Online welding process parameter setting method for lap welding of cold-rolled strip steel

Abstract

The invention discloses an online welding process parameter setting method for lap welding of cold-rolled strip steel, which adopts a random forest model to analyze welding historical data, and realizes the setting of welding parameters for known input parameters (strip steel type and thickness) and applicable welding machine process parameters output. According to the method, lap welding parameter recommendation and lap welding result optimization are completed by obtaining the steel grade, thickness chemical components (C, Mn, Cr and Cu) and the like of the running strip steel in advance, so that the welding process is truly digitalized, the situation that in the traditional welding quality evaluation process, subjective judgment and auxiliary manual tests are relied on to a large extent is changed, and the welding quality evaluation efficiency is greatly improved. Knowledge acquisition and inheritance are one-sided, and effective guidance cannot be provided for welding quality quantitative evaluation and welding machine maintenance.

Description

Technical Field

[0001] The invention relates to a thin strip steel welding control technology, and in particular to an online welding process parameter setting method for cold-rolled strip steel lap welding. Background Art

[0002] At present, there are two main types of cold-rolled strip welding machines: laser welding and lap welding. Laser welding is often used in pickling mills with thicker incoming materials, and this part of the production line is usually equipped with a relatively complete weld quality detection system. Most cold-rolled post-processing units outside the pickling mill, such as continuous annealing, hot-dip galvanizing, and electrogalvanizing (tin, chromium) units, mostly use narrow lap resistance welding machines. This type of lap welding machine is usually equipped with a relatively simple QCDS system (original welding equipment system).

[0003] There are two general ways to determine the welding parameters of lap welding: First, roughly locate the welding parameters according to the welding history in the established steel type (determine the welding parameter table), and then perform regression analysis. However, with this method, the model span is large, and the given welding process parameters have large deviations, which can easily lead to poor weld quality. Second, when the unit is shut down or directly shut down, the welding process parameters of the cold-rolled strip to be welded are simulated in real time, and then supplemented by manual inspection. However, this method is time-consuming, affects the production rhythm of the unit, and relies on manual labor, which is prone to deviations. Summary of the invention

[0004] In order to solve the above problems in the prior art, the present invention provides an online welding process parameter setting method for cold-rolled strip lap welding, which is based on a random forest algorithm and obtains optimized welding process parameters according to the steel grade, thickness and chemical composition of the strip to be welded.

[0005] A method for setting online welding process parameters for cold-rolled strip lap welding comprises the following steps:

[0006] S1. Collect historical welding performance data, including the thickness of the previous and subsequent coils, steel grade, material chemical composition as input parameters of the random forest algorithm to be screened, and corresponding welding process parameters as target parameters, wherein the welding process parameters include welding current, welding speed, welding wheel pressure, overlap amount, and overlap compensation amount;

[0007] S2. Based on Python, calculate the Pearson correlation coefficient and draw a heat map;

[0008] S3. Select thickness, steel grade, and material chemical composition that have high correlation with welding parameters according to the thermal map for random forest algorithm training;

[0009] S4. A welding parameter setting model is established based on a random forest algorithm, and the steel grade, thickness, and material chemical composition of the cold-rolled steel strip to be welded are used as input to calculate and obtain optimized welding process parameters, including welding current, welding wheel pressure, welding speed, overlap amount, and overlap compensation amount;

[0010] S5. Send the optimized welding process parameters to the welding machine and set them accordingly.

[0011] In step S1 or S3, the calculation formula of the random forest algorithm is:

[0012] In the formula, represents the expectation of random parameters of X and data set Dn; r n (X, Θ) is the nth tree, Θ is the expectation.

[0013] In step S2, the formula for calculating the row Pearson correlation coefficient is:

[0014]

[0015] Where, X i and Y i represents two different eigenvectors; n is the number of samples; and The eigenvectors X i and Y i Sample mean; r is the Pearson correlation coefficient, representing the eigenvector X i and Y i The linear correlation of r ranges from -1 to 1. The larger the absolute value, the stronger the X i and Y i The higher the correlation, the higher the correlation, r>0, indicating that X i and Y i Positive correlation, on the contrary, indicates that X i and Y i Negative correlation.

[0016] In step S3, the chemical composition of the material includes C, Mn, Cr, Cu, and Ni.

[0017] The method for setting online welding process parameters for cold-rolled strip lap welding of the present invention has the following advantages:

[0018] 1. By separating the welding history big data and optimizing the welding parameters of cold-rolled strip based on the computational forest algorithm, the welding process parameters of the online strip can be recommended and optimized.

[0019] 2. Provide a supplementary method for the current situation of welding process parameter determination relying on regression analysis and unit shutdown simulation experiment, and further explore the digitalization and unmanned operation of welding process.

[0020] 3. It will play a very positive role in improving production efficiency, product quality and product competitiveness. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 is a principle block diagram of a system adopted by the setting method of the present invention;

[0022] Figure 2 is a top view of the structure of a butt welding specimen of the present invention;

[0023] Figure 3 It is a schematic diagram of the relationship between the random forest model parameters adopted by the present invention and the prediction results. DETAILED DESCRIPTION

[0024] The following is a further description of an online welding process parameter setting method for cold-rolled strip lap welding of the present invention. The system used in the method is as follows: Figure 1 As shown, it includes welding machine system, supporting software system (OLW), random forest model, and corresponding data services and support (such as digital steel coil data platform).

[0025] Here’s how it works:

[0026] Using the on-site production data collected by the digital steel coil as the carrier, the steel grade, thickness, chemical composition (C, Mn, Cr, Cu, etc.) of the strip at the uncoiler (the strip at the rear coil) and the strip in operation (the strip at the front coil) are transmitted to the software system embedded with the random forest algorithm through the data interface, and the recommended welding parameters are output and transmitted to the welding machine through the software system for welding.

[0027] The random forest algorithm is:

[0028] The regression tree corresponds to the partition of the input space (composed of characteristic parameters) and the output of the partitioned unit (microstructure and mechanical properties). We assume that the input space is divided into M units k1, k2, km, then each unit km corresponds to an output target value Cm. When determining the partition of the input space, the minimum square error criterion between the output target value Cm and the true value yn is used to determine the optimal value of the regression decision tree, such as (f(x n ) is the estimate of the regression tree). The optimal value of Cm is the mean of the output yn of all inputs xn on km.

[0029] Take the jth variable xj and its corresponding yj as the segmentation variable and segmentation point. Define the set Ψ1={x n|y n <y j} and the set Ψ2={x n |y n ≥y j}. The formula for finding the optimal segmentation variable j and the optimal segmentation point yj is defined as

[0030]

[0031] In the formula is the optimal value on the set Ψ1, is the optimal value on cell Ψ2. Through the above steps, all input feature variables are traversed to find the optimal segmentation variable j*, form the first pair of segmentation feature space (j*, yj*), and divide the input space into two regions in turn. Then repeat the division process for each region until the stop condition is met, thus generating a decision tree.

[0032] However, a single regression tree is prone to overfitting and other problems. Random forest is a combination of multiple regression trees and belongs to ensemble learning. The random sampling of training samples and the random selection of features are the embodiment of the randomness of random forest. The weighted average model is obtained by calculating multiple different regression trees, just like multiple experts voting by raising their hands, which is the core idea of ​​random forest.

[0033] Therefore, the random forest model can be defined as:

[0034]

[0035] in, represents the expectation of random parameters of X and data set Dn, r n (X, Θ) is the nth tree, Θ is the expectation.

[0036] Based on the above principles, the present invention is achieved by the following measures:

[0037] First, based on the analysis of welding history data

[0038] In the welding history of cold-rolled strip steel, the welding process parameters required for strip steel of different steel grades and thicknesses are significantly different. Reasonable welding process parameters can improve the quality of welds and ensure the production rhythm of strip steel. Therefore, classifying cold-rolled strip steel by steel grade and thickness is helpful for analyzing welding history data. Then continue to sort out the welding history data, and find the welding process parameters of different thicknesses under each steel grade from the welding history data as the sample set of the random forest model. The high quality of sample data helps to improve the model recommendation effect.

[0039] In the existing technical framework, the method of determining welding parameters by regression analysis based on steel type and thickness is essentially a rough classification based on steel grade and thickness, and does not include the chemical composition of the material (C, Mn, Cu, Si, etc., which are closely related to welding quality) in the regression analysis. Based on correlation analysis, the relationship between steel grade, thickness, material chemical composition and welding parameters can be clarified.

[0040] Correlation analysis focuses on discovering the correlation between variables. If a factor is not correlated with welding parameters or has a very low correlation, then this factor is a redundant feature relative to the random forest model and can be excluded as the input of the random forest. This patent uses the Pearson correlation coefficient to explore the relationship between steel grade, thickness, chemical composition and welding parameters to screen the input of the random forest. The Pearson correlation coefficient calculation formula used is:

[0041]

[0042] Where, X i and Y i represents two different eigenvectors; n is the number of samples; and The eigenvectors X i and Y i Sample mean; r is the Pearson correlation coefficient, representing the eigenvector X i and Y i The range of r is from -1 to 1. The larger the absolute value, the higher the correlation between the two eigenvectors. When r>0, it means that the two eigenvectors are positively correlated. Otherwise, it means that the two eigenvectors are negatively correlated.

[0043] A heat map between each parameter and the welding parameters was drawn based on the Pearson correlation coefficient, and then each input parameter was manually traversed and screened, and the parameters with low correlation with the welding parameters were manually deleted, so as to screen the final input parameters of the random forest model.

[0044] The specific steps are as follows:

[0045] First, based on the historical (such as the past six months) welding performance data, including the thickness of the previous and subsequent coils, steel grade, material chemical composition (C, Si, P, S, Mn, Cu, Ni content), and welding process parameters (welding current, welding speed, welding wheel pressure, overlap, overlap compensation); among them, the thickness of the previous and subsequent coils, steel grade, and material chemical composition (C, Si, P, S, Mn, Cr, Cu, Ni content) are the random forest input parameters to be screened, and the welding process parameters of the corresponding welds (welding current, welding speed, welding wheel pressure, overlap, overlap compensation) are the target parameters.

[0046] Based on Python, calculate the Pearson correlation coefficient and draw a heat map as follows Figure 2 As shown, in Figure 2 In the figure, 1 represents the succeeding coil, 2 represents the preceding coil, Grd represents the steel grade, Thic represents the thickness, C, Si, Mn, P, S, Cr, Cu, and Ni represent the chemical composition of the material (carbon, silicon, manganese, phosphorus, sulfur, chromium, copper, and nickel), Curr represents the welding current, P represents the welding wheel pressure, V represents the welding speed, Lap represents the overlap amount, and ConLap represents the overlap compensation amount.

[0047] Depend on Figure 2 It can be seen that the thickness of the previous coil and the next coil has the highest correlation with the welding parameters. Except for the C content, the content of non-metallic elements such as Si has the lowest correlation with the welding parameters. The content of metal elements such as Mn has a high or low correlation with the welding process parameters. At the same time, according to the basic knowledge of welding: in metal welding, the C content and the content of metal elements such as Mn are directly related to the carbon equivalent, and the carbon equivalent directly affects the welding results. Therefore, from a physical point of view, the parameters of non-metallic and metal elements directly related to the carbon equivalent (C, Mn, Cr, Cu, Ni) should be retained.

[0048] According to the comprehensive judgment of Pearson correlation coefficient analysis and physical knowledge analysis (carbon equivalent), non-metallic elements such as Si have the lowest correlation with welding parameters and can be deleted. At the same time, parameters such as steel grade, thickness, chemical composition (C, Mn, Cr, Cu, Ni) are retained for random forest model training.

[0049] Secondly, a welding parameter recommendation model is established based on the random forest algorithm: After the welding parameter recommendation model based on the random forest algorithm is established, the steel grade, thickness, and chemical composition (C, Mn, Cr, Cu) of the cold-rolled strip are used as input, and the output is the welding process parameters (welding current, welding wheel pressure, welding speed, overlap, overlap compensation). The random forest algorithm has many parameters, and adjusting the model parameters helps to improve the accuracy of the model.

[0050] Communicate with the welding machine through OLW software and send optimized process parameters: Before welding starts, OLW software gives optimized welding process parameters of the cold-rolled steel strip to be welded based on the steel grade and thickness of the cold-rolled steel strip to be welded based on the random forest algorithm. OLW communicates directly with the welding machine and sends the optimized welding process parameters to the welding machine. After the welding machine PLC obtains the parameters, it is displayed on the welding machine operation terminal and the corresponding welding is performed.

[0051] Example 1

[0052] Analysis based on welding history data

[0053] In the history of cold-rolled strip welding, the welding process parameters required for strips of different steel grades and thicknesses are significantly different. This is because the carbon equivalent of cold-rolled steel strips of different steel grades is different, which directly affects the welding results. At the same time, the thickness directly affects the heat generation during welding. For example, if the thickness is large and the other parameters remain unchanged, it is easy to cause the defect of incomplete welding. Therefore, reasonable welding process parameters can improve the quality of the weld and ensure the production rhythm of strip steel.

[0054] As shown in Table 1 (partial sample set), the cold-rolled strip is classified by steel grade and thickness, which is helpful for the analysis of welding history data. Then the welding history data is sorted out, and the thickness and chemical composition of each steel grade are found from the welding history data as the sample set of the random forest model. The high quality of sample data helps to improve the model recommendation effect, so the input parameters need to be analyzed.

[0055] Table 1 Welding history data

[0056] ID Front thickness Back thickness Speed ​​m / min Pressure KN Welding current A Overlap mm Compensation mm 11 0.3 0.3 15 4 10 1 0.5 11 0.3 0.4 15 4 10 1 0.5 11 0.3 0.6 15 5 12 1 0.5 21 0.3 0.3 15 4 9 1 0.4 21 0.3 0.5 15 5 11 1 0.5 21 0.3 0.6 15 7 12 1 0.5 22 0.3 0.3 15 3 8 1 0.4 22 0.3 0.4 15 3 9 1 0.4 22 0.3 0.6 15 5 11 1 0.4 31 0.3 0.3 15 4 8 0.8 0.5 31 0.3 0.4 15 4 8 0.8 0.5 31 0.3 0.5 15 5 9 0.8 0.5 31 0.3 0.6 15 5 9 0.8 0.5 32 0.3 0.3 15 4 8 0.8 0.5 32 0.3 0.4 15 4 8 0.8 0.5 32 0.3 0.5 15 5 9 0.8 0.5 32 0.3 0.6 15 5 9 0.8 0.5 33 0.3 0.3 15 4 8 1 0.5 33 0.3 0.4 15 4 8 1 0.5 33 0.3 0.5 15 5 9 1 0.5 33 0.3 0.6 15 5 9 1 0.5

[0057] Note: ID is the steel grade classification number sorted by steel grade.

[0058] Depend on Figure 2 It can be seen that the thickness has the highest correlation with welding parameters. Non-metallic elements such as Si have the lowest correlation with welding parameters. The correlation between metal elements such as Mn and welding process parameters varies. This is because, in metal welding, the C content and the content of metal elements such as Mn are directly related to the carbon equivalent, and the carbon equivalent directly affects the welding results.

[0059] Establishing welding parameter recommendation model based on random forest algorithm

[0060] After establishing the welding parameter recommendation model based on the random forest algorithm, the steel grade and thickness of the cold-rolled strip are used as input, and the output is the welding process parameters (welding current, welding wheel pressure, welding speed, overlap, overlap compensation). The random forest algorithm has many parameters, and adjusting the model parameters helps to improve the accuracy of the model. As shown in Table 2 and Figure 3 The figure shows the physical meaning and range of the parameters of the random forest model in this embodiment (preliminarily given by artificial experience) and their impact on the prediction results. In the data set, a random sampling method is used to divide the training set and the validation set (the ratio is 4:1).

[0061] Table 2 Physical meaning of the parameters of the computational forest algorithm

[0062]

[0063] Use random search CV in Python to search for optimal parameters.

[0064] First, the parameters of the random forest model (the parameter range is given according to manual experience) are passed to the random search CV, and then all parameters are combined in the form of a dictionary (randomly combining the possible ways of each parameter). Secondly, in order to save model computing space, avoid overfitting and improve accuracy, the number of random combinations of parameters (n_iter) should be defined, which defaults to 10 (this is the parameter value initially given in the principle of the random forest model). Third, the root mean square error (RMSE) is used to evaluate the model results composed of random search parameters. The smaller their values, the smaller the error between the model prediction results and the true value. Fourth, the number of iterations is determined by the product of the number of cross-validation folds (CV) and n_iter. For example, if N_iter = 100, CV = 5, the number of iterations is 500. Fifth, the optimal parameters are returned.

[0065] After determining the optimal parameters, a random forest model is established, and then the model is used to recommend parameters. For example, in this embodiment, the steel strip to be welded has a steel grade ID of 11, a front coil thickness of 0.4 mm, and a rear coil thickness of 0.3 mm. The welding process parameters optimized by the random forest algorithm parameter recommendation model are: welding current 9.8KA, welding wheel pressure 4KN, welding speed 14.8m / min, overlap amount 1mm, and overlap compensation amount 0.5mm.

[0066] Compared with the actual welding parameters (welding current 10KA, welding wheel pressure 4KN, welding speed 15m / min, overlap 1mm, overlap compensation 0.5mm), the deviations of the model recommended parameters are within 4% (welding process requirements), which meets the welding process requirements.

[0067] Communicate with the welding machine and send optimized process parameters

[0068] Before welding begins, the OLW software, based on the random forest algorithm, provides the optimized welding process parameters (welding current 9.8KA, welding wheel pressure 4KN, welding speed 14.8m / min, overlap 1mm, overlap compensation 0.5mm) of the cold-rolled steel strip to be welded according to the steel grade (11) and thickness (the thickness of the front coil is 0.4mm and the thickness of the rear coil is 0.3mm) of the cold-rolled steel strip to be welded. The OLW software directly communicates with the welder and sends the optimized welding process parameters to the welder. After the welder PLC obtains the parameters, it is displayed on the welder operation terminal and the corresponding welding is performed.

[0069] After one month of on-site trial operation, the error between the line welding process parameters set by the present invention and the existing welding process parameters was within 4%, which provided a guarantee for the production of cold-rolled strip steel.

[0070] However, 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. As long as they are within the spirit of the present invention, any changes or modifications to the above embodiments will fall within the scope of the claims of the present invention.

Claims

1. A method for setting online welding process parameters for cold-rolled strip lap welding, characterized in that , including the following steps: S1. Collect historical welding performance data, including the thickness of the previous and subsequent coils, steel grade, material chemical composition as input parameters of the random forest algorithm to be screened, and corresponding welding process parameters as target parameters, wherein the welding process parameters include welding current, welding speed, welding wheel pressure, overlap amount, and overlap compensation amount; S2. Based on Python, calculate the Pearson correlation coefficient and draw a heat map; S3. Select thickness, steel grade, and material chemical composition that have high correlation with welding parameters according to the thermal map for random forest algorithm training; S4. A welding parameter setting model is established based on a random forest algorithm, taking the steel grade, thickness, and chemical composition of the cold-rolled strip to be welded as input, and calculating and obtaining optimized welding process parameters, including welding current, welding wheel pressure, welding speed, overlap amount, and overlap compensation amount; S5. Send the optimized welding process parameters to the welding machine and set them accordingly.

2. The method for setting online welding process parameters for cold-rolled strip lap welding according to claim 1, characterized in that: In step S1 or S3, the calculation formula of the random forest algorithm is: In the formula, represents the expectation of random parameters of X and data set Dn; r n (X, Θ) is the nth tree, Θ is the expectation.

3. The method for setting online welding process parameters for cold-rolled strip lap welding according to claim 1, characterized in that: In step S2, the formula for calculating the row Pearson correlation coefficient is: In the formula, X i and Y i represents two different eigenvectors; n is the number of samples; and The eigenvectors X i and Y i Sample mean; r is the Pearson correlation coefficient, representing the eigenvector X i and Y i The linear correlation of r ranges from -1 to 1. The larger the absolute value, the stronger the X i and Y i The higher the correlation, the higher the correlation, r>0, indicating that X i and Y i Positive correlation, on the contrary, indicates that X i and Y i Negative correlation.

4. The method for setting online welding process parameters for cold-rolled strip lap welding according to claim 1, characterized in that: In step S3, the chemical composition of the material includes C, Mn, Cr, Cu, and Ni.

Images