A method and system for on-line detection of plastic strain ratio of cold-rolled thin strip steel

Abstract

The application discloses a kind of cold-rolled thin strip steel plastic strain ratio on-line detection method and system, the cold-rolled thin strip steel plastic strain ratio on-line detection method includes the following steps: establishing the artificial neuron network algorithm model of cold-rolled thin strip steel;The plastic strain ratio of cold-rolled thin strip steel is obtained by artificial method;The artificial neuron network algorithm model of cold-rolled thin strip steel is modelled training;The artificial neuron network algorithm model of cold-rolled thin strip steel is used for on-line detection.The cold-rolled thin strip steel plastic strain ratio on-line detection method is applied to the comprehensive electromagnetic detection to running strip steel, and multiple electromagnetic signals are acquired in real time, and the electromagnetic signal is expanded, the cold-rolled thin strip steel plastic strain ratio on-line detection method is established by artificial neuron network algorithm model, realizes the purpose of on-line accurate measurement strip steel plastic strain ratio, scientific performance is strong, practical performance is high, reduces the waste of raw material and manpower cost.

Description

Technical Field

[0001] This invention relates to the field of nondestructive testing technology, and more specifically, to an online method and system for detecting the plastic strain ratio of cold-rolled thin strip steel. Background Technology

[0002] The plastic strain ratio, or R-value, is the most important parameter for evaluating the deep-drawing performance of thin metal sheets. When a strip specimen is subjected to a tensile test machine and subjected to 20% tensile deformation, the ratio of the strain in the width direction to the strain in the thickness direction is called the plastic strain ratio, also known as the Lankford value or r-value. It reflects the ability of a thin metal sheet to resist thinning or thickening when subjected to tensile or compressive forces in a certain plane. Currently, steel companies mainly use the offline cutting test method to test the plastic strain ratio (R) of cold-rolled thin strip steel. This method is widely used. It involves cutting a sample from a specific location on a coil of strip, such as the beginning or end, and then sending it to the laboratory for offline testing to obtain the plastic strain ratio of the sample, from which the yield elongation of the coil of strip can be inferred.

[0003] The ratio of the strain in the width direction to the strain in the thickness direction when a thin plate specimen is stretched is denoted by R: R = ε w / ε t In the formula ε w =lnb / b0 is the strain in the width direction; ε t =lnt / t0 is the strain in the thickness direction. It is also commonly referred to as the thickness anisotropy coefficient. The r value can be determined by measuring the width and length dimensions within the gauge length of the specimen before and after tensile deformation and calculating it using the following formula:

[0004]

[0005] In the formula: l0, b0 are the length and width dimensions of the specimen within the gauge length before deformation; if the specimen's orientation in the plate is different, please refer to the appendix of the instruction manual for details. Figure 1 The thickness anisotropy coefficients obtained from the experiments also differ. This is because the texture and mechanical properties of the sheet metal are not uniform in all directions. Here, R90 is selected as the main research object. The advantages of the offline sample cutting test method are its simplicity, direct results, and high accuracy. However, this method has the following drawbacks: First, the data has a large time lag, offering limited assistance to the production process, and online control is out of the question. Second, the data is incomplete, only reflecting the values ​​of the head and tail of a single coil of strip. Third, there is waste in shearing. During production, if the unit stops or operates at low speed for some reason, in order to maintain the empirical judgment that "if the head and tail are qualified, then the middle is also qualified," a section of "suspected unqualified" strip steel is usually cut off. There is no standard for how much to cut, so as much as possible is cut, which obviously leads to waste. Fourth, it requires 24 / 7 human intervention at the machine, resulting in high labor intensity and high labor costs. Summary of the Invention

[0006] The purpose of this invention is to provide an online detection method for the plastic strain ratio of cold-rolled thin strip steel. By manually measuring the input parameters of the cold-rolled thin strip steel online, multiple electromagnetic signals are acquired in real time. At the same time, the electromagnetic signals are expanded, the actual fluctuation values ​​of the spacing affecting the electromagnetic parameters are corrected, and the influence of the strip thickness is considered. The developed artificial neural network algorithm model does not depend on the real-time process parameters of the unit, and realizes accurate online measurement of the plastic strain ratio of the strip steel.

[0007] To achieve the above objectives, the present invention adopts the following technical solution:

[0008] According to one aspect of the present invention, an online method for detecting the plastic strain ratio of cold-rolled thin strip steel is provided, comprising the following steps:

[0009] S1. Establish an artificial neural network algorithm model for online detection of plastic strain ratio in cold-rolled thin strip steel;

[0010] S2. Obtain the parameters of the cold-rolled thin strip steel detected online, and use them as input to the artificial neural network algorithm model;

[0011] S3. Train the artificial neural network algorithm model;

[0012] S4. Based on the parameters of cold-rolled thin strip steel, the plastic strain ratio of cold-rolled thin strip steel is obtained by applying the artificial neural network algorithm model.

[0013] According to the above-mentioned aspects of the present invention, the online detection method for the plastic strain ratio of cold-rolled thin strip steel includes the following steps in S1:

[0014] S11. Select a BP multilayer feedforward neural network as the artificial neural network algorithm model;

[0015] S12. Select the parameters for the BP multilayer feedforward neural network.

[0016] According to the above-mentioned online detection method for plastic strain ratio of cold-rolled thin strip steel of the present invention, the BP multilayer feedforward neural network in S11 includes an input layer, a hidden layer and an output layer; wherein, the input layer has a total of 44 input parameters; the hidden layer has 8-12 layers and the output layer has 1 output parameter; the transfer function between the input layer and the hidden layer adopts the Sigmoid function, and the transfer function between the hidden layer and the output layer adopts the purelin function.

[0017] According to the above-mentioned online detection method for plastic strain ratio of cold-rolled thin strip steel of the present invention, the parameters of the BP multilayer feedforward neural network in S12 include transfer function, learning algorithm, number of iterations, learning rate, training target error, initial weights and threshold.

[0018] According to the online detection method for the plastic strain ratio of cold-rolled thin strip steel of the present invention described above, S2 includes the following steps:

[0019] S21. The parameters of the cold-rolled thin strip steel measured online include a set of electromagnetic parameters, the actual fluctuation value of the gap between the strip steel and the probe (Gap), the current strip steel thickness, and the current strip steel tension;

[0020] S22. Input the above parameters into the artificial neural network algorithm model to calculate the plastic strain ratio of the cold-rolled thin strip steel.

[0021] According to the above-mentioned method for online detection of plastic strain ratio of cold-rolled thin strip steel of the present invention, a set of electromagnetic parameters includes 41 electromagnetic response parameters, which correspond to tangential magnetic field harmonic response parameters EM1-EM11, Barkhausen noise detection response parameters EM12-EM18, incremental permeability electromagnetic detection response parameters EM19-EM25 and multi-frequency eddy current electromagnetic detection response parameters EM26-EM41, respectively.

[0022] According to the online detection method for plastic strain ratio of cold-rolled thin strip steel according to the above-mentioned aspects of the present invention, when the actual fluctuation value Gap between the strip steel and the probe satisfies 4mm≤Gap≤6mm, the actual fluctuation value Gap is introduced into the artificial neural network algorithm model. When Gap>6mm or Gap<4mm, the detection system is in an abnormal state and the detection is invalid.

[0023] According to the above-described online detection method for the plastic strain ratio of cold-rolled thin strip steel of the present invention, S3 includes the following steps:

[0024] S31. Using the plastic strain ratios produced by the production line over a period of time as training samples, provide the input vector and output vector;

[0025] S32. Initialize weights and thresholds, and calculate the output of each node in the hidden layer and output layer;

[0026] S33. Calculate the error E between the expected output and the actual output, and determine whether the error E meets the technical requirements;

[0027] S34. If the error E meets the technical requirements, the calculation ends. If the error E does not meet the technical requirements, the error of each unit in the hidden layer and the output layer is calculated, the error gradient is calculated, and the weights and thresholds are updated.

[0028] According to the online detection method for the plastic strain ratio of cold-rolled thin strip steel according to the above-mentioned aspects of the present invention, the execution process of the artificial neural network algorithm model training in S3 includes forward propagation of signal and backward propagation of error; for each training sample, during forward propagation of signal, the input vector is passed from the input layer to the output layer layer by layer; during backward propagation of error, the error is passed from the hidden layer to the input layer layer by layer; the two execution processes are repeated until the set termination condition is met.

[0029] According to the above-described method for online detection of the plastic strain ratio of cold-rolled thin strip steel of the present invention, S4 includes the following steps:

[0030] S41. Compare the calculated values ​​of the plastic strain ratio of the sample using the artificial neural network algorithm model with the offline test values ​​of the sample;

[0031] S42. Evaluate whether the artificial neural network algorithm model meets the requirements according to the online detection evaluation criteria. The online detection evaluation criteria are: for a given number of samples N, the measurement accuracy of 90% of the samples meets the requirement of relative error ≤10%.

[0032] According to another aspect of the present invention, an online detection system for the plastic strain ratio of cold-rolled thin strip steel is also provided, comprising:

[0033] The electromagnetic detection unit is located on the lifting device below the strip steel. It performs electromagnetic detection on the strip steel to obtain multiple electromagnetic response signals.

[0034] A rangefinder is mounted on the electromagnetic detection unit to obtain the distance G between the lower surface of the strip and the electromagnetic detection unit;

[0035] A tension detection unit is disposed on the electromagnetic detection unit and is used to acquire the tension of the strip steel;

[0036] A control computer is used to control the lifting and lateral movements of the lifting device, as well as the operation of the electromagnetic detection unit, the rangefinder, and the tension detection unit.

[0037] The online detection system obtains the plastic strain ratio of cold-rolled thin strip steel by performing the online detection method for the plastic strain ratio of cold-rolled thin strip steel.

[0038] By adopting the above technical solution, the present invention has the following advantages:

[0039] This invention provides an online detection method and system for the plastic strain ratio of cold-rolled thin strip steel. By manually measuring the input parameters of the cold-rolled thin strip steel online, multiple electromagnetic signals are acquired in real time. The electromagnetic signals are extended, and the input parameters are substituted into an artificial neural network algorithm model. The artificial neural network algorithm model is then trained and manually tested to achieve the purpose of accurately measuring the plastic strain ratio of the strip steel online. This method has strong scientific performance and high practicality, reducing the waste of raw materials and labor costs. Attached Figure Description

[0040] Other features, objects, and advantages of the present invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings.

[0041] Figure 1 This is a schematic diagram of the sampling location on the plate in the existing technology;

[0042] Figure 2 This is a flowchart of the online detection method for the plastic strain ratio of cold-rolled thin strip steel according to the present invention;

[0043] Figure 3 This is the BP neural network topology of the present invention;

[0044] Figure 4 This is a schematic diagram of the cold-rolled thin strip steel inspection system of the present invention;

[0045] Figure 5 This is a schematic diagram of the online working position of the hardware system for detecting cold-rolled thin strip steel of the present invention;

[0046] Figure 6 This is a schematic diagram of the descending position of the hardware system for detecting cold-rolled thin strip steel according to the present invention;

[0047] Figure 7 This is a flowchart of the BP multilayer feedforward neural network algorithm of the present invention;

[0048] Figure 8 This refers to the plastic strain ratio distribution along the entire length of the cold-rolled thin strip in this embodiment of the invention. Detailed Implementation

[0049] The technical solution of the present invention will be specifically described below with reference to the accompanying drawings. The detailed features and advantages of the present invention are described in detail in the specific embodiments. The content is sufficient to enable any person skilled in the art to understand the technical content of the present invention and implement it accordingly. Based on the specification, claims and drawings disclosed in this specification, those skilled in the art can easily understand the related objects and advantages of the present invention.

[0050] Figure 2The flowchart of the online detection method for the plastic strain ratio of cold-rolled thin strip steel of the present invention is shown; the online detection method for the plastic strain ratio of cold-rolled thin strip steel includes the following steps as detailed below. Figure 2 As shown:

[0051] S1. Establish an artificial neural network algorithm model for cold-rolled thin strip steel;

[0052] Figure 3 The BP neural network topology of this invention is shown;

[0053] S1 specifically includes the following steps:

[0054] S11. The BP multilayer feedforward neural network is selected as the artificial neural network algorithm model for cold-rolled thin strip steel;

[0055] S12. Select the parameters for the BP multilayer feedforward neural network.

[0056] The Back Propagation (BP) neural network algorithm is a multi-layer feedforward network that learns and trains using an error backpropagation algorithm. It was first proposed by scientists such as Rumelhart and McCelland in 1986. Its network topology is as follows: Figure 3 As shown, the network structure consists of three parts: an input layer, hidden layers, and an output layer. There is only one input layer and one output layer, while there can be multiple hidden layers. Each layer is composed of several neurons. Neurons within a layer are not connected to each other, but neurons between layers are fully connected. That is, any neuron in the input layer is connected to all nodes in the hidden layer, and any neuron in the hidden layer is connected to all nodes in the output layer.

[0057] The design of a complete BP neural network model needs to consider factors such as the design of the network structure and the selection and setting of relevant parameters.

[0058] (1) Network Structure Design: Since there is only one input layer and one output layer, and the number of neurons in each layer is determined by the specific problem being solved and the data representation method, the key to network structure design lies in the design of the hidden layer structure, including the selection of the number of layers and nodes. For the design of the number of hidden layers, it is common practice to first consider setting up one hidden layer, and only when one hidden layer cannot meet the relevant performance requirements should the number of hidden layers be increased. Experimental studies show that when using two hidden layers, if the number of neurons in the first hidden layer is greater than that in the second hidden layer, the performance of the BP network will be improved to some extent. However, for some practical problems, the number of neurons required by using two hidden layers may be less than that of a single hidden layer. The more hidden layers there are, the longer the training time of the BP network will be, and the greater the local minimum error will be, meaning the probability of the network getting trapped in a local optimum will increase. Therefore, increasing the number of hidden layers should only be considered when increasing the number of hidden layer nodes cannot significantly improve network performance.

[0059] Currently, there is a lack of mature theoretical guidance for determining the number of neurons in the hidden layer; it is generally determined based on experience. The following empirical estimation formulas can be used as a reference:

[0060]

[0061] m = log2 n

[0062]

[0063] Where m represents the number of hidden layer neurons, n represents the number of input layer nodes, l represents the number of output layer nodes, and a is a constant between [1, 10].

[0064] The selection of the number of hidden layer neurons is crucial to the success of network design. If the number of hidden layer nodes is too small, the network's learning ability from training samples will be poor, and the learned patterns will not be sufficient to reflect the patterns of the entire sample set, thus increasing the prediction error of test samples. However, if the number of hidden layer nodes is too large, it will greatly prolong the training time, and the network may also learn and record irregular content (such as noise) in the samples, resulting in "overfitting" and reducing the network's generalization ability.

[0065] (2) Selection of parameters: The parameters of the BP neural network include the transfer function, learning algorithm, number of iterations, learning rate, training target error, initial weights and threshold.

[0066] The transfer function is set according to the network requirements and the relationship between input and output. The commonly used transfer function is the S(sigmoid) type function, and its function expression is f(x) = 1 / (1+e^(-x)). -xThe sigmoid function possesses nonlinear and everywhere differentiable characteristics, and offers good gain control over the signal: when |x| is small, f(x) has a large gain; when |x| is large, f(x) has a small gain, which can prevent the network from entering a saturation state to a certain extent. Therefore, this function has been widely used. In addition, other common transfer functions include the logarithmic sigmoid transfer function (logsid), the hyperbolic tangent sigmoid transfer function (tangsig), and linear transfer functions. This invention selects the most commonly used sigmoid function.

[0067] In an embodiment of the present invention, the BP multilayer feedforward neural network in step S11 has a three-layer structure, including an input layer, a hidden layer, and an output layer. The input layer has a total of 44 parameters, including 41 electromagnetic parameters, one parameter each for the distance between the probe and the lower surface of the strip, and one parameter for the strip thickness. The hidden layer has 8-12 layers, specifically optimized based on data experiments during model training. The output layer has one parameter, namely the baking hardening value. The transfer function between the input layer and the hidden layer uses the Sigmoid function, and the transfer function between the hidden layer and the output layer uses the Purelin function.

[0068] Figure 4 A schematic diagram of the cold-rolled thin strip steel inspection system of the present invention is shown; Figure 5 A schematic diagram of the online working position of the cold-rolled thin strip steel inspection hardware system of the present invention is shown; Figure 6 A schematic diagram of the descending position of the hardware system for detecting cold-rolled thin strip steel of the present invention is shown.

[0069] The specific details of the cold-rolled thin strip steel inspection system of this invention are as follows: Figure 4 As shown, it includes an online detection hardware system, a supporting software system, a mathematical model, and corresponding data interfaces and computer networks. The online detection hardware system is specifically as follows: Figure 5 and Figure 6 As shown, Figure 3 The strip steel 1 typically runs at a speed of 0-300 m / min. The strip steel's trajectory is stabilized by two idler rollers 2 arranged one behind the other. An electromagnetic detection unit 3, capable of vertical movement and lateral shifting, is positioned between the idler rollers 2. The electromagnetic detection unit 3 is located below the running strip steel, and its vertical movement and lateral shifting are controlled by the control system. The hardware detection system also includes a distance measuring instrument 4, which measures the distance between the electromagnetic detection unit 3 and the lower surface of the strip steel 1 in real time and sends the data to the control computer. A probe lifting device 5 enables the vertical movement of the electromagnetic detection unit 3, and a mechanical limit device 6 ensures a safe distance between the electromagnetic detection unit 3 and the strip steel 1.

[0070] Figure 4The distance 7 between the lower surface of the strip and the probe surface is a key parameter. Due to the vibration of the strip during operation and the influence of external factors such as the inherent plate properties of thin strip, the spacing of the strip fluctuates slightly. It is measured in real time by the distance measuring instrument 4, with a target value of 5mm and an allowable error of ±1mm. This parameter is the actual fluctuation value Gap of the spacing between the strip and the probe, which serves as an input to the BP neural network model.

[0071] S2. Obtain the plastic strain ratio of cold-rolled thin strip steel by artificial methods;

[0072] S2 specifically includes the following steps:

[0073] S21. Input parameters for cold-rolled thin strip steel measured manually online, the input parameters including a set of electromagnetic parameters, the actual fluctuation value of the gap between the strip steel and the probe Gap, the current strip steel thickness, and the current strip steel tension;

[0074] Among them, the electromagnetic parameter group in S21 includes four detection methods and 41 electromagnetic parameters. The four detection methods include tangential magnetic field harmonic analysis, Barkhausen noise, incremental permeability and multi-frequency eddy current. Each detection method outputs a curve signal, and each curve signal is transformed into several electromagnetic parameters for characterization by definition.

[0075] In one specific embodiment, the detection system comprehensively applies four electromagnetic detection methods: tangential magnetic field harmonic analysis, Barkhausen noise, incremental permeability, and multi-frequency eddy current. The result curves of the above four electromagnetic detections are characterized by several quantitative parameters defined below, as shown in Tables 1, 2, 3, and 4:

[0076] Table 1. Tangential harmonic response parameters of the excitation magnetic field (11 items in total, EMi, i=1,...,11)

[0077]

[0078] Table 2 Barkhausen noise detection response parameters (7 items in total, EMi, i = 12, ..., 18)

[0079]

[0080] Table 3 Incremental permeability electromagnetic detection response parameters (7 items in total, EMi, i = 19, ..., 25)

[0081]

[0082] Table 4. Response parameters of multi-frequency eddy current electromagnetic detection (16 items in total, EMi, i = 20, ..., 41)

[0083]

[0084] In summary, the integrated detection system outputs a maximum of 41 electromagnetic parameters. Online prediction of the plastic strain ratio is achieved using a backpropagation (BP) neural network within an artificial neural network (ANN).

[0085] It should be noted that the actual fluctuation value Gap between the strip and the probe in S21 is valid when 4mm≤Gap≤6mm. When Gap is introduced into the artificial neural network algorithm model, the measurement is valid and the detection result can be corrected. When Gap>6mm or Gap<4mm, the detection system is in an abnormal state, the detection conditions are not met, and the detection is invalid.

[0086] S22. Input the above input parameters into the artificial neural network algorithm model to calculate the plastic strain ratio of cold-rolled thin strip steel.

[0087] In a specific embodiment, the basic principle of the BP algorithm is introduced using a BP neural network with one hidden layer as an example. Figure 3 As shown, x = (x0, x1, ..., x n-1 ) is the input vector, x i ,i=(0,1,…,n-1) represents the input quantity of the i-th neuron in the input layer, b j j = (0, 1, ..., l-1) represents the output of the hidden layer neuron node, y k k = (0, 1, ..., m-1) represents the output of the output layer nodes, n, l, and m represent the number of neurons in the input layer, hidden layer, and output layer, respectively, and v ij w represents the weights from the i-th neuron in the input layer to the j-th neuron in the hidden layer. jk Let represent the weights from the j-th node in the hidden layer to the k-th node in the output layer. Then:

[0088] The output of each neuron node in the hidden layer is

[0089]

[0090] The output of each neuron node in the output layer is

[0091]

[0092] Where, θ j φ k f1 and f2 represent the threshold values ​​for the hidden layer node and the output layer node, respectively; f1 and f2 represent the transfer functions for the hidden layer and the output layer, respectively.

[0093] Assume the expected output of the output layer node is O k The total error is...

[0094]

[0095] The purpose of error backpropagation is to continuously reduce the total error by adjusting the weights. Therefore, the weights should be adjusted along the negative gradient of the error, that is, the change in the weights should satisfy the following equation:

[0096]

[0097]

[0098] In the two equations above, η is a constant, called the learning rate, which is usually taken as 0 < η < 1.

[0099] Combining the above equations, we can obtain the following formulas for adjusting the weights of the hidden layer and the output layer:

[0100]

[0101]

[0102] Similarly, the threshold adjustment formulas for the hidden layer and output layer are as follows:

[0103]

[0104]

[0105] After calculating the new weights and thresholds according to the weight and threshold adjustment formula, a new round of forward propagation process begins.

[0106] The above derivation process is based on a single training sample and a single-hidden-layer BP neural network. The basic idea is equally applicable to multi-hidden-layer BP neural networks. When calculating the adjustment amounts for weights and thresholds, the process proceeds recursively from the output layer through each intermediate hidden layer back to the first hidden layer. The error across multiple training samples is calculated as the cumulative sum of the errors of each individual sample.

[0107] Figure 7 The flowchart of the BP multilayer feedforward neural network algorithm of the present invention is shown;

[0108] S3. Train the artificial neural network algorithm model for cold-rolled thin strip steel;

[0109] S3 includes the following specific steps, as follows: Figure 7 As shown:

[0110] S31. Using the plastic strain ratios produced by the production line over a period of time as training samples, provide the input vector and output vector;

[0111] S32. Initialize weights and thresholds, and calculate the output of each node in the hidden layer and output layer;

[0112] S33. Calculate the error E between the expected output and the actual output, and determine whether the error E meets the technical requirements;

[0113] S34. If the error E meets the technical requirements, the calculation ends. If the error E does not meet the technical requirements, the error of each unit in the hidden layer and the output layer is calculated, the error gradient is calculated, and the weights and thresholds are updated.

[0114] In S3, the artificial neural network algorithm model training process includes forward propagation of signals and backward propagation of errors. For each training sample, during forward propagation of signals, the input vector is passed from the input layer to the output layer layer by layer. During backward propagation of errors, the error is passed from the hidden layer to the input layer layer by layer. The two execution processes are repeated until the set termination condition is met.

[0115] Figure 8 This illustration shows the plastic strain ratio distribution along the entire length of the cold-rolled thin strip in an embodiment of the present invention.

[0116] S4. Apply the artificial neural network algorithm model of cold-rolled thin strip steel to online detection; wherein S4 includes the following specific steps:

[0117] S41. Compare the calculated values ​​of the plastic strain ratio of the sample using the artificial neural network algorithm model with the offline test values ​​of the sample;

[0118] S42. Evaluate whether the artificial neural network algorithm model meets the requirements according to the online detection evaluation criteria. The online detection evaluation criteria are: for a given number of samples N, the measurement accuracy of 90% of the samples meets the requirement of relative error ≤10%.

[0119] In a specific embodiment, the online detection technology for the plastic strain ratio of a coil of steel on a production line is described above. The main parameters of the artificial neural network are as follows: Model: BP multilayer feedforward neural network; Model structure: Three-layer structure: input layer, hidden layer, and output layer.

[0120] The model input layer contains 44 parameters, including electromagnetic parameters: EM1, ..., EM41, totaling 41 items; the distance between the probe and the lower surface of the strip: Gap (satisfying condition 4≤Gap≤6); strip thickness and tension.

[0121] The model has 10 hidden layers, with the optimal number to be determined during model training based on data experiments. The output layer has one parameter: the plastic strain ratio. The transfer function between the input layer and the hidden layers is the Sigmoid function, while the transfer function between the hidden layers and the output layer is the Purelin function.

[0122] During the model training phase, the plastic strain ratios produced by this production line over a past period were used as training samples. The specific input and output parameters of the model are shown in Tables 5 and 6.

[0123] Table 5. Training dataset for online detection of plastic strain ratio (R90) using a BP neural network model (partial).

[0124] Training-1 Training-2 Training-3 Training-4 Training-5 Training-6 Training-7 Training-8 Training-9 Training-10 EM01 1.5213 1.6067 1.5213 1.5827 1.6 1.5747 1.6067 1.508 1.508 1.74 EM02 0.434 0.4966 0.4586 0.4539 0.4605 0.4509 0.4591 0.4307 0.4417 0.4546 EM03 0.3513 0.3717 0.3637 0.3627 0.3668 0.3369 0.3527 0.3503 0.3529 0.3399 EM04 0.281 0.2766 0.2699 0.2758 0.2808 0.2671 0.2686 0.2692 0.2714 0.2751 EM05 2.6808 2.7576 2.6464 2.7108 2.7721 2.7689 2.7531 2.6809 2.6907 2.8609 EM06 2.8725 2.8383 2.8607 2.8728 2.883 2.8837 2.8765 2.8914 2.9075 2.7782 EM07 2.031 1.9844 2.0061 1.9826 2.0055 2.04 2.0427 2.0193 2.051 1.9533 EM08 1.2801 1.3542 1.2942 1.3067 1.328 1.2642 1.2967 1.2558 1.2793 1.2896 EM09 5.0239 5.126 5.1582 4.942 5.061 4.9664 4.9187 5.0239 5.1406 4.6113 EM10 -4.6135 11.577 -3.9271 -4.9482 -5.4328 -6.1176 -5.8523 -4.4457 -4.8331 12.291 EM11 0.2315 0.2395 0.2614 0.2326 0.2118 0.2119 0.2302 0.2283 0.236 0.1884 EM12 0.1417 0.1114 0.1495 0.109 0.1093 0.1158 0.1086 0.0995 0.0955 0.1177 EM13 0.1314 0.1033 0.1292 0.0977 0.0978 0.105 0.1011 0.0905 0.0854 0.1033 EM14 0.1264 0.1074 0.1348 0.0975 0.1034 0.1039 0.1041 0.0964 0.086 0.1056 EM15 -1.7945 10.36 14.043 -4.3057 -0.2372 2.3236 9.9613 5.4468 -0.5955 11.871 EM16 0.3672 25.232 27.33 13.828 14.809 19.195 0 17.498 6.9295 9.4715 EM17 1.9403 8.6383 2.1787 10.174 0.0926 7.0775 16.484 9.9616 1.0494 2.3835 EM18 1.2049 6.0115 1.4392 4.1875 0.1023 3.6889 5.1846 5.3022 0.8809 1.5954 EM19 0.019 0.0145 0.0171 0.0166 0.0191 0.0218 0.0192 0.0197 0.0207 0.0201 EM20 0.0077 0.007 0.0087 0.0072 0.008 0.0093 0.0076 0.0083 0.0084 0.0076 EM21 0.0188 0.0141 0.0169 0.0158 0.0181 0.0216 0.0188 0.0189 0.02 0.02 EM22 0.9558 1.8851 1.4886 2.1959 2.1786 1.1484 1.5803 2.0625 1.6373 0.5846 EM23 31.995 33.682 34.603 32.665 30.886 32.063 28.983 32.574 31.799 31.387 EM24 19.958 25.388 26.639 21.889 20.244 20.957 18.797 21.789 20.12 19.514 EM25 11.549 15.38 17.153 12.37 11.65 12.167 10.763 12.348 11.619 11.345 EM26 0.036 0.0301 0.034 0.0331 0.0301 0.0334 0.0294 0.0342 0.034 0.0304 EM27 -0.2431 -0.2405 -0.2408 -0.243 -0.2404 -0.2426 -0.2403 -0.2437 -0.2434 -0.2404 EM28 0.2458 0.2424 0.2432 0.2453 0.2423 0.2449 0.2421 0.246 0.2458 0.2423 EM29 -1.4237 -1.4461 -1.4304 -1.4353 -1.4461 -1.4342 -1.4491 -1.4314 -1.4322 -1.4451 EM30 0.4031 0.3912 0.3968 0.3979 0.3914 0.3979 0.3908 0.4006 0.3999 0.3908 EM31 -0.3426 -0.3494 -0.342 -0.3483 -0.3485 -0.3471 -0.3497 -0.3475 -0.3479 -0.3481 EM32 0.529 0.5245 0.5238 0.5288 0.5241 0.528 0.5244 0.5304 0.53 0.5234 EM33 -0.7045 -0.729 -0.7113 -0.719 -0.7274 -0.7172 -0.73 -0.7146 -0.7161 -0.7277 EM34 0.4248 0.446 0.4296 0.4421 0.4435 0.44 0.4474 0.4407 0.4409 0.4435 EM35 0.8323 0.83 0.825 0.8352 0.8286 0.8335 0.8298 0.838 0.8377 0.8264 EM36 0.9344 0.9422 0.9301 0.945 0.9398 0.9425 0.9427 0.9468 0.9467 0.9379 EM37 1.0989 1.0777 1.0907 1.084 1.0794 1.0851 1.0763 1.0866 1.0863 1.0783 EM38 -0.5073 -0.5054 -0.5016 -0.508 -0.5049 -0.5069 -0.5051 -0.5097 -0.5096 -0.5034 EM39 0.7764 0.7988 0.7807 0.7987 0.7947 0.7941 0.7996 0.7962 0.7967 0.7952 EM40 0.9274 0.9452 0.9279 0.9466 0.9415 0.942 0.9458 0.9454 0.9458 0.9412 EM41 2.1495 2.1349 2.1419 2.1372 2.1368 2.1389 2.1342 2.1402 2.1399 2.1352 THK 0.7 0.8 0.9 0.6 0.67 0.7 0.8 0.67 0.67 0.7 GAP 5.08 5.07 5.11 5.09 5.06 5.05 5.07 5.09 5.1 4.93 TEN 15.85 17.81 21.72 20.99 18.92 20.31 15.21 14.62 15.69 16 Specimen Tension 1.54 1.97 1.96 1.77 1.94 1.97 1.71 1.82 1.98 2.22

[0125] Table 6. Partial Model Input Parameters During the Training Phase

[0126] Test-1 Test-2 Test-3 Test-4 Test-5 EM01 1.5707 1.5547 1.58 1.6093 1.888 EM02 0.4691 0.5011 0.4335 0.4837 0.5039 EM03 0.3235 0.3726 0.3311 0.3658 0.3835 EM04 0.2568 0.2921 0.269 0.2886 0.2991 EM05 2.7739 2.6813 2.7895 2.6965 2.833 EM06 2.8243 2.9187 2.8071 2.8567 2.7152 EM07 1.9484 2.0701 1.953 2.012 1.8899 EM08 1.2563 1.3872 1.2392 1.3578 1.4174 EM09 5.0542 5.5532 4.9197 5.204 4.8753 EM10 10.805 -5.1106 -5.1737 -5.2669 12.798 EM11 0.2149 0.2575 0.1906 0.2549 0.2275 EM12 0.1164 0.1179 0.0778 0.0853 0.1028 EM13 0.1037 0.1063 0.0697 0.0783 0.0954 EM14 0.1019 0.1133 0.0735 0.0787 0.0993 EM15 2.3063 -6.2634 -1.3668 -0.4248 15.064 EM16 20.224 14.518 1.0816 20.881 14.85 EM17 10.579 9.1835 0.0006 9.3375 15.127 EM18 3.6796 3.8015 -0.0934 3.7069 3.2154 EM19 0.0209 0.0179 0.0198 0.019 0.0181 EM20 0.0084 0.0068 0.0081 0.0077 0.0069 EM21 0.0209 0.0172 0.0196 0.0181 0.018 EM22 -0.3317 1.7163 0.8424 2.0486 0.1862 EM23 30.697 29.365 31.492 30.467 33.395 EM24 19.563 18.033 20.724 19.704 20.78 EM25 11.254 10.362 12.102 10.962 11.945 EM26 0.0236 0.0221 0.0227 0.0214 0.0304 EM27 -0.2336 -0.2316 -0.2329 -0.2303 -0.2401 EM28 0.2347 0.2327 0.234 0.2313 0.2421 EM29 -1.4703 -1.4756 -1.4736 -1.4781 -1.4447 EM30 0.3732 0.3691 0.3713 0.3657 0.3901 EM31 -0.3475 -0.3473 -0.3478 -0.3458 -0.3469 EM32 0.5099 0.5068 0.5087 0.5033 0.5221 EM33 -0.7499 -0.755 -0.7526 -0.7573 -0.7268 EM34 0.4511 0.4494 0.4526 0.4524 0.4432 EM35 0.8072 0.8045 0.8057 0.798 0.8236 EM36 0.9247 0.9215 0.9241 0.9174 0.9353 EM37 1.0612 1.0614 1.059 1.0551 1.0771 EM38 -0.4905 -0.4906 -0.4904 -0.4856 -0.5004 EM39 0.7973 0.7932 0.7964 0.7938 0.797 EM40 0.9361 0.9327 0.9353 0.9306 0.941 EM41 2.1223 2.1247 2.1227 2.1198 2.1314 THK 1.2 0.6 0.7 0.67 0.6 GAP 4.92 4.86 4.88 4.8 4.43 TEN 24.38 18.26 21.28 18.13 17.18

[0127] The developed BP multilayer feedforward neural network was used for model verification, as shown in Table 7 below. The model-calculated values ​​of the plastic strain ratio of five samples were compared with the offline test values. According to the requirement of relative error ≤10%, the pass rate was 100%. This indicates that the BP multilayer feedforward neural network meets the technical requirements for online detection of plastic strain ratio (R90) of cold-rolled thin strip steel.

[0128] Table 7. Partial Model Input Parameters During the Training Phase

[0129] Test-1 Test-2 Test-3 Test-4 Test-5 Strength value 1.6 1.91 1.79 2.1 1.82 Non-destructive testing values 1.7 1.9 1.85 1.99 1.94 error 0.1 0.01 0.06 0.11 0.12 Relative error (%) 6.2 0.5 3.4 5.2 6.6

[0130] A BP multilayer feedforward neural network was used for real-time detection of a coil of steel strip. The specific detection results of the plastic strain ratio (R90) along the entire length are as follows: Figure 8 As shown. Figure 8 In this online method, the horizontal axis represents the strip length, and the vertical axis represents the plastic strain ratio (R90) along the entire length of a coil of strip. Compared to existing technologies that rely solely on shearing samples for testing cold-rolled thin strip, this method significantly improves both data volume and real-time performance. This online detection method for the plastic strain ratio of cold-rolled thin strip can be comprehensively applied to online quality monitoring systems for cold-rolled strip, enabling continuous detection, classification, and recording of steel plate production quality. This will play a very positive role in improving production efficiency, product quality, and product competitiveness.

[0131] This invention also provides an online detection system for the plastic strain ratio of cold-rolled thin strip steel, comprising:

[0132] The electromagnetic detection unit is located on the lifting device below the strip steel. It performs electromagnetic detection on the strip steel to obtain multiple electromagnetic response signals.

[0133] A rangefinder is mounted on the electromagnetic detection unit to obtain the distance G between the lower surface of the strip and the electromagnetic detection unit;

[0134] A tension detection unit is disposed on the electromagnetic detection unit and is used to acquire the tension of the strip steel;

[0135] A control computer is used to control the lifting and lateral movements of the lifting device, as well as the operation of the electromagnetic detection unit, the rangefinder, and the tension detection unit.

[0136] The online detection system obtains the plastic strain ratio of cold-rolled thin strip steel by performing the online detection method for the plastic strain ratio of cold-rolled thin strip steel.

[0137] It should be noted that although the present invention has been described with reference to specific embodiments, those skilled in the art should recognize that the above embodiments are only used to illustrate the present invention and are not intended to limit the present invention. Various equivalent changes or substitutions can be made without departing from the concept of the present invention. Therefore, any changes or modifications to the above embodiments within the essential spirit of the present invention will fall within the scope of the claims of the present invention.

Claims

1. A method for on-line detection of plastic strain ratio of cold-rolled thin strip steel, characterized in that, The method comprises the following steps: S1. Establishing an artificial neuron network algorithm model for online detection of plastic strain ratio of cold-rolled thin strip steel; S2. Obtaining parameters of the cold-rolled thin strip steel detected online as input of the artificial neuron network algorithm model, comprising the following steps: S21. Obtaining parameters of the cold-rolled thin strip steel detected online, including a group of electromagnetic parameters, actual fluctuation value Gap of the distance between the strip steel and the probe, current thickness of the strip steel and current tension of the strip steel; S22. Inputting the above parameters into the artificial neuron network algorithm model to calculate the plastic strain ratio of the cold-rolled thin strip steel, The group of electromagnetic parameters includes 41 electromagnetic response parameters, which respectively correspond to tangent magnetic field harmonic response parameters EM1-EM11, Barkhausen noise detection response parameters EM12-EM18, incremental magnetic permeability electromagnetic detection response parameters EM19-EM25 and multi-frequency eddy current electromagnetic detection response parameters EM26-EM41, When the actual fluctuation value Gap of the distance between the strip steel and the probe satisfies 4mm≤Gap≤6mm, the actual fluctuation value Gap is introduced into the artificial neuron network algorithm model, when Gap>6mm or Gap<4mm, the detection system is in an abnormal state and the detection is invalid; S3. Model training of the artificial neuron network algorithm model; S4. Obtaining the plastic strain ratio of the cold-rolled thin strip steel based on the parameters of the cold-rolled thin strip steel by applying the artificial neuron network algorithm model.

2. The method for detecting plastic strain ratio of cold-rolled thin strip steel in line according to claim 1, characterized in that, The S1 comprises the following steps: S11. Selecting a BP multilayer feedforward neural network as the artificial neuron network algorithm model; S12. Selecting parameters of the BP multilayer feedforward neural network.

3. A method for on-line detection of plastic strain ratio of cold-rolled thin strip steel according to claim 2, characterized in that, The BP multilayer feedforward neural network in the S11 comprises an input layer, a hidden layer and an output layer; wherein the total number of input parameters of the input layer is 44; the hidden layer is 8-12 layers, the number of output parameters of the output layer is 1; the transfer function between the input layer and the hidden layer adopts a Sigmoid function, and the transfer functions of the hidden layer and the output layer adopt purelin functions.

4. The method for detecting plastic strain ratio of cold-rolled thin strip steel on line according to claim 2, characterized in that, The parameters of the BP multilayer feedforward neural network in the S12 include transfer functions, learning algorithms, iteration times, learning rates, training target errors, initial weights and threshold values.

5. The method for detecting plastic strain ratio of cold-rolled thin strip steel in line according to claim 1, wherein, The S3 comprises the following steps: S31. Taking plastic strain ratios produced by the production line in a period of time as training samples to give input vectors and output vectors; S32. Initializing weights and threshold values to calculate node outputs of the hidden layer and the output layer; S33. Calculating errors E between expected outputs and actual outputs to judge whether the errors E meet technical requirements; S34. Ending the calculation when the errors E meet the technical requirements, and calculating errors of each unit of the hidden layer and the output layer, calculating error gradients and updating the weights and the threshold values when the errors E do not meet the technical requirements.

6. A method for on-line detection of plastic strain ratio of cold-rolled thin strip steel according to claim 5, characterized in that, The execution process of the model training of the artificial neuron network algorithm model in S3 includes forward propagation of signals and backward propagation of errors; for each training sample, when the forward propagation of signals is performed, the input vector is transmitted from the input layer to the output layer layer by layer; when the backward propagation of errors is performed, the error is transmitted from the hidden layer to the input layer layer by layer; the two execution processes are repeated until the set termination condition is met.

7. The method for detecting plastic strain ratio of cold-rolled thin strip steel in line according to claim 1, wherein, S4 includes the following steps: S41. Comparing the calculated value of the artificial neuron network algorithm model of the plastic strain ratio of the sample and the offline sample test value; S42. According to the online detection evaluation standard, it is judged whether the artificial neuron network algorithm model meets the requirements, and the online detection evaluation standard is: for a given sample quantity N, 90% of the sample measurement accuracy satisfies the relative error ≤10%.

8. An on-line system for detecting plastic strain ratio of cold-rolled thin strip steel, characterized by It comprises: An electromagnetic detection unit arranged on a lifting device below the strip steel, which obtains a plurality of electromagnetic response signals by performing electromagnetic detection on the strip steel; A range finder arranged on the electromagnetic detection unit, which is used to obtain the distance G between the lower surface of the strip steel and the electromagnetic detection unit; A tension detection unit arranged on the electromagnetic detection unit, which is used to obtain the tension of the strip steel; A control computer, which is used to control the lifting and transverse movement of the lifting device, and control the work of the electromagnetic detection unit, the range finder and the tension detection unit, The online detection system obtains the plastic strain ratio of the cold-rolled thin strip steel by performing the cold-rolled thin strip steel plastic strain ratio online detection method according to claim 1.

Images