Method for determining ultrasonic thickness measurement frequency of steel

Abstract

The invention discloses a method for determining the ultrasonic thickness measurement frequency of steel, which comprises the following steps of: 1, establishing a model: 1.1, determining various chemical elements influencing the thickness measurement frequency in the steel; step 1.2, determining corresponding thickness measuring frequencies for steel materials of different materials, and combining the thickness measuring frequencies with the chemical elements in the step 1.1 to establish a database; 1.3, establishing a neural network model between the chemical elements of the steel and the thickness measurement frequency, and obtaining the thickness measurement frequency of the steel according to the chemical element content of the steel; 1.4, training the neural network model in the step 3 by using the database in the step 1.2 so as to improve the precision of the neural network model; the method has the advantages that the most suitable ultrasonic thickness measuring frequency of different steel types is predicted by means of calculation of a large amount of data, a basis is provided for measurement of the pipe fitting, and the size measurement precision of the pipe fitting is improved.

Description

Technical Field

[0001] This invention relates to the field of steel thickness testing technology, and in particular to a method for determining the ultrasonic thickness measurement frequency of steel. Background Technology

[0002] As industrial manufacturing demands increasing efficiency, quality, and consistency, improving dimensional accuracy has become a crucial aspect of quality improvement in modern manufacturing industries (automobiles, ships, pressure vessels, construction machinery, aerospace, etc.). Strengthening raw material quality control is increasingly becoming a key control point in the manufacturing industry, and good raw material quality will lay a solid quality foundation for equipment manufacturing.

[0003] In the raw material acceptance process, dimensional measurement is a crucial basis for ensuring the shape requirements of raw materials. How to correctly and accurately complete the measurement work is a decisive factor in quality acceptance. Currently, the thickness of conventional round pipes can be measured using tools such as vernier calipers. However, for workpieces with more complex shapes, their shape and other factors limit measurement methods. Ultrasonic thickness gauges are commonly used for measuring the wall thickness of pipe fittings. In practical applications, it has been found that thickness gauges have errors in measuring different metal materials. The fundamental reason is that different metals have different ultrasonic frequencies. Currently, research on suitable ultrasonic frequencies for different metal materials is limited to the main material of the metal (such as aluminum, steel, copper, etc.). However, in reality, the optimal ultrasonic thickness measurement frequency is also different for different steel grades (with different elemental contents).

[0004] Currently, research on different steel grades and the most suitable ultrasonic thickness measurement frequencies for each steel grade is lacking. No relevant technical solutions have been found in existing patent documents either. Summary of the Invention

[0005] The purpose of this invention is to provide a method for determining the ultrasonic thickness measurement frequency of steel, which can predict the optimal ultrasonic measurement frequency for steel grades and continuously improve the prediction accuracy.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: A method for determining the ultrasonic thickness measurement frequency of steel includes the following steps: Step 1: Establish the model Step 1.1: Identify the various chemical elements in the steel that affect its thickness measurement frequency; Step 1.2: Confirm the corresponding thickness measurement frequency for steel of different materials, and establish a database by combining it with the chemical elements in Step 1.1; Step 1.3: Establish a neural network model between the chemical elements of steel and the thickness measurement frequency, and obtain the thickness measurement frequency by the chemical element content of steel; Step 1.4: Use the database from Step 1.2 to train the neural network model from Step 1.3, thereby improving the accuracy of the neural network model; Step 1.5: Based on the content of each chemical element in the new steel, use the neural network model in Step 1.3 to predict the thickness measurement frequency of the new steel; Step 1.6: Confirm the thickness measurement frequency of the new steel described in Step 1.5, and compare it with the predicted data in Step 1.5 to further improve the model accuracy; Step 2: Based on the model, input the chemical element content of the steel to obtain its most suitable thickness measurement frequency.

[0007] Preferably, in step 1.3, the calculation process for the thickness measurement frequency x is as follows: The input values ​​in the input layer are weighted and then introduced into the hidden layer: z_j = ∑(w_ij * x_i) + b_j The activation value is obtained through the activation function in the hidden layer, and the activation value of the hidden layer is passed to the output layer; a_j = tanh(z_j) = (e^ z_j - e^(- z_j)) / (e^ z_j + e^(- z_j)) The activation values ​​of the output layer are obtained through a linear function. y = ∑(w_j * a_j) + b_output The predicted thickness measurement frequency x is output through the output layer function; x = purelin(y) Where: x_i is the input value, that is, the content of the i-th chemical element in the steel; w_ij represents the weights from the input layer to the hidden layer; b_j is the hidden layer bias term; z_j is a weighted assignment; a_j is the activation value of the hidden layer; tanh(z_j) is the activation function of the hidden layer; w_j is the weight of the j-th neuron in the hidden layer; The constant term generated by the b_output model during data processing; y is the activation value of the output layer; purelin(y) is the output layer function.

[0008] Preferably, in step 1.4, the method to improve the accuracy of the neural network model mainly involves correcting w_ij, and the correction method is as follows: The backpropagation algorithm for updating w_ij is as follows: Δw = -η * ∂E / ∂w The chain rule for calculating the gradient is as follows, for weights w_ij: ∂E / ∂w_ij = ∂E / ∂y * ∂y / ∂z_j * ∂z_j / ∂w_ij Where: Δw is the change in weight; η is the learning rate; ∂E / ∂w is the partial derivative of the loss function with respect to the weights.

[0009] Preferably, in step 1.6, the method for improving model accuracy is the mean square error method, specifically: MSE = (1 / n) * ∑(y_pred - y_true)² Where: n is the number of samples used; y_pred is the model's predicted value; y_true is the true value used in the training data.

[0010] Preferably, in step 1.6, the method for improving model accuracy is the root mean square error method, specifically: RMSE = √[(1 / n) * ∑(y_pred - y_true)²].

[0011] The beneficial effects of this invention are as follows: 1. By using a large amount of data for calculation, the most suitable ultrasonic thickness measurement frequency for different steel grades can be predicted, providing a basis for pipe fitting measurement and improving the dimensional measurement accuracy of pipe fittings.

[0012] 2. Based on the prediction results, the thickness of the steel is measured, and the inner diameter shape model of the steel is established to provide data support for further machining and reduce waste generated in mechanical design. Attached Figure Description

[0013] Figure 1 This is the interface for running the neural network model in this method; Figure 2 for Figure 1 A schematic diagram of the left side of the image; Figure 3 for Figure 1 A schematic diagram of the input interface.

[0014] The accompanying drawings are for illustrative purposes only and should not be construed as limiting the scope of this patent. To better illustrate this embodiment, some components in the drawings may be omitted, enlarged, or reduced, and do not represent the actual dimensions of the product. It is understandable to those skilled in the art that some well-known structures and their descriptions may be omitted in the drawings. Detailed Implementation

[0015] The present invention will now be further described with reference to the accompanying drawings. Example

[0016] This embodiment of a method for determining the ultrasonic thickness measurement frequency of steel includes the following steps: Step 1: Establish the model Step 1.1: Identify the various chemical elements in the steel that affect its thickness measurement frequency, such as C, Si, Mn, etc.

[0017] Step 1.2: Determine the most suitable thickness measurement frequency for steel of different materials, and establish a database by combining it with the chemical elements in Step 1.1; This method is mainly for steel with cross-sectional shapes such as round and square. For example, for a round steel pipe, first measure its thickness with a vernier caliper; then measure its thickness with an ultrasonic thickness gauge (using different frequencies). When the thickness measured at a certain frequency is consistent with the thickness measured by the vernier caliper, it can be concluded that this frequency is the most suitable frequency for the ultrasonic thickness gauge to measure this type of steel.

[0018] Step 1.3: Use MATLAB software to establish a neural network model between the chemical elements of steel and the thickness measurement frequency, and obtain the thickness measurement frequency by the chemical element content of steel. In this neural network model, the calculation process for the thickness measurement frequency x is as follows: The input values ​​in the input layer are weighted and then introduced into the hidden layer: z_j = ∑(w_ij * x_i) + b_j The activation value is obtained through the activation function in the hidden layer, and the activation value of the hidden layer is passed to the output layer; a_j = tanh(z_j) = (e^ z_j - e^(- z_j)) / (e^ z_j + e^(- z_j)) The activation values ​​of the output layer are obtained through a linear function. y = ∑(w_j * a_j) + b_output The predicted thickness measurement frequency x is output through the output layer function; x = purelin(y) Where: x_i is the input value, that is, the content of the i-th chemical element in the steel; w_ij represents the weights from the input layer to the hidden layer; b_j is the hidden layer bias term; z_j is a weighted assignment, representing the net input of the j-th neuron. It is a linear combination of the input signals and represents the total stimulus intensity received by the neuron. a_j is the activation value of the hidden layer, which is the output of the j-th neuron. z_j is nonlinearly transformed by the activation function tanh(z_j) and becomes the input of the next layer. tanh(z_j) is the activation function of the hidden layer, which is the hyperbolic tangent function; w_j is the weight of the j-th neuron in the hidden layer; The constant term generated by the b_output model during data processing; y is the activation value of the output layer; purelin(y) is the output layer function.

[0019] The above model is mainly based on the mathematical principles of feedforward neural networks, using the network topology as the basic structure of the neural network. Nine neurons are introduced in the input layer to correspond to the nine chemical elements, the hidden layer contains ten neurons (a single hidden layer), and the output layer introduces one neuron to output the predicted value.

[0020] Step 1.4: Use the database from Step 1.2 to train the neural network model from Step 1.3, thereby improving the accuracy of the neural network model; One method to improve the accuracy of neural network models is to correct w_ij, and the correction method is as follows: The backpropagation algorithm for updating w_ij is as follows: Δw = -η * ∂E / ∂w The chain rule for calculating the gradient is as follows, for weights w_ij: ∂E / ∂w_ij = ∂E / ∂y * ∂y / ∂z_j * ∂z_j / ∂w_ij Where: Δw is the change in weight; η is the learning rate; ∂E / ∂w is the partial derivative of the loss function with respect to the weights.

[0021] Step 1.5: Based on the content of each chemical element in the new steel, use the neural network model in Step 1.3 to predict the thickness measurement frequency of the new steel; Step 1.6: Confirm the thickness measurement frequency of the new steel in Step 1.5 and compare it with the predicted data in Step 1.5 to further improve the model accuracy, which can improve the measurement accuracy from the current 1mm to 0.1mm.

[0022] The methods to further improve model accuracy are Mean Squared Error (MSE) and Root Mean Squared Error (RMSE), specifically: MSE = (1 / n) * ∑(y_pred - y_true)² RMSE = √[(1 / n) * ∑(y_pred - y_true)²].

[0023] Where: n is the number of samples used; y_pred is the model's predicted value; y_true is the true value used for training data, that is, the most suitable thickness measurement frequency obtained by using the method in step two.

[0024] After predicting the optimal thickness measurement frequency for a steel grade through the aforementioned steps, for more complex steel materials, the cross-sectional shape of the steel can be established through precise thickness measurement, providing a basis for subsequent machining of the workpiece.

[0025] Step 2: Based on the model established in Step 1, input the chemical element content of the steel to predict its most suitable thickness measurement frequency.

[0026] This embodiment utilizes extensive data calculations to predict the optimal ultrasonic thickness measurement frequency for different steel grades, providing a basis for pipe fitting measurement and improving the accuracy of pipe fitting dimensional measurements. Furthermore, based on this, an inner diameter shape model of the steel can be established, providing data support for further machining and reducing waste generated in mechanical design.

[0027] Experimental Example This experimental example demonstrates the calculation process of the neural network model using the calculation process of two different steel grades as examples, based on the method of the aforementioned embodiments.

[0028] The table below shows the chemical element content of two steel grades, in units of %.

[0029] In this embodiment, the chemical elements are mainly divided into nine types: C, Si, Mn, Cr, W, Mo, Ni, Cu, and Co. In other embodiments, the aforementioned nine chemical elements may be added to or subtracted accordingly.

[0030] Table 1 Table 2 Taking the above-mentioned steel grade 1 as an example, after inputting the chemical composition content (x_i) of the steel grade, the model assigns values ​​to the hidden layer neurons through the activation function z_j= ∑(w_ij * x_i) + b_j. In this embodiment, w_ij=0.1 and b_j=0.9.

[0031] z_j = ∑(w_ij * x_i) + b_j = (0.1*0.1+0.1*0.26+0.1*0.48+0.1*2.14+0+0.1*0.93+0.1*0.03+0.1*0.05+0)+0.9 =1.299.

[0032] a_j = tanh(z_j) = (e^ z_j - e^(- z_j)) / (e^ z_j + e^(- z_j)) =[EXP(1.299)-EXP(-1.299]) / [EXP(1.299)+EXP(-1.299)] =0.8615.

[0033] EXP(1.299) is the exponent with the natural logarithm e as the base.

[0034] In this embodiment, w_j=5.5 and b_output=0.

[0035] y = ∑(w_j * a_j) + b_output =5.5*0.8615 =4.7381.

[0036] x = purelin(y) =4.7381. In this embodiment, the unit of the thickness measurement frequency x is kHz.

[0037] The calculation results for each item in the aforementioned steel grade 1 are shown in Table 1.

[0038] Similar to the calculation process for steel grade one, in the calculation process for steel grade two, w_ij=0.1, b_j=0.9, w_j=5.5, and b_output=0. The final predicted value for steel grade two is 4.5398. The calculation results for each item are shown in Table 2.

[0039] Then, based on the formula in step four, repeated data training is performed by adjusting the value of w_ij. In this embodiment, Δw=0.01, then w_ij=0.11. Thus, the predicted value of steel grade one is 4.7927, and the predicted value of steel grade two is 4.5867.

[0040] Next, based on the formula in step six, the accuracy of the model is evaluated using mean squared error (MSE) and root mean square error (RMSE). In this embodiment, n=1, and the resulting data is shown in Table 3 below: Through the above comparison, the result of prediction value one is more accurate, that is, when w_ij=0.1 is used, the result is more accurate.

[0041] The neural network model continuously improves its prediction accuracy by repeating the above calculation process.

[0042] That is, by continuously assigning values ​​to the following constants w_ij, b_j, w_j, b_output, etc. and performing calculations, the predicted value obtained is made to continuously approach the true value y_true, thereby improving the prediction accuracy.

[0043] Once the optimal values ​​for the aforementioned constants are obtained, the most suitable thickness measurement frequency can be predicted by inputting the chemical element content of the steel based on the aforementioned model.

[0044] As per the instruction manual Figure 1 — Figure 3 The image shown is a screenshot of the model created by this software and its usage.

[0045] The above embodiments are not intended to limit the shape, material, structure, etc. of the present invention in any way. Any simple modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention shall fall within the protection scope of the present invention.

[0046] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only used to facilitate the description of this invention and to simplify the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limiting the scope of protection of this invention.

[0047] If the terms "first" or "second" are used in this document to define components, those skilled in the art should know that the use of "first" or "second" is merely for the convenience of describing the invention and simplifying the description, and unless otherwise stated, the above terms have no special meaning.

[0048] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. However, these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for determining the frequency of ultrasonic thickness measurement of steel, characterized in that, Includes the following steps: Step 1: Establish the model Step 1.1: Identify the various chemical elements in the steel that affect its thickness measurement frequency; Step 1.2: Confirm the corresponding thickness measurement frequency for steel of different materials, and establish a database by combining it with the chemical elements in Step 1.1; Step 1.3: Establish a neural network model between the chemical elements of steel and the thickness measurement frequency, and obtain the thickness measurement frequency by the chemical element content of steel; Step 1.4: Use the database from Step 1.2 to train the neural network model from Step 1.3, thereby improving the accuracy of the neural network model; Step 1.5: Based on the content of each chemical element in the new steel, use the neural network model in Step 1.3 to predict the thickness measurement frequency of the new steel; Step 1.6: Confirm the thickness measurement frequency of the new steel described in Step 1.5, and compare it with the predicted data in Step 1.5 to further improve the model accuracy; Step 2: Based on the model, input the chemical element content of the steel to obtain its most suitable thickness measurement frequency.

2. The method for determining the ultrasonic thickness measurement frequency of steel according to claim 1, characterized in that, In step 1.3, the calculation process for the thickness measurement frequency x is as follows: The input values ​​in the input layer are weighted and then introduced into the hidden layer: z_j = ∑(w_ij * x_i) + b_j The activation value is obtained through the activation function in the hidden layer, and the activation value of the hidden layer is passed to the output layer; a_j = tanh(z_j) = (e^ z_j - e^(- z_j)) / (e^ z_j + e^(- z_j)) The activation values ​​of the output layer are obtained through a linear function. y = ∑(w_j * a_j) + b_output The predicted thickness measurement frequency x is output through the output layer function; x = purelin(y) Where: x_i is the input value, that is, the content of the i-th chemical element in the steel; w_ij represents the weights from the input layer to the hidden layer; b_j is the hidden layer bias term; z_j is a weighted assignment; a_j is the activation value of the hidden layer; tanh(z_j) is the activation function of the hidden layer; w_j is the weight of the j-th neuron in the hidden layer; The constant term generated by the b_output model during data processing; y is the activation value of the output layer; purelin(y) is the output layer function.

3. The method for determining the ultrasonic thickness measurement frequency of steel according to claim 2, characterized in that, In step 1.4, the main method to improve the accuracy of the neural network model is to correct w_ij. The correction method is as follows: The backpropagation algorithm for updating w_ij is as follows: Δw = -η * ∂E / ∂w The chain rule for calculating the gradient is as follows, for weights w_ij: ∂E / ∂w_ij = ∂E / ∂y * ∂y / ∂z_j * ∂z_j / ∂w_ij Where: Δw is the change in weight; η is the learning rate; ∂E / ∂w is the partial derivative of the loss function with respect to the weights.

4. The method for determining the ultrasonic thickness measurement frequency of steel according to claim 1, characterized in that, In step 1.6, the method to improve model accuracy is the mean square error method, specifically: MSE = (1 / n) * ∑(y_pred - y_true)² Where: n is the number of samples used; y_pred is the model's predicted value; y_true is the true value used in the training data.

5. The method for determining the ultrasonic thickness measurement frequency of steel according to claim 1, characterized in that, In step 1.6, the method to improve model accuracy is the root mean square error method, specifically: RMSE = √[(1 / n) * ∑(y_pred - y_true)²].

Images