A Multivariate Process Logistic Regression-Based Classification Method for Laser Panel Weld Quality

Abstract

A kind of multi-element process logistic regression laser plate weld quality classification method, comprising: record the welding process parameters and quality results of each weld of daily debugging laser plate station;Data set is divided into training set and test set;The weld process parameters of training set and test set are normalized and centralized processing;Gradient descent method is used to solve and construct multi-element process logistic regression model;Set multi-element process logistic regression model hyperparameter and model parameter initial value;Train multi-element process logistic regression model;The loss and accuracy of multi-element process logistic regression model are visualized and output;Weight vector inverse normalization;Weld quality prediction classification is carried out to laser plate;According to the welding process parameters of weld, the weld process parameters of training set and test set are normalized and centralized processing, the loss and accuracy of multi-element process logistic regression model are visualized and output, and the accurate quality prediction classification is carried out to laser plate weld.

Description

Technical Field

[0001] This invention relates to the field of weld quality classification technology, and in particular to a laser-laminated weld quality classification method based on multivariate process logistic regression. Background Technology

[0002] During the process debugging or production stage of the laser splicing station in the thin plate segmentation production line, due to the numerous factors affecting welding quality, it is impossible to predict the weld quality by inputting a single process parameter. Through long-term field experience, it has been learned that the welding parameters are different depending on the plate thickness. Summary of the Invention

[0003] The purpose of this invention is to provide a method for classifying the quality of laser-laminated weld seams using multivariate process logistic regression.

[0004] To achieve the above objectives, the technical solution of the present invention is as follows:

[0005] A method for classifying the quality of laser-laminated weld seams using multivariate process logistic regression, characterized by the following steps:

[0006] Step S1) Record the welding process parameters of each weld seam at the laser panel assembly station during daily debugging or trial production. The welding process parameters of this weld seam are used as the input features of the dataset, and the welding quality of this weld seam is recorded as the label of the dataset.

[0007] Step S2) Divide the dataset into a training set and a test set according to the principle of independent and identically distributed datasets;

[0008] Step S3) Normalize and center the weld process parameters of the training set and the test set;

[0009] Step S4) Use gradient descent to solve and construct a multivariate process logistic regression model; construct a multivariate process linear model, and nest the output of the multivariate process linear model as an independent variable in the Sigmoid function, mapping the output of the multivariate process linear model to probability values ​​between 0 and 1, representing the probability that the laser panel weld quality is qualified;

[0010] Step S5) Set the hyperparameters and initial values ​​of the model parameters for the multivariate process logistic regression model;

[0011] Step S6) Train the multivariate process logistic regression model and calculate the minimum value of the average cross-entropy loss function Loss;

[0012] Step S7) Visualize the loss and accuracy of the multivariate process logistic regression model;

[0013] Step S8) Reverse normalize the weight vector W;

[0014] Step S9) Perform quality prediction and classification on the laser-welded panel seams.

[0015] Furthermore, in step S2, the dataset is divided into a training set and a test set in an 8:2 ratio according to the principle of independent and identically distributed datasets.

[0016] Furthermore, in step S4, the output of the multivariate linear process model is nested as an independent variable in the Sigmoid function, y = σ(z), mapping the output of the multivariate linear process model to probability values ​​between 0 and 1, representing the probability that the laser-welded joint is of acceptable quality.

[0017] The linear model of the multivariate process is z = w1x 1 +w2x 2 +…+w 11 x 11 +b;

[0018] x 1 x represents the thickness of the sheet metal. 2 x is the weld length. 3 x is the welding speed. 4 x is the wire feeding speed. 5 x is the pulse frequency. 6 Let x be the pulse duration. 7 For peak voltage, x 8 x is the base current. 9 x is the laser power. 10 x is the defocus amount. 11 The distance between the optical filaments;

[0019] w1~w 11 This represents the weight of each process parameter that influences the weld quality.

[0020] The Sigmoid function is b is a linear model constant;

[0021] Let b = w0, x0 = 1,

[0022] Transform the multi-factor linear process model into vector form

[0023]

[0024] Using the average cross-entropy loss function To evaluate the performance of a multivariate logistic regression model for process control,

[0025] Among them, y i The label representing the i-th sample. This represents the predicted probability of the i-th sample.

[0026] Further, in step S5, the hyperparameters include the learning rate η and the number of iterations, and the initial value is the weight vector W of the multivariate process linear model, initialized as (w0, w1, w2, ..., w...). 11 A random value in ).

[0027] Furthermore, in step S6, the method for calculating the minimum value of the average cross-entropy loss function Loss is as follows:

[0028] For w1~w 11 Find the partial derivative

[0029] Take the partial derivative with respect to b

[0030] Iteratively update the weight vector W of the multivariate process linear model.

[0031] Furthermore, in step S7, the performance of the average cross-entropy loss function and model accuracy on the test set data is observed, the number of training times i with the highest possible accuracy and lowest possible loss is found, the model is retrained, and the weight vector W of the better multivariate process linear model is calculated to obtain the better multivariate process logistic regression model.

[0032] Furthermore, in step S8, the weight vector W is reverse-normalized.

[0033]

[0034] Furthermore, in step S9, the characteristic data of the laser panel welding parameters of the new segment are input, the quality of the panel weld is predicted according to the multivariate process logistic regression model, and the results are classified.

[0035] This invention normalizes and centers the weld process parameters of the training and test sets based on the weld process parameters to improve the accuracy of the model. It uses the gradient descent method to solve and construct a multivariate process logistic regression model, and visualizes the loss and accuracy of the multivariate process logistic regression model. The scale of the input feature values ​​does not need to be changed, making it more user-friendly for input. Finally, it performs accurate quality prediction and classification of laser-laminated welds. Attached Figure Description

[0036] Figure 1 A flowchart illustrating the construction and use of the laser panel weld quality classifier of this invention;

[0037] Figure 2 Input sample features and label maps into the dataset of this invention;

[0038] Figure 3 This is a graph showing the relationship between plate thickness and welding parameters in this invention.

[0039] Figure 4 This is a diagram of the nested multivariate process linear model of the Sigmoid function of this invention;

[0040] Figure 5 This is a graph showing the decrease in training loss and testing loss as a function of the number of training iterations in this invention.

[0041] Figure 6 This is a visualization of how the training loss and test loss decrease with the number of iterations during the learning process of this invention.

[0042] Figure 7 This is a visualization of how the accuracy of training and testing increases with the number of iterations during the learning process of this invention. Detailed Implementation

[0043] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0044] This embodiment discloses a laser panel weld quality classifier based on multivariate process logistic regression, including the following steps:

[0045] Step S1) as follows Figure 2 , 3 As shown, the welding process parameters of each weld at the laser panel welding station during daily debugging / trial production are recorded as input features of the dataset, and the welding quality of that weld is recorded as the dataset label. The input features of the laser panel welding process include material thickness, weld length, welding speed, wire feed speed, pulse frequency, pulse time, peak voltage, base current, laser power, defocusing amount, and filament spacing. The welding quality pass / fail status (0 / 1) is used as the dataset label value. In this embodiment, a total of 189 valid samples were collected.

[0046] Step S2) Divide the dataset into training and test sets in an 8:2 ratio according to the principle of independent and identically distributed datasets. The training and test sets need to have approximately the same mean and variance.

[0047] In this embodiment, the training set contains 151 samples and the test set contains 38 samples, and they have similar means.

[0048] Training set:

[0049] [9.62251656e+00 2.22536424e+03 2.24072848e+03 1.15947020e+012.79582781e+02 1.73496689e+00 3.76125828e+01 1.15741722e+021.11622517e+01-3.17284768e+00 2.99668874e+00],

[0050] Test set:

[0051] [1.00263158e+01 1.96578947e+03 2.25131579e+03 1.15973684e+012.75842105e+02 1.73421053e+00 3.77631579e+01 1.17657895e+021.16184211e+01-3.47894737e+00 2.97368421e+00].

[0052] Step S3) Normalize and center the weld process parameters of the training set and test set to improve the accuracy of the model.

[0053] Step S4) Solve and construct a multivariate process logistic regression model using the gradient descent method. Construct a multivariate process linear model and nest the output of the multivariate process linear model as an independent variable in the Sigmoid function, y = σ(z), as shown below. Figure 4 As shown, the output of the multivariate linear process model is mapped to a probability value between 0 and 1, representing the probability that the laser-coated panel weld is of acceptable quality.

[0054] The linear model of the multivariate process is z = w1x 1 +w2x 2 +…+w n x n +w 11 x 11 +b;

[0055] x 1 x represents the thickness of the sheet metal. 2 x is the weld length. 3 x is the welding speed. 4 x is the wire feeding speed. 5 x is the pulse frequency. 6 Let x be the pulse duration. 7 For peak voltage, x 8 x is the base current. 9 x is the laser power. 10 x is the defocus amount. 11 The distance between the optical filaments is w1 to w2.11 This represents the weight of each process parameter influencing the weld quality.

[0056] The Sigmoid function is

[0057] b is a linear model constant.

[0058] Let b = w0, x0 = 1, and convert the multivariate linear process model into vector form.

[0059]

[0060] Using the average cross-entropy loss function

[0061]

[0062] To evaluate the performance of a multivariate logistic regression model for process control, where y i The label representing the i-th sample. This represents the predicted probability of the i-th sample.

[0063] Step S5) Set the hyperparameters and initial values ​​of the model parameters for the multivariate process logistic regression model. The hyperparameters include the learning rate η and the number of iterations. The initial value refers to the weight vector W of the multivariate process linear model. The weight vector W of the linear model is initialized as (w0, w1, w2, ..., w...). 11 A random value.

[0064] Step S6) Train the multivariate process logistic regression model and calculate the minimum value of the average cross-entropy loss function;

[0065] The method for calculating the minimum value of the average cross-entropy loss function (Loss) is as follows:

[0066] For w1~w 11 Find the partial derivative

[0067] Take the partial derivative with respect to b

[0068] Iteratively update the weight vector W of the multivariate process linear model.

[0069] Step S7) Visualize the loss and accuracy of the multivariate process logistic regression model. By visualizing the results, observe the performance of the average cross-entropy loss function and the model accuracy on the test set data, and find the number of training times i with the highest possible accuracy and the lowest possible loss.

[0070] like Figure 5 As shown, after 550 training iterations, the accuracy decreased. Find an i value of around 550 and retrain the model.

[0071] like Figure 6 , 7 As shown, a better weight vector W is obtained for the multivariate process linear model and the multivariate process logistic regression model.

[0072]

[0073] Step S8) Invert and normalize the weight vector W.

[0074]

[0075] We get (w0, w1, w2, ..., w 11 ) are respectively

[0076] (-9.9775,-6.5707e-02,7.8393e-06,5.1349e-03,-2.5739e-02,4.6374e-03,-5.93 13e-01,-5.5637e-02,-9.314688877e-03,4.5847e-02,-4.2401e-01,1.7445e-01),

[0077] Substituting this coefficient into the multivariate process logistic regression model eliminates the need to change the scale of the input feature values, making it more input-friendly.

[0078] Step S9) Input the characteristic data of the laser panel welding parameters of the new segment, predict the quality of the panel weld according to the multivariate process logistic regression model, and classify the results.

[0079] 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 or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for classifying the quality of laser-laminated panel welds using multivariate process logistic regression, characterized in that, Includes the following steps: Step S1) Record the welding process parameters of each weld seam at the laser panel assembly station during daily debugging or trial production. The welding process parameters of this weld seam are used as the input features of the dataset, and the welding quality of this weld seam is recorded as the label of the dataset. Step S2) Divide the dataset into a training set and a test set according to the principle of independent and identically distributed datasets; Step S3) Normalize and center the weld process parameters of the training set and the test set; Step S4) Use gradient descent to solve and construct a multivariate process logistic regression model; A multivariate linear process model is constructed, and the output of the multivariate linear process model is nested as an independent variable in the Sigmoid function, y = σ(z). The output of the multivariate linear process model is mapped to a probability value between 0 and 1, which represents the probability that the laser splicing weld seam quality is qualified. The linear model of the multivariate process is z = w1x 1 +w2x 2 +…+w 11 x 11 +b; x 1 x represents the thickness of the sheet metal. 2 x is the weld length. 3 x is the welding speed. 4 x is the wire feeding speed. 5 x is the pulse frequency. 6 Let x be the pulse duration. 7 For peak voltage, x 8 x is the base current. 9 x is the laser power. 10 x is the defocus amount. 11 The distance between the optical filaments; w1~w 11 This represents the weight of each process parameter that influences the weld quality. The Sigmoid function is b is a linear model constant; Let b = w0, x0 = 1, Transform the multi-factor linear process model into vector form Using the average cross-entropy loss function To evaluate the performance of a multivariate logistic regression model for process control, Among them, y i The label representing the i-th sample. Represents the predicted probability of the i-th sample. Step S5) Set the hyperparameters and initial values ​​of the model parameters for the multivariate process logistic regression model; The hyperparameters include the learning rate η and the number of iterations. The initial value is the weight vector W of the multivariate linear process model, initialized as (w0, w1, w2, ..., w...). 11 A random value from ) Step S6) Train the multivariate process logistic regression model and calculate the minimum value of the average cross-entropy loss function Loss; The method for calculating the minimum value of the average cross-entropy loss function (Loss) is as follows: For w1~w 11 Find the partial derivative Take the partial derivative with respect to b Iteratively update the weight vector W of the multivariate process linear model; Step S7) Visualize the loss and accuracy of the multivariate process logistic regression model; Step S8) Reverse normalize the weight vector W; Step S9) Perform quality prediction and classification on the laser-welded panel seams.

2. The laser panel weld quality classification method based on multivariate process logistic regression according to claim 1, characterized in that, In step S2, the dataset is divided into a training set and a test set in an 8:2 ratio according to the principle of independent and identically distributed datasets.

3. The laser panel weld quality classification method based on multivariate process logistic regression according to claim 1, characterized in that, In step S7, the average cross-entropy loss function and model accuracy are observed on the test set data. The number of training times i with the highest possible accuracy and lowest possible loss is found. The model is retrained, and the weight vector W of the better multivariate process linear model is calculated to obtain the better multivariate process logistic regression model.

4. The laser panel weld quality classification method based on multivariate process logistic regression according to claim 3, characterized in that, In step S8, the weight vector W is reversed and normalized.

5. The laser panel weld quality classification method based on multivariate process logistic regression according to claim 4, characterized in that, In step S9, the characteristic data of the laser panel welding parameters of the new segment are input, the quality of the panel weld is predicted according to the multivariate process logistic regression model, and the results are classified.

Images