Fabricated beam bridge finished product quality evaluation method fusing fuzzy reasoning and neural network

Abstract

The invention discloses a fabricated beam bridge finished product quality evaluation method fusing fuzzy reasoning and a neural network, and belongs to the field of comprehensive evaluation, and the method comprises the steps: constructing a fabricated beam bridge finished product quality intelligent evaluation system which comprises a first-level index and a second-level index; adopting a fuzzy number to describe quantitative and qualitative indexes, and constructing a fuzzy judgment matrix; calculating an initial weight vector by using an analytic hierarchy process and a defuzzification technology; calculating an improved regularization item balancing data driving and expert priori; and performing weight optimization and weighted summation on the parameters of each component by using a weighted neural network, and outputting a finished product quality score of the fabricated beam bridge. The invention aims to construct an intelligent evaluation system for the quality of the finished product of the fabricated beam bridge, realizes collaborative evaluation of expert knowledge and a data driving method by fusing a deep learning technology, improves the accuracy and robustness of the quality evaluation of the finished product of the fabricated beam bridge, and provides an efficient and reliable technical means for the intelligent evaluation of the quality of the fabricated beam bridge.

Description

Technical Field

[0001] The present invention belongs to the field of comprehensive evaluation, and in particular relates to a method for evaluating the quality of a finished product of an assembled beam bridge by integrating fuzzy reasoning and a neural network. Background Art

[0002] With the continuous development of infrastructure construction, prefabricated beam bridges have been widely used in engineering practice due to their advantages such as short construction period, high engineering quality and low environmental impact. However, the quality of its finished product is directly related to the safety and service life of the bridge, so quality evaluation has always been a key issue in engineering management.

[0003] Traditional quality evaluation methods mainly rely on expert experience and manual inspection, such as the analytic hierarchy process and the fuzzy comprehensive evaluation method. These methods determine the importance of each indicator by constructing a judgment matrix and expert scoring, and then calculate the comprehensive evaluation score. However, they are highly subjective, unstable, and difficult to fully consider the complex relationship between indicators and data uncertainty. In recent years, the development of data-driven technology and deep learning has gradually attracted attention to evaluation methods based on neural networks. Such methods can automatically learn the implicit laws in the data and realize nonlinear modeling and prediction of complex systems. However, in actual engineering, there is often insufficient sample data, and relying solely on neural networks is prone to overfitting. At the same time, it is difficult to fully utilize the valuable experience accumulated by experts in building evaluation index systems. Summary of the invention

[0004] In order to solve the above technical problems, the present invention provides a method for evaluating the quality of finished products of prefabricated beam bridges by integrating fuzzy reasoning and neural network, comprising:

[0005] Constructing an intelligent evaluation system for the quality of finished products of prefabricated beam bridges, the intelligent evaluation system including primary indicators and secondary indicators;

[0006] According to the evaluation system, fuzzy numbers are used to describe quantitative and qualitative indicators and a fuzzy judgment matrix is ​​constructed;

[0007] According to the fuzzy judgment matrix, an initial weight vector is calculated using a hierarchy analysis method and a defuzzification technique;

[0008] According to the initial weight vector, an improved regularization term that balances data-driven and expert priors is calculated;

[0009] According to the improved regularization term, a weighted neural network is used to perform weight optimization and weighted summation on the parameters of each component, and a finished product quality score of the prefabricated beam bridge is output.

[0010] Preferably, the first-level indicators include precast concrete components, ordinary rubber bearings, skateboard rubber bearings, spherical bearings, pot-type bearings, anchors, unit-type multi-directional displacement comb-shaped plate bridge expansion devices, modular expansion devices, metal bellows and plastic bellows;

[0011] The secondary indicators are specific evaluation parameters under each primary indicator.

[0012] Preferably, the step of using fuzzy numbers to describe quantitative and qualitative indicators and constructing a fuzzy judgment matrix includes:

[0013] Constructing a fuzzy judgment matrix of the primary indicators, wherein each element in the fuzzy judgment matrix uses a triangular fuzzy number to represent the importance of the indicators;

[0014] A fuzzy judgment matrix of the secondary indicators is constructed, and each element in the fuzzy judgment matrix uses a triangular fuzzy number to represent the importance between the indicators.

[0015] Preferably, the step of calculating the initial weight vector using the analytic hierarchy process and defuzzification technology comprises:

[0016] Using the fuzzy geometric mean method to calculate each row in the fuzzy judgment matrix to obtain a fuzzy weight;

[0017] Defuzzifying the fuzzy weights using a centroid method to obtain deterministic weights;

[0018] The deterministic weights are normalized to obtain an initial weight vector.

[0019] Preferably, the process of calculating the improved regularization term that balances data-driven and expert priors includes:

[0020] Construct an L1 regularization term to constrain the deviation between the weight and the initial weight;

[0021] Construct an L2 regularization term to promote the sparsity of weights;

[0022] Construct a dynamic regularization term to balance the impact of the output score on the weight optimization process;

[0023] The L1 regularization term, the L2 regularization term and the dynamic regularization term are combined to obtain an improved regularization term.

[0024] Preferably, the process of using a weighted neural network to perform weight optimization and weighted summation on the parameters of each component and outputting a quality score of the finished assembled beam bridge comprises:

[0025] Initialize the neural network parameters according to the initial weight vector, and generate the current weight through the Softmax function;

[0026] Calculate a prediction score based on the current weight;

[0027] Constructing a loss function, and combining the improved regularization term, updating the network parameters through a back propagation algorithm;

[0028] According to the optimized weights, the new input is predicted and the quality score of the finished prefabricated beam bridge is output.

[0029] Preferably, the secondary indicators of the precast concrete components in the primary indicators include appearance, geometric dimensions, concrete compressive strength, steel bar protective layer thickness, deflection, strain and cracks.

[0030] Preferably, the secondary indicators of ordinary rubber bearings in the said primary indicators include appearance quality, dimensional deviation, compressive elastic modulus, shear elastic modulus, shear adhesion performance, ultimate compressive strength, shear elastic modulus after aging, tangent value, internal quality and internal quality.

[0031] Preferably, the secondary indicators of the skateboard rubber bearing in the said primary indicators include appearance quality, dimensional deviation, compressive elastic modulus, ultimate compressive strength, surface friction coefficient between the skateboard and the stainless steel plate, internal quality and internal quality.

[0032] Preferably, the secondary indicators of the spherical bearing in the primary indicators include size and deviation, appearance quality of the bearing material and bearing assembly.

[0033] Compared with the prior art, the present invention has the following advantages and technical effects:

[0034] The present invention constructs a finished product quality evaluation system for prefabricated beam bridges that covers key evaluation indicators such as prefabricated concrete components, ordinary rubber bearings, skateboard rubber bearings, spherical bearings, pot bearings, anchors, and unit-type multi-directional displacement comb-shaped plate bridge expansion devices, which can accurately and comprehensively reflect the quality of finished bridges. At the same time, by using expert knowledge to construct a fuzzy judgment matrix, and combining the analytic hierarchy process and defuzzification technology to obtain the initial weights, and then introducing them into the weighted neural network for data-driven learning, it ensures that rapid convergence and accurate prediction can be achieved when sample data is limited. This method effectively balances the advantages of expert priors and data training, improves the robustness and interpretability of the model, and provides an efficient and reliable technical means for the intelligent evaluation of the quality of finished prefabricated beam bridges. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] The drawings constituting a part of the present application are used to provide a further understanding of the present application. The illustrative embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0036] Figure 1The figure is a schematic diagram of a method flow of an embodiment of the present invention. DETAILED DESCRIPTION

[0037] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0038] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0039] Embodiment 1

[0040] like Figure 1 As shown, in this embodiment, a method for evaluating the quality of a finished assembled beam bridge integrating fuzzy reasoning and neural network is provided, comprising:

[0041] Step S1, constructing an intelligent evaluation system for the quality of finished products of prefabricated beam bridges, the system including primary indicators and secondary indicators;

[0042] Step S2, using fuzzy numbers to describe quantitative and qualitative indicators and constructing a fuzzy judgment matrix;

[0043] Step S3, calculating the initial weight vector using the analytic hierarchy process and defuzzification technology;

[0044] Step S4, calculating an improved regularization term that balances data-driven and expert priors;

[0045] Step S5, using a weighted neural network to perform weight optimization and weighted summation on the parameters of each component, and output the quality score of the finished assembled beam bridge.

[0046] Furthermore, the first-level indicators include precast concrete components, ordinary rubber bearings, skateboard rubber bearings, spherical bearings, pot bearings, anchors, unit-type multi-directional displacement comb-shaped plate bridge expansion devices, modular expansion devices, metal bellows and plastic bellows.

[0047] Furthermore, the first-level indicators for precast concrete components include appearance, geometric dimensions, concrete compressive strength, steel cover thickness, deflection, strain and cracks;

[0048] The first-level indicators of ordinary rubber bearings include appearance quality, dimensional deviation, compressive elastic modulus, shear elastic modulus, shear bonding performance, ultimate compressive strength, shear elastic modulus after aging, tangent value, internal quality (anatomy) and internal quality (peeling);

[0049] The first-level indicators of skateboard rubber bearings include appearance quality, dimensional deviation, compressive elastic modulus, ultimate compressive strength, friction coefficient between skateboard and stainless steel plate surface, internal quality (dissection) and internal quality (peeling);

[0050] The first-level indicators for spherical bearings include secondary indicators such as size and deviation, appearance quality of bearing materials, and bearing assembly;

[0051] The first-level indicators of pot-type bearings include second-level indicators such as appearance, size and bearing assembly height;

[0052] The first-level indicators of anchors include appearance (size, cracks), static load anchoring performance, cyclic load performance, fatigue load performance, Rockwell hardness, steel strand shrinkage, friction loss rate of anchor mouth (including anchor pad) and tensioning process performance;

[0053] The first-level index of the unitized multi-directional displacement comb-shaped plate bridge expansion device includes the second-level indexes such as appearance, size deviation and waterproof performance;

[0054] The first-level index of modular telescopic devices includes secondary indexes such as tensile strength, bending performance and impact performance;

[0055] The first-level indicators of metal bellows include appearance, size, external load resistance (uniformly distributed load and local transverse load) and leakage resistance (after bending and local transverse load);

[0056] The first-level indicators of plastic corrugated pipes include appearance, ring stiffness, local lateral load, flexibility, impact resistance, specifications, tensile properties, longitudinal load, ash content, anti-aging performance, oxidation induction time, pull-out force and sealing.

[0057] Furthermore, in step S2, fuzzy numbers are used to describe quantitative and qualitative indicators, and a fuzzy judgment matrix is ​​constructed as follows:

[0058] Step S21, constructing a fuzzy judgment matrix of the first-level indicators

[0059] The finished product quality evaluation system of prefabricated beam bridges has a total of m first-level indicators, denoted as , , …, , construct the fuzzy judgment matrix of the first-level index , where each element represents the importance of the indicators and is represented by triangular fuzzy numbers. Fuzzy judgment matrix for:

[0060]

[0061] Among them, the diagonal elements (1,1,1) indicate that the indicator is equally important to itself; the off-diagonal elements Indicates the expert's Relative to the indicator fuzzy evaluation.

[0062] Step S22: construct the fuzzy judgment matrix of the secondary index

[0063] For each first-level indicator (i=1,2,…,m), below which there are Secondary indicators, denoted as , , …, The second-level index under the i-th first-level index, the fuzzy judgment matrix for:

[0064]

[0065] Among them, the diagonal elements (1,1,1) indicate that the indicator is equally important to itself; the off-diagonal elements Indicates the expert's Relative to the indicator fuzzy evaluation.

[0066] Furthermore, in step S3, the initial weight vector is calculated using the hierarchical analysis method and defuzzification technology as follows:

[0067] The fuzzy geometric mean method is used to calculate each row in the fuzzy judgment matrix of the primary and secondary indicators.

[0068] Fuzzy weight of each first-level indicator :

[0069]

[0070] For each first-level indicator The j-th secondary index under , its fuzzy weight for:

[0071]

[0072] The centroid method is used to defuzzify the fuzzy weights of the primary indicators and the secondary indicators, and the fuzzy weights are converted into deterministic weights.

[0073] The i-th level indicator The deblurring formula is:

[0074] ,

[0075] in, , , Respectively The lower, median and upper bounds of .

[0076] The i-th level indicator The deblurring formula is:

[0077] ,

[0078] in, , , Respectively The lower, median and upper bounds of .

[0079] The certainty weights of the primary and secondary indicators are normalized to satisfy:

[0080] ,

[0081] The comprehensive initial weight of each secondary indicator is composed of two levels of weights, namely:

[0082] ,

[0083] Calculate the initial weight vector based on the comprehensive initial weights of the secondary indicators:

[0084]

[0085] Where N is the total number of all secondary indicators.

[0086] Furthermore, in step S4, the improved regularization term that balances data-driven and expert priors is calculated as follows:

[0087] In order to balance data-driven learning and expert prior knowledge in the neural network training process and prevent the model from overfitting, an improved regularization term is constructed. . Improve the regularization term It consists of three parts: the L1 regularization term is used to penalize the deviation between the weight and the initial weight during the optimization process, the L2 regularization term is used to promote the sparsity of the weight, and the dynamic regularization term is used to balance the impact of the output score on the weight optimization process.

[0088] Step S41, construct L1 regularization term:

[0089]

[0090] in, is the L1 regularization hyperparameter.

[0091] Step S42, constructing an L2 regularization term:

[0092]

[0093] in, is the L2 regularization hyperparameter.

[0094] Step S43, construct a dynamic regularization term:

[0095]

[0096] in, is a dynamic regularization hyperparameter, and y is the predicted quality score.

[0097] Step S44, construct an improved regularization term:

[0098]

[0099] Furthermore, in step S5, the weighted neural network is used to perform weight optimization and weighted summation on the parameters of each component, and the quality score of the finished assembled beam bridge is output as follows:

[0100] Step S51, generating current weight;

[0101] The input vector of the finished product quality evaluation system of prefabricated beam bridge is:

[0102]

[0103] According to the initial weight vector , use logarithmic changes to initialize the neural network parameters:

[0104] ,

[0105] Generate the current weight through the Softmax function:

[0106] ,

[0107] Step S52, calculating the output layer;

[0108] Based on the generated weight , the prediction score Q is calculated based on the weighted sum:

[0109]

[0110] Step S53: construct loss function and update parameters

[0111] The original loss function is the mean square error:

[0112]

[0113] in, and They represent the predicted score and true score of the kth sample respectively, and m is the number of samples.

[0114] Combined with the improved regularization term constructed in step S4 , the total loss function is:

[0115]

[0116] Using the back-propagation algorithm, update the network parameters by gradient descent , the update formula is:

[0117] ,

[0118] in, is the learning rate.

[0119] The calculation of the gradient relies on the chain rule:

[0120]

[0121] Using the properties of the Softmax function,

[0122] , ,

[0123] Iteratively update to the total loss function Convergence, get the optimized parameters and the corresponding weights .

[0124] Step S54, calculate the final score

[0125] Using the optimized weights Predict the new input X and output the final score Q of the finished quality of the prefabricated beam bridge:

[0126]

[0127] The indicator classification is shown in Table 1 and Table 2-3:

[0128] Table 1

[0129]

[0130] Table 2

[0131]

[0132] Table 3

[0133]

[0134] The above are only preferred specific implementations of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A method for evaluating the quality of finished products of prefabricated beam bridges by integrating fuzzy reasoning and neural network, characterized in that: include: Constructing an intelligent evaluation system for the quality of finished products of prefabricated beam bridges, the intelligent evaluation system including primary indicators and secondary indicators; According to the evaluation system, fuzzy numbers are used to describe quantitative and qualitative indicators and a fuzzy judgment matrix is ​​constructed; According to the fuzzy judgment matrix, an initial weight vector is calculated using a hierarchy analysis method and a defuzzification technique; According to the initial weight vector, an improved regularization term that balances data-driven and expert priors is calculated; According to the improved regularization term, a weighted neural network is used to perform weight optimization and weighted summation on the parameters of each component, and a finished product quality score of the prefabricated beam bridge is output.

2. The method according to claim 1, characterized in that The first-level indicators include precast concrete components, ordinary rubber bearings, skateboard rubber bearings, spherical bearings, pot bearings, anchors, unit-type multi-directional displacement comb-shaped plate bridge expansion devices, modular expansion devices, metal bellows and plastic bellows; The secondary indicators are specific evaluation parameters under each primary indicator.

3. The method according to claim 1, characterized in that The steps of using fuzzy numbers to describe quantitative and qualitative indicators and constructing a fuzzy judgment matrix include: Constructing a fuzzy judgment matrix of the primary indicators, wherein each element in the fuzzy judgment matrix uses a triangular fuzzy number to represent the importance of the indicators; A fuzzy judgment matrix of the secondary indicators is constructed, and each element in the fuzzy judgment matrix uses a triangular fuzzy number to represent the importance between the indicators.

4. The method according to claim 1, characterized in that: The step of calculating the initial weight vector by using the hierarchical analysis method and the defuzzification technology comprises: Using the fuzzy geometric mean method to calculate each row in the fuzzy judgment matrix to obtain a fuzzy weight; Defuzzifying the fuzzy weights using a gravity center method to obtain deterministic weights; The deterministic weights are normalized to obtain an initial weight vector.

5. The method according to claim 1, characterized in that The process of calculating the improved regularization term that balances data-driven and expert priors includes: Construct an L1 regularization term to constrain the deviation between the weight and the initial weight; Construct an L2 regularization term to promote the sparsity of weights; Construct a dynamic regularization term to balance the impact of the output score on the weight optimization process; The L1 regularization term, the L2 regularization term and the dynamic regularization term are combined to obtain an improved regularization term.

6. The method according to claim 1, characterized in that The process of using a weighted neural network to perform weight optimization and weighted summation on the parameters of each component and outputting the quality score of the finished assembled beam bridge includes: Initialize the neural network parameters according to the initial weight vector, and generate the current weight through the Softmax function; Calculate a prediction score based on the current weight; Constructing a loss function, and combining the improved regularization term, updating the network parameters through a back propagation algorithm; According to the optimized weights, the new input is predicted and the quality score of the finished prefabricated beam bridge is output.

7. The method according to claim 1, characterized in that The secondary indicators of precast concrete components in the primary indicators include appearance, geometric dimensions, concrete compressive strength, steel bar protective layer thickness, deflection, strain and cracks.

8. The method according to claim 1, characterized in that The secondary indicators of ordinary rubber bearings in the said first-level indicators include appearance quality, dimensional deviation, compressive elastic modulus, shear elastic modulus, shear adhesion performance, ultimate compressive strength, shear elastic modulus after aging, tangent value, internal quality and internal quality.

9. The method according to claim 1, characterized in that: The secondary indicators of the skateboard rubber bearing in the said primary indicators include appearance quality, dimensional deviation, compressive elastic modulus, ultimate compressive strength, surface friction coefficient between the skateboard and the stainless steel plate, internal quality and internal quality.

10. The method according to claim 1, characterized in that The secondary indicators of the spherical bearing in the said primary indicators include size and deviation, appearance quality of the bearing material and bearing assembly.

Images