Concrete quality prediction method based on artificial intelligence

Abstract

The invention belongs to the technical field of machine learning application, and particularly relates to a concrete quality prediction method based on artificial intelligence. The method comprises the following steps: firstly, acquiring historical concrete quality and related parameters, fusing proportioning, environment and production process parameters, independently processing the steel fiber content, and verifying the representation capability of the fused parameters; fourthly, constructing four groups of BP neural network prediction sub-models, and establishing a weight correlation model to determine a final prediction model weight; and finally, inputting weight parameters to obtain concrete surface hardness and compressive strength predicted values, then judging the prediction reasonability by establishing a statistical relationship model of the two, and correcting the result if necessary. Compared with the prior art, the method has the advantages that the internal relation of parameters is deeply explored, the synergistic effect of multiple factors is considered, the concrete quality prediction precision is improved, and accurate control over the concrete quality can be better achieved.

Description

Technical Field

[0001] The present invention belongs to the technical field of machine learning applications, and particularly relates to a method for predicting the quality of concrete based on artificial intelligence. Background Art

[0002] In the field of construction engineering, as a key building material, the quality of concrete directly affects the safety and durability of the project. Existing concrete quality control methods are mostly based on simple algorithms such as random forests and support vector machines, which only simply search for prediction results without deeply exploring the internal relationships between different parameters. For example, there are complex and subtle associations between parameters such as the fineness of cement, the gradation of aggregates, and the types of admixtures. At the same time, the synergistic effects of multiple factors are also ignored. Factors such as temperature and humidity during construction will interact with the components of the concrete itself and jointly determine the final quality of the concrete. This simplistic approach results in a large deviation between the prediction results and the actual situation, making it difficult to truly achieve precise control of the quality of concrete. Summary of the Invention

[0003] In view of the technical problems existing in the above background art, the present invention proposes a method for predicting the quality of concrete based on artificial intelligence.

[0004] To achieve the above object, the technical solution adopted by the present invention is as follows, including the following steps:

[0005] S1. Obtain a plurality of first historical concrete quality parameters and quality correlation parameters. The first concrete quality parameters include surface hardness and compressive strength, and the quality correlation parameters include mix ratio, environment, production process, and steel fiber content.

[0006] S2. Perform data fusion processing. Combine cement, sand, stone, water, and admixtures in the mix ratio into one parameter, combine temperature and humidity in the environment into one parameter, combine mixing time, mixing speed, and pouring method in the production process into one parameter, and keep the steel fiber content as a single data;

[0007] S3. Construct four groups of BP neural network prediction models, and respectively train to obtain prediction sub-models under the constraint condition of keeping other three types of parameters unchanged;

[0008] S4. Establish a weight correlation model:

[0009] a. Obtain the weight parameters of each sub-model;

[0010] b. Establish a main prediction model and train to obtain weight parameters;

[0011] c. Construct a combined objective function through regularization term constraints;

[0012] d. Solve to obtain the weight parameters of the final prediction model;

[0013] S5. Input the optimized weight parameters into the final prediction model, and output the predicted values of the surface hardness and compressive strength of the concrete.

[0014] Preferably, the mixture ratio includes the dosage ratios of cement, sand, stone, water, and admixture, the environmental data is temperature and humidity, and the process data is mixing time, mixing speed, and pouring method.

[0015] Preferably, in step S2, the data is subjected to fusion processing:

[0016] The calculation formula for fusing the mixture ratio parameters is: where x i is the actual dosage of the i-th group, ρ i is the design value of the reference mix ratio, n is the effective component threshold, and μ i is the dynamic weight distribution function;

[0017] The calculation method for fusing the environmental parameters is: where T0 and H0 are the reference temperature and humidity values, v1 and v2 are the components of the feature vector, and v1 2 + v2 2 = 1;

[0018] The calculation method for fusing the process parameters is: where ω k is the weight parameter, and z k is the normalized process parameter.

[0019] Preferably, by combining the fused mixture ratio, environment, and production process with the independent steel fiber content, verify the characterization ability of the fused parameters, construct a verification mechanism, first construct a parameter reconstruction constraint through a bidirectional mapping autoencoder, and require that the cosine similarity between the feature of the fused parameters after reverse decoding and the original data is ≥ 90%, otherwise re-perform data fusion.

[0020] Preferably, in step S3, the mixture ratio is changed, and the data for prediction with the environment, production process, and steel fiber content fixed is recorded as the first prediction data sub-model;

[0021] The data for prediction with the environment changed and the mixture ratio, production process, and steel fiber content fixed is recorded as the second prediction data sub-model;

[0022] The data for prediction with the production process changed and the environment, mixture ratio, and steel fiber content fixed is recorded as the third prediction data sub-model;

[0023] The data for prediction with the steel fiber content changed and the environment, production process, and mixture ratio fixed is recorded as the fourth prediction data sub-model.

[0024] Preferably, the combined objective function constructed in step S4 is as follows: where N is the training sample, y n is the true quality data, τ main is the training weight parameter of the main model, γ q is the dynamic decay coefficient of the q-th sub-model, KL(p main ||p q ) is the KL divergence between the prediction distributions of the main model and the q-th sub-model.

[0025] Preferably, after the final prediction model makes a prediction, a statistical relationship model between surface hardness and compressive strength is established as: D(Y) = a·Y b + c, where D is the compressive strength, Y is the surface hardness, a represents the empirical coefficient, b represents the non-linear exponent, and c represents the correction term; and based on this, a judgment on the prediction rationality is made, and the calculation formula for the judgment is: where, Y pred , D pred are the predicted values of surface hardness and compressive strength respectively. When HL is 0, a correction mechanism is triggered to obtain the final prediction result.

[0026] Preferably, the calculation formula of the correction mechanism is:

[0027] Compared with the prior art, the advantages and positive effects of the present invention are as follows: By fusing multiple parameters, the data structure is simplified and the synergistic effect is highlighted, enhancing the model learning ability. Four groups of BP neural network prediction sub-models and weight correlation models are constructed to comprehensively consider the influence of various factors and optimize the prediction effect. A statistical relationship model between surface hardness and compressive strength is established, which can judge and correct the rationality of the prediction result. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0029] Figure 1 It is a schematic diagram of the overall structural flow of a concrete quality prediction method based on artificial intelligence. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0030] To better understand the above objects, features, and advantages of the present invention, the present invention will be further described below in conjunction with the accompanying drawings and embodiments. It should be noted that, without conflict, the embodiments of the present application and the features in the embodiments may be combined with each other.

[0031] In the following description, many specific details are set forth in order to provide a thorough understanding of the present invention. However, the present invention may be implemented in other ways different from those described herein. Therefore, the present invention is not limited by the specific embodiments disclosed in the following specification.

[0032] Embodiment. In the construction industry, the precise control of concrete quality is a core element to ensure project safety and long-term stability. Traditional concrete quality prediction methods have large deviations between the prediction results and the actual situation due to simple algorithms and insufficient mining of parameter correlations, making it difficult to meet the requirements of modern complex construction projects. The present invention proposes an artificial intelligence-based concrete quality prediction method, and the process of the method is as Figure 1 shown, and the specific implementation steps are as follows:

[0033] First, in the data collection stage, a plurality of first historical concrete quality parameters and quality correlation parameters are obtained; the first concrete quality parameters include surface hardness and compressive strength, and the quality correlation parameters include mix ratio, environment, production process, and steel fiber content. Among them, the mix ratio includes the dosage ratios of cement, sand, stone, water, and admixture, the environmental data is temperature and humidity, and the process data is mixing time, mixing speed, and pouring method.

[0034] Considering that in concrete quality prediction, there are complex interactions among the mix ratio components such as cement, sand, stone, water, and admixture, the temperature and humidity are closely related to the setting and hardening process of concrete, the mixing time, mixing speed, and pouring method jointly act on the uniformity and density of concrete in the production process, and the relationships between these factors and concrete quality do not exist in isolation but are intertwined. At the same time, traditional prediction methods often result in inaccurate prediction results due to the inability to effectively handle these complex relationships of multiple factors. To more accurately capture the comprehensive influence of these factors on concrete quality, the data is fused. The cement, sand, stone, water, and admixture in the mix ratio are combined into one parameter, the temperature and humidity in the environment are combined into one parameter, the mixing time, mixing speed, and pouring method in the production process are combined into one parameter, and the steel fiber content remains as a single data.

[0035] Specifically, the fusion calculation formula for the mix ratio parameter is: where x i is the actual dosage of the i-th group, ρ i is the design value of the reference mix ratio, n is the effective component threshold, μ iIt is a dynamic weight allocation function. The calculation method of environmental parameter fusion is as follows: Among them, T0 and H0 are the reference temperature and humidity values, v1 and v2 are the components of the feature vector, and v1 2 +v2 2 = 1. The calculation method of process parameter fusion is as follows: Among them, ω k is the weight parameter, and z k is the normalized process parameter. By combining the above-mentioned fusion ratios, environment, and production processes with the independent steel fiber content, the characterization ability of the fused parameters is verified, and a verification mechanism is constructed. First, a parameter reconstruction constraint is constructed through a bidirectional mapping autoencoder, requiring that the cosine similarity of the fused parameters after reverse decoding with the features of the original data be ≥ 90%. Otherwise, the data fusion is performed again. This method highlights the synergistic effect within each factor, enabling the fused parameters to more comprehensively reflect the comprehensive impact on the quality of concrete. Based on this, the model can learn more accurate features and rules, greatly improving the prediction accuracy. The simplified parameter system is convenient for constructing and interpreting the model.

[0036] Next, four groups of BP neural network prediction models are constructed. Under the constraint of keeping other three types of parameters unchanged, the prediction sub-models are trained respectively. Specifically, the data for prediction with the mixture ratio changed and the environment, production process, and steel fiber content fixed is recorded as the first prediction data sub-model; the data for prediction with the environment changed and the mixture ratio, production process, and steel fiber content fixed is recorded as the second prediction data sub-model; the data for prediction with the production process changed and the environment, mixture ratio, and steel fiber content fixed is recorded as the third prediction data sub-model; the data for prediction with the steel fiber content changed and the environment, production process, and mixture ratio fixed is recorded as the fourth prediction data sub-model.

[0037] Then, a weight correlation model is constructed: the weight parameters of each sub-model are obtained, a main prediction model is established and trained to obtain the weight parameters, and a joint objective function is constructed through the regularization term constraint. The constructed joint objective function is: where N is the training sample, y n is the true quality data, τ main is the training weight parameter of the main model, γ q is the dynamic attenuation coefficient of the q-th sub-model, and KL(p main ||p q ) is the KL divergence between the prediction distributions of the main model and the q-th sub-model. The weight parameters of the final prediction model are obtained by solving. The finally obtained weight parameters integrate the advantages of the main and sub-models, enabling the model to take into account both the individual and synergistic effects of each factor.

[0038] Then, the optimized weight parameters are input into the final prediction model to output the predicted values of the surface hardness and compressive strength of the concrete. After the output, considering the internal relationship between the surface hardness and the compressive strength, a statistical relationship model between the two is established, and based on this, the rationality of the prediction results is judged. If the judgment does not meet the requirements, the correction mechanism is triggered to further optimize the prediction results to ensure that they can more accurately reflect the actual quality of the concrete. Specifically, the statistical relationship model between the surface hardness and the compressive strength is established as: D(Y) = a·Y b + c, where D is the compressive strength, Y is the surface hardness, a represents the empirical coefficient, b represents the non-linear exponent, and c represents the correction term; and based on this, the judgment of the prediction rationality is carried out, and the calculation formula for the judgment is: where, Y pred , D pred are the predicted values of the surface hardness and the compressive strength respectively. When HL is 0, the correction mechanism is triggered to obtain the final prediction result. The calculation formula of the correction mechanism is: After being processed by the statistical relationship model, the potential relationship between the surface hardness and the compressive strength can be mined, and based on this, the unreasonable parts in the prediction data can be accurately identified. Targeted processing of these data can effectively correct the deviation, make the prediction results more in line with the actual situation, and improve the accuracy and reliability of the prediction.

[0039] The above is only the preferred embodiment of the present invention, and it is not a limitation of the present invention in other forms. Any person skilled in the art may use the technical content disclosed above to make changes or modifications into equivalent embodiments with equivalent changes and apply them to other fields. However, as long as it does not depart from the technical solution content of the present invention, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention still belong to the protection scope of the technical solution of the present invention.

Claims

1. A method for predicting the quality of concrete based on artificial intelligence, characterized in that, It includes the following steps: S1. Obtain multiple first historical concrete quality parameters and quality correlation parameters. The first concrete quality parameters include surface hardness and compressive strength, and the quality correlation parameters include mix ratio, environment, production process, and steel fiber content. S2. Perform data fusion processing. Combine cement, sand, stone, water, and admixtures in the mix ratio into one parameter, combine temperature and humidity in the environment into one parameter, combine mixing time, mixing speed, and pouring method in the production process into one parameter, and keep the steel fiber content as a single data. S3. Construct four groups of BP neural network prediction models, and respectively train to obtain prediction sub-models under the constraint condition of keeping other three types of parameters unchanged. S4. Establish a weight correlation model: a. Obtain the weight parameters of each sub-model. b. Establish a main prediction model and train to obtain the weight parameters. c. Construct a joint objective function through regularization term constraints. d. Solve to obtain the weight parameters of the final prediction model. S5. Input the optimized weight parameters into the final prediction model and output the predicted values of the surface hardness and compressive strength of the concrete.

2. The concrete quality prediction method based on artificial intelligence according to claim 1, characterized in that, In step S1, the mix ratio includes the dosage ratios of cement, sand, stone, water, and admixtures, the environmental data are temperature and humidity, and the process data are mixing time, mixing speed, and pouring method.

3. The method for predicting the quality of concrete based on artificial intelligence according to claim 1, characterized in that, In step S2, data fusion processing is performed: Among them, the fusion calculation formula of the proportioning parameters is as follows: where x i is the actual dosage of the i-th group, ρ i is the design value of the reference mix proportion, n is the effective component threshold, μ i is the dynamic weight distribution function; Among them, the calculation method of environmental parameter fusion is as follows: Among them, T0 and H0 are reference temperature and humidity values, v1 and v2 are characteristic vector components, and v1 2 + v2 2 = 1; The calculation method of process parameter fusion is as follows: where ω k is the weight parameter, and z k is the normalized process parameter.

4. The concrete quality prediction method based on artificial intelligence according to claim 4, characterized in that Verify the parameter characterization ability of the fused parameters by combining the above-mentioned fused mix ratio, environment, and production process with the independent steel fiber content, and construct a verification mechanism. First, construct a parameter reconstruction constraint through a bidirectional mapping autoencoder, requiring that the feature cosine similarity between the fused parameters after reverse decoding and the original data is ≥ 90%. Otherwise, re-perform data fusion.

5. A method for predicting the quality of concrete based on artificial intelligence according to claim 1, characterized in that, In step S3, when the mix ratio is changed and the environment, production process, and steel fiber content are fixed for prediction, the data is recorded as the first prediction data sub-model. When the environment is changed and the mix ratio, production process, and steel fiber content are fixed for prediction, the data is recorded as the second prediction data sub-model. When the production process is changed and the environment, mix ratio, and steel fiber content are fixed for prediction, the data is recorded as the third prediction data sub-model. When the steel fiber content is changed and the environment, production process, and mix ratio are fixed for prediction, the data is recorded as the fourth prediction data sub-model.

6. The method for predicting the quality of concrete based on artificial intelligence according to claim 1, characterized in that, The combined objective function constructed in step S4 is as follows: where N is the training sample, y n is the true quality data, τ main is the training weight parameter of the main model, γ q is the dynamic decay coefficient of the q-th sub-model, KL(p main ||p q ) is the KL divergence between the prediction distributions of the main model and the q-th sub-model.

7. A method for predicting the quality of concrete based on artificial intelligence according to claim 1, characterized in that, After the final prediction model makes a prediction, a statistical relationship model between surface hardness and compressive strength is established as: D(Y) = a·Y b + c, where D is the compressive strength, Y is the surface hardness, a represents the empirical coefficient, b represents the non-linear exponent, and c represents the correction term; and based on this, a judgment on the prediction rationality is made, and the calculation formula for the judgment is: where, Y pred , D pred are the predicted values of surface hardness and compressive strength respectively. When HL is 0, the correction mechanism is triggered to obtain the final prediction result.

8. The method for predicting the quality of concrete based on artificial intelligence according to claim 8, characterized in that, The calculation formula of the correction mechanism is as follows:

Images