An online detection method for bake hardening value of bake hardened cold rolled sheet

Abstract

The present invention discloses an on-line detection method for the baking hardening value (BH2) of cold-rolled sheets of bake hardening steel, which includes: on-line measuring a group of electromagnetic parameters of the strip, the distance between the probe and the lower surface of the strip, and the thickness of the strip; using a multi-layer neural network model to calculate the baking hardening value of the strip. The model has a three-layer structure: an input layer, a hidden layer, and an output layer. The input layer of the model has 43 parameters, the hidden layer has 8 to 15 layers, and the output layer has one parameter. The transfer function between the layers of the model uses the Sigmoid function to measure the baking hardening value of the cold-rolled strip in real time on-line. High-density data values of the entire length of the strip are obtained, and within the relative error accuracy range of 10%, the qualified rate of the samples is more than 90%.

Description

Technical Field

[0001] The invention relates to the field of nondestructive testing, and more specifically to an online testing method for a bake hardening value of a bake hardened cold rolled sheet. Background Art

[0002] Bake-hardened high-strength steel refers to steel that retains a certain amount of dissolved carbon and nitrogen atoms, and can also be strengthened by adding strengthening elements such as phosphorus and manganese. After processing and forming, it is baked at a certain temperature, and the yield strength of the steel is further increased due to aging hardening. Bake-hardened high-strength cold-rolled steel sheets and strips with a thickness of 0.50mm to 2.5mm are usually used in automotive exterior panels.

[0003] At present, the domestic steel enterprises mainly use the off-line test method of cutting samples to test the bake hardening value (BH2) of cold-rolled bake hardened steel, which is the widely used method at present. That is, a sample is cut from a certain part of a roll of strip steel, such as the head and tail, and then sent to the laboratory for offline testing to obtain the bake hardening value of the sample, thereby inferring the bake hardening value of a roll of strip steel. The existing measurement methods of bake hardening value (BH2) are briefly introduced as follows:

[0004] Sample preparation

[0005] The size and sampling direction of the specimen shall be prepared in accordance with the requirements for mechanical properties specimens.

[0006] Test conditions

[0007] When measuring the bake hardening value, according to the provisions of GB / T228, the sample is first pre-stretched with a total extension of 2%, and the R t2.0 After the 2% pre-stretched specimen has completed the prescribed heat treatment, the specimen is subjected to a tensile test again and ReL or Rp0.2 is measured.

[0008] In order to better maintain the consistency of the test results, it is advisable to control the stretching speed by displacement or strain, and it is recommended to set the stretching speed at a rate of 5% / min of the parallel length of the specimen. Do not switch the speed from the beginning of stretching until the above indicators are measured.

[0009] Rt2=Ft2.0 / A0

[0010] Rp0.2=Fp0.2 / A1

[0011] R eL =F eL / A1

[0012] Among them: F t2.0 : tensile force (N) when the specimen is stretched to a total extension of 2%;

[0013] F p0.2: Tensile strength of the heat-treated sample when the non-proportional extension is 0.2% (when there is no obvious yield)

[0014] (N);

[0015] F eL - Tensile force at which the heat-treated specimen shows lower yield (N);

[0016] A0: original cross-sectional area of ​​the sample (mm 2 );

[0017] A1: The cross-sectional area of ​​the sample after 2% pre-strain (mm 2 ).

[0018] Heat treatment conditions

[0019] After the temperature of the heating device reaches 170°C, put the sample that has been pre-strained by 2%. After the heating device reaches 170°C again, keep it warm for (20±0.5) minutes. The temperature control accuracy is maintained at ±2°C, and the maximum resolution of the temperature measuring device does not exceed 1°C. After heating, the sample is cooled to room temperature in the air.

[0020] Calculation of bake hardening value (BH2)

[0021] The bake hardening value (BH2) is the difference between the lower yield strength of the sample after baking or the yield strength corresponding to 0.2% non-proportional extension (when there is no obvious yield) and the yield strength corresponding to 2% total extension of the same sample before baking. The calculation diagram of BH2 is as follows Figure 1 As shown, Figure 1 In the figure, 1 is the stress-strain curve of 2% pre-strain; 2 is the stress-strain curve of the same sample after baking. The calculation formula of BH2 is as follows:

[0022] BH2=R eL (or R p0.2 )(After baking)-R t2.0 (Before baking)

[0023] The advantages of the offline test method of cutting samples are simplicity, direct results, and high precision. However, this method has the following disadvantages: First, the data has a large time lag, which is of limited help to the production process, and online control is out of the question. Second, the data is incomplete and can only reflect the values ​​of the head and tail of a roll of strip steel. Third, shearing waste. When the unit is in production, it is shut down or produced at a low speed for some reason. In order to maintain the empirical judgment that "if the head and tail are qualified, the middle is also qualified", a section of "suspected unqualified" strip steel is usually cut off at this time. There is no standard for how much to cut, so you can only cut as much as possible, which obviously causes waste. Fourth, someone needs to work beside the machine around the clock, which is labor-intensive and labor-intensive. Summary of the invention

[0024] The present invention aims to develop a method for online measurement of bake hardening value of bake hardening steel cold rolled thin plate. The method applies comprehensive electromagnetic detection to the running strip steel, obtains multiple electromagnetic signals in real time, and takes into account the correction of the spacing affecting the electromagnetic parameters and the influence of the strip steel thickness. The developed method does not rely on the real-time process parameters of the unit, and achieves the purpose of online accurate measurement of the bake hardening value of the strip steel.

[0025] The present invention provides a mathematical method for online detection of bake hardening values ​​of cold-rolled thin strip steel, in which an artificial neural network model is adopted. After the data model is established, the adaptability of the model needs to be trained, that is, the bake hardening values ​​of a group of materials are obtained by artificial methods, and the electromagnetic parameter set at the corresponding position of the above materials is obtained at the same time, the spacing and the thickness of the strip are measured, and then the model is trained to obtain a trained neural network, and then the model is used for online detection.

[0026] The present invention provides an online detection method for the bake hardening value of a bake hardened cold rolled sheet, comprising the following steps:

[0027] Online measurement of a set of electromagnetic parameters of the strip, the distance between the probe and the lower surface of the strip, and the thickness of the strip;

[0028] A multi-layer neural network model is used to calculate the bake hardening value of strip steel.

[0029] Wherein, the set of electromagnetic parameters includes tangent magnetic field harmonic detection response parameters, Barkhausen noise detection response parameters, incremental permeability detection response parameters, and multi-frequency eddy current detection response parameters; and / or

[0030] The multi-layer neural network model has a three-layer structure, including an input layer, a hidden layer, and an output layer.

[0031] Preferably, a set of electromagnetic parameters includes 41 electromagnetic 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.

[0032] The input of each neuron node in the input layer of the multi-layer neural network model is

[0033] x=(x0,x1,…,x n-1 )

[0034] The output of each neuron node in the hidden layer is

[0035]

[0036] The output of each neuron node in the output layer is

[0037]

[0038] Among them, x i ,i=(0,1,…,n-1) represents the input of the ith neuron in the input layer, b j ,j=(0,1,…,l-1) is the output of the hidden layer neuron node, y k , k = (0, 1, ..., m-1) is the output of the output layer node, n, l, m represent the number of neuron nodes in the input layer, hidden layer, and output layer respectively;

[0039] v ij Represents the weight from the i-th neuron node in the input layer to the j-th node in the hidden layer;

[0040] w jk Represents the weight from the jth node in the hidden layer to the kth node in the output layer;

[0041] θ j ,φ k Respectively represent the thresholds of hidden layer nodes and output layer nodes;

[0042] f1 and f2 represent the transfer functions of the hidden layer and the output layer respectively.

[0043] The learning algorithm of the multi-layer neural network model is:

[0044] Set the input vector and output vector;

[0045] Initialize the weights and thresholds of hidden layer nodes and output layer nodes;

[0046] Calculate the output of each node in the hidden layer and output layer;

[0047] Calculate the error between the expected output and the actual output;

[0048] Determine whether the error satisfies the requirement. If yes, end the learning process. If no, execute the following process.

[0049] Calculate the error of each unit in the hidden layer and output layer;

[0050] Calculate the error gradient;

[0051] Update the weights and thresholds of hidden layer nodes and output layer nodes.

[0052] Preferably, the input amount of the input layer of the multi-layer neural network model is 43 parameters, including electromagnetic parameters: EM1, ..., EM41, a total of 41 items; the distance between the probe and the lower surface of the strip: Gap; and the thickness of the strip;

[0053] The multi-layer neural network model has 8 to 15 hidden layers.

[0054] The output of the output layer of the multi-layer neural network model is 1 parameter, namely the bake hardening value BH2.

[0055] The transfer function f1 of the hidden layer and the transfer function f2 of the output layer are both S-type transfer functions or linear transfers.

[0056] The S-type function is a logarithmic S-type function or a hyperbolic tangent S-type function.

[0057] The online detection method for the bake hardening value of the bake hardened cold-rolled thin plate provided by the present invention can be used to perform online detection on the bake hardening value of the online running strip steel, and obtain high-density data values ​​of the entire length of the strip steel. Within the relative error accuracy range of 10%, the sample qualification rate is above 90%. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Other features, objectives and advantages of the present invention will become more apparent from a reading of the detailed description of non-limiting embodiments made with reference to the following accompanying drawings.

[0059] Figure 1 It is a schematic diagram of BH2 calculation according to the prior art;

[0060] Figure 2 It is a system structure for the on-line detection method of the bake hardening value of the bake hardened cold rolled sheet of the present invention;

[0061] FIG3 is an electromagnetic detection unit arranged at a detection site, wherein FIG3(a) corresponds to an online working position, and FIG3(b) corresponds to a descending position;

[0062] Figure 4 is the topological structure of the BP neural network according to the present invention;

[0063] Figure 5 It is the learning algorithm flow chart of BP neural network;

[0064] Figure 6 It is the bake hardening value (BH2) distribution of a coil of steel along its entire length. DETAILED DESCRIPTION

[0065] The technical solution of the present invention is described in detail below in conjunction with the accompanying drawings.

[0066] The detailed features and advantages of the present invention are described in detail in the specific implementation manner, and the content is sufficient to enable any technical personnel in this field to understand the technical content of the present invention and implement it accordingly. According to the description, claims and drawings disclosed in this specification, those skilled in the art can easily understand the relevant purposes and advantages of the present invention.

[0067] System Configuration

[0068] Figure 2 The system structure of the method for online detection of bake hardening value of bake hardened cold rolled sheet of the present invention is shown. The detection system includes an online detection hardware system, a matching software system, a mathematical model, and corresponding data interfaces and computer networks.

[0069] The system includes a detection sensor unit and a calculation and control unit, wherein the detection sensor unit is connected to the calculation and control unit via a data communication interface.

[0070] The detection sensing unit includes an electromagnetic detection unit, a distance meter and a thickness meter. The electromagnetic detection unit detects the electromagnetic response parameters of the strip and transmits the detected electromagnetic parameter data to the calculation and control unit; the distance meter detects the distance between the electromagnetic detection unit and the lower surface of the strip and sends the detected distance data to the calculation and control unit; the thickness meter detects the thickness of the strip and sends the detected thickness data to the calculation and control unit.

[0071] The calculation and control unit also includes a detection sensor unit controller for controlling the electromagnetic detection unit, and a probe lifting and lateral movement control device for controlling the lifting and lateral movement of the electromagnetic detection unit.

[0072] FIG3 shows an electromagnetic detection unit installed at the detection site. As shown in FIG3 , the steel strip 1 usually runs at a speed of 0-300 m / min, and the steel strip passes through two rollers 2 arranged in front and behind to achieve a stable running trajectory of the steel strip. An electromagnetic detection unit 3 that can be raised and lowered and moved transversely in the width direction is arranged between the rollers 2. The electromagnetic detection unit 3 is placed under the running steel strip, and its raising and lowering and transverse movement are achieved by the control system. The detection system also includes a distance meter 4, which measures the distance between the electromagnetic detection unit 3 and the lower surface of the steel strip 1 in real time and sends it to the control computer. The probe lifting device 5 realizes the up and down movement of the electromagnetic detection unit 3, and the mechanical limit device 6 ensures a safe distance between the electromagnetic detection unit 3 and the steel strip 1.

[0073] In particular, the distance 7 between the lower surface of the strip and the probe surface is a key parameter. Due to the jitter of the strip during operation and the fluctuation of the inherent plate properties of the thin strip, the spacing of the strip fluctuates slightly. It is measured in real time by the distance meter 4. Its target value is 4mm and the allowable fluctuation range is ±2mm. This parameter is called Gap, which is used as an input of the detection mathematical model. It should be particularly noted that when the measurement spacing G is 2mm≤G≤6mm, the measurement is valid and the detection result can be corrected. When G>6mm or G<2mm, the system is in an abnormal state, the detection conditions are not met, and the detection is invalid. In Figure 3, Figure 3(a) and Figure 3(b) respectively show that the electromagnetic detection unit is in the online position and the descending position.

[0074] How it works

[0075] The electromagnetic detection unit detects the physical parameters of the strip. In this technical solution, the detection system comprehensively applies four electromagnetic detection methods, namely, tangent magnetic field harmonic analysis, Barkhausen noise detection, incremental magnetic permeability, and multi-frequency eddy current. The above four detection devices are mature products and can be developed and applied after being purchased from the market. Since the output of each electromagnetic method is a curve signal, in order to facilitate application, the result curves of the above four electromagnetic detections are defined and converted into several quantitative parameters for characterization. The details are shown in Table 1-4.

[0076] Table 1 Excitation magnetic field tangential harmonic response parameters (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]

[0081] Table 3 Incremental permeability electromagnetic detection response parameters (7 items in total, EMi, i = 19, ..., 25)

[0082]

[0083] Table 4 Multi-frequency eddy current electromagnetic testing response parameters (16 items in total, EMi, i = 20, ..., 41)

[0084]

[0085] In summary, the integrated detection system outputs up to 41 electromagnetic parameters.

[0086] The elongation after break is represented by "BH2". The calculation process, online detection method and mathematical model of BH2 are described as follows:

[0087] According to the present invention, the online prediction of the bake hardening value of the steel strip is realized based on the BP neural network in the artificial neural network (ANN).

[0088] BP neural network algorithm principle

[0089] BP (Back Propagation) neural network algorithm is a multi-layer feedforward network that learns and trains according to the error back propagation algorithm. It was first proposed by scientists such as Rumelhart and McCelland in 1986. Its network topology is as follows: Figure 6 As shown in the figure, the network structure consists of three parts: input layer, hidden layer, and output layer. There is only one input layer and output layer, but there can be multiple hidden layers. Each layer is composed of several neurons. The neuron nodes in the layer are not connected to each other, while the neuron nodes between layers are fully connected, that is, any neuron node in the input layer is connected to all nodes in the hidden layer, and any neuron node in the hidden layer is connected to all nodes in the output layer.

[0090] The following will take a hidden layer BP neural network as an example to introduce the basic principles of the BP algorithm.

[0091] Figure 6 Shows the topological structure of the BP neural network.

[0092] Figure 6 In the input layer, the input of each neuron node is:

[0093] x=(x0,x1,…,x n-1 )

[0094] That is, x=(x0,x1,…,x n-1 ) is the input vector, x i ,i=(0,1,…,n-1) represents the input of the ith neuron in the input layer, b j ,j=(0,1,…,l-1) is the output of the hidden layer neuron node, y k , k = (0, 1, ..., m-1) is the output of the output layer node, n, l, m represent the number of neuron nodes in the input layer, hidden layer, and output layer respectively, v ij represents the weight from the i-th neuron node in the input layer to the j-th node in the hidden layer, w jk Represents the weight from the jth node in the hidden layer to the kth node in the output layer.

[0095] The output of each neuron node in the hidden layer is

[0096]

[0097] The output of each neuron node in the output layer is

[0098]

[0099] Among them, θ j ,φ k Respectively represent the thresholds of hidden layer nodes and output layer nodes;

[0100] f1 and f2 represent the transfer functions of the hidden layer and the output layer respectively.

[0101] Assume that the expected output result of the output layer node is O k , then the total error

[0102]

[0103] The purpose of error back propagation is to continuously reduce the total error by adjusting the weights. Therefore, the weights should be adjusted along the negative gradient direction of the error, that is, the change in the weights should satisfy the following formula:

[0104]

[0105]

[0106] In the above two formulas, η is a constant, called the learning rate, which is usually 0<η<1.

[0107] Combining the above formulas, the weight adjustment formulas for the hidden layer and output layer are:

[0108]

[0109] Similarly, the threshold adjustment formulas for the hidden layer and output layer are:

[0110]

[0111] After calculating the new weights and thresholds according to the weight and threshold adjustment formula, a new round of forward propagation process begins.

[0112] The above derivation process is for a single training sample, based on a single hidden layer BP neural network. Its basic idea is also applicable to multi-hidden layer BP neural networks. When calculating the adjustment of weights and thresholds, the output layer is recursively pushed forward to the first hidden layer through the intermediate hidden layers. The error calculation method for multiple training samples is the cumulative sum of the errors of each single sample.

[0113] In summary, the execution process of the BP neural network learning algorithm can be summarized into two main parts: forward propagation of signals and back propagation of errors. For each training sample, during forward propagation, the input vector is transmitted from the input layer to the output layer layer by layer; during back propagation, the error is transmitted from the hidden layer to the input layer layer by layer in some form. These two processes are repeated until the set termination condition is met.

[0114] Figure 5 Shows the BP learning algorithm flow chart.

[0115] The factors that need to be considered in the design of a complete BP neural network model include the design of the network structure, the selection and setting of relevant parameters, etc.

[0116] (1) Design of network structure

[0117] There is only one input layer and output layer, and the number of neuron nodes in the input layer and output layer is determined by the actual problem to be solved and the data representation method. Therefore, 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.

[0118] When designing the number of hidden layers, we usually consider setting up one hidden layer first, and only consider increasing the number of hidden layers when one hidden layer cannot meet the relevant performance requirements. Experimental studies have shown that when two hidden layers are used, if the number of neuron nodes in the first hidden layer is greater than the number of nodes in the second hidden layer, the performance of the BP network will be improved to a certain extent. However, for some practical problems, the number of neuron nodes required for double hidden layers may be less than the number of nodes in a single hidden layer. The more hidden layers there are, the longer the BP network training and learning time will be, and the local minimum error will also increase, that is, the probability of the network falling into a local optimal solution will increase. Therefore, only when increasing the number of hidden layer nodes cannot significantly improve network performance, should we consider increasing the number of hidden layers.

[0119] There is currently no mature theoretical basis for determining the number of hidden layer neuron nodes, and it is generally set based on experience. The following empirical estimation formulas can be used as reference:

[0120]

[0121] m=log2 n

[0122]

[0123] Where m represents the number of hidden layer neuron nodes, n represents the number of input layer nodes, l represents the number of output layer nodes, and a is a constant between [1,10].

[0124] The selection of the number of hidden layer neuron nodes is the key to the success of network design. If the number of hidden layer nodes is too small, the network's learning ability from the training samples will be poor, and the learned rules will not be enough to reflect the rules of the entire sample set, and the prediction error of the test sample will increase; but if the number is too large, the training time will be greatly extended, and the network may also learn to record irregular content in the sample (such as noise), resulting in "overfitting" problems and reducing the generalization ability of the network.

[0125] (2) Parameter selection

[0126] The selection of BP neural network parameters includes transfer function, learning algorithm, initial weights and thresholds, as well as parameters such as number of iterations, learning rate, and training target error.

[0127] 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 the function expression is f(x) = 1 / (1+e -x ). The S-type function has the characteristics of nonlinearity and being differentiable everywhere, and has a good gain control on the signal: when the |x| value is small, f(x) has a large gain; when the |x| value is large, f(x) has a small gain, which can prevent the network from entering a saturation state to a certain extent, so this function has been widely used. In addition, the more common transfer functions include the logarithmic S-type transfer function (logsid), the hyperbolic tangent S-type transfer function (tangsig), and the linear transfer function. This patent selects the most commonly used sigmoid function.

[0128] Technical requirements for a neural network model for online inspection of bake-hardened steel

[0129] Model: BP multi-layer feedforward neural network:

[0130] Model structure: three-layer structure: input layer, hidden layer, output layer, among which

[0131] Model input layer: 43 parameters, including electromagnetic parameters: EM1, ..., EM41, a total of 41 items; the distance between the probe and the lower surface of the strip: Gap (satisfying the condition 4≤Gap≤6), and the thickness of the strip.

[0132] The number of hidden layers is 8-15, which is selected based on the data during model training.

[0133] Output layer one parameter: bake hardening layer

[0134] The transfer function between model layers uses the Sigmoid function.

[0135] The model training phase includes error analysis and algorithm flow. Figure 5 After the model training is completed and the error analysis meets the technical requirements, the model can be installed in the online system for online calculation.

[0136] The above method can be used to detect the bake hardening value of the online running strip steel on-line, and obtain the high-density data value of the full length of the strip steel. Within the relative error accuracy range of 10%, the sample qualification rate is above 90%.

[0137] Example

[0138] The method of the present invention was applied on a production line, and the bake hardening value of a coil of steel strip was obtained by using the online detection technical solution of the present invention. The main parameters of the artificial neural network are as follows:

[0139] Model: BP multi-layer feedforward neural network:

[0140] Model structure: three-layer structure: input layer, hidden layer, output layer, among which

[0141] Model input layer: 43 parameters, including electromagnetic parameters: EM1, ..., EM41, a total of 41 items; the distance between the probe and the lower surface of the strip: Gap (satisfying the condition 4≤Gap≤6); strip thickness.

[0142] There are 12 hidden layers, which are selected based on the data during model training.

[0143] Output layer one parameter: bake hardening value

[0144] The transfer function between model layers uses the Sigmoid function.

[0145] First, in the model training phase, the bake-hardened steel produced by the production line in the past period of time is used as the training sample. The specific input parameters and output parameters of the model are expressed as follows:

[0146] Table 5 Bake hardening value (BH2) online detection BP neural network model training data set (partial)

[0147]

[0148] The developed model is used for model testing, as shown in the following table.

[0149] Table 6 Model input parameters in the training phase (partial)

[0150] Test-1 Test-2 Test-3 Test-4 Test-5 EM01 1.5867 1.74 1.508 1.6427 1.7067 EM02 0.4264 0.5114 0.3935 0.4131 0.488 EM03 0.334 0.3928 0.3173 0.3413 0.3722 EM04 0.268 0.3069 0.2499 0.2785 0.2894 EM05 2.7518 2.7913 2.7454 2.8146 2.7912 EM06 2.8738 2.8464 2.8677 2.822 2.8524 EM07 2.0196 2.0366 1.9925 1.9664 2.0251 EM08 1.2356 1.4543 1.1575 1.2378 1.3694 EM09 4.739 5.1476 4.6327 4.6176 4.9726 EM10 -5.2785 -6.7428 -4.8621 -5.1216 -6.2742 EM11 0.2009 0.2448 0.1916 0.1654 0.2181 EM12 0.1131 0.1223 0.1194 0.1243 0.114 EM13 0.1017 0.112 0.1082 0.108 0.1006 EM14 0.1103 0.1127 0.1074 0.1165 0.0991 EM15 4.0538 5.0019 -11.546 4.0147 2.1656 EM16 20.841 26.321 11.802 6.0136 4.735 EM17 10.625 16.61 4.6146 2.8907 2.7001 EM18 4.6048 3.9271 3.4466 1.8103 1.6999 EM19 0.022 0.0198 0.0204 0.0188 0.0209 EM20 0.0086 0.0081 0.0083 0.0078 0.0089 EM21 0.0217 0.0196 0.0197 0.0183 0.0207 EM22 0.9207 0.7451 1.6482 1.3598 0.5917 EM23 30.905 34.105 31.529 32.722 34.39 EM24 19.425 21.835 20.448 21.297 22.469 EM25 11.287 12.416 11.644 12.102 13.001 EM26 0.0305 0.0306 0.0343 0.0305 0.0292 EM27 -0.2401 -0.2404 -0.244 -0.2405 -0.2395 EM28 0.242 0.2424 0.2464 0.2424 0.2413 EM29 -1.4443 -1.4442 -1.4309 -1.4446 -1.4494 EM30 0.391 0.3914 0.4015 0.3915 0.3887 EM31 -0.3481 -0.348 -0.3481 -0.3481 -0.3483 EM32 0.5235 0.5237 0.5313 0.5239 0.5219 EM33 -0.7274 -0.7268 -0.7143 -0.7267 -0.7307 EM34 0.4455 0.4433 0.4409 0.4432 0.4451 EM35 0.8279 0.827 0.8392 0.8273 0.8255 EM36 0.9401 0.9383 0.9479 0.9385 0.9378 EM37 1.0771 1.0787 1.0871 1.079 1.0763 EM38 -0.5036 -0.5039 -0.5103 -0.5041 -0.5028 EM39 0.7974 0.7955 0.7971 0.7953 0.7962 EM40 0.9431 0.9417 0.9465 0.9416 0.9417 EM41 2.1341 2.1354 2.1403 2.1358 2.1341 GAP 5.08 4.94 5.1 4.98 4.94 THK 0.7 0.6 0.67 0.6 0.65

[0151] Table 7 Model input parameters in the training phase (partial)

[0152] Test-1 Test-2 Test-3 Test-4 Test-5 Model calculated value (MPa) 37.8 28.4 43.5 29.2 36.0 Sample tensile value (MPa) 38.4 29.3 41.2 32.2 34.5 Error(MPa) 0.6 0.9 -2.3 3.0 -1.5 Relative error (%) 1.6 3.1 5.5 9.2 4.4

[0153] Comparison of the model calculation values ​​and offline test values ​​of five samples of bake hardening steel shows that the qualified rate is 100% according to the requirement of 10% relative error, indicating that the model meets the technical requirements for online detection of bake hardening value (BH2).

[0154] The model was used for real-time testing of a coil of steel strip, and the test results of the bake hardening value (BH2) in the full length direction are as follows. Compared with the existing technology that can only test by shearing samples, the data volume and real-time performance are greatly improved.

[0155] The invention can be fully extended to the online quality detection system of the baking hardening value of cold-rolled strip steel, realizing the continuous detection, classification and recording of the production quality of steel plates, which will play a very positive role in improving production efficiency, product quality and product competitiveness.

[0156] Finally, it should be pointed out that although the present invention has been described with reference to the current specific embodiments, ordinary technicians in this technical field should realize that the above embodiments are only used to illustrate the present invention and are not used as a limitation of the present invention. Various equivalent changes or substitutions can be made without departing from the concept of the present invention. Therefore, as long as the changes and modifications to the above embodiments are within the essential spirit of the present invention, they will fall within the scope of the claims of the present invention.

Claims

1. An online detection method for the bake hardening value of a bake hardened cold rolled sheet, characterized in that: The following steps are involved: Online measurement of a set of electromagnetic parameters of the strip, the distance between the probe and the lower surface of the strip, and the thickness of the strip; The set of electromagnetic parameters includes 41 electromagnetic 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; The bake hardening value BH2 of the strip steel is calculated using a multi-layer neural network model. The multi-layer neural network model has a three-layer structure, including an input layer, a hidden layer, and an output layer. The input quantity of the input layer is 43 parameters, including electromagnetic parameters: EM1, ..., EM41, a total of 41 items; the distance between the probe and the lower surface of the strip: Gap; and strip thickness; The input of each neuron node in the input layer is x=(x0,x1,…,x n-1 ) The output of each neuron node in the hidden layer is The output of each neuron node in the output layer is Among them, x i ,i=(0,1,…,n-1) represents the input of the ith neuron in the input layer, j＝(0,1,…,l-1)， k=(0,1,…,m-1), n, l, and m represent the number of neuron nodes in the input layer, hidden layer, and output layer, respectively; v ij Represents the weight from the i-th neuron node in the input layer to the j-th node in the hidden layer; w jk Represents the weight from the jth node in the hidden layer to the kth node in the output layer; θ j ,φ k Respectively represent the thresholds of hidden layer nodes and output layer nodes; f1 and f2 represent the transfer functions of the hidden layer and the output layer respectively.

2. The method for online detection of bake hardening value according to claim 1, characterized in that: The learning algorithm of the multi-layer neural network model is: Set the input vector and output vector; Initialize the weights and thresholds of hidden layer nodes and output layer nodes; Calculate the output of each node in the hidden layer and output layer; Calculate the error between the expected output and the actual output; Determine whether the error satisfies the requirement. If yes, end the learning process. If no, execute the following process. Calculate the error of each unit in the hidden layer and output layer; Calculate the error gradient; Update the weights and thresholds of hidden layer nodes and output layer nodes.

3. The method for online detection of bake hardening value according to claim 1, characterized in that: The hidden layer is 8-15 layers; The output of the output layer is one parameter, namely, the bake hardening value BH2.

4. The method for online detection of bake hardening value according to claim 1, characterized in that: The transfer function f1 of the hidden layer and the transfer function f2 of the output layer are both S-type transfer functions.

5. The method for online detection of bake hardening value according to claim 4, characterized in that: The S-type transfer function is a logarithmic S-type function or a hyperbolic tangent S-type function.

Images