Prefabricated beam piece quality evaluation model modeling method based on machine learning
Through the machine learning-based prefabricated beam sheet quality evaluation model model model model, the lack of scientific analysis and machine learning evaluation in the quality control of bridge beam sheet prefabricated process is solved, and the precise improvement of beam sheet quality and efficient detection of finished product quality is achieved. Extend the service life of the bridge.
Patent Information
- Application Number
- CN202411072352.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-06
- Publication Date
- 2024-11-08
AI Technical Summary
The existing technology lacks scientific mathematical analysis methods in the quality control of bridge beam sheet prefabrication process, and does not use machine learning to perform quality evaluation, resulting in deviations in quality inspection results and inaccurate quality control of finished beams.
Using machine learning-based prefabricated beam sheet quality evaluation model model model, we collect engineering quality data from key processes, perform data normalization processing and outlier filtering, establish a quality correlation analysis model, select relevant factors for multi-factor machine learning Training, establish a process quality evaluation model, and realize accurate control of the beam sheet prefabrication process and finished product quality prediction.
The overall improvement of the quality of beam sheets is achieved, the prefabrication process is accurately controlled, and the blindness and high cost of random sampling of finished beam sheets is reduced. It is found that potential artificial fraud of concrete strength reports is found, which extends the service life of the bridge.
Smart Images

Figure CN118917202A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of expressways, and in particular to a modeling method for a prefabricated beam quality evaluation model based on machine learning. Background Art
[0002] In the construction of highway bridge projects, with the development of highway digital transformation and highway intelligent construction technology, the traditional method focusing on the quality evaluation of finished beams can no longer meet the requirements of building high-quality highway bridge projects. In the era of increasingly mature development of new generation information technologies such as the Internet of Things, big data, and artificial intelligence, how to establish a more scientific, effective, and accurate beam prefabrication process quality evaluation model through data mining analysis and machine learning technology to achieve accurate control of key beam prefabrication processes and accurate sampling of finished beams is an important problem that needs to be solved. Although there are some studies on beam process quality control and information application in the literature, no machine learning method has been found to scientifically evaluate and study the quality of beam prefabrication process. Therefore, this study is of great value in promoting the improvement of beam prefabrication quality.
[0003] In the research on quality control of bridge beam prefabrication process. Liu Qiang et al. [1] A quality evaluation method for the prefabricated beam production process based on information management and control technology is proposed. It mainly collects and cleans the data of key indicators in the beam prefabrication process, uses the mean-range control chart to analyze and judge the data of key control indicators of a single process, and evaluates the quality of each production process through a grade evaluation model. This method mainly evaluates the quality control accuracy deviation, and has not yet adopted data correlation analysis and machine learning modeling for evaluation; Wang Huilin et al. proposed an application method for intelligent high-speed rail prefabricated beam yard management based on BIM technology, constructed an intelligent high-speed rail prefabricated beam yard management platform based on BIM technology, and prospected the intelligent control of beam prefabrication process quality in the future; Yang In order to solve the quality control problems of T-beam prefabrication and erection during bridge construction, Lu proposed the key points of construction quality control during T-beam prefabrication and erection based on previous engineering construction experience; Xiang Hong and others conducted research on the quality collection and monitoring of concrete in highway beam yards. In view of the lack of monitoring in the production of concrete raw material aggregates and the concrete pouring stages, they proposed a new method of online monitoring of raw material aggregate particle size and sand and gravel moisture content using new technologies; Luo Ruyi and others conducted research on the full life cycle production of smart highway beam yards, and applied information technology to the full life cycle processes of prefabricated beam reinforcement cotton binding, concrete pouring, steam curing, etc., realizing "intelligent monitoring" and "digital archiving" of beam section production.
[0004] In terms of the quality inspection technology of finished bridge beams, the beam inspection technology has gradually changed from manual inspection to intelligent and automated inspection, mainly including ultrasonic inspection technology, fiber optic sensing inspection technology, spectrum analysis inspection technology, image non-destructive inspection technology, etc. Ultrasonic inspection technology is an advanced non-destructive inspection technology. The advantages of this technology are simple operation and rapid feedback. The disadvantage is that if there is water or air inside the bridge beam, it will affect the propagation of ultrasonic waves, resulting in a large deviation in the inspection results. Fiber optic sensing detection technology converts some physical quantities such as cable stress, internal stress and strain of prestressed concrete beams into optical signals, and bridge beam inspection can be completed by measuring the optical signals. This technology has the advantages of high sensitivity, strong flexibility, large dynamic response range, and portability, but it has high cost of use, and the fiber optic technology is not perfect, so it cannot achieve ideal inspection results. Spectral analysis detection technology is a technology that determines the characteristics of the propagation medium by analyzing waves of specific frequencies. Rayleigh wave analysis is a widely used spectrum analysis technology. Compared with the manual tapping method, it has higher efficiency and accuracy, but this analysis method is more sensitive to environmental noise, temperature, and humidity, and has a high cost, so it is not very practical at present. Image non-destructive testing technology mainly uses laser holographic imaging photography technology, and performs digital signal processing and image analysis on the image data collected in real time to obtain the status information of the bridge. This technology has the characteristics of high accuracy, comprehensive detection, and strong intuitiveness. The method is simple to operate, has no radiation, and can quickly detect large areas, but the disadvantage is that it mainly focuses on the detection of appearance.
[0005] In terms of technical specifications and standards for bridge beam inspection. The American "AASHTO Manual for Bridge Evaluation" (AASHTOMBE-3-I2:2020) mainly puts forward index requirements from the aspects of geometric parameters, material parameters, surface conditions and load-bearing capacity of finished beams. Such as the width, length, thickness, cross-sectional shape, concrete strength, flatness, surface defects, design load, etc. of the beam. At present, my country's highway bridge beam quality evaluation standards mainly focus on the requirements for qualified quality inspection of prefabricated beams, mainly including the concrete strength, beam length, cross-sectional dimensions, flatness, cross slope, etc. of the finished prefabricated beams. Judging from the current domestic and foreign relevant beam quality inspection standards, they mainly focus on the evaluation of the finished beams themselves, and have not yet proposed specific methods and requirements for the quality evaluation of the beam prefabrication process control.
[0006] The above beam quality inspection technology and evaluation standards play an important role in beam quality inspection, but the current evaluation technical indicators mainly focus on the requirements for the quality inspection of prefabricated beam products. There are few related analysis and research and specific requirements for the quality control of the beam prefabrication process, especially the use of machine learning methods to establish evaluation models through learning and training for quality evaluation. There are mainly the following problems:
[0007] 1. The current evaluation technology method does not adopt a scientific mathematical analysis method to analyze the impact of the quality control deviation rate of the beam prefabrication process on the quality of the finished beam.
[0008] 2. The current evaluation technology methods mainly focus on the quality inspection of finished beams, and have not yet used machine learning modeling to achieve differentiated evaluation of beam prefabrication quality through quality control deviation data of the beam prefabrication process. Summary of the invention
[0009] The object of the present invention is to provide a method for modeling a precast beam quality evaluation model based on machine learning.
[0010] The technical solution adopted by the present invention is:
[0011] A method for modeling a precast beam quality evaluation model based on machine learning comprises the following steps:
[0012] Step 1: Collect engineering quality data of key processes and analyze various quality evaluation indicators collected in each key process link;
[0013] Furthermore, the project quality data includes process quality control data and finished product quality inspection data; the process quality control data includes process quality control data of raw materials, steel bar processing and binding, concrete mixing, concrete test blocks, pouring and vibration, steaming, tensioning and grouting; the finished product quality inspection data includes the finished product quality inspection data of concrete test blocks and beam pieces.
[0014] Furthermore, the collected engineering quality data are processed by data normalization, outlier filtering and outlier data processing.
[0015] Specifically, in terms of data processing, we mainly use methods such as data normalization, filtering outliers and outlier data to achieve data noise reduction. The main steps are: first, through data normalization, standardize the data of each eigenvalue; second, through scatter plot analysis and deviation, filter outliers and outlier data to make the data conform to business logic and expected range.
[0016] Step 2: Establish a quality correlation analysis model, and establish corresponding quality correlation analysis models for different indicators of beam quality; select at least three correlation factors for each indicator for machine learning training through correlation analysis;
[0017] Furthermore, different indicators of beam segment quality include concrete specimen strength, steel bar cover thickness, and steel bar spacing.
[0018] Specifically, the complete quality evaluation index of the finished beam includes steel bar testing (yield strength, elongation), concrete test strength, concrete mixing, steel cage production size, pouring, steaming, tensioning, rebound strength, protective layer thickness, steel bar spacing, structural size and other contents.
[0019] Furthermore, the quality correlation analysis model includes a concrete specimen strength correlation analysis model, a steel bar cover thickness correlation analysis model, and a steel bar spacing correlation analysis model.
[0020] Furthermore, at least three data indicators whose correlation factors are greater than set values are selected through correlation analysis for machine learning training.
[0021] Furthermore, the correlation analysis of the Pearson correlation coefficient is used to perform the correlation analysis of the beam concrete strength. Specifically, the steps are as follows:
[0022] Step 2-1, calculate the standard deviation σ of the independent variable x of n samples of precast beams x and the standard deviation σ of the dependent variable y y , the standard deviation is calculated as follows:
[0023]
[0024] Where n is the total number of precast beam samples, x i is the value of the ith independent variable x, is the mean value of the independent variable x; y i is the value of the ith dependent variable y, is the mean value of the dependent variable y;
[0025] Step 2-2, calculate the precast beam sample for each independent variable x i Deviation Right now Each dependent variable y i Deviation Right now And calculate the independent variable x i With the dependent variable y i The correlation coefficient value R is in the interval [-1,1], -1≤R<0 indicates negative correlation, 0<R≤1 indicates positive correlation, and the closer it is to 1 or -1, the stronger the correlation. The calculation formula is as follows:
[0026]
[0027] Where R is the Pearson correlation coefficient, is the deviation, σ is the standard deviation of n samples of precast beam pieces, and n is the total number of n samples of precast beam pieces;
[0028] Step 2-3, using the collected beam sample data and combining it with the Pearson algorithm correlation analysis algorithm, the following scatter plot and correlation analysis graph are generated;
[0029] Step 2-4, based on the scatter plot and the correlation analysis diagram, select at least three correlation factors for each evaluation indicator whose correlation factor ratio is greater than the set value.
[0030] Furthermore, establishing a correlation analysis model for the strength of a concrete specimen includes establishing correlation analysis models for the strength of a concrete specimen and concrete raw materials, concrete mixing ratio, concrete mixing time, and beam steaming process.
[0031] Furthermore, the data collected for modeling the beam concrete strength quality evaluation model include concrete mixing data, beam steaming data, and concrete test block data; among them, the concrete mixing data collects sand, aggregate, cement, fly ash, and water indicators; the beam steaming data collects temperature, humidity, and temperature control time indicators; the concrete test block data collects test block strength, test temperature, and test humidity indicators.
[0032] Furthermore, in the modeling of the concrete strength quality evaluation model, each beam is assigned a unique beam number to realize the data association of the specified beam concrete mixing, beam steaming, concrete test block, etc., laying the foundation for data correlation analysis.
[0033] Step 3, establish a process quality evaluation model: based on the relevant factors selected by the quality correlation analysis model, select a multi-factor machine learning algorithm model that supports machine learning training, and establish a process quality evaluation model for prefabricated beams; use the process quality evaluation model and sample data testing to predict, and perform convergence and differentiation analysis on the quality results of sample predictions and the measured quality index results, so as to optimize the machine learning algorithm model and parameters until the established model prediction results and the measured quality results achieve maximum convergence, thereby completing the final modeling of the process quality evaluation model for the corresponding indicators, and using the final process quality evaluation model to control and evaluate the beam prefabrication production process.
[0034] Furthermore, the criterion for achieving maximum convergence in step 3 is: the minimum root mean square error is less than the allowable error.
[0035] The present invention adopts the above technical scheme to analyze the correlation of various factors through the quality of finished beam pieces and the deviation of process quality control, find out the key processes and key indicators that affect the quality, and use this to accurately control and achieve the overall improvement of beam piece quality. Collect prefabricated beam piece process control and finished product quality sample data, establish beam piece process control and finished product quality prediction and evaluation models through a variety of multi-factor machine learning methods, and analyze the deviation between the sample measured value and the predicted value, so as to perform machine learning algorithm model and parameter tuning, until the established model prediction results and the measured quality results achieve maximum convergence, and generate the optimal prefabricated beam piece quality evaluation model. According to the result value of the beam piece quality evaluation of the model, it can be used for the accurate sampling of the quality of the finished beam piece, that is, focusing on the sampling of beam pieces with low evaluation scores, which can effectively avoid the blindness of random sampling of finished beam pieces and the high cost and low efficiency investment caused by comprehensive quality inspection, and can also effectively discover the possibility of artificial fraud in the concrete strength report of the beam piece. The present invention can effectively promote the homogenization of prefabricated beam pieces, improve the overall quality of beam piece prefabrication, help build high-quality bridge projects, and extend the service life of bridges. BRIEF DESCRIPTION OF THE DRAWINGS
[0036] The present invention is further described in detail below with reference to the accompanying drawings and specific embodiments;
[0037] Figure 1 A flow chart of a method for modeling a precast beam quality evaluation model based on machine learning according to the present invention;
[0038] Figure 2 It is a scatter plot of the correlation between concrete mixing deviation value and concrete strength;
[0039] Figure 3 It is the correlation matrix analysis diagram between concrete mixing deviation value and concrete strength;
[0040] Figure 4 It is a scatter plot of linear regression predicted values and measured values;
[0041] Figure 5 This is a schematic diagram showing the difference between the linear regression prediction and the measured value;
[0042] Figure 6 It is a scatter plot of decision tree regression predicted values and measured values;
[0043] Figure 7 This is a schematic diagram comparing the difference between the decision tree regression prediction and the measured value;
[0044] Figure 8 It is a scatter plot of XGB regression predicted value and measured value;
[0045] Fig. 9 Schematic diagram for comparison of the difference between XGB regression prediction and measured values. Implementation
[0047] In order to make the purpose, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application.
[0048] like Figures 1 to 9 As shown in one, the present invention discloses a modeling method for prefabricated beam piece quality evaluation model based on machine learning. According to the theory of quality control in the production process, the quality of each process is reflected in the results of the process. High process quality reflects good results and high pass rate. In the prefabrication of beam production, the prefabricated beam pieces are formed by various processes such as raw materials, steel bar processing and binding, concrete mixing, pouring and vibrating, steaming, tensioning, grouting, etc., so the quality control of each process determines the quality of the final finished beam. The present invention specifically includes the following steps:
[0049] Step 1: Collect engineering quality data of key processes and analyze various quality evaluation indicators collected in each key process link;
[0050] Furthermore, the project quality data includes process quality control data and finished product quality inspection data; the process quality control data includes process quality control data of raw materials, steel bar processing and binding, concrete mixing, concrete test blocks, pouring and vibration, steaming, tensioning and grouting; the finished product quality inspection data includes the finished product quality inspection data of concrete test blocks and beam pieces.
[0051] Furthermore, the collected engineering quality data are processed by data normalization, outlier filtering and outlier data processing.
[0052] Specifically, in terms of data processing, we mainly use methods such as data normalization, filtering outliers and outlier data to achieve data noise reduction. The main steps are: first, through data normalization, standardize the data of each eigenvalue; second, through scatter plot analysis and deviation, filter outliers and outlier data to make the data conform to business logic and expected range.
[0053] Step 2: Establish a quality correlation analysis model, and establish corresponding quality correlation analysis models for different indicators of beam quality; select at least three correlation factors for each indicator for machine learning training through correlation analysis;
[0054] Furthermore, different indicators of beam segment quality include concrete specimen strength, steel bar cover thickness, and steel bar spacing.
[0055] Specifically, the complete quality evaluation index of the finished beam includes steel bar testing (yield strength, elongation), concrete test strength, concrete mixing, steel cage production size, pouring, steaming, tensioning, rebound strength, protective layer thickness, steel bar spacing, structural size and other contents.
[0056] Furthermore, the quality correlation analysis model includes a concrete specimen strength correlation analysis model, a steel bar cover thickness correlation analysis model, and a steel bar spacing correlation analysis model.
[0057] Furthermore, at least three data indicators whose correlation factors are greater than set values are selected through correlation analysis for machine learning training.
[0058] Furthermore, the correlation analysis of the Pearson correlation coefficient is used to perform the correlation analysis of the beam concrete strength. Specifically, the steps are as follows:
[0059] Step 2-1, calculate the standard deviation σ of the independent variable x of n samples of precast beams x and the standard deviation σ of the dependent variable y y , the standard deviation is calculated as follows:
[0060]
[0061] Where n is the total number of precast beam samples, x i is the value of the ith independent variable x, is the mean value of the independent variable x; y i is the value of the ith dependent variable y, is the mean value of the dependent variable y;
[0062] Step 2-2, calculate the precast beam sample for each independent variable x i Deviation Right now Each dependent variable y i Deviation Right now And calculate the independent variable x i With the dependent variable y i The correlation coefficient value R is in the interval [-1,1], -1≤R<0 indicates negative correlation, 0<R≤1 indicates positive correlation, and the closer it is to 1 or -1, the stronger the correlation. The calculation formula is as follows:
[0063]
[0064] Where R is the Pearson correlation coefficient, is the deviation, σ is the standard deviation of n samples of precast beam pieces, and n is the total number of n samples of precast beam pieces;
[0065] Step 2-3, using the collected beam sample data and combining it with the Pearson algorithm correlation analysis algorithm, the following scatter plot and correlation analysis graph are generated;
[0066] Step 2-4, based on the scatter plot and the correlation analysis diagram, select at least three correlation factors for each evaluation indicator whose correlation factor ratio is greater than the set value.
[0067] Furthermore, establishing a correlation analysis model for the strength of a concrete specimen includes establishing correlation analysis models for the strength of a concrete specimen and concrete raw materials, concrete mixing ratio, concrete mixing time, and beam steaming process.
[0068] Furthermore, the data collected for modeling the beam concrete strength quality evaluation model include concrete mixing data, beam steaming data, and concrete test block data; among them, the concrete mixing data collects sand, aggregate, cement, fly ash, and water indicators; the beam steaming data collects temperature, humidity, and temperature control time indicators; the concrete test block data collects test block strength, test temperature, and test humidity indicators.
[0069] Furthermore, in the modeling of the concrete strength quality evaluation model, each beam is assigned a unique beam number to realize the data association of the specified beam concrete mixing, beam steaming, concrete test block, etc., laying the foundation for data correlation analysis.
[0070] Step 3, establish a process quality evaluation model: based on the relevant factors selected by the quality correlation analysis model, select a multi-factor machine learning algorithm model that supports machine learning training, and establish a process quality evaluation model for prefabricated beams; use the process quality evaluation model and sample data testing to predict, and perform convergence and differentiation analysis on the quality results of sample predictions and the measured quality index results, so as to optimize the machine learning algorithm model and parameters until the established model prediction results and the measured quality results achieve maximum convergence, thereby completing the final modeling of the process quality evaluation model for the corresponding indicators, and using the final process quality evaluation model to control and evaluate the beam prefabrication production process.
[0071] Furthermore, the criterion for achieving maximum convergence in step 3 is: the minimum root mean square error is less than the allowable error.
[0072] The present invention uses the Pearson algorithm to perform multi-factor correlation analysis on the quality of finished beam pieces and the deviation data of process quality control to find out the key processes and key indicators that affect the quality. The multi-factor machine learning method is mainly used to establish a beam piece process control and finished product quality prediction and evaluation model, and the sample measured value and the predicted value are subjected to deviation analysis, so as to perform machine learning algorithm model and parameter tuning, until the established model prediction results and the measured quality results are maximized and converged, and the optimal prefabricated beam piece quality evaluation model is generated. The result of the model evaluation of the present invention is the result value of each indicator, such as the beam piece concrete evaluation, which is the concrete strength (MPa) result value.
[0073] The specific principle of the present invention is described in detail below:
[0074] First, process quality data collection. Combined with the smart beam site to collect quality control data of raw materials, steel bar processing and binding, concrete mixing, concrete test blocks, pouring and vibration, steaming, tensioning, grouting, etc., as well as quality inspection data of concrete test blocks and beam pieces, the quality indicators collected in each key process are analyzed.
[0075] Second, establish a quality correlation analysis model. This includes establishing quality correlation analysis models for concrete test block strength, steel bar protective layer thickness, steel bar spacing, etc. Taking the concrete test block strength analysis as an example, establish a correlation analysis model between the concrete test block strength and concrete raw materials, concrete mixing ratio, concrete mixing time, beam steaming and curing processes, etc. Through correlation analysis, select several data indicators with relatively large correlation factors for machine learning training.
[0076] Third, establish a process quality evaluation model. Taking the strength of concrete test blocks as an example, select a model that supports multi-factor machine learning algorithm, conduct machine learning training, establish a concrete strength process quality evaluation model, and use the generated model and sample data test for prediction. The quality results of sample prediction and the measured quality index results are analyzed for convergence and differentiation. The machine learning algorithm model and parameter tuning are performed until the prediction results of the established model and the measured quality results achieve maximum convergence (minimum root mean square error), thereby completing the effective modeling of the concrete strength process quality evaluation model, and this model can be used for beam prefabrication production process control and quality evaluation.
[0077] Modeling method and evaluation of quality evaluation model: The complete quality evaluation index of beam finished product includes steel bar test (yield strength, elongation), concrete test strength, concrete mixing, steel cage production size, pouring, steaming, tensioning, rebound strength, cover thickness, steel bar spacing, structural size, etc. Taking the modeling of beam concrete strength quality evaluation model as an example, the modeling implementation method is as follows:
[0078] (I) Sample data collection and processing: The data collected for the modeling of beam concrete strength quality evaluation model include concrete mixing data, beam steaming data, concrete test block data and other data. Among them, the concrete mixing data mainly collects sand, aggregate, cement, fly ash, water and other indicators; the beam steaming data mainly collects temperature, humidity, temperature control time and other indicators; the concrete test block data mainly collects test block strength, test temperature, test humidity and other indicators. In the modeling of concrete strength quality evaluation model, each beam is assigned a unique beam number to realize the association of data such as concrete mixing of the specified beam, steaming of the beam, and concrete test block, laying the foundation for data correlation analysis.
[0079] In terms of data processing, data normalization, filtering of outliers and outlier data are mainly used to achieve data noise reduction. The main steps are: first, through data normalization, standardization of each eigenvalue data is achieved; second, through scatter plot analysis and deviation, outliers and outlier data are filtered out to make the data consistent with business logic and expected range. The changes in sample data before and after data normalization are shown in Tables 1 and 2.
[0080] Table 1 Precast beam concrete mix ratio deviation data table
[0081]
[0082] Table 2 Normalized data of precast beam concrete mix ratio
[0083]
[0084]
[0085] The unit of concrete strength in Table 1 and Table 2 is MPa. The standard value of deviation rate in Table 1 is 1. A deviation less than 1 is a negative deviation, and a deviation greater than 1 is a positive deviation. The data in Table 2 are the normalized data in Table 1. The Min-Max Normalization method is used to scale the data to the interval [0,1], see formula (1), where x′ is the normalized value of each parameter column, and max(x) and min(x) are used to find the maximum and minimum values of the parameter column.
[0086]
[0087] (II) Data correlation analysis: In the prefabrication production of the beam yard, concrete mixing is mainly carried out according to the mix ratio indicators provided by the laboratory. The degree of influence of the deviation of each raw material mix on the concrete strength is still unknown, which also makes it impossible for on-site personnel to accurately control the key indicators of concrete mixing and feeding. Therefore, it is necessary to conduct a correlation analysis between the deviation of each raw material mix ratio of concrete and the concrete strength, so as to find out the key influencing indicators of the deviation of each raw material mix on the concrete strength, and then achieve targeted and precise control of the raw material mix ratio.
[0088] The mainstream correlation analysis methods currently include Pearson, Spearman correlation coefficient, CovGc correlation and distance correlation
[31] . Since the Pearson correlation coefficient can be used to measure the linear correlation between two or more continuous variables, it is more in line with the requirements of the correlation analysis between the concrete mix ratio and strength of beam slices. Therefore, this study uses the Pearson correlation coefficient to analyze the correlation between the concrete strength of beam slices.
[0089] The main materials for mixing beam concrete include sand, aggregate (4.75-9.5), aggregate (9.5-19), cement, fly ash, additives, water and other 7 components. The following are AR1-AR7 to represent the type code, and STH represents the concrete strength. AR1-AR7 (independent variables) and STH (dependent variable). In this study, the mathematical formula of the Pearson correlation coefficient is as follows:
[0090] First, calculate the standard deviation σ of the independent variable x of n samples of precast beams x and the standard deviation σ of the dependent variable y y , the standard deviation is calculated as follows:
[0091]
[0092] Where n is the total number of precast beam samples, x i is the value of the ith independent variable x, is the mean value of the independent variable x; y i is the value of the ith dependent variable y, is the mean value of the dependent variable y;
[0093] Then, calculate the precast beam sample for each independent variable x i Deviation Right now Each dependent variable y i Deviation Right now And calculate the independent variable x i With the dependent variable y i The correlation coefficient value R is in the interval [-1,1], -1≤R<0 indicates negative correlation, 0<R≤1 indicates positive correlation, and the closer it is to 1 or -1, the stronger the correlation. The calculation formula is as follows:
[0094]
[0095] Where R is the Pearson correlation coefficient, is the deviation, σ is the standard deviation of n samples of precast beams, and n is the total number of n samples of precast beams.
[0096] Then, the following scatter plot and correlation analysis diagram are generated by combining the collected beam sample data of the smart beam field with the Pearson algorithm correlation analysis algorithm ( Figure 2 , Figure 3 ).
[0097] like Figure 2 The last line of the scatter plot shows that concrete strength is positively correlated with AR4, AR6, and AR7, and negatively correlated with AR1 and AR5. Figure 3As shown in the correlation coefficient R value in the last row of the matrix diagram, the correlation between concrete strength (STH) and AR1-AR7 is from strong to weak, namely AR4 (cement), AR1 (sand), AR6 (additive), AR7 (water), AR5 (fly ash), AR3 (aggregate 9.5-19), and AR2 (aggregate 4.75-9.5). This shows that the concrete strength is highly correlated with the deviation values of AR4 (cement), AR1 (sand), AR7 (water), AR6 (additive), and AR5 (fly ash). Among them, the concrete strength is positively correlated with AR4 (cement), AR6 (additive), and AR7 (water), and negatively correlated with AR1 (sand) and AR5 (fly ash). The correlation between concrete strength and AR2 (aggregate 4.75-9.5) and AR3 (aggregate 9.5-19) is significantly smaller, which can be explained by the fact that the aggregate particle size is large, the bonding degree between cement and aggregate is low, and it has little effect on concrete strength.
[0098] Through the above analysis, we can draw a conclusion that when mixing concrete, we should focus on the deviation control of cement, sand, water, additives, and fly ash, so as not to affect the strength index of concrete; on the other hand, since sand and fly ash are negatively correlated, the mixing ratio should be controlled as little as possible from the theoretical mix ratio of the test. In addition, in the process of modeling the intelligent evaluation model of beam quality, the AR2 and AR3 independent variable factors with significantly weaker correlation are eliminated to achieve eigenvalue dimensionality reduction analysis modeling and improve modeling accuracy.
[0099] (III) Comparison and evaluation of the construction of beam process quality evaluation models: The current mainstream regression model construction algorithms include linear regression, Lasso regression, decision tree regression, random forest regression, support vector regression, XGBoost regression (extreme gradient boosting) and other algorithms. In this study, based on the characteristics of sample data that focus on linear and multi-independent variable factor changes, linear regression, decision tree regression, and XGBoost machine learning algorithms suitable for this feature were selected for comparison. Mainly based on Python development tools, the corresponding regression prediction model is generated through relevant machine learning algorithm training and parameter tuning. In machine learning training and evaluation, in order to reasonably evaluate the applicability of each machine learning algorithm and the accuracy of model evaluation, the training samples and test samples are uniformly trained and modeled in a ratio of 8:2 and the model is tested and evaluated.
[0100] In terms of the accuracy assessment of the beam quality evaluation model, the root mean square error (RMSE) indicator is used to find the model corresponding to the minimum RMSE as the optimal model for beam quality evaluation. The RMSE uses formula (4).
[0101]
[0102] Formula Description: xi is the independent variable data collected for the i-th beam prefabrication test sample, y i is the actual measured value of the test sample (dependent variable data), f(x i ) is x i The predicted value data of the test samples, n is the total number of test samples for beam prefabrication.
[0103] In addition, in order to more intuitively evaluate the prediction error rate deviation range of the generated evaluation model, the mean absolute percentage error (MAPE) is used here, that is, the average absolute percentage error rate between the predicted value and the measured value of the tested sample data. The error rate result value is expressed in percentage, see formula (5):
[0104]
[0105] Formula description: The symbols in the formula are the same as those in formula (4).
[0106] The linear regression model is a regression analysis method in statistics and machine learning, which is used to establish a linear relationship model between independent variables (features) and dependent variables (targets). The least squares method is mainly used to obtain the optimal coefficients to establish a linear regression model. Figure 4 , Figure 5 Sample chart for comparative analysis of predicted and actual values of beam concrete strength test sample data.
[0107] The decision tree regression algorithm adopts a top-down greedy recursive method based on a binary tree, that is, each division only considers the current optimal solution, and then reaches the optimal solution through multiple recursive tuning. In this study, the independent variable (feature) value and the division point of the precast beam segment are selected to minimize the root mean square error result of the tree after division, thereby achieving the optimal model. Figure 6 , Figure 7 A sample chart is provided for comparison and analysis of the predicted values and actual values of the beam concrete strength test sample data generated by the algorithm.
[0108] XGB (eXtreme Gradient Boosting) is an implementation of the boosting algorithm, which is a further improvement on the decision tree regression algorithm. In this study, multiple base learners are mainly used for training. The next learner learns the results of the previous base learner until the model generated by the training reaches the minimum root mean square error value. Figure 8 , Fig. 9 A sample chart is provided for comparison and analysis of the predicted values and actual values of the beam concrete strength test sample data generated by the algorithm.
[0109] Through the training and tuning of the above algorithm models, the RMSE and MAPE values are obtained respectively, and the most suitable regression algorithm is selected for comparison and establishment. The research shows that the quality evaluation model generated by the machine learning algorithm of XGB is the relatively optimal model. The result value of the beam quality evaluation based on this model can be used for the accurate sampling of the quality of the finished beams, that is, focusing on the beams with low evaluation scores, which can effectively avoid the blindness of random sampling of finished beams and the high cost and low efficiency caused by comprehensive quality inspection, and can also effectively discover the possibility of artificial falsification of beam concrete strength reports. In addition, through the precise control of the quality deviation rate index of the key process of beam prefabrication, the overall quality of the beams has been homogenized, which can effectively avoid the losses caused by the occurrence of unqualified waste beams.
[0110] The present invention adopts the above technical scheme to analyze the correlation of various factors through the quality of finished beam pieces and the deviation of process quality control, find out the key processes and key indicators that affect the quality, and use this to accurately control and achieve the overall improvement of beam piece quality. Collect prefabricated beam piece process control and finished product quality sample data, establish beam piece process control and finished product quality prediction and evaluation models through a variety of multi-factor machine learning methods, and analyze the deviation between the sample measured value and the predicted value, so as to perform machine learning algorithm model and parameter tuning, until the established model prediction results and the measured quality results achieve maximum convergence, and generate the optimal prefabricated beam piece quality evaluation model. According to the result value of the beam piece quality evaluation of the model, it can be used for the accurate sampling of the quality of the finished beam piece, that is, focusing on the sampling of beam pieces with low evaluation scores, which can effectively avoid the blindness of random sampling of finished beam pieces and the high cost and low efficiency investment caused by comprehensive quality inspection, and can also effectively discover the possibility of artificial fraud in the concrete strength report of the beam piece. The present invention can effectively promote the homogenization of prefabricated beam pieces, improve the overall quality of beam piece prefabrication, help build high-quality bridge projects, and extend the service life of bridges.
[0111] Obviously, the described embodiments are part of the embodiments of the present application, rather than all of the embodiments. In the absence of conflict, the embodiments in the present application and the features in the embodiments can be combined with each other. The components of the embodiments of the present application generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the detailed description of the embodiments of the present application is not intended to limit the scope of the application claimed for protection, but merely represents the selected embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians of the art without making creative work are within the scope of protection of the present application.
Claims
1. A method for modeling a precast beam quality evaluation model based on machine learning, characterized in that: It includes the following steps: Step 1: Collect engineering quality data of key processes and analyze various quality evaluation indicators collected in each key process link; Step 2: Establish a quality correlation analysis model, and establish corresponding quality correlation analysis models for different evaluation indicators of beam quality; select at least three correlation factors for each indicator for machine learning training through correlation analysis; Step 3, establish a process quality evaluation model: based on the relevant factors selected by the quality correlation analysis model, select a multi-factor machine learning algorithm model that supports machine learning training, and establish a process quality evaluation model for prefabricated beams; use the process quality evaluation model and sample data testing to predict, and perform convergence and differentiation analysis on the quality results of sample predictions and the measured quality index results, so as to optimize the machine learning algorithm model and parameters until the established model prediction results and the measured quality results achieve maximum convergence, thereby completing the final modeling of the process quality evaluation model for the corresponding indicators, and using the final process quality evaluation model to control and evaluate the beam prefabrication production process.
2. The method for modeling a precast beam quality evaluation model based on machine learning according to claim 1, characterized in that: Project quality data include process quality control data and finished product quality inspection data; process quality control data include process quality control data of raw materials, steel bar processing and binding, concrete mixing, concrete test blocks, pouring and vibration, steaming, tensioning and grouting; finished product quality inspection data include finished product quality inspection data of concrete test blocks and beam pieces.
3. The method for modeling a precast beam quality evaluation model based on machine learning according to claim 1, characterized in that: In step 1, the collected engineering quality data is normalized, outliers and outlier data are filtered out.
4. The method for modeling a precast beam quality evaluation model based on machine learning according to claim 1, characterized in that: Different evaluation indicators of beam quality include concrete specimen strength, steel bar protection layer thickness and steel bar spacing; that is, the quality correlation analysis model includes concrete specimen strength correlation analysis model, steel bar protection layer thickness correlation analysis model and steel bar spacing correlation analysis model.
5. The method for modeling a precast beam quality evaluation model based on machine learning according to claim 1, characterized in that: Through correlation analysis, at least three data indicators with correlation factors greater than the set value are selected for machine learning training.
6. The method for modeling a precast beam quality evaluation model based on machine learning according to claim 4, characterized in that: The Pearson correlation coefficient correlation analysis is used to perform the beam concrete strength correlation analysis. The specific steps are as follows: Step 2-1, calculate the standard deviation σ of the independent variable x of n samples of precast beams x and the standard deviation σ of the dependent variable y y , the standard deviation is calculated as follows: Where n is the total number of precast beam samples, x i is the value of the ith independent variable x, is the mean value of the independent variable x; y i is the value of the ith dependent variable y, is the mean value of the dependent variable y; Step 2-2, calculate the precast beam sample for each independent variable x i Deviation Right now Each dependent variable y i Deviation Right now And calculate the independent variable x i With the dependent variable y i The correlation coefficient value R is in the interval [-1,1], -1≤R<0 indicates negative correlation, 0<R≤1 indicates positive correlation, and the closer it is to 1 or -1, the stronger the correlation. The calculation formula is as follows: Where R is the Pearson correlation coefficient; Step 2-3, generating a scatter plot and a correlation analysis graph by combining the collected beam sample data with the Pearson algorithm correlation analysis algorithm; Step 2-4, based on the scatter plot and the correlation analysis diagram, select at least three correlation factors for each evaluation indicator whose correlation factor ratio is greater than the set value.
7. The method for modeling a precast beam quality evaluation model based on machine learning according to claim 4, characterized in that: Establishing a correlation analysis model for the strength of concrete test blocks includes establishing correlation analysis models for the strength of concrete test blocks and concrete raw materials, concrete mixing ratio, concrete mixing time, and beam steaming process.
8. The method for modeling a precast beam quality evaluation model based on machine learning according to claim 4, characterized in that: The data collected for building the beam concrete strength quality evaluation model include concrete mixing data, beam steaming data, and concrete test block data; among them, the concrete mixing data collects sand, aggregate, cement, fly ash, and water indicators; the beam steaming data collects temperature, humidity, and temperature control time indicators; the concrete test block data collects test block strength, test temperature, and test humidity indicators.
9. The method for modeling a precast beam quality evaluation model based on machine learning according to claim 4, characterized in that: In the modeling of concrete strength quality evaluation model, each beam is assigned a unique beam number to realize the association of data such as concrete mixing, beam steaming, concrete test block, etc., which lays the foundation for data correlation analysis.
10. The method for modeling a precast beam quality evaluation model based on machine learning according to claim 1, characterized in that: The criterion for achieving maximum convergence in step 3 is: the minimum root mean square error is less than the allowable error.
Citation Information
Cited By
Five-degree evaluation method for production quality of precast beam
CN119692835A
Intelligent quality inspection system and method for precast beam field, electronic equipment and storage medium
CN121678672A
Assembly type prefabricated beam bridge construction quality intelligent evaluation method based on integrated tree model
CN121960994A
