Machine learning-based method and system for predicting blast response of ultra-high performance concrete slab
By constructing a machine learning-based adversarial network for blasting data and a virtual data generation method, the problems of insufficient accuracy and data in the blast resistance response prediction model of ultra-high performance concrete slabs were solved, and high-precision blast resistance performance prediction and comprehensive analysis were achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- CHINA UNIV OF MINING & TECH
- Filing Date
- 2025-02-24
- Publication Date
- 2026-06-26
AI Technical Summary
Existing technologies struggle to construct high-precision prediction models for the blast-resistant response of ultra-high-performance concrete slabs. Furthermore, high experimental costs and insufficient data limit the generalization ability of the models, and there is a lack of comprehensive analysis of the response.
By conducting explosion experiments on ultra-high performance concrete slabs, an adversarial network for blasting data is generated. This network is then used to train and generate a virtual blasting dataset. In conjunction with various physical indicators, residual coefficients and blasting damage coefficients are generated. A blasting response prediction model is constructed, and comprehensive analysis is performed to obtain the comprehensive blasting response coefficient.
It significantly enhances the diversity and coverage of data samples, improves the model's adaptability to complex scenarios and prediction accuracy, and comprehensively reflects the blast resistance performance of ultra-high performance concrete slabs.
Smart Images
Figure CN120012606B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of blasting effect prediction technology, specifically to a method and system for predicting the blast resistance response of ultra-high performance concrete slabs based on machine learning. Background Technology
[0002] Ultra-high performance concrete (UHVPC) is increasingly widely used in national defense engineering and safety protection fields due to its excellent strength, toughness, and impact resistance. However, the blast resistance performance of UHVPC slabs is complex when faced with explosive impact loads, and its dynamic response involves multiple physical, mechanical, and failure mechanisms within the material. Currently, traditional blast resistance research relies heavily on experimental testing and finite element simulation technology. However, experimental testing is costly, time-consuming, and difficult to comprehensively cover multi-dimensional parameter combinations. While finite element simulation can partially compensate for the limitations of experiments, its accuracy and reliability are highly dependent on the material model and parameter settings, thus having certain limitations.
[0003] With the development of machine learning technology, its application to predicting the dynamic response of materials has become a trend. However, there is currently no systematic method that can combine multidimensional experimental data with generative adversarial networks to build a high-precision, ultra-high-performance concrete slab blast-resistant response prediction model, thus filling the gap in experimental data.
[0004] In the prior art, CN117610407A discloses a method and apparatus for predicting the impact response of steel-concrete composite slabs based on machine learning. This method involves constructing a first dataset based on factors influencing the impact response; filling the first dataset; training an improved Gaussian process regression algorithm model using the first dataset; and calculating and visualizing the results of the improved Gaussian process regression algorithm model. This prior art predicts the impact response capability of steel-concrete composite slabs with different structures in multiple scenarios by incorporating features of objects that may fall in urban areas and waveform steel structure features, thus reducing the limitation of the scope and increasing the upper limit of the impact resistance of steel-concrete composite slabs. It also improves the model's accuracy and reduces computation time by using Bayesian optimization to optimize the hyperparameters of the Gaussian process regression model used for prediction and finding the hyperparameter combination that minimizes the loss function. However, the prior art still has shortcomings. Due to the high cost of impact response experiments, it is difficult to obtain a sufficient amount of training data, making it difficult to guarantee the accuracy of the results. Furthermore, relying on raw data may limit the generalization ability of the trained model. Additionally, the prior art only analyzes existing responses and lacks a comprehensive analysis of the responses.
[0005] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention
[0006] The purpose of this invention is to provide a method and system for predicting the blast resistance response of ultra-high performance concrete slabs based on machine learning, so as to solve the problems mentioned in the background art.
[0007] To achieve the above objectives, the present invention provides the following technical solution:
[0008] A machine learning-based method for predicting the blast resistance response of ultra-high performance concrete slabs, comprising the following steps:
[0009] Step 1: Conduct N explosion experiments on the ultra-high performance concrete slab to obtain explosion experiment data, including explosion data, pre-explosion data, post-explosion data, and images of the ultra-high performance concrete slab after explosion.
[0010] Step 2: Analyze the images of the ultra-high performance concrete slab after the blast to obtain post-blast image feature data. Combine the post-blast data and post-blast image feature data to generate blast response data.
[0011] Step 3: Construct an explosion experiment set by combining the blasting data, pre-blasting data, and corresponding blasting response data from the N explosion experiments in Step 1; build an explosion data adversarial network, train the explosion data adversarial network using the explosion experiment set, and generate a virtual explosion dataset using the trained explosion data adversarial network.
[0012] Step 4: Using the blasting data and pre-blasting data in the virtual blasting dataset as input and the explosion response data as labels, train the blasting response prediction model, obtain the pre-blasting data and blasting data of the current blasting operation, and input them into the blasting response model to obtain the predicted blasting response data;
[0013] Step 5: Perform a comprehensive analysis of the pre-blasting data and explosion response data corresponding to the current blasting operation to obtain the property residual coefficient and blasting damage coefficient; generate a comprehensive blasting response coefficient based on the property residual coefficient and blasting damage coefficient.
[0014] Furthermore, the blasting data includes the pressure and duration of the applied blasting shock wave, the distance between the blasting shock wave source and the plane of the ultra-high performance concrete slab, the angle between the line connecting the blasting shock wave source and the center of the ultra-high performance concrete slab and the plane of the ultra-high performance concrete slab, and the pre-blasting data includes the thickness, compressive strength, tensile strength, elastic modulus, mass, and density of the ultra-high performance concrete slab; the post-blasting data includes the compressive strength, elastic modulus, and mass of the ultra-high performance concrete slab after blasting, as well as the stress distribution on the back of the ultra-high performance concrete slab.
[0015] The explosion response data includes images, compressive strength, elastic modulus and mass of the ultra-high performance concrete slab after the explosion, as well as the stress distribution on the back side of the ultra-high performance concrete slab, which is the opposite side of the ultra-high performance concrete slab facing the explosion.
[0016] Furthermore, the explosion response data includes the crack ratio, maximum crack depth ratio, deep crack ratio, back stress distribution variance, compressive strength, elastic modulus, and mass of the ultra-high performance concrete slab after the explosion.
[0017] The specific logic for calculating the proportion of cracks, the relative ratio of maximum crack depth, and the proportion of deep cracks is as follows: In the image of ultra-high performance concrete slab, the frequency of each pixel in 256 gray levels is used to form a gray level histogram, and the inter-class variance of each threshold in the histogram is calculated. The optimal separation threshold is selected by the inter-class variance, and the optimal separation threshold is used to perform threshold segmentation on the image of the front of ultra-high performance concrete slab to identify the crack part and the concrete part.
[0018] Calculate the inter-class variance corresponding to the threshold using the following formula:
[0019]
[0020] in, Indicates the threshold is Inter-class variance over time ,and , , The thresholds are respectively The number of pixels in the cracked and concrete sections at that time. , The thresholds are respectively The average grayscale values of pixels in the background and foreground regions at that time;
[0021] Calculate the threshold value at which the inter-class variance reaches its maximum, and define this value as the optimal separation threshold;
[0022] The crack percentage is calculated by dividing the number of pixels in the cracked area by the total number of pixels. The average grayscale value of the concrete portion is calculated, and the pixel with the lowest grayscale value in the cracked area is compared to the average grayscale value of the concrete portion to obtain the maximum crack depth. A crack grayscale threshold is preset, and pixels with grayscale values less than this threshold are defined as deep pixels. The number of deep pixels is counted, and the percentage of deep cracks is calculated by dividing this number by the total number of pixels in the cracked area. The crack grayscale threshold is 50% of the average grayscale value of the concrete portion.
[0023] The specific formulas used to calculate the crack ratio, the relative ratio of maximum crack depth, the proportion of deep cracks, and the variance of stress distribution on the back of the plate are as follows:
[0024]
[0025]
[0026]
[0027]
[0028] in, This represents the average gray value of the concrete portion. For the concrete part grayscale value of each pixel This represents the total number of pixels in the concrete section. This is the index of the pixels in the concrete section. The percentage of cracks, This represents the total number of pixels in the cracked area. This represents the total number of pixels. For comparison of maximum crack depth, The gray value of the pixel with the smallest gray value in the cracked area; The percentage of deep cracks, This represents the total number of deep pixels.
[0029] Furthermore, the explosive data adversarial network includes a generator and a discriminator. The generator comprises an input layer, an output layer, and a hidden layer. The input layer of the generator receives the explosive experiment set and adds random noise to the explosive data, pre-explosion data, and corresponding explosive response data as initial input. The hidden layer of the generator consists of multiple layers, each containing several nodes, and each node is connected to the previous layer through weights. The ReLU activation function is used to perform a nonlinear transformation on the input data, enabling the model to generate complex distributions. An independent neuron is set in the output layer of the generator, responsible for converting the data transformed by the hidden layer into virtual data generated by the generator.
[0030] The discriminator comprises an input layer, an output layer, and a hidden layer. The input layer receives the explosion test set and virtual data generated by the generator. The hidden layer consists of multiple layers, each containing several nodes, with each node connected to the previous layer via weights. The ReLU activation function is used to perform a non-linear transformation on the input data, enabling the model to generate complex distributions. The output layer contains an independent neuron that uses the Sigmoid activation function, with an output range of [0,1] reflecting the discriminator's output probabilities for real data and generated virtual data.
[0031] The formula for the loss function is:
[0032]
[0033] in, For the data brute-force adversarial network loss function, The output probability of the discriminator for the real data; To obtain the mathematical expectation operation, for From real data A sample was obtained from it. The output probability of the discriminator for the generated virtual data; for From random noise distribution A sample was obtained from it.
[0034] Furthermore, the specific logic for obtaining the residual coefficients of the properties is as follows: a comprehensive analysis of the compressive strength, elastic modulus, and mass of the ultra-high performance concrete slab is conducted to obtain the residual strength ratio and residual mass ratio;
[0035] The specific formula used to calculate the residual coefficient of the aforementioned property is as follows:
[0036]
[0037] in, The residual coefficient is the property coefficient. The compressive strength of the ultra-high performance concrete slab before blasting. The compressive strength of the ultra-high performance concrete slab after blasting. The elastic modulus of the ultra-high performance concrete slab before blasting. The elastic modulus of the ultra-high performance concrete slab after blasting; The quality of the ultra-high performance concrete slab before blasting. This refers to the quality of the ultra-high performance concrete slab after the blasting.
[0038] Furthermore, the blasting failure coefficient is generated based on the crack ratio, the relative ratio of maximum crack depth, the proportion of deep cracks, and the variance of stress distribution on the back of the plate. The specific formula used to generate the blasting failure coefficient is as follows:
[0039]
[0040] in, The destructive coefficient of blasting. The percentage of cracks; For comparison of maximum crack depth, The percentage of deep cracks, The variance of the stress distribution on the back of the plate;
[0041] The comprehensive blasting response coefficient is generated based on the residual coefficient and the blasting damage coefficient. The specific formula for generating the comprehensive blasting response coefficient is as follows:
[0042]
[0043] in, To achieve a comprehensive blasting response coefficient, This is the residual coefficient of the property.
[0044] This invention further provides a machine learning-based system for predicting the blast resistance response of ultra-high performance concrete slabs. The system is used to implement the aforementioned machine learning-based method for predicting the blast resistance response of ultra-high performance concrete slabs, specifically including:
[0045] The data acquisition module is used to conduct N explosion experiments on ultra-high performance concrete slabs and acquire explosion experiment data, including blasting data, pre-blasting data, post-blasting data, and images of the ultra-high performance concrete slabs after blasting.
[0046] The response analysis module is used to analyze the images of the ultra-high performance concrete slab after blasting, obtain the post-blast image feature data, and combine the post-blast data and the post-blast image feature data to generate blast response data.
[0047] The data virtual module is used to construct an explosion experiment set by combining the blasting data, pre-blasting data and corresponding blasting response data of the N explosion experiments in step 1; to build a blasting data adversarial network, to train the blasting data adversarial network using the explosion experiment set, and to generate a virtual blasting dataset using the trained blasting data adversarial network.
[0048] The model building module is used to train a blasting response prediction model by taking blasting data and pre-blasting data from the virtual blasting dataset as input and blasting response data as labels. It obtains the pre-blasting data and blasting data of the current blasting operation and inputs them into the blasting response model to obtain the predicted blasting response data.
[0049] The comprehensive analysis module is used to comprehensively analyze the pre-blasting data and explosion response data corresponding to the current blasting operation to obtain the property residual coefficient and blasting damage coefficient; and to generate the comprehensive blasting response coefficient based on the property residual coefficient and blasting damage coefficient.
[0050] Compared with the prior art, the beneficial effects of the present invention are:
[0051] According to the above scheme, the present invention generates virtual experimental data through a data adversarial network, which significantly enhances the diversity and coverage of data samples, thereby improving the model's adaptability to complex scenarios and prediction accuracy.
[0052] The present invention also uses the above-mentioned scheme to generate a property residual coefficient and a blasting failure coefficient by integrating multiple physical indicators, and further analyzes them to obtain a comprehensive blasting response coefficient, which can comprehensively reflect the blast resistance performance of ultra-high performance concrete slabs and has stronger applicability. Attached Figure Description
[0053] Figure 1 This is a schematic diagram of the overall method flow of the present invention.
[0054] Figure 2 This is a schematic diagram of the overall system structure of the present invention. Detailed Implementation
[0055] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.
[0056] It should be noted that, unless otherwise defined, the technical or scientific terms used in this invention should have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.
[0057] Example:
[0058] Please see Figure 1 The present invention provides a technical solution:
[0059] A machine learning-based method for predicting the blast resistance response of ultra-high performance concrete slabs, comprising the following steps:
[0060] Step 1: Conduct N explosion experiments on the ultra-high performance concrete slab to obtain explosion experiment data, including explosion data, pre-explosion data, post-explosion data, and images of the ultra-high performance concrete slab after explosion.
[0061] Furthermore, the blasting data includes the pressure and duration of the applied blasting shock wave, the distance between the blasting shock wave source and the plane of the ultra-high performance concrete slab, the angle between the line connecting the blasting shock wave source and the center of the ultra-high performance concrete slab and the plane of the ultra-high performance concrete slab, and the pre-blasting data includes the thickness, compressive strength, tensile strength, elastic modulus, mass, and density of the ultra-high performance concrete slab; the post-blasting data includes the compressive strength, elastic modulus, and mass of the ultra-high performance concrete slab after blasting, as well as the stress distribution on the back of the ultra-high performance concrete slab.
[0062] The pre-blast data is obtained by selecting one ultra-high performance concrete slab from the explosion experiment as a sample and sending it to the laboratory for measurement. If the ultra-high performance concrete slab does not break into multiple pieces after the explosion, the ultra-high performance concrete slab after the explosion is directly sent to the laboratory to obtain the post-blast data. If the ultra-high performance concrete slab breaks into multiple pieces, the largest piece is selected and sent to the laboratory to obtain the post-blast data.
[0063] Step 2: Analyze the images of the ultra-high performance concrete slab after the blast to obtain post-blast image feature data. Combine the post-blast data and post-blast image feature data to generate blast response data.
[0064] Furthermore, the explosion response data includes the crack ratio, maximum crack depth ratio, deep crack ratio, back stress distribution variance, compressive strength, elastic modulus, and mass of the ultra-high performance concrete slab after the explosion.
[0065] The specific logic for calculating the proportion of cracks, the relative ratio of maximum crack depth, and the proportion of deep cracks is as follows: In the image of ultra-high performance concrete slab, the frequency of each pixel in 256 gray levels is used to form a gray level histogram, and the inter-class variance of each threshold in the histogram is calculated. The optimal separation threshold is selected by the inter-class variance, and the optimal separation threshold is used to perform threshold segmentation on the image of the front of ultra-high performance concrete slab to identify the crack part and the concrete part.
[0066] Calculate the inter-class variance corresponding to the threshold using the following formula:
[0067]
[0068] in, Indicates the threshold is Inter-class variance over time ,and , , The thresholds are respectively The number of pixels in the cracked and concrete sections at that time. , The thresholds are respectively The average grayscale values of pixels in the background and foreground regions at that time;
[0069] Calculate the threshold value at which the inter-class variance reaches its maximum, and define this value as the optimal separation threshold;
[0070] The crack percentage is calculated by dividing the number of pixels in the cracked area by the total number of pixels. The average grayscale value of the concrete portion is calculated, and the maximum crack depth is obtained by comparing the pixel with the lowest grayscale value in the cracked area with the average grayscale value of the concrete portion. A crack grayscale threshold is preset, and pixels with grayscale values less than this threshold are defined as deep pixels. The number of deep pixels is counted, and the percentage of deep cracks is obtained by dividing this number by the total number of pixels in the cracked area. The crack grayscale threshold is 50% of the average grayscale value of the concrete portion.
[0071] The specific formulas used to calculate the crack ratio, the relative ratio of maximum crack depth, the proportion of deep cracks, and the variance of stress distribution on the back of the plate are as follows:
[0072]
[0073]
[0074]
[0075]
[0076] in, This represents the average gray value of the concrete portion. For the concrete part grayscale value of each pixel This represents the total number of pixels in the concrete section. This is the index of the pixels in the concrete section. The percentage of cracks, This represents the total number of pixels in the cracked area. This represents the total number of pixels. For comparison of maximum crack depth, The gray value of the pixel with the smallest gray value in the cracked area; The percentage of deep cracks, This represents the total number of deep pixels.
[0077] In grayscale images, generally, a higher grayscale value indicates a shallower crack. This is because cracked areas typically reflect more light, resulting in higher grayscale values. Conversely, the areas surrounding the crack appear darker. Therefore, in calculating the maximum crack depth, the minimum grayscale value of the cracked area is used to represent the maximum depth; pixels with grayscale values in the cracked area that are less than the crack grayscale threshold are defined as deep pixels.
[0078] Step 3: Construct an explosion experiment set by combining the blasting data, pre-blasting data, and corresponding blasting response data from the N explosion experiments in Step 1; build an explosion data adversarial network, train the explosion data adversarial network using the explosion experiment set, and generate a virtual explosion dataset using the trained explosion data adversarial network.
[0079] The explosive data adversarial network includes a generator and a discriminator. The generator comprises an input layer, an output layer, and a hidden layer. The input layer of the generator receives the explosive experiment set and adds random noise to the explosive data, pre-explosion data, and corresponding explosive response data as initial input. The hidden layer of the generator consists of multiple layers, each containing several nodes, and each node is connected to the previous layer through weights. The ReLU activation function is used to perform a nonlinear transformation on the input data, enabling the model to generate complex distributions. An independent neuron is set in the output layer of the generator, responsible for converting the data transformed by the hidden layer into virtual data generated by the generator.
[0080] The discriminator comprises an input layer, an output layer, and a hidden layer. The input layer receives the explosion test set and virtual data generated by the generator. The hidden layer consists of multiple layers, each containing several nodes, with each node connected to the previous layer via weights. The ReLU activation function is used to perform a non-linear transformation on the input data, enabling the model to generate complex distributions. The output layer contains an independent neuron that uses the Sigmoid activation function, with an output range of [0,1] reflecting the discriminator's output probabilities for real data and generated virtual data.
[0081] The formula for the loss function is:
[0082]
[0083] in, For the data brute-force adversarial network loss function, The output probability of the discriminator for the real data; To obtain the mathematical expectation operation, for From real data A sample was obtained from it. The output probability of the discriminator for the generated virtual data; for From random noise distribution A sample was obtained from it.
[0084] Because explosion experiments are costly and it is impossible to obtain a large amount of explosion experiment data to train the subsequent blast response prediction model, the accuracy of the blast response prediction model will be insufficient without a large amount of explosion experiment data for training. Therefore, it is necessary to generate a large amount of virtual explosion experiment data through blast data adversarial networks to complete the training of the blast response prediction model.
[0085] Step 4: Using the blasting data and pre-blasting data in the virtual blasting dataset as input and the explosion response data as labels, train the blasting response prediction model, obtain the pre-blasting data and blasting data of the current blasting operation, and input them into the blasting response model to obtain the predicted blasting response data;
[0086] The blast response prediction model employs a feedforward neural network, using blast response data as labels. Existing techniques can be used to train and optimize the model using blast data and pre-blast data from the virtual blast dataset. Specifically, it includes an input layer, hidden layers, an output layer, and an activation function. The input layer receives historical environmental data; the hidden layer processes this data; each layer consists of multiple layers with four nodes. Nodes in each hidden layer are connected to the previous layer via weights, performing feature abstraction and nonlinear transformation on the input historical environmental data. The ReLU activation function introduces nonlinear relationships, enabling the model to fit complex feature relationships. An independent neuron in the output layer transforms the local and high-level feature representations extracted by the hidden layer into the output air purifier response value. A root mean square error loss function is used. The input data is processed through the network once to obtain the output result. The loss function is calculated based on the predicted and true values. The gradient of the loss function with respect to each weight and bias is calculated using the chain rule. The gradient descent algorithm is used to update the network weights and biases to minimize the loss function.
[0087] Step 5: Perform a comprehensive analysis of the pre-blasting data and explosion response data corresponding to the current blasting operation to obtain the property residual coefficient and blasting damage coefficient; generate a comprehensive blasting response coefficient based on the property residual coefficient and blasting damage coefficient.
[0088] The explosion response data includes images, compressive strength, elastic modulus and mass of the ultra-high performance concrete slab after the explosion, as well as the stress distribution on the back side of the ultra-high performance concrete slab, which is the opposite side of the ultra-high performance concrete slab facing the explosion.
[0089] The specific logic for obtaining the residual coefficients of the properties is as follows: a comprehensive analysis of the compressive strength, elastic modulus and mass of the ultra-high performance concrete slab is conducted to obtain the residual strength ratio and residual mass ratio;
[0090] The specific formula used to calculate the residual coefficient of the aforementioned property is as follows:
[0091]
[0092] in, The residual coefficient is the property coefficient. The compressive strength of the ultra-high performance concrete slab before blasting. The compressive strength of the ultra-high performance concrete slab after blasting. The elastic modulus of the ultra-high performance concrete slab before blasting. The elastic modulus of the ultra-high performance concrete slab after blasting; The quality of the ultra-high performance concrete slab before blasting. This refers to the quality of the ultra-high performance concrete slab after the blasting.
[0093] The property residual coefficient represents the degree to which the overall performance of ultra-high performance concrete slabs is retained after blasting relative to their pre-blast state. A higher value indicates that the overall performance of the concrete slab after blasting is closer to its initial state, and its blast resistance is better. The property residual coefficient directly compares the retention ratio of various material properties before and after blasting, reflecting changes in load-bearing capacity and also considering the weakening of material deformation capacity, providing a more comprehensive damage response index. The compressive strength of ultra-high performance concrete slabs reflects their load-bearing capacity under axial force, while the elastic modulus reflects their resistance to deformation under stress. It reflects the degree to which the compressive strength of ultra-high performance concrete slabs is retained after blasting relative to that before blasting. The larger the value, the closer the compressive strength of the concrete slab is to its initial state after blasting, and the better its blast resistance. It reflects the degree to which the elastic modulus of ultra-high performance concrete slabs are retained after blasting relative to their initial state. The larger the value, the closer the elastic modulus of the concrete slab is to its initial state after blasting, and the better its blast resistance. This value reflects the degree to which the mass of the ultra-high performance concrete slab is preserved after blasting relative to its initial state. The higher the value, the closer the mass of the concrete slab is to its initial state after blasting, and the better its blast resistance.
[0094] The blasting failure coefficient is generated based on the crack ratio, the relative ratio of maximum crack depth, the proportion of deep cracks, and the variance of stress distribution on the back of the plate. The specific formula for generating the blasting failure coefficient is as follows:
[0095]
[0096] in, The destructive coefficient of blasting. The percentage of cracks; For comparison of maximum crack depth, The percentage of deep cracks, The variance of the stress distribution on the back of the plate;
[0097] The blasting failure factor is a coefficient that reflects the degree of damage to a plate material under blasting, taking into account factors such as the proportion of cracks, crack depth, and stress distribution. It is an indicator for measuring the degree of damage to a plate material. The smaller the value, the more severe the damage to the board under explosive force. Crack percentage refers to the proportion of the entire board area with cracks, measuring the breadth or extent of crack propagation; the larger the value, the wider the crack distribution and the more severe the damage to the board. More cracks mean the board is more susceptible to explosive or impact damage. Deep crack percentage indicates the proportion of deeper cracks in the board. A higher percentage of deep cracks indicates a higher proportion of deep cracks in the slab, meaning these deep cracks are more likely to cause slab failure, as deep cracks typically propagate or fracture more easily than shallow cracks. Since both the crack percentage and deep crack percentage reflect the degree of damage to the concrete slab through the proportion of cracks, this... The two are coupled together to describe the degree of damage to the concrete slab; the gray value of the pixel with the smallest gray value in the crack area represents the pixel with the smallest gray value in the crack area, which usually indicates the deepest part of the crack or the most vulnerable area. The larger the value, the shallower the deepest point, indicating a less severe crack or less surface damage. However, if... Smaller cracks indicate deeper cracks and a greater threat to structural stability. Before the blast, the stress distribution on the back of the concrete slab is relatively uniform due to its relatively regular structure, and the variance of the stress distribution on the back of the slab is also relatively small. However, after the blast, the concrete slab is damaged, and its structure is no longer regular. Therefore, the stress distribution on the back of the slab is also uneven. Generally, the greater the damage to the concrete slab, the higher the degree of irregularity in its structure, and the greater the variance of the stress distribution on the back of the slab.
[0098] The comprehensive blasting response coefficient is generated based on the residual coefficient and the blasting damage coefficient. The specific formula for generating the comprehensive blasting response coefficient is as follows:
[0099]
[0100] in, To achieve a comprehensive blasting response coefficient, The residual coefficient represents the overall performance of the ultra-high performance concrete slab. The comprehensive blast response coefficient reflects the overall response of the slab; the higher the value, the better the performance of the slab during blasting. The generation of this coefficient provides an important basis for evaluating the comprehensive response performance of the slab. The blast damage coefficient reflects the degree of damage to the slab under blasting. It is an indicator for measuring the degree of damage to the slab. The larger the value, the less damage the slab will suffer under blasting, and the better the properties of the ultra-high performance concrete slab in response to blasting. The property residual coefficient indicates the degree to which the overall performance of the ultra-high performance concrete slab is retained after blasting relative to its pre-blast state. The larger the value, the closer the overall performance of the concrete slab is to its initial state after blasting; the better the performance of the ultra-high performance concrete slab in response to blasting.
[0101] Please see Figure 2 The present invention further provides a machine learning-based system for predicting the blast resistance response of ultra-high performance concrete slabs. This system is used to implement the aforementioned machine learning-based method for predicting the blast resistance response of ultra-high performance concrete slabs, specifically including:
[0102] The data acquisition module is used to conduct N explosion experiments on ultra-high performance concrete slabs and acquire explosion experiment data, including blasting data, pre-blasting data, post-blasting data, and images of the ultra-high performance concrete slabs after blasting.
[0103] The response analysis module is used to analyze the images of the ultra-high performance concrete slab after blasting, obtain the post-blast image feature data, and combine the post-blast data and the post-blast image feature data to generate blast response data.
[0104] The data virtual module is used to construct an explosion experiment set by combining the blasting data, pre-blasting data and corresponding blasting response data of the N explosion experiments in step 1; to build a blasting data adversarial network, to train the blasting data adversarial network using the explosion experiment set, and to generate a virtual blasting dataset using the trained blasting data adversarial network.
[0105] The model building module is used to train a blasting response prediction model by taking blasting data and pre-blasting data from the virtual blasting dataset as input and blasting response data as labels. It obtains the pre-blasting data and blasting data of the current blasting operation and inputs them into the blasting response model to obtain the predicted blasting response data.
[0106] The comprehensive analysis module is used to comprehensively analyze the pre-blasting data and explosion response data corresponding to the current blasting operation to obtain the property residual coefficient and blasting damage coefficient; and to generate the comprehensive blasting response coefficient based on the property residual coefficient and blasting damage coefficient.
[0107] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.
[0108] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented in software, the above embodiments can be implemented, in whole or in part, as a computer program product. Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution.
[0109] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.
[0110] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that cannot be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.
Claims
1. A method for predicting the blast resistance response of ultra-high performance concrete slabs based on machine learning, characterized in that, The specific steps include: Step 1: Conduct N explosion experiments on the ultra-high performance concrete slab to obtain explosion experiment data, including explosion data, pre-explosion data, post-explosion data, and images of the ultra-high performance concrete slab after explosion. Step 2: Analyze the images of the ultra-high performance concrete slab after the blast to obtain post-blast image feature data. Combine the post-blast data and post-blast image feature data to generate blast response data. Step 3: Construct an explosion experiment set by combining the blasting data, pre-blasting data, and corresponding blasting response data from the N explosion experiments in Step 1; build an explosion data adversarial network, train the explosion data adversarial network using the explosion experiment set, and generate a virtual explosion dataset using the trained explosion data adversarial network. The explosive data adversarial network includes a generator and a discriminator. The generator comprises an input layer, an output layer, and a hidden layer. The input layer of the generator receives the explosive experiment set and adds random noise to the explosive data, pre-explosion data, and corresponding explosive response data as initial input. The hidden layer of the generator consists of multiple layers, each containing several nodes, and each node is connected to the previous layer through weights. The ReLU activation function is used to perform a nonlinear transformation on the input data, enabling the model to generate complex distributions. An independent neuron is set in the output layer of the generator, responsible for converting the data transformed by the hidden layer into virtual data generated by the generator. The discriminator comprises an input layer, an output layer, and a hidden layer. The input layer receives the explosion test set and virtual data generated by the generator. The hidden layer consists of multiple layers, each containing several nodes, with each node connected to the previous layer via weights. The ReLU activation function is used to perform a non-linear transformation on the input data, enabling the model to generate complex distributions. The output layer contains an independent neuron that uses the Sigmoid activation function, with an output range of [0,1] reflecting the discriminator's output probabilities for real data and generated virtual data. The formula for the loss function is: in, For the data brute-force adversarial network loss function, The output probability of the discriminator for the real data; To obtain the mathematical expectation operation, for From real data A sample was obtained from it. The output probability of the discriminator for the generated virtual data; for From random noise distribution A sample was obtained from it; Step 4: Using the blasting data and pre-blasting data in the virtual blasting dataset as input and the explosion response data as labels, train the blasting response prediction model, obtain the pre-blasting data and blasting data of the current blasting operation, and input them into the blasting response model to obtain the predicted blasting response data; Step 5: Perform a comprehensive analysis of the pre-blasting data and explosion response data corresponding to the current blasting operation to obtain the property residual coefficient and blasting damage coefficient; generate a comprehensive blasting response coefficient based on the property residual coefficient and blasting damage coefficient.
2. The method for predicting the blast resistance response of ultra-high performance concrete slabs based on machine learning according to claim 1, characterized in that: The blasting data includes the pressure and duration of the applied blasting shock wave, the distance between the blasting shock wave source and the plane of the ultra-high performance concrete slab, the angle between the line connecting the blasting shock wave source and the center of the ultra-high performance concrete slab and the plane of the ultra-high performance concrete slab, and the pre-blasting data including the thickness, compressive strength, tensile strength, elastic modulus, mass, and density of the ultra-high performance concrete slab; the post-blasting data includes the compressive strength, elastic modulus, and mass of the ultra-high performance concrete slab after blasting, as well as the stress distribution on the back of the ultra-high performance concrete slab.
3. The method for predicting the blast resistance response of ultra-high performance concrete slabs based on machine learning according to claim 1, characterized in that: The explosion response data includes the crack ratio, maximum crack depth ratio, deep crack ratio, back stress distribution variance, compressive strength, elastic modulus, and mass of the ultra-high performance concrete slab after blasting. The specific logic for calculating the proportion of cracks, the relative ratio of maximum crack depth, and the proportion of deep cracks is as follows: In the image of ultra-high performance concrete slab, the frequency of each pixel in 256 gray levels is used to form a gray level histogram, and the inter-class variance of each threshold in the histogram is calculated. The optimal separation threshold is selected by the inter-class variance, and the optimal separation threshold is used to perform threshold segmentation on the image of the front of ultra-high performance concrete slab to identify the crack part and the concrete part. Calculate the inter-class variance corresponding to the threshold using the following formula: in, Indicates the threshold is Inter-class variance over time ,and , , The thresholds are respectively The number of pixels in the cracked and concrete sections at that time. , The thresholds are respectively The average grayscale values of pixels in the background and foreground regions at that time; Calculate the threshold value at which the inter-class variance reaches its maximum, and define this value as the optimal separation threshold; The crack percentage is calculated by dividing the number of pixels in the cracked area by the total number of pixels. The average grayscale value of the concrete portion is calculated, and the maximum crack depth is obtained by comparing the pixel with the lowest grayscale value in the cracked area with the average grayscale value of the concrete portion. A crack grayscale threshold is preset, and pixels with grayscale values less than this threshold are defined as deep pixels. The number of deep pixels is counted, and the percentage of deep cracks is obtained by dividing this number by the number of pixels in the cracked area. The crack grayscale threshold is 50% of the average grayscale value of the concrete portion. The specific formulas used to calculate the crack ratio, the relative ratio of maximum crack depth, the proportion of deep cracks, and the variance of stress distribution on the back of the plate are as follows: in, The average gray value of the concrete portion. For the concrete part grayscale value of each pixel This represents the total number of pixels in the concrete section. This is the index of the pixels in the concrete section. The percentage of cracks, This represents the total number of pixels in the cracked area. This represents the total number of pixels. For comparison of maximum crack depth, The gray value of the pixel with the lowest gray value in the cracked area; The percentage of deep cracks, This represents the total number of deep pixels.
4. The method for predicting the blast resistance response of ultra-high performance concrete slabs based on machine learning according to claim 1, characterized in that: The specific logic for obtaining the residual coefficients of the properties is as follows: a comprehensive analysis of the compressive strength, elastic modulus and mass of the ultra-high performance concrete slab is conducted to obtain the residual strength ratio and residual mass ratio; The specific formula used to calculate the residual coefficient of the aforementioned property is as follows: in, The residual coefficient is the property coefficient. The compressive strength of the ultra-high performance concrete slab before blasting. The compressive strength of the ultra-high performance concrete slab after blasting. The elastic modulus of the ultra-high performance concrete slab before blasting. Elastic modulus of ultra-high performance concrete slab after blasting; The quality of the ultra-high performance concrete slab before blasting. This refers to the quality of the ultra-high performance concrete slab after the blasting.
5. The method for predicting the blast resistance response of ultra-high performance concrete slabs based on machine learning according to claim 2, characterized in that: The blasting failure coefficient is generated based on the crack ratio, the relative ratio of maximum crack depth, the proportion of deep cracks, and the variance of stress distribution on the back of the plate. The specific formula for generating the blasting failure coefficient is as follows: in, The destructive coefficient of blasting. The percentage of cracks; For comparison of maximum crack depth, The percentage of deep cracks, The variance of the stress distribution on the back of the plate; The comprehensive blasting response coefficient is generated based on the residual coefficient and the blasting damage coefficient. The specific formula for generating the comprehensive blasting response coefficient is as follows: in, To achieve a comprehensive blasting response coefficient, This is the residual coefficient of the property.
6. A machine learning-based prediction system for the anti-blast response of ultra-high performance concrete slabs, characterized in that: The system is used to implement the machine learning-based method for predicting the blast resistance response of ultra-high performance concrete slabs as described in any one of claims 1-5, specifically including: The data acquisition module is used to conduct N explosion experiments on ultra-high performance concrete slabs and acquire explosion experiment data, including blasting data, pre-blasting data, post-blasting data, and images of the ultra-high performance concrete slabs after blasting. The response analysis module is used to analyze the images of the ultra-high performance concrete slab after blasting, obtain the post-blast image feature data, and combine the post-blast data and the post-blast image feature data to generate blast response data. The data virtual module is used to construct an explosion experiment set by combining the blasting data, pre-blasting data and corresponding blasting response data of the N explosion experiments in step 1; to build a blasting data adversarial network, to train the blasting data adversarial network using the explosion experiment set, and to generate a virtual blasting dataset using the trained blasting data adversarial network. The model building module is used to train a blasting response prediction model by taking blasting data and pre-blasting data from the virtual blasting dataset as input and blasting response data as labels. It obtains the pre-blasting data and blasting data of the current blasting operation and inputs them into the blasting response model to obtain the predicted blasting response data. The comprehensive analysis module is used to comprehensively analyze the pre-blasting data and explosion response data corresponding to the current blasting operation to obtain the property residual coefficient and blasting damage coefficient; and to generate the comprehensive blasting response coefficient based on the property residual coefficient and blasting damage coefficient.