A method for recycling of road concrete

Abstract

The application discloses a recycling method for road concrete, and relates to the technical field of concrete recycling. The application fuses service environment perception and intelligent dynamic regulation and control mechanism, effectively breaks the adaptability bottleneck of traditional process to climate difference demolition objects. Deep coupling modeling of environmental parameters and aggregate characteristics enables dynamic optimization of crushing process parameters according to the service history of raw materials, significantly improving the performance stability of recycled aggregates in freeze-thaw / salt and moisture alternating scenarios. The introduction of the attention mechanism strengthens the association mapping of extreme climate characteristics and process parameters, making the grading adjustment more consistent with the material microscopic damage characteristics.

Description

Technical Field

[0001] This invention relates to the field of concrete recycling technology, and in particular to a method for recycling and reusing road concrete. Background Technology

[0002] Concrete recycling technology, as an important means of resource recycling, often adopts a unified crushing, screening and grading method. Jaw crushers, impact crushers and other equipment are used to crush demolition concrete in multiple stages. After magnetic separation to remove steel bars and impurities, recycled aggregates are prepared according to particle size.

[0003] However, the application of uniformly processed recycled aggregates in cross-climate road engineering is not effective. In the frigid northern regions, freeze-thaw cycles cause the proliferation of microcracks inside the aggregates, while in the hot and humid southern regions, salt crystallization erosion significantly alters the pore structure and water absorption characteristics of recycled aggregates. Existing production lines use fixed-parameter crushing processes, which cannot dynamically adjust the crushing intensity and gradation strategy according to the service environment of the raw materials. For example, when processing concrete from bridge demolition that has undergone many years of freeze-thaw cycles, constant crushing parameters easily produce ultrafine aggregates, while chloride residues on the surface of salt-eroded aggregates in coastal areas interfere with the accuracy of gradation testing, leading to problems such as inaccurate foaming agent dosage and unbalanced porosity distribution during recycled concrete preparation.

[0004] To address the impact of climate differences, some solutions establish regional pretreatment standards, classify crushing strength thresholds according to climate type, or use acid washing processes to remove crystals from the surface of salt-corroded aggregates. However, static parameter settings are difficult to be compatible with different service lives within the same climate zone, such as the performance differences between 20-year-old cross-sea bridges and 5-year-old curb stones, or between load-bearing beams and non-load-bearing retaining walls in different structural parts. Therefore, there is an urgent need for a method for recycling and reusing road concrete to solve such problems. Summary of the Invention

[0005] In view of the aforementioned existing problems, the present invention is proposed.

[0006] This invention provides a method for recycling and reusing road concrete, which solves the problems of traditional crushing processes using fixed parameters to process demolished concrete in different climate zones, making it unable to adapt to the degradation of aggregate performance caused by freeze-thaw and salt corrosion, and the large fluctuations in the water absorption rate of recycled materials restricting engineering applications.

[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0008] This invention provides a method for recycling and reusing road concrete, comprising:

[0009] Step S1: Collect service environment parameters of the concrete to be demolished, including annual freeze-thaw cycles, salt spray exposure duration, and extreme temperature difference threshold.

[0010] Step S2: The demolished material is subjected to particle size classification using multi-stage crushing equipment, and the distribution data of microcracks on the aggregate surface is obtained simultaneously.

[0011] Step S3: Input environmental parameters and aggregate characteristic data into the pre-trained dynamic mix design model to generate crushing strength compensation coefficient and gradation optimization parameters.

[0012] Step S4: Adjust the hammer speed of the impact crusher and the screen inclination angle of the vibrating screen according to the gradation optimization parameters.

[0013] In a preferred embodiment of the method for recycling and reusing road concrete according to the present invention, in step S1: service environment parameters are collected by a distributed sensor array deployed at the feed inlet of the crushing line, the array including temperature and humidity sensors and a near-infrared spectrometer, the sensor spacing being 1 / 8 to 1 / 6 of the width of the crusher inlet.

[0014] In a preferred embodiment of the method for recycling and reusing road concrete according to the present invention, the dynamic mix design model construction method in step S3 includes:

[0015] A training dataset containing demolition samples from different climate zones was established, with each sample associated with aggregate water absorption rate and Cl- content detection data;

[0016] A deep residual network is used to extract the mapping relationship between environmental parameters and aggregate performance, and a temperature-salt spray coupled attention module is set between network layers.

[0017] In a preferred embodiment of the method for recycling and reusing road concrete according to the present invention, the temperature-salt spray coupled attention module is specifically implemented as follows:

[0018] Learnable weight adjustment coefficients are embedded in the feature fusion layer of the neural network. These coefficients are generated by nonlinear transformation of extreme temperature values ​​and salt spray exposure duration.

[0019] In a preferred embodiment of the method for recycling and reusing road concrete according to the present invention, in step S3, the method for generating the crushing strength compensation coefficient and gradation optimization parameters is as follows:

[0020] The service environment parameter vector a is concatenated with the aggregate surface microcrack feature vector b, and an initial representation is generated through linear mapping:

[0021]

[0022] Where x0 represents the initial feature vector after fusion, W1 is a weight matrix of size p×(3+m), d1 is a bias vector of dimension p, a1 is the number of freeze-thaw cycles per year, a2 is the salt spray exposure duration, a3 is the extreme temperature difference threshold, and b i Let m represent the microcrack characteristics of the i-th type of aggregate, m be the dimension of the microcrack characteristics, and p be the dimension of the mapped characteristics.

[0023] x0 is fed into the L-level residual module, and the recursive formula is:

[0024] x l =x l-1 +F l (x l-1 ),l=1,…,L,

[0025] Where, x l Let x represent the output feature of the l-th layer. l-1 F represents the output of layer (l-1). l Let L be the residual mapping function of the l-th layer, and L be the number of residual module layers.

[0026] Embedding temperature-salt spray coupled attention in the residual mapping, specifically:

[0027] F l (x)=A l ⊙(W l+1 x+d l+1 ),

[0028] Among them, F l (x) represents the residual mapping over input x, A l Let W be the attention weight vector of the l-th layer, where ⊙ represents element-wise multiplication. l+1 Let d be a weight matrix of size p×p. l+1 Let x be a bias vector of length p, x be the input feature vector, and l be the layer index.

[0029] Attention vector A l Perform explicit normalization calculation, the formula is:

[0030]

[0031] Among them, A l,i Let w be the i-th element of the vector, where i is the index, 1 ≤ i ≤ n, and n is the length of the attention vector. 2,i For the i-th row of the weight matrix W2, d 2,i w is the i-th element of the bias vector d2 3,i For the i-th row of the weight matrix W3, d 3,iLet be the i-th element of the bias vector d3, a3 be the extreme temperature difference threshold, a2 be the salt spray exposure time, φ and ψ be ReLU activation functions, exp(·) be the exponential operation, and ∑· be the summation operation;

[0032] Network end feature x L The crushing strength compensation coefficient α and the gradation optimization parameter vector β are generated by linear projection:

[0033] β=W4x L +d5,

[0034] Where α is the crushing strength compensation coefficient, w α Let x be a mapping vector of length p. L Let d4 be the output feature of the Lth layer, β be the mapping bias, β be the gradation optimization parameter vector, W4 be the weight matrix of size q×p, and d5 be the bias vector of length q, where q is the dimension of the gradation parameter.

[0035] In a preferred embodiment of the method for recycling and reusing road concrete according to the present invention, step S3, which involves extracting environmental parameters using a deep residual network and embedding learnable weight adjustment coefficients in the neural network feature fusion layer, is as follows:

[0036] Environmental parameters and aggregate performance characteristics are first fused together, and learnable weight adjustment coefficients are introduced into the fusion layer. Then, the data is fed into a deep residual network to extract the mapping relationship layer by layer.

[0037] Let the environmental parameter vector a∈R 3 With aggregate performance feature vector b∈R m Concatenate into a fusion vector:

[0038]

[0039] Where z is the fusion vector, a1 represents the annual number of freeze-thaw cycles, a2 represents the salt spray exposure duration, a3 represents the extreme temperature difference threshold, and b i Let m represent the performance characteristics of the i-th type of aggregate, where m is the dimension of the performance characteristics.

[0040] Design a learnable weight adjustment module to perform nonlinear mapping and normalization on the fusion vector z, generating adjustment coefficient γ:

[0041] γ = Sigmoid(W γ z+d γ ),

[0042] Where, γ∈R p W is the weight adjustment vector. γ ∈R p×(3+m) To adjust the mapping weight matrix, d γ∈R p To adjust the bias vector, Sigmoid(·) is the element-wise Sigmoid activation function, and p is the adjustment vector dimension.

[0043] In a preferred embodiment of the method for recycling and reusing road concrete according to the present invention, step S3, the step of embedding learnable weight adjustment coefficients, further includes:

[0044] The linear mapping result is scaled bitwise by γ to construct the input features of the residual network:

[0045]

[0046] in, Let W be the input feature vector, where ⊙ represents element-wise product. f ∈R p×(3+m) For the fusion mapping weight matrix, d f ∈R p For the fusion bias vector;

[0047] Will The input is a deep network consisting of L residual units, each of which consists of two fully connected layers and ReLU:

[0048]

[0049] in, For the output features of the l-th layer, For the output of layer (l-1), W l,1 W l,2 ∈R p×p These are the first and second layer mapping matrices of the l-th layer, respectively, d l,1 ,d l,2 ∈R p These are the corresponding bias vectors, ReLU(·) is the element-wise ReLU activation function, and L is the total number of residual units.

[0050] In a preferred embodiment of the method for recycling and reusing road concrete according to the present invention, step S4 further includes:

[0051] A moisture content feedback system is set up in the aggregate mixing stage. The moisture content of the aggregate is monitored in real time by a dual-frequency dielectric sensor. When the deviation between the detected value and the model prediction value exceeds the set threshold, the parameters are re-optimized.

[0052] As a preferred embodiment of the method for recycling and reusing road concrete according to the present invention, the feedback system is implemented in the following ways:

[0053] A microwave moisture detector was installed at the inlet of the mixer, and a moisture content calibration model was established using the difference in dielectric constant between 2.45 GHz and 5.8 GHz.

[0054] The discharge rate of the crusher is adjusted by setting a dynamic compensation factor, and the compensation factor is positively correlated with the moisture content deviation.

[0055] In a preferred embodiment of the method for recycling and reusing road concrete according to the present invention, in step S4, the method of setting the dynamic compensation factor to adjust the crusher discharge rate is as follows:

[0056] During the aggregate mixing stage, the moisture content deviation is assumed to be:

[0057] Δw=w meas -w pred ,

[0058] Where Δw represents the moisture content deviation, w meas This indicates real-time detection of moisture content, w pred This indicates the model's predicted moisture content;

[0059] Based on the requirements of positive correlation and saturation characteristics, the compensation factor γ(Δw) is defined as:

[0060]

[0061] Where γ(Δw) is the compensation factor, k1 is the linear gain of the first interval, w1 is the saturation threshold of the first interval, k2 is the linear gain of the second interval, w2 is the saturation threshold of the second interval, and γ max This represents the maximum value of the compensation factor.

[0062] Then, γ is applied to the online adjustment of the crusher's baseline discharge rate v0:

[0063] v out =v0(1+γ(Δw)),

[0064] Among them, v out This indicates the adjusted discharge rate, while v0 represents the calibrated baseline discharge rate.

[0065] The beneficial effects of this invention are as follows: This invention integrates service environment perception and intelligent dynamic control mechanisms, effectively breaking through the bottleneck of traditional processes in adapting to climatically different demolition materials; the deep coupling modeling of environmental parameters and aggregate characteristics enables the crushing process parameters to be dynamically optimized according to the service history of raw materials, significantly improving the performance stability of recycled aggregates in freeze-thaw / salt-wet alternating scenarios; the introduction of an attention mechanism strengthens the correlation mapping between extreme climate characteristics and process parameters, making the gradation adjustment more in line with the microscopic damage characteristics of materials; the closed-loop control system formed by the moisture content feedback system fundamentally overcomes the defects of moisture content mismatch caused by traditional static proportioning. Attached Figure Description

[0066] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0067] Figure 1 This is a schematic flowchart of the method for recycling and reusing road concrete in Example 1. Detailed Implementation

[0068] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0069] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0070] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0071] Example 1, referring to Figure 1 This embodiment provides a method for recycling and reusing road concrete, comprising:

[0072] Step S1: Collect service environment parameters of the concrete to be demolished, including annual freeze-thaw cycles, salt spray exposure duration, and extreme temperature difference threshold.

[0073] In step S1: Service environment parameters are collected by a distributed sensor array deployed at the feed inlet of the crushing line. The array includes temperature and humidity sensors and a near-infrared spectrometer. The sensor spacing is 1 / 8 to 1 / 6 of the width of the crusher inlet.

[0074] Step S2: The demolished material is subjected to particle size classification using multi-stage crushing equipment, and the distribution data of microcracks on the aggregate surface is obtained simultaneously.

[0075] Step S3: Input environmental parameters and aggregate characteristic data into the pre-trained dynamic mix design model to generate crushing strength compensation coefficient and gradation optimization parameters.

[0076] The method for constructing the dynamic proportioning model in step S3 includes:

[0077] A training dataset containing demolition samples from different climate zones was established, with each sample associated with aggregate water absorption rate and Cl- content detection data;

[0078] A deep residual network is used to extract the mapping relationship between environmental parameters and aggregate performance, and a temperature-salt spray coupled attention module is set between network layers;

[0079] The temperature-salt spray coupled attention module is specifically implemented as follows:

[0080] Learnable weight adjustment coefficients are embedded in the feature fusion layer of the neural network. These coefficients are generated by nonlinear transformation of extreme temperature values ​​and salt spray exposure duration.

[0081] In step S3, the method for generating the crushing strength compensation coefficient and gradation optimization parameters is as follows:

[0082] The service environment parameter vector a is concatenated with the aggregate surface microcrack feature vector b, and an initial representation is generated through linear mapping:

[0083]

[0084] Where x0 represents the initial feature vector after fusion, W1 is a weight matrix of size p×(3+m), d1 is a bias vector of dimension p, a1 is the number of freeze-thaw cycles per year, a2 is the salt spray exposure duration, a3 is the extreme temperature difference threshold, and b i Let m represent the microcrack characteristics of the i-th type of aggregate, m be the dimension of the microcrack characteristics, and p be the dimension of the mapped characteristics.

[0085] x0 is fed into the L-level residual module, and the recursive formula is:

[0086] x l =x l-1 +F l (x l-1 ),l=1,…,L,

[0087] Where, x l Let x represent the output feature of the l-th layer. l-1 F represents the output of layer (l-1). l Let L be the residual mapping function of the l-th layer, and L be the number of residual module layers.

[0088] Embedding temperature-salt spray coupled attention in the residual mapping, specifically:

[0089] F l (x)=A l ⊙(W l+1 x+d l+1 ),

[0090] Among them, F l (x) represents the residual mapping over input x, A l Let W be the attention weight vector of the l-th layer, where ⊙ represents element-wise multiplication. l+1 Let d be a weight matrix of size p×p. l+1 Let x be a bias vector of length p, x be the input feature vector, and l be the layer index.

[0091] Attention vector A l Perform explicit normalization calculation, the formula is:

[0092]

[0093] Among them, A l,i Let w be the i-th element of the vector, where i is the index, 1 ≤ i ≤ n, and n is the length of the attention vector. 2,i For the i-th row of the weight matrix W2, d 2,i w is the i-th element of the bias vector d2 3,i For the i-th row of the weight matrix W3, d 3,i Let be the i-th element of the bias vector d3, a3 be the extreme temperature difference threshold, a2 be the salt spray exposure time, φ and ψ be ReLU activation functions, exp(·) be the exponential operation, and ∑· be the summation operation;

[0094] Network end feature x L The crushing strength compensation coefficient α and the gradation optimization parameter vector β are generated by linear projection:

[0095] β=W4x L +d5,

[0096] Where α is the crushing strength compensation coefficient, w α Let x be a mapping vector of length p. L Let d4 be the output feature of the Lth layer, β be the mapping bias, β be the gradation optimization parameter vector, W4 be the weight matrix of size q×p, and d5 be the bias vector of length q, where q is the dimension of the gradation parameter.

[0097] Specifically, firstly, environmental parameters and aggregate microcrack features are uniformly encoded to achieve deep fusion of multi-source information. Then, a temperature difference-salt spray coupled attention module is embedded in the residual network to automatically allocate feature weights according to the service environment, highlighting key performance under extreme conditions. The explicit normalized attention mechanism not only ensures the stability of model training but also improves the ability to capture nonlinear features. Finally, a lightweight linear projection is used to generate the crushing strength compensation coefficient and gradation optimization parameters in one step, balancing computational efficiency and output accuracy.

[0098] In step S3, the step of extracting environmental parameters using a deep residual network and embedding learnable weight adjustment coefficients into the feature fusion layer of the neural network is as follows:

[0099] Environmental parameters and aggregate performance characteristics are first fused together, and learnable weight adjustment coefficients are introduced into the fusion layer. Then, the data is fed into a deep residual network to extract the mapping relationship layer by layer.

[0100] Let the environmental parameter vector a∈R 3 With aggregate performance feature vector b∈R m Concatenate into a fusion vector:

[0101]

[0102] Where z is the fusion vector, a1 represents the annual number of freeze-thaw cycles, a2 represents the salt spray exposure duration, a3 represents the extreme temperature difference threshold, and b i Let m represent the performance characteristics of the i-th type of aggregate, where m is the dimension of the performance characteristics.

[0103] Design a learnable weight adjustment module to perform nonlinear mapping and normalization on the fusion vector z, generating adjustment coefficient γ:

[0104] γ = Sigmoid(W γ z+d γ ),

[0105] Where, γ∈R p W is the weight adjustment vector. γ ∈R p×(3+m) To adjust the mapping weight matrix, d γ ∈R p To adjust the bias vector, Sigmoid(·) is the element-wise Sigmoid activation function, and p is the dimension of the adjustment vector;

[0106] Step S3, the step of embedding learnable weight adjustment coefficients, further includes:

[0107] The linear mapping result is scaled bitwise by γ to construct the input features of the residual network:

[0108]

[0109] in, Let W be the input feature vector, where ⊙ represents element-wise product. f ∈R p×(3+m) For the fusion mapping weight matrix, d f ∈R p For the fusion bias vector;

[0110] Will The input is a deep network consisting of L residual units, each of which consists of two fully connected layers and ReLU:

[0111]

[0112] in, For the output features of the l-th layer, For the output of layer (l-1), W l,1 W l,2 ∈R p×p These are the first and second layer mapping matrices of the l-th layer, respectively, d l,1 ,d l,2 ∈R p These are the corresponding bias vectors, ReLU(·) is the element-wise ReLU activation function, and L is the total number of residual units;

[0113] Specifically, through a learnable weight adjustment module, environmental parameters and aggregate performance characteristics acquire dynamic, non-linear normalized weights before entering the residual network, which helps to highlight key features and suppress redundant information. The bit-scaled fusion method maintains the independent adjustability of each feature channel, providing differentiated input for subsequent residual blocks. The residual network structure effectively alleviates the gradient vanishing problem in deep network training. Through cross-layer direct connection, the identity flow of information is achieved - which not only retains the basic information of the original fused features, but also learns complex non-linear mappings.

[0114] Step S4: Adjust the hammer speed of the impact crusher and the screen inclination angle of the vibrating screen according to the gradation optimization parameters;

[0115] Step S4 also includes:

[0116] A moisture content feedback system is set up in the aggregate mixing stage. The moisture content of the aggregate is monitored in real time by a dual-frequency dielectric sensor. When the deviation between the detected value and the model prediction value exceeds the set threshold, the parameters are re-optimized.

[0117] The implementation methods of a feedback system include:

[0118] A microwave moisture detector was installed at the inlet of the mixer, and a moisture content calibration model was established using the difference in dielectric constant between 2.45 GHz and 5.8 GHz.

[0119] The discharge rate of the crusher is adjusted by setting a dynamic compensation factor. The compensation factor is positively correlated with the moisture content deviation.

[0120] In step S4, the method for adjusting the crusher discharge rate using the dynamic compensation factor is as follows:

[0121] During the aggregate mixing stage, the moisture content deviation is assumed to be:

[0122] Δw=w meas -wpred ,

[0123] Where Δw represents the moisture content deviation, w meas This indicates real-time detection of moisture content, w pred This indicates the model's predicted moisture content;

[0124] Based on the requirements of positive correlation and saturation characteristics, the compensation factor γ(Δw) is defined as:

[0125]

[0126] Where γ(Δw) is the compensation factor, k1 is the linear gain of the first interval, w1 is the saturation threshold of the first interval, k2 is the linear gain of the second interval, w2 is the saturation threshold of the second interval, and γ max This represents the maximum value of the compensation factor.

[0127] Then, γ is applied to the online adjustment of the crusher's baseline discharge rate v0:

[0128] v out =v0(1+γ(Δw)),

[0129] Among them, v out This indicates the adjusted discharge rate, while v0 represents the calibrated baseline discharge rate.

[0130] Specifically, this piecewise function allows for rapid compensation with a larger gain k1 when the moisture content deviation is small, ensuring uniform aggregate moisture content. When the deviation enters the intermediate range, a smaller gain k2 is used for smooth transition, avoiding drastic fluctuations in the discharge rate. Once the threshold w2 is exceeded, the compensation factor saturates to γ. max To prevent overcompensation from causing equipment overload or oscillation, and to balance real-time responsiveness with system stability, parameters w1, w2, k1, k2, γ are used. max Adjustments can be made based on on-site testing.

[0131] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for recycling and reusing road concrete, characterized in that, include, Step S1: Collect service environment parameters of the concrete to be demolished, including annual freeze-thaw cycles, salt spray exposure duration, and extreme temperature difference threshold. Step S2: The demolished material is subjected to particle size classification using multi-stage crushing equipment, and the distribution data of microcracks on the aggregate surface is obtained simultaneously. Step S3: Input environmental parameters and aggregate characteristic data into the pre-trained dynamic mix design model to generate crushing strength compensation coefficient and gradation optimization parameters. Step S4: Adjust the hammer speed of the impact crusher and the screen inclination angle of the vibrating screen according to the gradation optimization parameters; The method for constructing the dynamic proportioning model in step S3 includes: A training dataset containing demolition samples from different climate zones was established, with each sample associated with aggregate water absorption rate and Cl- content detection data; A deep residual network is used to extract the mapping relationship between environmental parameters and aggregate performance, and a temperature-salt spray coupled attention module is set between network layers; The temperature-salt spray coupled attention module is specifically implemented as follows: Learnable weight adjustment coefficients are embedded in the feature fusion layer of the neural network. These coefficients are generated by nonlinear transformation of extreme temperature values ​​and salt spray exposure duration. In step S3, the method for generating the crushing strength compensation coefficient and gradation optimization parameters is as follows: Service environment parameter vector With the characteristic vector of microcracks on the aggregate surface Concatenate the data and generate an initial representation using a linear mapping: ， in, This represents the initial feature vector after fusion. For size The weight matrix, For dimension The bias vector, The number of freeze-thaw cycles per year. This refers to the duration of salt spray exposure. The extreme temperature difference threshold, For the first Microcrack characteristics of aggregate-like materials For microcrack characteristic dimensions, The feature dimension after mapping; Will Send in The recursive formula for the layer residual module is: ， in, Indicates the first Layer output features, Indicates the first Layer output, For the first Layer residual mapping function, This represents the number of residual module layers. Embedding temperature-salt spray coupled attention in the residual mapping, specifically: ， in, Indicates input The residual mapping, For the first The attention weight vector of the layer, Represents element-wise product. For size The weight matrix, For length is The bias vector, For the input feature vector, For layer index; attention vector Perform explicit normalization calculation, the formula is: ， in, For vector number 1 element, For indexing, The length of the attention vector. Weight matrix The OK, For bias vector The element, Weight matrix The OK, For bias vector The element, The extreme temperature difference threshold, This refers to the duration of salt spray exposure. and Both are ReLU activation functions. For exponential operations, For summation operations; Network endpoint characteristics The fracture strength compensation coefficient is generated by linear projection. and gradation optimization parameter vector : ， ， in, This is the crushing strength compensation coefficient. For length is The mapping vector, For the first Layer output features, For mapping bias, To optimize the gradation parameter vector, For size The weight matrix, For length is The bias vector, For gradation parameter dimensions.

2. The method for recycling and reusing road concrete as described in claim 1, characterized in that, In step S1: service environment parameters are collected by a distributed sensor array deployed at the feed inlet of the crushing line. The array includes temperature and humidity sensors and a near-infrared spectrometer. The sensor spacing is 1 / 8 to 1 / 6 of the width of the crusher inlet.

3. The method for recycling and reusing road concrete as described in claim 2, characterized in that, In step S3, the step of extracting environmental parameters using a deep residual network and embedding learnable weight adjustment coefficients into the neural network feature fusion layer is as follows: Environmental parameters and aggregate performance characteristics are first fused together, and learnable weight adjustment coefficients are introduced into the fusion layer. Then, the data is fed into a deep residual network to extract the mapping relationship layer by layer. Environmental parameter vector With aggregate performance eigenvector Concatenate into a fusion vector: ， in, For the fusion vector, Indicates the number of freeze-thaw cycles per year. Indicates the duration of salt spray exposure. Indicates the extreme temperature difference threshold. Indicates the first Performance characteristics of aggregates For performance characteristics; Design a learnable weight adjustment module to adjust the fused vector. Perform nonlinear mapping and normalization to generate adjustment coefficients. : ， in, For weight adjustment vector, To adjust the mapping weight matrix, To adjust the bias vector, It is an element-wise Sigmoid activation function. To adjust the vector dimension.

4. The method for recycling and reusing road concrete as described in claim 3, characterized in that, In step S3, the step of embedding learnable weight adjustment coefficients further includes: use The linear mapping result is scaled bitwise to construct the input features for the residual network: ， in, For the input feature vector, Represents element-wise product. To fuse the mapping weight matrix, For the fusion bias vector; Will Sent by A deep network consisting of residual units, each unit consisting of two fully connected layers and ReLU: ， in, For the first Layer output features, For the first Layer output, The first The mapping matrices of the first and second layers, These are the corresponding bias vectors. It is an element-wise ReLU activation function. This represents the total number of residual units.

5. A method for recycling and reusing road concrete as described in claim 1, characterized in that, Step S4 further includes: A moisture content feedback system is set up in the aggregate mixing stage. The moisture content of the aggregate is monitored in real time by a dual-frequency dielectric sensor. When the deviation between the detected value and the model prediction value exceeds the set threshold, the parameters are re-optimized.

6. A method for recycling and reusing road concrete as described in claim 5, characterized in that, The feedback system can be implemented in the following ways: A microwave moisture detector was installed at the inlet of the mixer, and a moisture content calibration model was established using the difference in dielectric constant between 2.45 GHz and 5.8 GHz. The discharge rate of the crusher is adjusted by setting a dynamic compensation factor, and the compensation factor is positively correlated with the moisture content deviation.

7. A method for recycling and reusing road concrete as described in claim 6, characterized in that, In step S4, the method of setting the dynamic compensation factor to adjust the crusher discharge rate is as follows: During the aggregate mixing stage, the moisture content deviation is assumed to be: ， in, Indicates moisture content deviation. This indicates real-time detection of moisture content. This indicates the model's predicted moisture content; Based on the requirements of positive correlation and saturation characteristics, a compensation factor is defined. for: ， in, As a compensation factor, For the first interval linear gain, The saturation threshold for the first interval. For the second interval linear gain, The second interval saturation threshold, This represents the maximum value of the compensation factor. Then Applied to the benchmark discharge rate of crushers Online adjustments: ， in, This indicates the adjusted discharge rate. This indicates the calibrated baseline discharge rate.

Images