehensive scoring of the safety state. The combination of the two realizes scientific and dynamic safety evaluation: the deep learning model captures the complex relationships of multi-dimensional data through end-to-end learning, solving the limitations of traditional methods in dealing with non-linear and high-dimensional data; the fuzzy comprehensive evaluation method introduces the membership function and the fuzzy matrix, combines various quantitative indicators and weights, and outputs clear comprehensive scoring results; compared with traditional single methods, the present invention has significantly improved in terms of evaluation accuracy and practicality. It can not only accurately evaluate the current safety state of the node, but also, based on the long-term prediction ability of the deep learning model, predict the performance change trend of the node under different working conditions in the future, providing a scientific basis for formulating node maintenance and reinforcement strategies, and significantly improving the predictability and accuracy of maintenance. For example, it can predict the long-term performance of concrete-filled steel tubular joints such as the fatigue crack propagation rate and the material degradation rate, predict the fatigue crack propagation rate, and quantify the whole process from crack initiation to failure; evaluate the material degradation rate and predict the remaining life of the node under different environmental conditions; provide the long-term performance prediction curve of the node, providing a reliable basis for structural design, maintenance planning and life extension decision-making. Through dynamic performance prediction, the present invention effectively solves the defect that traditional static experimental methods and finite element numerical simulation methods cannot evaluate long-term dynamic performance.

[0098] 6. Compared with the traditional analytic hierarchy process (AHP) and fuzzy comprehensive evaluation method, the present invention uses a deep learning model to process high-dimensional and multi-nonlinear data, and has stronger computing power and the ability to process complex data. Especially under dynamic working conditions, it can quickly respond to high-frequency real-time monitoring data, ensuring the efficiency and timeliness of the safety evaluation of concrete-filled steel tubular joints. By introducing deep learning technology and being driven by multi-dimensional real-time data, the objectivity and scientificity of the evaluation are significantly improved, avoiding the situation that the existing safety evaluation methods of concrete-filled steel tubular joints mostly rely on expert experience, have strong subjectivity, and are prone to distortion of evaluation results due to human errors.

[0099] In summary, the present invention has significant advantages in terms of evaluation scientificity, data-drivenness, dynamic response ability, long-term performance prediction and hierarchical maintenance suggestions, and can provide comprehensive technical support for the design, use and maintenance of concrete-filled steel tubular joints, effectively improving the safety and economy of engineering structures. BRIEF DESCRIPTION OF THE DRAWINGS

[0100] Figure 1 It is a flowchart of the safety state evaluation method for concrete-filled steel tubular joints based on deep learning of the present invention.

[0101] Figure 2 is Figure 1 the safety evaluation index system of the concrete-filled steel tubular joint in DETAILED DESCRIPTION OF THE EMBODIMENTS

[0102] The present invention will be further explained and illustrated below in conjunction with the accompanying drawings and specific embodiments. It should be noted that the specific embodiments are not used to limit the scope of the rights of the present invention.

[0103] As Figure 1 and Figure 2 shown, this specific embodiment provides a safety status evaluation method for concrete-filled steel tubular joints based on deep learning, including the following steps:

[0104] Step 1: Construct a comprehensive safety evaluation index system for concrete-filled steel tubular joints. The comprehensive safety evaluation index system includes bearing capacity index, deformation performance index, fatigue life index, durability index, and joint stability index;

[0105] The judgment criteria for constructing the comprehensive safety evaluation index system of concrete-filled steel tubular joints are as follows:

[0106] (1) Bearing capacity index

[0107] The bearing capacity index includes ultimate bearing capacity and yield bearing capacity.

[0108] Ultimate bearing capacity: It refers to the bearing capacity of the concrete-filled steel tubular joint under the action of the maximum load. Through experimental tests, under the action of gradually increasing loads, record the maximum bearing capacity when the joint reaches the ultimate state. The calculation formula is:

[0109]

[0110] In the formula: F max is the maximum load in the test of the concrete-filled steel tubular joint, and A is the cross-sectional area of the concrete-filled steel tubular joint.

[0111] Yield bearing capacity: It refers to the bearing capacity of the concrete-filled steel tubular joint in the yield state. Measure the load when the joint yields through experiments and calculate it in combination with the geometric parameters of the joint. The calculation formula for the yield bearing capacity is:

[0112]

[0113] In the formula: Fyield is the yield load, and A is the cross-sectional area of the concrete-filled steel tubular joint.

[0114] Record the design load as L d ', and the evaluation criteria for the ultimate bearing capacity P u are shown in Table 1.

[0115] Table 1 Evaluation criteria for ultimate bearing capacity

[0116]

[0117] Record the design load as Ld ', and the evaluation criteria for the yield bearing capacity Py are shown in Table 2.

[0118] Table 2 Evaluation Criteria for Yield Bearing Capacity

[0119]

[0120] (2) Deformation performance indicators

[0121] The deformation performance indicators include the ultimate deformation and the stiffness degradation rate.

[0122] Ultimate deformation: The maximum deformation of the concrete-filled steel tubular joint under the ultimate load. The displacement or deformation of the joint under the ultimate load is measured through experiments, and the percentage of the deformation relative to the joint length is calculated. The calculation formula is:

[0123]

[0124] In the formula: δ max is the ultimate deformation, and A is the length of the concrete-filled steel tubular joint.

[0125] Stiffness degradation rate: It refers to the rate of decrease in the stiffness of the concrete-filled steel tubular joint under the action of the load. Through multiple cyclic loading experiments, the stiffness values of each loading are recorded, and the rate of decrease in the stiffness is calculated. The calculation formula is:

[0126]

[0127] In the formula: K i is the initial stiffness, and K f is the final stiffness.

[0128] Denote the joint length as L k , and the evaluation criteria for the ultimate deformation Δ max are shown in Table 3.

[0129] Table 3 Evaluation Criteria for Ultimate Deformation

[0130]

[0131] The evaluation criteria for the stiffness degradation rate D s are shown in Table 4.

[0132] Table 4 Evaluation Criteria for Stiffness Degradation Rate

[0133]

[0134] (3) Fatigue life indicators

[0135] The fatigue life indicators include the fatigue limit and the fatigue crack growth rate.

[0136] Fatigue limit: The maximum stress amplitude at which a concrete-filled steel tubular joint does not experience fatigue failure under cyclic loading. Through fatigue experiments, the cyclic load is gradually increased, and the maximum stress amplitude before the joint undergoes fatigue fracture is recorded. The calculation formula is:

[0137]

[0138] In the formula: ΔF fatigue is the fatigue limit load, and A is the cross-sectional area.

[0139] Fatigue crack growth rate: The rate at which cracks in the material of a concrete-filled steel tubular joint expand under fatigue. Use non-destructive testing methods (such as ultrasonic testing) to measure the change in crack length over time and calculate the fatigue crack growth rate. The calculation formula is:

[0140]

[0141] In the formula: a f is the final crack length, a i is the initial crack length, and N is the number of cycles.

[0142] Define the fatigue load as L fd , and the evaluation criteria for the fatigue limit σ f are shown in Table 5.

[0143] Table 5 Evaluation criteria for fatigue limit

[0144]

[0145] The evaluation criteria for the fatigue crack growth rate v c are shown in Table 6.

[0146] Table 6 Evaluation criteria for fatigue crack growth rate

[0147]

[0148] (4) Durability index

[0149] The durability index includes environmental durability and material deterioration rate.

[0150] Environmental durability: The service life of a concrete-filled steel tubular joint under the action of a specific environment. Through long-term experiments or environmental simulations, combined with the experimental data of the weather resistance of the joint material, the service life of the joint in different environments is estimated.

[0151] Material deterioration rate: The rate at which the performance of the joint material decreases during long-term use. Through the aging experiment of the material, measure the decrease in the performance of the material during use and calculate the annual deterioration rate. The calculation formula is:

[0152] ​

[0153] Where: P 0 is the initial performance value, and P t is the performance value after t years, where t is the time (in years).

[0154] The evaluation criteria for environmental durability are shown in Table 7.

[0155] Table 7 Evaluation Criteria for Environmental Durability

[0156]

[0157] The evaluation criteria for the material deterioration rate R d are shown in Table 8.

[0158] Table 8 Evaluation Criteria for Material Deterioration Rate

[0159]

[0160] (4) Node stability index

[0161] The said node stability index includes a stability coefficient.

[0162] Stability coefficient: It reflects the overall stability of the concrete-filled steel tube node under various load conditions. Through simulation with structural analysis software (such as ABAQUS), combined with the actual load situation, the stability coefficient of the node is calculated. The calculation formula is:

[0163]

[0164] Where: F cr is the critical load, and F applied is the actual applied load.

[0165] The evaluation index of the node stability index K s is shown in Table 9.

[0166] Table 9 Evaluation Criteria for Stability Coefficient

[0167]

[0168] Step 2: According to the comprehensive safety evaluation index system constructed in Step 1, collect the dynamic performance data of the displacement, strain, crack propagation, material aging, and environmental parameters of the concrete-filled steel tube node under various complex loads and environmental conditions. The dynamic performance data is multi-dimensional data, and perform denoising, normalization, and feature extraction on the collected multi-dimensional data to form multi-dimensional normalized time series data;

[0169] (1) Selection of acquisition equipment

[0170] Laser displacement sensor: The accuracy is 0.01μm, which monitors the displacement change of the concrete-filled steel tube node and is used to capture the deformation of the concrete-filled steel tube node.

[0171] Multi-channel strain gauge: with a frequency of 1 kHz, it collects the strain changes of concrete-filled steel tube joints for real-time monitoring of the strain behavior under load;

[0172] Non-destructive testing equipment: an ultrasonic detector is used to measure the crack propagation rate of the joint; an X-ray scanner is used to detect the crack distribution inside the joint;

[0173] Environmental monitoring equipment: it records the changes in temperature, humidity, concentration of corrosive medium, etc. of the environment where the joint is located in real time.

[0174] (2) Acquisition process

[0175] Install laser displacement sensors: laser displacement sensors are arranged at the key parts of the concrete-filled steel tube joint, and the key parts of the concrete-filled steel tube joint include the connection part and the bearing part of the concrete-filled steel tube joint;

[0176] Loading conditions: for static loads, load step by step and record the deformation and bearing capacity of the joint under each load level; for dynamic loads, apply cyclic loads and record the change of strain with time;

[0177] Environmental simulation experiment: set environments such as humid heat, high salt fog, low temperature, etc. in the laboratory and record the deterioration rate and durability data of the joint material.

[0178] (3) Data storage and backup

[0179] The collected multi-dimensional data is stored in the formats of CSV and JSON in a unified format, recording information such as sampling time, load value, displacement, crack length, etc.; the multi-dimensional data is stored in a highly reliable database, and a regular backup mechanism is set up to prevent data loss.

[0180] The dynamic performance data is denoised, normalized and feature extracted to ensure data quality and the consistency of model input. The specific steps are as follows:

[0181] Step 21, Denoising: Use the wavelet denoising algorithm to denoise the dynamic performance data described in Step 2. The purpose of using the wavelet denoising algorithm is to decompose the signal into sub-bands of different frequencies and perform threshold processing on the high-frequency sub-bands to remove the noise components. The dynamic performance data is multi-dimensional data. Remove the random noise in the signal, detect the outliers in the multi-dimensional data. The outlier is defined as the data point that exceeds the normal range. Usually, the mean plus or minus 3 times the standard deviation is used to identify the outliers, and the detected outliers in the multi-dimensional data are removed to avoid the influence of outliers on subsequent analysis. Since the linear interpolation method is applicable to time series data and fills the missing values through the linear relationship between adjacent data points; the mean filling method fills the missing values through the mean of the data and is applicable to non-time series data. Therefore, use the linear interpolation method or the mean filling method to process the missing values in the multi-dimensional data;

[0182] Step 22, Normalization: Perform normalization on the multi-dimensional data described in Step 21 to form multi-dimensional normalized time series data, ensuring that the influence of the bearing capacity index, deformation performance index, fatigue life index, durability index, and joint stability index on the input of the deep learning model is consistent.

[0183] The calculation formula for normalization is as follows:

[0184]

[0185] In the formula, X normalized represents the normalized dynamic performance data; X represents the dynamic performance data; X min represents the minimum value of the dynamic performance data; X max represents the maximum value of the dynamic performance data;

[0186] Step 23, Feature Extraction: Use principal component analysis (PCA) to extract the main features of the multi-dimensional normalized time series data described in Step 22, reducing redundant information; Set the cumulative contribution rate to be retained at 99%, determine the number of principal components, and extract the key feature values from the multi-dimensional normalized time series data. The key feature values include the maximum value, minimum value, and change rate. Classify the processed dynamic performance data according to time and working conditions (such as short-term static experiments, long-term fatigue experiments, environmental simulation experiments) to provide a hierarchical data set for subsequent analysis.

[0187] The main features include:

[0188] 1. Safety evaluation index system: It includes five major categories of indicators: bearing capacity, deformation performance, fatigue life, durability, and joint stability, covering the performance of joints in multiple dimensions such as short-term bearing capacity, long-term fatigue life, and durability.

[0189] 2. Data acquisition and processing: Combine laser displacement sensors, strain gauges, non-destructive testing equipment, and environmental monitoring equipment to collect the dynamic performance data of the nodes and perform preprocessing such as denoising, normalization, and feature extraction.

[0190] 3. Deep learning model: Adopt a deep learning model based on long short-term memory network or convolutional neural network to analyze the dynamic performance data of the nodes and make predictions.

[0191] 4. Fuzzy comprehensive evaluation method: Combine the prediction results of the deep learning model and the index weights to conduct a comprehensive safety score, and divide the node safety into five levels: excellent, good, medium, poor, and extremely poor.

[0192] 5. Hierarchical maintenance suggestions: Provide specific hierarchical maintenance suggestions according to the safety level of the nodes, from routine inspections to emergency reinforcement or replacement, to ensure structural safety.

[0193] The main components include:

[0194] 1. Bearing capacity index: including ultimate bearing capacity and yield bearing capacity, reflecting the maximum bearing capacity of the node.

[0195] 2. Deformation performance index: including ultimate deformation and stiffness degradation rate, measuring the deformation ability of the node under load and the stiffness decay speed.

[0196] 3. Fatigue life index: including fatigue limit and crack growth rate, evaluating the durability of the node under fatigue load and the crack growth situation.

[0197] 4. Durability index: including environmental durability and material deterioration rate, reflecting the long-term use performance of the node in different environments.

[0198] 5. Node stability index: including stability coefficient, measuring the overall stability of the node under different load conditions.

[0199] These main components together constitute the safety evaluation system of concrete-filled steel tubular joints, which can comprehensively evaluate the comprehensive safety and long-term performance of the joints.

[0200] Step 3, calculate the weights of the bearing capacity index, deformation performance index, fatigue life index, durability index, and node stability index described in Step 1, and optimize the weight distribution to ensure the scientific nature of the comprehensive evaluation;

[0201] The steps for optimizing the weight distribution are as follows:

[0202] Step 31, construct an index comparison matrix by the analytic hierarchy process (AHP), and invite 5 - 10 experts in the field of civil engineering to score;

[0203] Step 32: Calculate the weights through matrix normalization and verify the consistency ratio CR, ensuring that CR < 0.1; otherwise, adjust the scores.

[0204] Step 33: Use the Delphi method to conduct multiple rounds of expert consultation and adjustment on the preliminary weight results to form the weight allocation results for the middle layer and bottom layer indicators, and finally form an authoritative index weight system; verify the weight allocation results and experimental data to ensure that the index weights can reflect the safety of actual nodes.

[0205] Step 4: Input the multi-dimensional normalized time series data described in Step 2 into the deep learning model, aiming to construct a dynamic safety prediction model based on deep learning to analyze the comprehensive performance of concrete-filled steel tube joints under complex loads. Combine the optimized weight allocation in Step 3 to perform model training, verification, and optimization, and output the preliminary comprehensive safety scores and long-term performance prediction results of concrete-filled steel tube joints. The long-term performance prediction results include fatigue life, crack propagation rate, and material deterioration rate.

[0206] (1) The structure of the deep learning model includes an input layer, a hidden layer, and an output layer.

[0207] The input layer is used to receive multi-dimensional normalized time series data. The format of the input time series data is (T, F), where T represents the time step and F represents the number of features.

[0208] The hidden layer contains a three-layer network structure, with the number of nodes in each layer being 64, 32, and 16. The first layer is a long short-term memory network layer for extracting long-term dependence features of multi-dimensional normalized time series data. The second layer is a convolutional neural network layer for extracting local features of time series data. The third layer is a fully connected layer for integrating the extracted features and mapping them to the output results.

[0209] The output layer is used to output the comprehensive safety scores and long-term performance prediction values of the nodes. The scoring range of the comprehensive safety scores of the nodes is 0 - 1.

[0210] (2) Training process

[0211] Dataset division: Divide the training set and test set in a ratio of 8:2.

[0212] Optimization algorithm: Use the Adam optimizer with a learning rate of 0.001.

[0213] Loss function: Select the mean squared error (MSE) to calculate the deviation between the predicted value and the true value.

[0214]

[0215] Training parameter settings: The number of training iterations is 1000 times, and the batch size is 32.

[0216] (3) Model Validation

[0217] Use 10-fold cross-validation to evaluate the model performance, and the model performance includes mean squared error and coefficient of determination R 2 . The specific steps are as follows:

[0218] 1) Dataset Partitioning

[0219] Randomly shuffle the dataset to avoid bias caused by uneven data distribution. Evenly divide the dataset into 10 subsets (folds), ensuring that the number of samples in each subset is similar. For example, assuming the total dataset contains 1000 samples, then each subset contains 100 samples. Ensure balanced data distribution, such that different working conditions and different node characteristics are evenly distributed among the folds.

[0220] 2) Cross-Validation Process

[0221] The cross-validation is divided into 10 rounds, and the following are the specific steps for each round:

[0222] A. Select the training set and the validation set. In each round, select one subset as the validation set and the remaining 9 subsets as the training set. For example, in the first round, select fold1 as the validation set and the remaining fold2-fold10 as the training set; in the second round, select fold2 as the validation set and the other folds as the training set, and so on.

[0223] B. Model Training. Use the current training set to train the deep learning model: input multi-dimensional normalized time series data into the deep learning model; use the Adam optimizer with a learning rate of 0.001; adopt the mean squared error (MSE) as the loss function, and train the model within 1000 iterations with a batch size of 32.

[0224] C. Model Validation: Input the validation set into the trained deep learning model and calculate the deviation between the predicted value and the true value. Record the performance metrics of the current validation set:

[0225] Mean Squared Error (MSE): Measure the average error size of the predicted values of the deep learning model.

[0226]

[0227] Coefficient of Determination (R 2 ): Measure the goodness of fit of the predictions of the deep learning model.

[0228]

[0229] 3) Preliminary Comprehensive Safety Score

[0230] A. Calculate the average performance: In 10-fold cross-validation, record the MSE and R 2 values for each fold; calculate the average of the 10-fold MSE and R 2 as the overall performance metric of the deep learning model.

[0231]

[0232] B. Calculate the standard deviation

[0233] Calculate the standard deviation of the 10-fold validation results to measure the stability of the deep learning model's performance:

[0234]

[0235] 4) Output the validation results

[0236] A. Model performance results: Output the MSE and R 2 values for each fold; output the average MSE and R 2 values for 10 folds, as well as their standard deviations, to evaluate the overall performance of the deep learning model.

[0237] B. Result analysis:

[0238] Low MSE and R 2 close to 1: indicating good performance and high prediction accuracy of the deep learning model.

[0239] High MSE or low R 2 : indicating that there may be overfitting or underfitting problems in the deep learning model, and it is necessary to adjust the model hyperparameters, such as the learning rate, the number of hidden layer nodes, or the number of iterations.

[0240] 5) Evaluate the fitting effect of the deep learning model by calculating the deviation between the actual values and the predicted values in the validation set. The performance of the model can be quantified by metrics such as the mean squared error (MSE) and the coefficient of determination (R 2 ). Low MSE and R 2 values close to 1 indicate high prediction accuracy of the model.

[0241] 6) Adjust the model optimization

[0242] According to the results of 10-fold cross-validation, if the performance of the deep learning model is not good, the following adjustments can be made:

[0243] Adjust the model hyperparameters: increase or decrease the number of hidden layer nodes, such as changing from 64, 32, 16 to 128, 64, 32. Adjust the learning rate, such as changing from 0.001 to 0.0005 or 0.01.

[0244] Increase the amount of training data: If the amount of data is insufficient, data augmentation can be carried out, such as collecting more working condition data to improve the generalization ability of the model.

[0245] Optimize the loss function: According to the characteristics of the problem, try other loss functions, such as Huber loss or logarithmic loss.

[0246] The steps of the long-term performance prediction are as follows:

[0247] Step 41, Use a deep learning model to obtain the prediction results of the fatigue life, crack propagation rate, and material deterioration rate of the concrete-filled steel tubular joints within a specific future time period, and output the time series prediction curve of the performance change of the concrete-filled steel tubular joints;

[0248] Step 42, According to the prediction results in Step 41, formulate a maintenance and reinforcement plan for the concrete-filled steel tubular joints.

[0249] Step 5, Based on the preliminary comprehensive safety score described in Step 4, use the fuzzy comprehensive evaluation method to further comprehensively evaluate the safety status of the concrete-filled steel tubular joints, determine the final comprehensive safety score, divide the comprehensive safety into grades according to the final comprehensive safety score, and formulate corresponding hierarchical maintenance suggestions according to the divided grades.

[0250] The steps of determining the final comprehensive safety score are as follows:

[0251] Step 51, Construct a membership function: According to the evaluation criteria of each underlying index of the comprehensive safety evaluation index, define the membership function respectively; the membership function of each underlying index is used to map the actual value of the index to the membership value corresponding to the fuzzy evaluation grade. Specifically, divide the evaluation grade of each underlying index into fuzzy evaluation grades of excellent, good, medium, poor, and extremely poor, and the corresponding membership values are [1, 0.75, 0.5, 0.25, 0]; the underlying indexes include ultimate bearing capacity, yield bearing capacity, ultimate deformation, stiffness degradation rate, fatigue limit, fatigue crack propagation rate, environmental durability, material deterioration rate, and stability coefficient.

[0252] Step 52, Calculate the membership values: According to the membership function defined in Step 51, calculate the membership values of each underlying index respectively, and form a fuzzy evaluation matrix with the membership values of all underlying indexes; each row of the fuzzy evaluation matrix corresponds to the membership value of an underlying index, and each column corresponds to a fuzzy evaluation grade.

[0253] Step 53, comprehensively calculate the safety score: Combine the fuzzy evaluation matrix obtained in Step 52 with the weights determined in Step 3, and use the fuzzy comprehensive evaluation method to determine the final comprehensive safety score of the concrete-filled steel tubular joint. The comprehensive safety score is used to reflect the overall safety status of the joint and serve as the basis for subsequent safety level classification and maintenance suggestions.

[0254] The scoring range of the final comprehensive safety score is 0 to 1. The comprehensive safety score of the final concrete-filled steel tubular joint is divided into five grades according to the interval. The grade division is Grade A, Grade B, Grade C, Grade D, and Grade E;

[0255] The comprehensive safety score range of Grade A is 0.9 to 1.0, indicating that the comprehensive safety performance reaches an excellent level. The maintenance suggestion is that the performance of the concrete-filled steel tubular joint is good, and only a routine inspection is required every six months;

[0256] The comprehensive safety score range of Grade B is 0.7 to 0.9, indicating that the comprehensive safety performance reaches a good level. The maintenance suggestion is to check every quarter and perform local maintenance as appropriate;

[0257] The comprehensive safety score range of Grade C is 0.5 to 0.7, indicating that the comprehensive safety performance is at a medium level. The maintenance suggestion is that there are certain potential hazards and comprehensive reinforcement is required during the next major overhaul;

[0258] The comprehensive safety score range of Grade D is 0.3 to 0.5, indicating that the comprehensive safety performance is at a poor level. The maintenance suggestion is that the safety is poor and reinforcement or partial replacement is required as soon as possible;

[0259] The comprehensive safety score range of Grade E is 0 to 0.3, indicating that the comprehensive safety performance is at an extremely poor level. The maintenance suggestion is that there are serious potential hazards and it is necessary to immediately stop using and conduct large-scale replacement.

[0260] Step 6, based on the long-term performance prediction results of the deep learning model described in Step 4, further dynamically predict the future performance change trend of the concrete-filled steel tubular joint, quantify the fatigue life, crack propagation rate, and material degradation of the joint, and provide a scientific reference for long-term maintenance. The specific steps are as follows:

[0261] (1) Data preparation

[0262] ① Collect long-term monitoring data: Collect long-term monitoring data from laser displacement sensors, multi-channel strain gauges, non-destructive testing equipment, etc., including dynamic loads, crack length changes, material aging conditions, and environmental change parameters such as humidity, temperature, and corrosion concentration. Organize the historical data of the joint performance containing time series to provide input data for the dynamic prediction model.

[0263] ② Data preprocessing

[0264] Cleaning: Remove data noise, handle outliers and missing values (using wavelet denoising or linear interpolation).

[0265] Time series normalization: Perform Min - Max normalization on time series data, standardize the data to the interval [0, 1] to avoid the impact of different dimensions on model prediction.

[0266] Data windowing: Divide time series data into sliding windows of fixed length. For example, the data of the past 30 days is used to predict the performance of the next 7 days, forming input - output sample pairs for the model.

[0267] (2) Model selection and training

[0268] ① Select a model

[0269] Based on time series data, select a deep learning model suitable for dynamic performance prediction:

[0270] Long Short - Term Memory Network (LSTM): Suitable for capturing long - term dependencies in time series.

[0271] Recurrent Neural Network (RNN): Used to model the performance changes of time - related nodes.

[0272] Convolutional Neural Network (CNN): Extract local features of time series.

[0273] Hybrid model (such as LSTM - CNN): Combine short - term features and long - term dependencies of time series to improve prediction accuracy.

[0274] ② Model structure design

[0275] Input layer: Receive time series data, with dimensions (T, F), where T is the time step, i.e., the sliding window size, and F is the number of features.

[0276] Hidden layer: For the LSTM layer, extract long - term dependency features of time series. For the fully connected layer (Dense), map the extracted features to the prediction results of node performance.

[0277] Output layer: Output the predicted performance values for a future period, such as fatigue life, crack growth rate, etc.

[0278] ③ Model training

[0279] Dataset division: Divide the training set and test set according to time series (such as 80% for training and 20% for testing).

[0280] Optimization algorithm: Use the Adam optimizer with a learning rate set to 0.001.

[0281] Loss function: Mean Squared Error (MSE): Used for regression problems to measure the deviation between predicted values and true values.

[0282]

[0283] Training parameters: Batch size: 32. Number of iterations: 1000 times.

[0284] Model validation: Use cross-validation (such as 10-fold cross-validation) to evaluate the model performance and record the Mean Squared Error (MSE) and coefficient of determination (R 2 ).

[0285] (3) Performance prediction and result output

[0286] ① Input prediction data: Use the dynamically monitored data of the preprocessed time series as the input of the deep learning model. The input dimension is the time series length T and the number of features F.

[0287] ② Output predicted values

[0288] Output the performance change trend of future nodes, including:

[0289] Crack growth rate: Predict the change trend of crack length over time.

[0290] Material degradation rate: Predict the rate of decline of material properties over time.

[0291] Fatigue life: Predict the remaining fatigue life of the node under the existing working conditions.

[0292] The output is the performance predicted values of the time series, such as the changes in crack growth rate and fatigue life in the next 30 days.

[0293] ③ Generate a time series prediction curve through calculation and analysis, compare the predicted values with the true values, and verify the accuracy of the model prediction.

[0294] (4) Application of dynamic prediction results

[0295] ① Formulate a maintenance plan:

[0296] According to the crack growth prediction results, determine the time when the crack growth reaches the critical value, and arrange for node reinforcement or replacement.

[0297] According to the fatigue life prediction results, formulate a structural maintenance and overhaul plan in advance.

[0298] According to the material degradation rate prediction results, evaluate the service life of the node under different environmental conditions and optimize the material selection.

[0299] ②Real-time update of the prediction model: During the actual monitoring process, new data is dynamically obtained, the model input is continuously updated, and the model is retrained to adapt to the latest changes in the node performance.

[0300] ③Generate a performance report: Output a quantitative analysis report on the future performance of the node, including the remaining fatigue life, crack propagation trend, and material deterioration, to assist engineers in making decisions.

[0301] (5) Model optimization and iteration

[0302] ①Continuous monitoring and feedback: Compare the predicted value with the actual value, and feedback the deviation to the model to optimize the prediction accuracy.

[0303] ②Model parameter optimization: Adjust the hyperparameters of the deep learning model, such as the learning rate and the number of nodes in the hidden layer, to improve the fitting ability and generalization ability of the deep learning model.

[0304] ③Expand the prediction range: Train specific prediction models for different working conditions, such as load types and environmental conditions, to enhance adaptability.

[0305] Through the dynamic performance prediction steps, the present invention can accurately quantify the long-term performance change trend of concrete-filled steel tubular joints under complex working conditions, providing reliable technical support for the design, operation and maintenance, and life management of the joints. This process significantly improves the scientific nature and engineering practicability of structural maintenance and has important application value.

Claims

1. A safety status evaluation method for concrete-filled steel tube nodes based on deep learning, characterized in that: The following steps are involved: Step 1, constructing a comprehensive safety evaluation index system for steel tube concrete nodes, wherein the comprehensive safety evaluation index system includes a bearing capacity index, a deformation performance index, a fatigue life index, a durability index, and a node stability index; Step 2: According to the comprehensive safety evaluation index system constructed in step 1, the dynamic performance data of displacement, strain, crack propagation, material aging and environmental parameters of the steel tube concrete node are collected, and the collected dynamic performance data are denoised, normalized and feature extracted to form multidimensional normalized time series data; Step 3, calculating the weight of the comprehensive safety evaluation index system described in step 1, and optimizing the weight distribution; Step 4, inputting the multidimensional normalized time series data described in step 2 into the deep learning model, and combining the weight distribution optimized in step 3, performing model training, verification and optimization, and outputting the preliminary comprehensive safety score and long-term performance prediction results of the steel tube concrete node, the long-term performance prediction results including fatigue life, crack growth rate and material degradation rate; Step 5: Based on the preliminary comprehensive safety score described in step 4, the safety status of the steel tube concrete node is further comprehensively evaluated by using a fuzzy comprehensive evaluation method to determine the final comprehensive safety score. According to the final comprehensive safety score, the comprehensive safety is divided into levels, and corresponding graded maintenance recommendations are formulated according to the divided levels. Step 6: Based on the long-term performance prediction results described in step 4, further dynamically predict the future performance change trend of the steel tube concrete node to quantify the fatigue life, crack growth rate and material degradation of the node.

2. The method according to claim 1, characterized in that The bearing capacity index in step 1 includes the ultimate bearing capacity and the yield bearing capacity, the deformation performance index includes the ultimate deformation and the stiffness degradation rate, the fatigue life index includes the fatigue limit and the fatigue crack growth rate, the durability index includes the environmental durability and the material degradation rate, and the node stability index includes the stability coefficient.

3. The method according to claim 2, characterized in that The calculation formula of the ultimate bearing capacity is as follows: Where: Fmax is the maximum load of the steel tube concrete node in the test; A is the cross-sectional area of ​​the steel tube concrete node; The calculation formula of the yield bearing capacity is as follows: Where: Fyield is the yield load; A is the cross-sectional area of ​​the steel tube concrete node; The calculation formula of the limit deformation is as follows: Where: δmax is the limit deformation; L is the length of the steel tube concrete node; The calculation formula of the stiffness degradation rate is as follows: Where: Ki is the initial stiffness; Kf is the final stiffness; The calculation formula of the fatigue limit is as follows: Where: ΔFfatigue is the fatigue limit load, A is the cross-sectional area; The calculation formula of the fatigue crack growth rate is as follows: Where: af is the final crack length; ai is the initial crack length; N is the number of cycles; The calculation formula of the material degradation rate is as follows: Where: P0 is the initial performance value; Pt is the performance value after t years; t is time; The calculation formula of the stability coefficient is as follows: Where: Fcr is the critical load and Fapplied is the actual applied load.

4. The method according to claim 1, characterized in that: The steps of denoising, normalizing and feature extraction of dynamic performance data in step 2 are as follows: Step 21, denoising: using a wavelet denoising algorithm to denoise the dynamic performance data in step 2, the dynamic performance data is multidimensional data, removing random noise in the signal, eliminating outliers in the multidimensional data, and using linear interpolation or mean filling method to process missing values ​​in the multidimensional data; Step 22, normalization processing: normalize the multidimensional data in step 21 to form multidimensional normalized time series data. The calculation formula for normalization processing is as follows: In the formula, Xnormalized represents the normalized dynamic performance data; X represents the dynamic performance data; Xmin represents the minimum value of the dynamic performance data; Xmax represents the maximum value of the dynamic performance data; Step 23, feature extraction: Use principal component analysis to extract the main features of the multidimensional normalized time series data described in step 22, set the cumulative contribution rate to retain 99%, determine the number of principal components, and extract key eigenvalues ​​of the multidimensional normalized time series data, the key eigenvalues ​​include maximum value, minimum value and change rate, and classify the processed dynamic performance data according to time and operating conditions.

5. The method according to claim 1, characterized in that The steps for optimizing weight distribution described in step 3 are as follows: Step 31, constructing an indicator comparison matrix through the hierarchical analysis method, and inviting experts in the field of civil engineering to score; Step 32, calculate the weight and verify the consistency ratio to ensure that CR < 0.1, otherwise adjust the score; Step 33, perform multiple rounds of adjustments on the preliminary weights to form the weight distribution results of the middle layer and bottom layer indicators, and verify the weight distribution results with the experimental data.

6. The method according to claim 1, characterized in that The structure of the deep learning model in step 4 includes an input layer, a hidden layer and an output layer, wherein the input layer is used to receive multidimensional normalized time series data, and the format of the input time series data is (T, F), where T represents the time step and F represents the number of features; The hidden layer includes a three-layer network structure, the first layer is a long short-term memory network layer for extracting long-term dependency features of multi-dimensional normalized time series data, the second layer is a convolutional neural network layer for extracting local features of time series data, and the third layer is a fully connected layer for integrating the extracted features and mapping them to output results; The output layer is used to output the comprehensive safety score and long-term performance prediction value of the steel tube concrete node. The comprehensive safety score of the steel tube concrete node ranges from 0 to 1.

7. The method according to claim 1, characterized in that The steps for long-term performance prediction described in step 4 are as follows: Step 41, using a deep learning model to obtain prediction results of fatigue life, crack growth rate, and material degradation rate of the steel tube concrete node in a specific future time period, and output a time series prediction curve of performance changes of the steel tube concrete node; Step 42, based on the prediction results of step 41, formulate a maintenance and reinforcement plan for the steel tube concrete node.

8. The method according to claim 1, characterized in that The steps for determining the final comprehensive safety score described in Step 5 are as follows: Step 51, constructing a membership function: defining a membership function according to the evaluation criteria of each underlying indicator of the comprehensive safety evaluation index system; the underlying indicators include ultimate bearing capacity, yield bearing capacity, ultimate deformation, stiffness degradation rate, fatigue limit, fatigue crack growth rate, environmental durability, material degradation rate and stability coefficient; Step 52, calculating the membership value: according to the membership function defined in step 51, calculating the membership value of each underlying indicator, and forming a fuzzy evaluation matrix with the membership values ​​of all underlying indicators; Step 53, comprehensive calculation of safety score: combining the fuzzy evaluation matrix obtained in step 52 with the weight determined in step 3, and determining the final comprehensive safety score of the steel tube concrete node through the fuzzy comprehensive evaluation method.

9. The method according to claim 1, characterized in that: The final comprehensive safety score in step 5 ranges from 0 to 1, and the grades are divided into grade A, grade B, grade C, grade D and grade E; The comprehensive safety score range of grade A is 0.9 to 1.0, indicating that the comprehensive safety performance has reached an excellent level. The maintenance recommendation is that the node performance is good and only regular inspections are required every six months. The comprehensive safety score range of grade B is 0.7 to 0.9, indicating that the comprehensive safety performance has reached a good level. The maintenance recommendation is to check every quarter and perform partial maintenance in a timely manner; The comprehensive safety score range of grade C is 0.5 to 0.7, indicating that the comprehensive safety performance is at a medium level. The maintenance recommendation is that there are certain hidden dangers and comprehensive reinforcement is required in the next overhaul; The comprehensive safety score range of grade D is 0.3 to 0.5, indicating that the comprehensive safety performance is at a poor level. The maintenance recommendation is that the safety is poor and reinforcement or partial replacement is required as soon as possible. The comprehensive safety score range of grade E is 0 to 0.3, indicating that the comprehensive safety performance is at an extremely poor level. The maintenance recommendation is that there are serious hidden dangers and the vehicle must be stopped immediately and replaced on a large scale.

Images