Reinforced concrete detection method and device based on machine learning, and electronic equipment
By using generative adversarial models in eddy current detection technology to expand the data set and training the neural network model, the problem of insufficient data samples in eddy current detection is solved, and the accuracy of the identification of reinforced concrete defects is improved.
Patent Information
- Application Number
- CN202410992570.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-23
- Publication Date
- 2024-11-12
Smart Images

Figure CN118937472A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of industrial detection technology or other related fields, and in particular to a reinforced concrete detection method based on machine learning, a device thereof, and an electronic device. Background Art
[0002] In the field of industrial inspection, eddy current testing technology is a strategy that uses the principle of electromagnetic induction for inspection. It is mainly used to analyze the hardness, density, internal composition, etc. of building materials to find defects. This strategy can detect the deep and subtle differences in building materials by analyzing and comparing the electromagnetic responses of the materials, thereby evaluating their quality. In eddy current testing technology, different forms of coils are usually required to achieve accurate detection. By using coils for detection, conductive materials such as steel and metal products can be inspected. These coils can generate a changing magnetic field. When there are defects or changes in the material, eddy currents will be generated, thereby generating different electromagnetic responses. By analyzing these electromagnetic responses, the defects in the material can be determined.
[0003] In related technologies, it is usually difficult to obtain inspection data for industrial scenarios, so when conducting industrial inspections, we face the problem of small number of data set samples and imbalanced categories. In deep learning processing, the quality and quantity of data are important supports for model training, which will directly affect the performance of the model. Eddy current inspection strategies are limited by limited data sets, especially when it comes to the detection of specific types of defects. The data usually has insufficient quantity or imbalanced categories, which will lead to unstable performance of deep learning models or inability to accurately identify all types of defects.
[0004] To address the above-mentioned problems, no effective solution has been proposed yet. Summary of the invention
[0005] The embodiments of the present invention provide a reinforced concrete detection method and device and electronic device based on machine learning, so as to at least solve the technical problem in the related technology that insufficient data samples lead to the inability to accurately identify defect categories.
[0006] According to one aspect of an embodiment of the present invention, a reinforced concrete detection method based on machine learning is provided, comprising: using electromagnetic eddy current detection equipment to detect intact steel bars and multiple different types of defective steel bars, obtaining waveform data of the intact steel bars and the multiple different types of defective steel bars, and matching the waveform data with corresponding defect category labels; collecting feature data of all waveform data and basic data of all steel bars, and combining the feature data, basic data and defect category labels into original steel bar parameters, wherein the waveform data includes a waveform peak value, a width corresponding to a specified ratio of a waveform peak drop, and a quartic term coefficient and a quadratic term coefficient obtained by fitting data within a specified width range on both sides of the waveform peak value, and all steel bars are matched. The steel bars include intact steel bars and various types of defective steel bars, and the basic data include steel bar diameter and protective layer thickness; the combined original steel bar parameters are input into the generative adversarial model, and the steel bar parameters of different types of defective steel bars are synthesized by the generative adversarial model, and the original steel bar parameters and the steel bar parameters synthesized by the generative adversarial model are combined into an expanded defective steel bar parameter set; the feature data of the intact steel bars and the defective steel bar feature data in the defective steel bar parameter set are input into the neural network model, and the neural network model is trained to obtain a trained neural network model; the steel bar feature data detected in real time is input into the trained neural network model to obtain the basic steel bar data and defect category of the detected steel bar object.
[0007] Optionally, when using electromagnetic eddy current detection equipment to detect multiple different types of defective steel bars, it also includes: using electromagnetic eddy current detection equipment to detect concrete around the steel bars to obtain waveform data of the concrete around the steel bars; merging the waveform data of the concrete around the steel bars into the waveform data of multiple different types of defective steel bars.
[0008] Optionally, the step of collecting characteristic data of all waveform data and basic data of all steel bars includes: extracting data within a specified width range on both sides of the waveform peak from the waveform data, and fitting the data in the range to a first function; collecting the waveform peak value and the width corresponding to a specified proportion of the peak value drop from the waveform data, and combining the waveform peak value, the width corresponding to the specified proportion of the waveform peak value drop, and the quartic term coefficient and the quadratic term coefficient of the first function as characteristic data of the waveform data; collecting the steel bar diameter and protective layer thickness of all intact steel bars and multiple different types of defective steel bars, and combining the steel bar diameter and protective layer thickness as the basic data of all steel bars.
[0009] Optionally, the step of synthesizing the steel bar parameters of different categories of defective steel bars by generating an adversarial model includes: initializing the generator and the discriminator in the generative adversarial model, and setting the network parameters of the generator and the discriminator; fixing the network parameters of the generator, randomly sampling a set of noise vectors from a normal distribution, inputting the noise vectors into the generator to obtain a set of synthetic false data, inputting the synthetic false data and the steel bar parameters of different categories of defective steel bars into the discriminator, obtaining the loss function value and the target optimization function value of the discriminator according to the output result of the discriminator, and updating the network parameters of the discriminator using a gradient descent strategy according to the loss function value and the target optimization function value of the discriminator; fixing the network parameters of the discriminator, randomly sampling a set of noise vectors from a normal distribution, inputting the noise vectors into the generator to obtain a set of new synthetic false data, inputting the new synthetic false data and the steel bar parameters of different categories of defective steel bars into the discriminator, obtaining the loss function value and the target optimization function value of the generator according to the output result of the discriminator, and updating the network parameters of the generator using a gradient descent strategy according to the loss function value and the target optimization function value of the generator; repeating the process of alternating training of the discriminator and the generator until the output result of the generator is the same as the output result of the discriminator.
[0010] Optionally, the step of synthesizing steel bar parameters of different categories of defective steel bars by generating an adversarial model also includes: inputting the output results of the discriminator for true data and the output results for false data into a second function to obtain a loss function of the discriminator, wherein the true data are the steel bar parameters of different categories of defective steel bars; inputting the output results of the discriminator for false data into a third function to obtain a loss function of the generator; and inputting the output results of the discriminator for true data and the output results for false data into a fourth function to obtain a target optimization function.
[0011] Optionally, before training the neural network model, the method includes: normalizing the feature data of intact steel bars and the defective steel bar feature data in the defective steel bar parameter set; initializing the network structure of the neural network model, setting an input layer, multiple hidden layers and an output layer; initializing the network parameters of the neural network model, wherein the network parameters include the number of iterations, the root mean square error rate and the learning rate.
[0012] Optionally, the step of training the neural network model includes: inputting the characteristic data of intact steel bars and the defective steel bar characteristic data in the defective steel bar parameter set into the input layer of the neural network model to obtain the output data of the input layer; inputting the output data of the input layer and the weights between the input layer neurons and the first hidden layer neurons into the first formula to obtain the input data of the first hidden layer; inputting the input data of the first hidden layer into the second formula to obtain the output data of the first hidden layer; a hidden layer iterative output step, inputting the output data of the previous hidden layer and the weights between the previous hidden layer neurons and the next hidden layer neurons into the first formula to obtain the input data of the next hidden layer, and inputting the input data of the next hidden layer into the second formula to obtain the output data of the next hidden layer; repeating the hidden layer iterative output step until the output data of the last hidden layer is obtained; inputting the output data of the last hidden layer and the weights between the last hidden layer neurons and the output layer neurons into the third formula to obtain the input data of the output layer; inputting the input data of the output layer into the activation function to obtain the output data of the output layer, that is, the output data of the neural network model.
[0013] Optionally, the step of training the neural network model also includes: inputting the output data of the output layer and the expected output data into the fourth formula to obtain the root mean square error rate, wherein the expected output data includes the basic data and defect category labels of intact steel bars and defective steel bars; when the root mean square error rate is less than or equal to the preset root mean square error rate, exiting the iterative training to obtain a trained neural network model; when the root mean square error rate is greater than the preset root mean square error rate, calculating the error gradient values of the output layer and all hidden layers, updating the network weights according to the error gradient values, and training the neural network model again according to the new network weights.
[0014] According to another aspect of an embodiment of the present invention, a reinforced concrete detection device based on machine learning is also provided, including: a detection unit, used to use electromagnetic eddy current detection equipment to detect intact steel bars and various different types of defective steel bars, obtain waveform data of intact steel bars and various different types of defective steel bars, and match corresponding defect category labels to the waveform data; an acquisition unit, used to acquire characteristic data of all waveform data and basic data of all steel bars, and combine the characteristic data, basic data and defect category labels into original steel bar parameters, wherein the waveform data includes a waveform peak value, a width corresponding to a specified proportion of a waveform peak drop, and a quartic term coefficient and a quadratic term coefficient obtained by fitting data within a specified width range on both sides of the waveform peak value. All steel bars including intact good steel bars and various different types of defective steel bars, the basic data includes steel bar diameter and protective layer thickness; an expansion unit, which is used to input the combined original steel bar parameters into the generative adversarial model, synthesize the steel bar parameters of different types of defective steel bars through the generative adversarial model, and combine the original steel bar parameters and the steel bar parameters synthesized by the generative adversarial model into an expanded defective steel bar parameter set; a training unit, which is used to input the feature data of the intact steel bars and the defective steel bar feature data in the defective steel bar parameter set into the neural network model, train the neural network model, and obtain a trained neural network model; a real-time acquisition unit, which is used to input the steel bar feature data detected in real time into the trained neural network model, and obtain the basic steel bar data and defect category of the detected steel bar object.
[0015] Optionally, the detection unit also includes: a concrete detection module, which is used to use electromagnetic eddy current detection equipment to detect the concrete around the steel bars to obtain waveform data of the concrete around the steel bars; and a merging module, which is used to merge the waveform data of the concrete around the steel bars into waveform data of multiple different types of defective steel bars.
[0016] Optionally, the acquisition unit includes: a fitting module, used to extract data within a specified width range on both sides of the waveform peak from the waveform data, and fit the data in the range into a first function; a first acquisition module, used to collect the waveform peak value and the width corresponding to a specified proportion of the peak value drop from the waveform data, and combine the waveform peak value, the width corresponding to the specified proportion of the waveform peak value drop, and the fourth-order coefficient and the quadratic coefficient of the first function into characteristic data of the waveform data; a second acquisition module, used to collect the steel bar diameter and protective layer thickness of all intact steel bars and various different types of defective steel bars, and combine the steel bar diameter and protective layer thickness into basic data of all steel bars.
[0017] Optionally, the expansion unit includes: a first initialization module, used to initialize the generator and the discriminator in the generative adversarial model, and set the network parameters of the generator and the discriminator; a discriminator training module, used to fix the network parameters of the generator, randomly sample a set of noise vectors from a normal distribution, input the noise vectors into the generator, obtain a set of synthetic false data, input the synthetic false data and the steel bar parameters of different categories of defective steel bars into the discriminator, obtain the loss function value and the target optimization function value of the discriminator according to the output result of the discriminator, and update the network parameters of the discriminator using the gradient descent strategy according to the loss function value and the target optimization function value of the discriminator. number; a generator training module, which is used to fix the network parameters of the discriminator, randomly sample a set of noise vectors from a normal distribution, input the noise vectors into the generator to obtain a set of new synthetic false data, input the new synthetic false data and the steel bar parameters of different categories of defective steel bars into the discriminator, and obtain the loss function value and the target optimization function value of the generator according to the output result of the discriminator. According to the loss function value and the target optimization function value of the generator, the network parameters of the generator are updated using the gradient descent strategy; an alternating training module, which is used to repeat the process of alternating training of the discriminator and the generator until the output result of the generator is the same as the output result of the discriminator.
[0018] Optionally, the expansion unit also includes: a first loss acquisition module, used to input the output results of the discriminator for true data and the output results for false data into a second function to obtain the loss function of the discriminator, wherein the true data are the steel bar parameters of different categories of defective steel bars; a second loss acquisition module, used to input the output results of the discriminator for false data into a third function to obtain the loss function of the generator; and an objective function acquisition module, used to input the output results of the discriminator for true data and the output results for false data into a fourth function to obtain a target optimization function.
[0019] Optionally, the training unit includes: a normalization module, used to normalize the feature data of intact steel bars and the defective steel bar feature data in the defective steel bar parameter set; a second initialization module, used to initialize the network structure of the neural network model, setting an input layer, multiple hidden layers and an output layer; a third initialization module, used to initialize the network parameters of the neural network model, wherein the network parameters include the number of iterations, the root mean square error rate and the learning rate.
[0020] Optionally, the training unit includes: a first processing module, which is used to input the characteristic data of intact steel bars and the defective steel bar characteristic data in the defective steel bar parameter set into the input layer of the neural network model to obtain the output data of the input layer; a second processing module, which is used to input the output data of the input layer and the weights between the input layer neurons and the first hidden layer neurons into the first formula to obtain the input data of the first hidden layer; a third processing module, which is used to input the input data of the first hidden layer into the second formula to obtain the output data of the first hidden layer; and a fourth processing module, which is used for the hidden layer iterative output step, to input the output data of the previous hidden layer and the weights between the previous hidden layer neurons and the next hidden layer neurons. The weights between neurons in a hidden layer are input into the first formula to obtain input data of the next hidden layer, and the input data of the next hidden layer is input into the second formula to obtain output data of the next hidden layer; the fifth processing module is used to repeatedly execute the hidden layer iterative output step until the output data of the last hidden layer is obtained; the sixth processing module is used to input the output data of the last hidden layer and the weights between the last hidden layer neuron and the output layer neuron into the third formula to obtain input data of the output layer; the seventh processing module is used to input the input data of the output layer into the activation function to obtain output data of the output layer, that is, the output data of the neural network model.
[0021] Optionally, the training unit also includes: an error calculation module, which is used to input the output data of the output layer and the expected output data into the fourth formula to obtain the root mean square error rate, wherein the expected output data includes the basic data and defect category labels of intact steel bars and defective steel bars; a first judgment module, which is used to exit the iterative training when the root mean square error rate is less than or equal to a preset root mean square error rate, and obtain a trained neural network model; a second judgment module, which is used to calculate the error gradient values of the output layer and all hidden layers when the root mean square error rate is greater than the preset root mean square error rate, update the network weights according to the error gradient values, and train the neural network model again according to the new network weights.
[0022] According to another aspect of an embodiment of the present invention, a computer-readable storage medium is also provided, wherein the computer-readable storage medium includes a stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute any one of the above-mentioned reinforced concrete detection methods based on machine learning.
[0023] According to another aspect of an embodiment of the present invention, there is also provided an electronic device, comprising one or more processors and a memory, wherein the memory is used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement any one of the above-mentioned reinforced concrete detection methods based on machine learning.
[0024] According to another aspect of an embodiment of the present invention, a computer program product is provided, including a computer program, which, when executed by a processor, implements the steps of any one of the above-mentioned reinforced concrete detection methods based on machine learning.
[0025] In the present disclosure, electromagnetic eddy current detection equipment is used to detect intact steel bars and various types of defective steel bars, waveform data of the intact steel bars and various types of defective steel bars are obtained, corresponding defect category labels are matched to the waveform data, and then characteristic data of all waveform data and basic data of all steel bars are collected, the characteristic data, basic data and defect category labels are combined into original steel bar parameters, and then the combined original steel bar parameters are input into a generative adversarial model, and steel bar parameters of different types of defective steel bars are synthesized through the generative adversarial model. The original steel bar parameters and the steel bar parameters synthesized by the generative adversarial model are combined into an expanded defective steel bar parameter set, the characteristic data of the intact steel bars and the defective steel bar characteristic data in the defective steel bar parameter set are input into a neural network model, the neural network model is trained to obtain a trained neural network model, and finally the steel bar characteristic data detected in real time is input into the trained neural network model to obtain the steel bar basic data and defect category of the detected steel bar object. The present invention adopts electromagnetic eddy current detection equipment to detect intact steel bars and various types of defective steel bars, inputs the original steel bar parameters into a generative adversarial model, and synthesizes steel bar parameters of more different types of defective steel bars through the generative adversarial model to achieve the effect of expanding the steel bar data set, and then inputs the expanded steel bar data into a neural network model to train the neural network model, which can effectively improve the ability of the neural network model to identify the defect categories of steel bars, thereby solving the technical problem in the related technology that insufficient data samples lead to the inability to accurately identify the defect categories. BRIEF DESCRIPTION OF THE DRAWINGS
[0026] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of this application. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:
[0027] Figure 1 is a flow chart of an optional reinforced concrete detection method based on machine learning according to an embodiment of the present invention;
[0028] Figure 2 is a schematic diagram of an optional reinforced concrete detection method based on machine learning according to an embodiment of the present invention;
[0029] Figure 3 is a schematic diagram of an optional reinforced concrete detection device based on machine learning according to an embodiment of the present invention;
[0030] Figure 4 It is a hardware structure block diagram of an electronic device (or mobile device) of a reinforced concrete detection method based on machine learning according to an embodiment of the present invention. DETAILED DESCRIPTION
[0031] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.
[0032] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.
[0033] It should be noted that the reinforced concrete detection method and device based on machine learning in the present disclosure can be used in the field of industrial detection technology. When reinforced concrete is detected based on machine learning, it can also be used in any field outside the field of industrial detection technology. When reinforced concrete is detected based on machine learning, the present disclosure does not limit the application field of the reinforced concrete detection method and device based on machine learning.
[0034] The following embodiments of the present invention can be applied to various systems / applications / devices for reinforced concrete detection based on machine learning. The present invention detects intact steel bars and various types of defective steel bars by using electromagnetic eddy current detection equipment to obtain waveform data of intact steel bars and various types of defective steel bars, matches the waveform data with corresponding defect category labels, and then collects feature data of all waveform data and basic data of all steel bars, combines the feature data, basic data and defect category labels into original steel bar parameters, and then inputs the combined original steel bar parameters into a generative adversarial model, synthesizes steel bar parameters of different types of defective steel bars through the generative adversarial model, combines the original steel bar parameters and the steel bar parameters synthesized by the generative adversarial model into an expanded defective steel bar parameter set, inputs the feature data of intact steel bars and the defective steel bar feature data in the defective steel bar parameter set into a neural network model, trains the neural network model, obtains a trained neural network model, and finally inputs the steel bar feature data detected in real time into the trained neural network model to obtain the steel bar basic data and defect category of the detected steel bar object. The present invention uses electromagnetic eddy current detection equipment to detect intact steel bars and various types of defective steel bars, inputs the original steel bar parameters into a generative adversarial model, and synthesizes steel bar parameters of more different types of defective steel bars through the generative adversarial model to achieve the effect of expanding the steel bar data set, and then inputs the expanded steel bar data into a neural network model to train the neural network model, which can effectively improve the ability of the neural network model to identify the steel bar defect category, thereby solving the technical problem in the related technology that insufficient data samples lead to the inability to accurately identify the defect category.
[0035] The present invention is described in detail below in conjunction with various embodiments.
[0036] Embodiment 1
[0037] According to an embodiment of the present invention, an embodiment of a reinforced concrete detection method based on machine learning is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.
[0038] Figure 1 is a flow chart of an optional reinforced concrete detection method based on machine learning according to an embodiment of the present invention, such as Figure 1 As shown, the method comprises the following steps:
[0039] Step S101, using electromagnetic eddy current testing equipment to test intact steel bars and various types of defective steel bars, obtaining waveform data of the intact steel bars and various types of defective steel bars, and matching corresponding defect category labels to the waveform data.
[0040] Step S102, collecting characteristic data of all waveform data and basic data of all steel bars, and combining the characteristic data, basic data and defect category labels into original steel bar parameters, wherein the waveform data includes the waveform peak value, the width corresponding to the specified proportion of the waveform peak value drop, and the quartic term coefficient and quadratic term coefficient obtained by fitting the data within the specified width range on both sides of the waveform peak value. All steel bars include intact steel bars and various types of defective steel bars, and the basic data include the steel bar diameter and the protective layer thickness.
[0041] Step S103, inputting the combined original steel bar parameters into the generative adversarial model, synthesizing the steel bar parameters of different categories of defective steel bars through the generative adversarial model, and combining the original steel bar parameters and the steel bar parameters synthesized by the generative adversarial model into an expanded defective steel bar parameter set.
[0042] Step S104, inputting the characteristic data of the intact steel bars and the defective steel bar characteristic data in the defective steel bar parameter set into the neural network model, training the neural network model, and obtaining a trained neural network model.
[0043] Step S105, inputting the steel bar feature data detected in real time into the trained neural network model to obtain the steel bar basic data and defect category of the detected steel bar object.
[0044] The steps provided by the above-mentioned reinforced concrete detection method based on machine learning are to detect intact steel bars and various types of defective steel bars by using electromagnetic eddy current detection equipment to obtain waveform data of intact steel bars and various types of defective steel bars, match corresponding defect category labels to the waveform data, and then collect feature data of all waveform data and basic data of all steel bars, combine the feature data, basic data and defect category labels into original steel bar parameters, and then input the combined original steel bar parameters into a generative adversarial model, synthesize steel bar parameters of different types of defective steel bars through the generative adversarial model, combine the original steel bar parameters and the steel bar parameters synthesized by the generative adversarial model into an expanded defective steel bar parameter set, input the feature data of intact steel bars and the defective steel bar feature data in the defective steel bar parameter set into a neural network model, train the neural network model to obtain a trained neural network model, and finally input the steel bar feature data detected in real time into the trained neural network model to obtain the steel bar basic data and defect category of the detected steel bar object. The embodiment of the present invention uses electromagnetic eddy current detection equipment to detect intact steel bars and various types of defective steel bars, inputs the original steel bar parameters into a generative adversarial model, and synthesizes steel bar parameters of more different types of defective steel bars through the generative adversarial model to achieve the effect of expanding the steel bar data set. Then, by inputting the expanded steel bar data into a neural network model and training the neural network model, the ability of the neural network model to identify the steel bar defect category can be effectively improved, thereby solving the technical problem in the related technology that insufficient data samples lead to the inability to accurately identify the defect category.
[0045] The embodiment of the present invention is described in detail below in combination with the above steps.
[0046] Step S101, using electromagnetic eddy current testing equipment to test intact steel bars and various types of defective steel bars, obtaining waveform data of the intact steel bars and various types of defective steel bars, and matching corresponding defect category labels to the waveform data.
[0047] The electromagnetic eddy current detection device is a device used to detect surface and near-surface defects of metal parts, which can detect various defects such as cracks, corrosion, holes, etc. in metal parts. This detection device is based on the principle of electromagnetics and realizes the detection of defects by generating eddy currents on the object to be detected. When detecting steel bars, the coil in the electromagnetic eddy current detection device will generate an alternating magnetic field. When the magnetic field passes through the steel bars, eddy currents will be induced inside the steel bars. The generation of eddy currents is related to the electrical conductivity, magnetic permeability, shape and size of the steel bars. When there are defects inside the steel bars, the distribution of eddy currents will be affected, thereby changing the distribution of the magnetic field. The sensor in the electromagnetic eddy current detection device can detect the changes in the magnetic field and convert these changes into electrical signals. The electrical signals are in the shape of waves and are a kind of waveform data. In an embodiment of the present invention, by using the electromagnetic eddy current device to detect intact steel bars and various types of defective steel bars, waveform data of intact steel bars and various types of defective steel bars can be obtained. By analyzing these waveform data, it can be determined whether the steel bars have defects and the location and size of the defects. In order to expand the data set of steel bars, an embodiment of the present invention matches corresponding defect category labels for all detected waveform data. Subsequently, the defect category labels need to be input into a generative adversarial model together with the steel bar feature data and basic data.
[0048] Optionally, when using electromagnetic eddy current detection equipment to detect multiple different types of defective steel bars, it also includes: using electromagnetic eddy current detection equipment to detect concrete around the steel bars to obtain waveform data of the concrete around the steel bars; merging the waveform data of the concrete around the steel bars into the waveform data of multiple different types of defective steel bars.
[0049] In the embodiment of the present invention, when electromagnetic eddy current detection equipment is used to detect reinforced concrete, the reinforced concrete may contain defects such as exposed steel bars, holes, and steel bar corrosion. In addition to the different degrees of steel bar corrosion causing different detected waveform data, when the steel bars are intact, holes, which are defects involving concrete properties, will also cause different waveform data detected by the electromagnetic eddy current detection equipment. Therefore, the embodiment of the present invention not only involves defects in steel bars, but also involves defects in concrete around steel bars. By using electromagnetic eddy current detection equipment to detect concrete around steel bars, waveform data of concrete around steel bars is obtained, and the waveform data of concrete around steel bars is merged into waveform data of multiple different types of defective steel bars.
[0050] Step S102, collecting characteristic data of all waveform data and basic data of all steel bars, and combining the characteristic data, basic data and defect category labels into original steel bar parameters, wherein the waveform data includes the waveform peak value, the width corresponding to the specified proportion of the waveform peak value drop, and the quartic term coefficient and quadratic term coefficient obtained by fitting the data within the specified width range on both sides of the waveform peak value. All steel bars include intact steel bars and various types of defective steel bars, and the basic data include the steel bar diameter and the protective layer thickness.
[0051] In an embodiment of the present invention, in order to expand the data set of steel bars, it is necessary to collect the characteristic data of steel bar waveform data and the basic data of steel bars. The steel bar characteristic data includes the peak value of steel bar waveform data, the width corresponding to the specified ratio of the waveform peak value drop, and the fourth-order coefficient and the quadratic coefficient obtained by fitting the data within the specified width range on both sides of the waveform peak value. Among them, the specified ratio and the specified width are determined according to the application scenario of the embodiment. In the embodiment of the present invention, the specified ratio is set to 10%, and the specified width is set to 80 mm. The basic data of steel bars includes the diameter of the steel bar and the thickness of the protective layer. The embodiment of the present invention combines the collected steel bar characteristic data and the basic data of steel bars together with the defect category label matched to the waveform data into the original steel bar data.
[0052] Optionally, the step of collecting characteristic data of all waveform data and basic data of all steel bars includes: extracting data within a specified width range on both sides of the waveform peak from the waveform data, and fitting the data in the range to a first function; collecting the waveform peak value and the width corresponding to a specified proportion of the peak value drop from the waveform data, and combining the waveform peak value, the width corresponding to the specified proportion of the waveform peak value drop, and the quartic term coefficient and the quadratic term coefficient of the first function as characteristic data of the waveform data; collecting the steel bar diameter and protective layer thickness of all intact steel bars and multiple different types of defective steel bars, and combining the steel bar diameter and protective layer thickness as the basic data of all steel bars.
[0053] In the embodiment of the present invention, when collecting characteristic data from waveform data, it is first necessary to extract data within a specified width range on both sides of the waveform peak, and fit the data within the range to a first function f(x), the first function being:
[0054] f(x)=Ax 4 +Bx 2 +C
[0055] In the first function, A is the coefficient of the quartic term, B is the coefficient of the quadratic term, C is the constant term, and x is the model output data within the specified width range on both sides of the peak value of the waveform data. The embodiment of the present invention adopts a data fitting strategy, and by establishing the mathematical model of the first function, the output data of the model is made as equal as possible to the data within the specified width range on both sides of the peak value of the steel bar waveform data. After obtaining the fitting function, the embodiment of the present invention extracts the quartic term coefficient A and the quadratic term coefficient B of the first function, together with the peak value of the steel bar waveform data and the width corresponding to the specified ratio of the peak value drop as the characteristic data of the steel bar.
[0056] In the embodiment of the present invention, it is also necessary to obtain the steel bar diameters and protective layer thicknesses of all intact steel bars and various types of defective steel bars, and combine the steel bar diameters and protective layer thicknesses into basic data for all steel bars.
[0057] Step S103, inputting the combined original steel bar parameters into the generative adversarial model, synthesizing the steel bar parameters of different categories of defective steel bars through the generative adversarial model, and combining the original steel bar parameters and the steel bar parameters synthesized by the generative adversarial model into an expanded defective steel bar parameter set.
[0058] The generative adversarial model is a deep learning architecture that mainly includes two main network structures: the generator and the discriminator. The goal of the generator is to generate fake data that is as close to the real data as possible. It usually uses a random noise vector as input and generates data samples through a neural network. The task of the discriminator is to distinguish between the fake data generated by the generator and the real data. It is responsible for receiving input data and trying to determine whether the data is real data or generated data.
[0059] In an embodiment of the present invention, the combined original steel bar parameters are input into a generative adversarial model, and the steel bar parameters of different categories of defective steel bars are synthesized by the generative adversarial model. The original steel bar parameters are real data, and the synthesized steel bar parameters are false data. In the training process of the generative adversarial network, the discriminator and the generator confront each other until a balance point is reached. At this balance point, the data generated by the generator is realistic enough that the discriminator cannot effectively distinguish between real data and generated data. In an embodiment of the present invention, the steel bar parameters of different categories of defective steel bars are synthesized by a generative adversarial model. The synthesized steel bar parameters can be mistaken for the real ones, so that the discriminator cannot distinguish between real data and generated data. In an embodiment of the present invention, the original steel bar parameters and the steel bar parameters synthesized by the generative adversarial model are combined into an expanded defective steel bar parameter set, and the expanded defective steel bar parameter set will be used for the training of subsequent neural network models.
[0060] Optionally, the step of synthesizing the steel bar parameters of different categories of defective steel bars by generating an adversarial model includes: initializing the generator and the discriminator in the generative adversarial model, and setting the network parameters of the generator and the discriminator; fixing the network parameters of the generator, randomly sampling a set of noise vectors from a normal distribution, inputting the noise vectors into the generator to obtain a set of synthetic false data, inputting the synthetic false data and the steel bar parameters of different categories of defective steel bars into the discriminator, obtaining the loss function value and the target optimization function value of the discriminator according to the output result of the discriminator, and updating the network parameters of the discriminator using a gradient descent strategy according to the loss function value and the target optimization function value of the discriminator; fixing the network parameters of the discriminator, randomly sampling a set of noise vectors from a normal distribution, inputting the noise vectors into the generator to obtain a set of new synthetic false data, inputting the new synthetic false data and the steel bar parameters of different categories of defective steel bars into the discriminator, obtaining the loss function value and the target optimization function value of the generator according to the output result of the discriminator, and updating the network parameters of the generator using a gradient descent strategy according to the loss function value and the target optimization function value of the generator; repeating the process of alternating training of the discriminator and the generator until the output result of the generator is the same as the output result of the discriminator.
[0061] In an embodiment of the present invention, when training a generative adversarial model, it is first necessary to initialize the generator and discriminator in the generative adversarial model, set the network parameters of the generator and the discriminator, and then fix the network parameters of the generator, and first train the discriminator. The discriminator will be iteratively trained multiple times. During each iterative training process, a set of noise vectors will be randomly sampled from a normal distribution, and the noise vectors will be input into the generator to obtain a set of synthetic false data, and then the synthetic false data and the steel bar parameters of different categories of defective steel bars will be input into the discriminator, and the discriminator will determine whether the input data is real data or synthetic data, and the loss function value and target optimization function value of the discriminator will be obtained according to the output result of the discriminator. According to the loss function value and target optimization function value of the discriminator, the network parameters of the discriminator are updated using a gradient descent strategy, and the gradient descent strategy gradually finds the minimum value of the loss function by moving along the negative direction of the gradient of the loss function to obtain better network parameters, and a discriminator with better performance can be obtained at this time.
[0062] After training to obtain a discriminator with better performance, the network parameters of the discriminator are fixed and the discriminator is used to train the generator. The generator usually only needs to iterate once, by inputting a set of random noise vectors into the generator to obtain a new set of false data, inputting the false data into the discriminator, and obtaining the loss function value and target optimization function value of the generator based on the output of the discriminator. Based on the loss function value and target optimization function value of the generator, the network parameters of the generator are updated using the gradient descent strategy, and a generator with better performance is obtained.
[0063] The embodiment of the present invention will alternately train the discriminator and the generator multiple times until the output results of the discriminator and the generator are the same, that is, the discriminator and the generator reach a balance point. At this balance point, the data generated by the generator is realistic enough that the discriminator cannot effectively distinguish between real data and generated data.
[0064] Optionally, the step of synthesizing steel bar parameters of different categories of defective steel bars by generating an adversarial model also includes: inputting the output results of the discriminator for true data and the output results for false data into a second function to obtain a loss function of the discriminator, wherein the true data are the steel bar parameters of different categories of defective steel bars; inputting the output results of the discriminator for false data into a third function to obtain a loss function of the generator; and inputting the output results of the discriminator for true data and the output results for false data into a fourth function to obtain a target optimization function.
[0065] In the embodiment of the present invention, the loss function of the discriminator is represented by a second function, which is:
[0066] Loss D =-(E x∈q(x) [logD(x)]+E z∈p(z) log[1-D(G(z))])
[0067] In the second function, D(x) is the output result of the discriminator for true data, D(G(z)) is the output result of the discriminator for false data, x is the steel bar parameter of different categories of defective steel bars, and z is the noise vector.
[0068] The loss function of the generator is expressed by the third function, which is:
[0069] Loss G =-E z∈p(z) log[1-D(G(z))]
[0070] In the third function, D(G(z)) is the output result of the discriminator for false data, and z is the noise vector.
[0071] The target optimization function is represented by the fourth function, which is:
[0072] V(G,D)=m G inm D axE x∈q(x) [logD(x)]+E z∈p(z) log[1-D(G(z))]
[0073] In the fourth function, D(x) is the output result of the discriminator for true data, D(G(z)) is the output result of the discriminator for false data, x is the steel bar parameter of different categories of defective steel bars, and z is the noise vector.
[0074] When the discriminator is trained, the embodiment of the present invention calculates the loss function of the discriminator through the second function and the target optimization function through the fourth function, and updates the network parameters of the discriminator according to the loss value of the discriminator and the value of the target optimization function. When the generator is trained, the loss function of the generator is calculated through the third function and the target optimization function is calculated through the fourth function, and updates the network parameters of the generator according to the loss value of the generator and the value of the target optimization function, and finally obtains the optimal generative adversarial model.
[0075] Step S104, inputting the characteristic data of the intact steel bars and the defective steel bar characteristic data in the defective steel bar parameter set into the neural network model, training the neural network model, and obtaining a trained neural network model.
[0076] In an embodiment of the present invention, the neural network model is trained by inputting the characteristic data of intact steel bars and the defective steel bar characteristic data in the defective steel bar parameter set into the neural network model, so that the neural network acquires the ability to identify whether the steel bar is defective and the type of steel bar defect. The embodiment of the present invention adopts a commonly used back propagation neural network, which is composed of an input layer, multiple hidden layers and an output layer. The training process of the neural network model mainly includes two stages. The first stage is the forward transmission of the steel bar characteristic data. The steel bar characteristic data passes through multiple hidden layers from the input layer and finally reaches the output layer to obtain the output data of the output layer. The second stage is the back propagation of the error. The error is calculated according to the output data of the output layer. The error passes through multiple hidden layers from the output layer and finally reaches the input layer. The network weights from the last hidden layer to the output layer, the network weights from the previous hidden layer to the next hidden layer, and the network weights from the input layer to the first hidden layer are adjusted in turn. After multiple iterations of training, a trained neural network model is obtained.
[0077] Optionally, before training the neural network model, the method includes: normalizing the feature data of intact steel bars and the defective steel bar feature data in the defective steel bar parameter set; initializing the network structure of the neural network model, setting an input layer, multiple hidden layers and an output layer; initializing the network parameters of the neural network model, wherein the network parameters include the number of iterations, the root mean square error rate and the learning rate.
[0078] In an embodiment of the present invention, before training the neural network model, the neural network model needs to be initialized. First, the characteristic data of the intact steel bars and the defective steel bar characteristic data in the defective steel bar parameter set are normalized, and then the network structure of the neural network model is set, an input layer, multiple hidden layers and an output layer are set, and the number of hidden layers is usually set to 2n+1, where n is the number of characteristic parameters of the input layer. In an embodiment of the present invention, the characteristic parameters of the input layer are the waveform peak, the width corresponding to the specified ratio of the waveform peak drop, and the quartic term coefficient and quadratic term coefficient obtained by fitting the data within the specified width range on both sides of the waveform peak. That is to say, the characteristic parameters of the input layer are four, that is, the value of n is 4, so the embodiment of the present invention sets 9 hidden layers. After determining the network structure of the neural network, it is also necessary to initialize the network parameters, including setting the number of iterations, the root mean square error rate and the learning rate, where the number of iterations determines the number of training times of the neural network model, the root mean square error rate will be used to calculate the network error in back propagation, and the learning rate is used to obtain the error gradient. The neural network model will update the network weight according to the error gradient.
[0079] Optionally, the step of training the neural network model includes: inputting the characteristic data of intact steel bars and the defective steel bar characteristic data in the defective steel bar parameter set into the input layer of the neural network model to obtain the output data of the input layer; inputting the output data of the input layer and the weights between the input layer neurons and the first hidden layer neurons into the first formula to obtain the input data of the first hidden layer; inputting the input data of the first hidden layer into the second formula to obtain the output data of the first hidden layer; a hidden layer iterative output step, inputting the output data of the previous hidden layer and the weights between the previous hidden layer neurons and the next hidden layer neurons into the first formula to obtain the input data of the next hidden layer, and inputting the input data of the next hidden layer into the second formula to obtain the output data of the next hidden layer; repeating the hidden layer iterative output step until the output data of the last hidden layer is obtained; inputting the output data of the last hidden layer and the weights between the last hidden layer neurons and the output layer neurons into the third formula to obtain the input data of the output layer; inputting the input data of the output layer into the activation function to obtain the output data of the output layer, that is, the output data of the neural network model.
[0080] The training of the neural network model includes two stages: forward propagation and back propagation. In the process of forward propagation, the characteristic data of intact steel bars and the defective steel bar characteristic data in the defective steel bar parameter set are first input into the input layer of the neural network model to obtain the output data of the input layer. Then, the output data of the input layer and the weights between the input layer neurons and the first hidden layer neurons are input into the first formula to obtain the input data of the first hidden layer. The first formula is:
[0081]
[0082] In the first formula, net j is the input data of the jth neuron in the first hidden layer, X i is the output data of the i-th neuron in the input layer, V ij and the weight between the i-th neuron in the input layer and the j-th neuron in the first hidden layer, n is the total number of neurons in the input layer, and p is the total number of neurons in the first hidden layer.
[0083] After obtaining the input data of the first hidden layer, the input data of the first hidden layer is input into the second formula to obtain the output data of the first hidden layer. The second formula is:
[0084] y j =f(net j ); j = 1, 2, ..., p
[0085] In the second formula, y j is the output data of the jth neuron in the first hidden layer, net j is the input data of the jth neuron in the first hidden layer, p is the total number of neurons in the first hidden layer, and the f function is an activation function. In the embodiment of the present invention, the activation function is set to the sigmoid function, that is:
[0086]
[0087] In the sigmoid activation function, x is the input data. In this step, x is the input data of the jth neuron in the first hidden layer. j .
[0088] After obtaining the output data of the first hidden layer, the output data needs to pass through the subsequent eight hidden layers according to the same steps from the input layer to the first hidden layer. That is to say, the hidden layer iterative output step is to input the output data of the previous hidden layer and the weights between the neurons of the previous hidden layer and the neurons of the next hidden layer into the first formula to obtain the input data of the next hidden layer, input the input data of the next hidden layer into the second formula to obtain the output data of the next hidden layer, and repeat the hidden layer iterative output step until the output data of the last hidden layer is obtained.
[0089] After obtaining the output data of the last hidden layer, the output data of the last hidden layer and the weights between the neurons of the last hidden layer and the neurons of the output layer are input into the third formula to obtain the input data of the output layer, where the third formula is:
[0090]
[0091] In the third formula, net k is the input data of the kth neuron in the output layer, w jk is the weight between the jth neuron in the last hidden layer and the kth neuron in the output layer, Y j is the output data of the jth neuron in the last hidden layer, p is the total number of neurons in the last hidden layer, and m is the total number of neurons in the output layer.
[0092] Finally, the input data of the output layer net k Input into the activation function to obtain the output data of the output layer, that is, the output data of the neural network model, which is the output result of the forward propagation of the neural network.
[0093] Optionally, the step of training the neural network model also includes: inputting the output data of the output layer and the expected output data into the fourth formula to obtain the root mean square error rate, wherein the expected output data includes the basic data and defect category labels of intact steel bars and defective steel bars; when the root mean square error rate is less than or equal to the preset root mean square error rate, exiting the iterative training to obtain a trained neural network model; when the root mean square error rate is greater than the preset root mean square error rate, calculating the error gradient values of the output layer and all hidden layers, updating the network weights according to the error gradient values, and training the neural network model again according to the new network weights.
[0094] The training of the neural network model also includes a second stage, namely the error back propagation stage. After obtaining the output data of the forward propagation, the output data of the output layer and the expected output data are input into the fourth formula to obtain the root mean square error rate E. The fourth formula is:
[0095]
[0096] In the fourth formula, Z is the output vector of the output layer, and d is the expected output vector. The expected output data includes the basic data of intact steel bars and defective steel bars and the defect category labels. Since both Z and d are vectors, the calculation of the root mean square error rate can be simplified to calculating the root mean square of each corresponding element in the Z and d vectors. In the fourth formula, m is the number of elements in the vector, and d is k is the kth element in the expected output vector, z k is the kth element in the output vector of the output layer.
[0097] After obtaining the root mean square error rate, it is necessary to perform a threshold judgment on the root mean square error rate. The embodiment of the present invention sets an initial value of the root mean square error rate when initializing the neural network model, that is, the preset root mean square error rate. When the root mean square error rate is less than or equal to the preset root mean square error rate, the iterative training is exited to obtain a trained neural network model. When the root mean square error rate is greater than the preset root mean square error rate, it is necessary to calculate the error gradient value of the output layer and all hidden layers, wherein the error gradient value of the output layer is calculated by the fifth formula, and the fifth formula is:
[0098]
[0099] In the fifth formula, Δw jk is the error gradient between the jth neuron in the last hidden layer and the kth neuron in the output layer, E is the root mean square error rate, and w jk is the weight between the jth neuron in the last hidden layer and the kth neuron in the output layer, p is the total number of neurons in the last hidden layer, m is the total number of neurons in the output layer, and η is the preset learning rate.
[0100] The error gradient value of the hidden layer is calculated by the sixth formula, which is:
[0101]
[0102] In the sixth formula, Δv jk is the error gradient between the i-th neuron in the previous hidden layer and the j-th neuron in the next hidden layer, E is the root mean square error rate, and v jk is the weight between the i-th neuron in the previous hidden layer and the j-th neuron in the next hidden layer, l is the total number of neurons in the previous hidden layer, p is the total number of neurons in the next hidden layer, and η is the preset learning rate.
[0103] After obtaining the error gradient values of the output layer and all hidden layers, the network weights are updated according to the error gradient values, where the hidden layer updates the network parameters according to the seventh formula, which is:
[0104] W jk =(n+1)=W jk (n)+ΔW jk
[0105] In the seventh formula, Δw jk is the error gradient between the jth neuron in the last hidden layer and the kth neuron in the output layer, w jk is the weight between the jth neuron in the last hidden layer and the kth neuron in the output layer, and n is the current number of iterations.
[0106] The output layer updates the network parameters according to the eighth formula, which is:
[0107] V jk (n+1)=V jk (n)+ΔV jk
[0108] In the eighth formula, Δv jk is the error gradient between the i-th neuron in the previous hidden layer and the j-th neuron in the next hidden layer, v jk is the weight between the i-th neuron in the previous hidden layer and the j-th neuron in the next hidden layer, and n is the current number of iterations.
[0109] The back propagation of the error calculates the error based on the output data of the output layer. The error passes through multiple hidden layers from the output layer and finally reaches the input layer. The network parameters between layers are adjusted in turn. The network parameters include the network weights between the last hidden layer and the output layer, the network weights between the previous hidden layer and the next hidden layer, and the network weights between the input layer and the first hidden layer. After meeting the preset number of iterations, the trained neural network model is obtained.
[0110] Step S105, inputting the steel bar feature data detected in real time into the trained neural network model to obtain the steel bar basic data and defect category of the detected steel bar object.
[0111] In an embodiment of the present invention, after obtaining a trained neural network model, the neural network model can be used in an actual detection scenario. By using electromagnetic eddy current detection equipment to detect actual steel bars, the steel bar feature data detected in real time is input into the trained neural network model, and the basic steel bar data and defect category of the detected steel bar object can be obtained.
[0112] Figure 2 is a schematic diagram of an optional reinforced concrete detection method based on machine learning according to an embodiment of the present invention, Figure 2 In the process, the detection data of intact steel bars and various different types of defective steel bars in reinforced concrete are input into the generative adversarial model after data preprocessing. The generative adversarial model synthesizes more detection data of defective steel bars of different categories. Then, the detection data of intact steel bars and various different types of defective steel bars are divided into training samples, verification samples and test samples. The reinforced concrete steel bar defect detection neural network model is trained, tested and verified, and finally a mature defect detection neural network model is obtained. Then, the real-time detection data is input into the mature defect detection neural network model to obtain the defect category of the detected steel bar object.
[0113] The embodiment of the present invention synthesizes defect data through a generative adversarial network model to achieve the effect of expanding the steel bar defect data set. By introducing deep learning and data enhancement strategies, the problems of insufficient data set samples and imbalanced categories can be more effectively handled. The embodiment of the present invention synthesizes defect data through a generative adversarial network model and then trains the neural network model, which can effectively expand the data set, improve the performance of the neural network model and the accuracy of identifying defect types, thereby solving the technical problem in the related technology that insufficient data samples lead to the inability to accurately identify defect categories.
[0114] The following is a detailed description in conjunction with another embodiment.
[0115] Embodiment 2
[0116] A reinforced concrete detection device based on machine learning provided in this embodiment includes multiple implementation units, each implementation unit corresponds to each implementation step in the above-mentioned embodiment one.
[0117] Figure 3 is a schematic diagram of an optional reinforced concrete detection device based on machine learning according to an embodiment of the present invention, such as Figure 3 As shown, the reinforced concrete detection device based on machine learning may include: a detection unit 30, a collection unit 31, an expansion unit 32, a training unit 33, and a real-time acquisition unit 34.
[0118] Among them, the detection unit 30 is used to use electromagnetic eddy current detection equipment to detect intact steel bars and various types of defective steel bars, obtain waveform data of intact steel bars and various types of defective steel bars, and match corresponding defect category labels for the waveform data; the acquisition unit 31 is used to collect characteristic data of all waveform data and basic data of all steel bars, and combine the characteristic data, basic data and defect category labels into original steel bar parameters, wherein the waveform data includes waveform peak value, width corresponding to a specified proportion of waveform peak drop, and quartic term coefficients and quadratic term coefficients obtained by fitting data within a specified width range on both sides of the waveform peak. All steel bars include intact steel bars and various types of defective steel bars, and basic data The data include the steel bar diameter and the protective layer thickness; an expansion unit 32, which is used to input the combined original steel bar parameters into the generative adversarial model, synthesize the steel bar parameters of different categories of defective steel bars through the generative adversarial model, and combine the original steel bar parameters and the steel bar parameters synthesized by the generative adversarial model into an expanded defective steel bar parameter set; a training unit 33, which is used to input the feature data of the intact steel bars and the defective steel bar feature data in the defective steel bar parameter set into the neural network model, train the neural network model, and obtain a trained neural network model; a real-time acquisition unit 34, which is used to input the steel bar feature data detected in real time into the trained neural network model, and obtain the basic steel bar data and defect category of the detected steel bar object.
[0119] Optionally, the detection unit 30 also includes: a concrete detection module, which is used to use electromagnetic eddy current detection equipment to detect the concrete around the steel bars to obtain waveform data of the concrete around the steel bars; and a merging module, which is used to merge the waveform data of the concrete around the steel bars into waveform data of multiple different types of defective steel bars.
[0120] Optionally, the acquisition unit 31 includes: a fitting module, which is used to extract data within a specified width range on both sides of the waveform peak from the waveform data, and fit the data in the range into a first function; a first acquisition module, which is used to collect the waveform peak value and the width corresponding to the specified proportion of the peak value drop from the waveform data, and combine the waveform peak value, the width corresponding to the specified proportion of the waveform peak value drop, and the fourth-order coefficient and the quadratic coefficient of the first function into characteristic data of the waveform data; a second acquisition module, which is used to collect the steel bar diameter and protective layer thickness of all intact steel bars and various different types of defective steel bars, and combine the steel bar diameter and protective layer thickness into basic data of all steel bars.
[0121] Optionally, the expansion unit 32 includes: a first initialization module, which is used to initialize the generator and the discriminator in the generative adversarial model and set the network parameters of the generator and the discriminator; a discriminator training module, which is used to fix the network parameters of the generator, randomly sample a set of noise vectors from a normal distribution, input the noise vectors into the generator, obtain a set of synthetic false data, input the synthetic false data and the steel bar parameters of different categories of defective steel bars into the discriminator, obtain the loss function value and the target optimization function value of the discriminator according to the output result of the discriminator, and update the network of the discriminator using the gradient descent strategy according to the loss function value and the target optimization function value of the discriminator. Parameters; a generator training module, which is used to fix the network parameters of the discriminator, randomly sample a set of noise vectors from a normal distribution, input the noise vectors into the generator to obtain a new set of synthetic false data, input the new synthetic false data and the steel bar parameters of different categories of defective steel bars into the discriminator, and obtain the loss function value and the target optimization function value of the generator according to the output result of the discriminator. According to the loss function value and the target optimization function value of the generator, the network parameters of the generator are updated using the gradient descent strategy; an alternating training module, which is used to repeat the process of alternating training of the discriminator and the generator until the output result of the generator is the same as the output result of the discriminator.
[0122] Optionally, the expansion unit 32 also includes: a first loss acquisition module, used to input the output results of the discriminator for true data and the output results for false data into a second function to obtain the loss function of the discriminator, wherein the true data are the steel bar parameters of different categories of defective steel bars; a second loss acquisition module, used to input the output results of the discriminator for false data into a third function to obtain the loss function of the generator; and an objective function acquisition module, used to input the output results of the discriminator for true data and the output results for false data into a fourth function to obtain a target optimization function.
[0123] Optionally, the training unit 33 includes: a normalization module, used to normalize the feature data of intact steel bars and the defective steel bar feature data in the defective steel bar parameter set; a second initialization module, used to initialize the network structure of the neural network model, setting an input layer, multiple hidden layers and an output layer; a third initialization module, used to initialize the network parameters of the neural network model, wherein the network parameters include the number of iterations, the root mean square error rate and the learning rate.
[0124] Optionally, the training unit 33 includes: a first processing module, which is used to input the characteristic data of the intact steel bars and the defective steel bar characteristic data in the defective steel bar parameter set into the input layer of the neural network model to obtain the output data of the input layer; a second processing module, which is used to input the output data of the input layer and the weights between the input layer neurons and the first hidden layer neurons into the first formula to obtain the input data of the first hidden layer; a third processing module, which is used to input the input data of the first hidden layer into the second formula to obtain the output data of the first hidden layer; and a fourth processing module, which is used for the hidden layer iterative output step, to input the output data of the previous hidden layer and the weights between the previous hidden layer neurons and the first hidden layer neurons into the first formula to obtain the output data of the first hidden layer. The weights between the neurons of the next hidden layer are input into the first formula to obtain the input data of the next hidden layer, and the input data of the next hidden layer is input into the second formula to obtain the output data of the next hidden layer; the fifth processing module is used to repeatedly execute the hidden layer iterative output step until the output data of the last hidden layer is obtained; the sixth processing module is used to input the output data of the last hidden layer and the weights between the neurons of the last hidden layer and the neurons of the output layer into the third formula to obtain the input data of the output layer; the seventh processing module is used to input the input data of the output layer into the activation function to obtain the output data of the output layer, that is, the output data of the neural network model.
[0125] Optionally, the training unit 33 also includes: an error calculation module, which is used to input the output data of the output layer and the expected output data into the fourth formula to obtain the root mean square error rate, wherein the expected output data includes the basic data and defect category labels of intact steel bars and defective steel bars; a first judgment module, which is used to exit the iterative training when the root mean square error rate is less than or equal to a preset root mean square error rate to obtain a trained neural network model; a second judgment module, which is used to calculate the error gradient values of the output layer and all hidden layers when the root mean square error rate is greater than the preset root mean square error rate, update the network weights according to the error gradient values, and train the neural network model again according to the new network weights.
[0126] The above-mentioned reinforced concrete detection device based on machine learning can detect intact steel bars and various types of defective steel bars by using electromagnetic eddy current detection equipment through the detection unit 30, obtain waveform data of intact steel bars and various types of defective steel bars, match corresponding defect category labels for the waveform data, and then collect feature data of all waveform data and basic data of all steel bars through the acquisition unit 31, combine the feature data, basic data and defect category labels into original steel bar parameters, and then input the combined original steel bar parameters into the generative adversarial model through the expansion unit 32, synthesize steel bar parameters of different types of defective steel bars through the generative adversarial model, combine the original steel bar parameters and the steel bar parameters synthesized by the generative adversarial model into an expanded defective steel bar parameter set, input the feature data of intact steel bars and the defective steel bar feature data in the defective steel bar parameter set into the neural network model through the training unit 33, train the neural network model to obtain a trained neural network model, and finally input the steel bar feature data detected in real time into the trained neural network model through the real-time acquisition unit 34 to obtain the steel bar basic data and defect category of the detected steel bar object. The embodiment of the present invention uses electromagnetic eddy current detection equipment to detect intact steel bars and various types of defective steel bars, inputs the original steel bar parameters into a generative adversarial model, and synthesizes steel bar parameters of more different types of defective steel bars through the generative adversarial model to achieve the effect of expanding the steel bar data set. Then, by inputting the expanded steel bar data into a neural network model and training the neural network model, the ability of the neural network model to identify the steel bar defect category can be effectively improved, thereby solving the technical problem in the related technology that insufficient data samples lead to the inability to accurately identify the defect category.
[0127] The above-mentioned reinforced concrete detection device based on machine learning can also include a processor and a memory. The above-mentioned detection unit 30, acquisition unit 31, expansion unit 32, training unit 33, real-time acquisition unit 34, etc. are all stored in the memory as program units, and the processor executes the above-mentioned program units stored in the memory to realize corresponding functions.
[0128] The processor includes a kernel, which retrieves the corresponding program unit from the memory. One or more kernels can be set, and the defect type of reinforced concrete can be identified by adjusting kernel parameters.
[0129] The above-mentioned memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM, and the memory includes at least one storage chip.
[0130] According to another aspect of an embodiment of the present invention, a computer-readable storage medium is further provided, the computer-readable storage medium including a stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute any one of the machine learning-based reinforced concrete detection methods in the above-mentioned embodiment 1.
[0131] According to another aspect of an embodiment of the present invention, there is also provided an electronic device, comprising one or more processors and a memory, wherein the memory is used to store one or more programs, wherein when the one or more programs are executed by one or more processors, the one or more processors implement the reinforced concrete detection method based on machine learning of any one of the above-mentioned embodiments one.
[0132] The present application also provides a computer program product, including a computer program, which, when executed by a processor, implements the steps of the reinforced concrete detection method based on machine learning described in each embodiment of the present application.
[0133] The present application also provides a computer program product, including a non-volatile computer-readable storage medium, wherein the non-volatile computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the reinforced concrete detection method based on machine learning described in each embodiment of the present application are implemented.
[0134] Figure 4 is a hardware structure block diagram of an electronic device (or mobile device) for a reinforced concrete detection method based on machine learning according to an embodiment of the present invention. Figure 4 As shown, the electronic device may include one or more ( Figure 4402a, 402b, ..., 402n are used to illustrate) processor 402 (processor 402 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA), memory 404 for storing data. In addition, it may also include: a display, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of the I / O interface), a network interface, a keyboard, a power supply and / or a camera. It can be understood by those skilled in the art that Figure 4 The structure shown is only for illustration and does not limit the structure of the above electronic device. Figure 4 More or fewer components as shown, or with Figure 4 Different configurations are shown.
[0135] The serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages or disadvantages of the embodiments.
[0136] In the above embodiments of the present invention, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.
[0137] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only schematic. For example, the division of the units can be a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.
[0138] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.
[0139] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.
[0140] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, magnetic disk or optical disk and other media that can store program codes.
[0141] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.
Claims
1. A reinforced concrete detection method based on machine learning, characterized in that: include: Using electromagnetic eddy current detection equipment to detect intact steel bars and multiple types of defective steel bars, obtaining waveform data of the intact steel bars and the multiple types of defective steel bars, and matching corresponding defect category labels to the waveform data; Collecting characteristic data of all the waveform data and basic data of all the steel bars, and combining the characteristic data, the basic data and the defect category label into original steel bar parameters, wherein the waveform data includes the waveform peak value, the width corresponding to the specified ratio of the waveform peak value drop, and the quartic term coefficient and the quadratic term coefficient obtained by fitting the data within the specified width range on both sides of the waveform peak value, all the steel bars include the intact steel bars and the multiple types of defective steel bars, and the basic data includes the steel bar diameter and the protective layer thickness; Inputting the combined original steel bar parameters into a generative adversarial model, synthesizing steel bar parameters of different types of defective steel bars through the generative adversarial model, and combining the original steel bar parameters and the steel bar parameters synthesized by the generative adversarial model into an expanded defective steel bar parameter set; Inputting the characteristic data of the intact steel bars and the defective steel bar characteristic data in the defective steel bar parameter set into a neural network model, training the neural network model, and obtaining a trained neural network model; The steel bar feature data detected in real time is input into the trained neural network model to obtain the steel bar basic data and defect category of the detected steel bar object.
2. The method according to claim 1, characterized in that When electromagnetic eddy current testing equipment is used to test various types of defective steel bars, it also includes: The electromagnetic eddy current detection device is used to detect the concrete around the steel bars to obtain waveform data of the concrete around the steel bars; The waveform data of the concrete around the steel bar is merged into the waveform data of the multiple different types of defective steel bars.
3. The method according to claim 1, characterized in that: The step of collecting the characteristic data of all the waveform data and the basic data of all the steel bars comprises: Extracting data within a specified width range on both sides of the waveform peak from the waveform data, and fitting the data within the range to a first function; Collecting the waveform peak value and the width corresponding to the specified ratio of the peak value drop from the waveform data, and combining the waveform peak value, the width corresponding to the specified ratio of the waveform peak value drop, and the quartic term coefficient and the quadratic term coefficient of the first function into characteristic data of the waveform data; The steel bar diameters and protective layer thicknesses of all the intact steel bars and the multiple types of defective steel bars are collected, and the steel bar diameters and protective layer thicknesses are combined into basic data of all the steel bars.
4. The method according to claim 1, characterized in that: The step of synthesizing steel bar parameters of different types of defective steel bars by using the generative adversarial model comprises: Initialize the generator and the discriminator in the generative adversarial model, and set the network parameters of the generator and the discriminator; The network parameters of the generator are fixed, a group of noise vectors are randomly sampled from a normal distribution, the noise vectors are input into the generator to obtain a group of synthetic false data, the synthetic false data and the steel bar parameters of the different types of defective steel bars are input into the discriminator, the loss function value and the target optimization function value of the discriminator are obtained according to the output result of the discriminator, and the network parameters of the discriminator are updated using a gradient descent strategy according to the loss function value and the target optimization function value of the discriminator; The network parameters of the discriminator are fixed, a set of noise vectors are randomly sampled from the normal distribution, the noise vectors are input into the generator to obtain a set of new synthetic false data, the new synthetic false data and the steel bar parameters of the different types of defective steel bars are input into the discriminator, the loss function value and the target optimization function value of the generator are obtained according to the output result of the discriminator, and the network parameters of the generator are updated using the gradient descent strategy according to the loss function value and the target optimization function value of the generator; The process of alternating training of the discriminator and the generator is repeated until the output result of the generator is the same as the output result of the discriminator.
5. The method according to claim 1, characterized in that The step of synthesizing steel bar parameters of different types of defective steel bars by generating an adversarial model also includes: Inputting the output result of the discriminator for true data and the output result of the discriminator for false data into a second function to obtain a loss function of the discriminator, wherein the true data are the steel bar parameters of the different categories of defective steel bars; Inputting the output result of the discriminator on the false data into a third function to obtain a loss function of the generator; The output result of the discriminator for the true data and the output result of the discriminator for the false data are input into a fourth function to obtain the target optimization function.
6. The method according to claim 1, characterized in that Before training the neural network model, the following steps are included: Normalizing the characteristic data of the intact steel bars and the defective steel bar characteristic data in the defective steel bar parameter set; Initializing the network structure of the neural network model, setting an input layer, multiple hidden layers and an output layer; Initialize the network parameters of the neural network model, wherein the network parameters include the number of iterations, the root mean square error rate, and the learning rate.
7. The method according to claim 6, characterized in that The step of training the neural network model comprises: Inputting the characteristic data of the intact steel bars and the defective steel bar characteristic data in the defective steel bar parameter set into the input layer of the neural network model to obtain output data of the input layer; Inputting the output data of the input layer and the weights between the neurons of the input layer and the neurons of the first hidden layer into a first formula to obtain the input data of the first hidden layer; Inputting the input data of the first hidden layer into the second formula to obtain the output data of the first hidden layer; Hidden layer iterative output step, inputting the output data of the previous hidden layer and the weights between the neurons of the previous hidden layer and the neurons of the next hidden layer into the first formula to obtain the input data of the next hidden layer, and inputting the input data of the next hidden layer into the second formula to obtain the output data of the next hidden layer; Repeat the hidden layer iterative output step until the output data of the last hidden layer is obtained; Inputting the output data of the last hidden layer and the weights between the last hidden layer neurons and the output layer neurons into a third formula to obtain input data of the output layer; The input data of the output layer is input into the activation function to obtain the output data of the output layer, that is, the output data of the neural network model.
8. The method according to claim 7, characterized in that The step of training the neural network model further includes: Input the output data of the output layer and the expected output data into a fourth formula to obtain a root mean square error rate, wherein the expected output data includes the basic data and defect category labels of the intact steel bars and the defective steel bars; When the root mean square error rate is less than or equal to a preset root mean square error rate, exiting the iterative training to obtain the trained neural network model; When the root mean square error rate is greater than the preset root mean square error rate, the error gradient values of the output layer and all the hidden layers are calculated, the network weights are updated according to the error gradient values, and the neural network model is trained again according to the new network weights.
9. A reinforced concrete detection device based on machine learning, characterized in that: include: A detection unit, used to detect intact steel bars and multiple types of defective steel bars using electromagnetic eddy current detection equipment, obtain waveform data of the intact steel bars and the multiple types of defective steel bars, and match corresponding defect category labels to the waveform data; A collection unit, used for collecting characteristic data of all the waveform data and basic data of all the steel bars, and combining the characteristic data, the basic data and the defect category label into original steel bar parameters, wherein the waveform data includes a waveform peak value, a width corresponding to a specified ratio of a waveform peak value drop, and a quartic term coefficient and a quadratic term coefficient obtained by fitting data within a specified width range on both sides of the waveform peak value, all the steel bars include the intact steel bars and the multiple types of defective steel bars, and the basic data includes a steel bar diameter and a protective layer thickness; An expansion unit, used for inputting the combined original steel bar parameters into a generative adversarial model, synthesizing steel bar parameters of different types of defective steel bars through the generative adversarial model, and combining the original steel bar parameters and the steel bar parameters synthesized by the generative adversarial model into an expanded defective steel bar parameter set; A training unit, used for inputting the characteristic data of the intact steel bar and the defective steel bar characteristic data in the defective steel bar parameter set into a neural network model, training the neural network model, and obtaining a trained neural network model; The real-time acquisition unit is used to input the steel bar feature data detected in real time into the trained neural network model to obtain the steel bar basic data and defect category of the detected steel bar object.
10. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored computer program, wherein when the computer program is executed, the device where the computer-readable storage medium is located is controlled to execute the reinforced concrete detection method based on machine learning according to any one of claims 1 to 7.
Citation Information
Cited By
Online detection method for small and special motor winding defects based on pulsed eddy current and deep learning
CN120334348A
