CN116921817B - Automatic TIG welding arc concentration online monitoring and intelligent early warning method - Google Patents

Abstract

The invention discloses an automatic TIG welding arc concentration online monitoring and intelligent early warning method. First, a mapping model is established for welding parameters, pipeline materials and protective gas types and arc image exposure to obtain a clear arc image; secondly, a welding model is established. The mapping model of parameters and edge detection dual thresholds accurately detects and intercepts the arc area; then, based on the relationship between the arc concentration and the three factors of current, protective gas flow rate and arc length, the arc concentration is normalized; Then, based on the ratio of the normalized arc concentration and the standard arc concentration, an objective arc image classification standard is established; finally, an arc concentration prediction model for robot automatic TIG welding is established and trained, and the results and processing plans output by the prediction model are , execute different warnings. The invention can realize online monitoring and intelligent early warning of arc concentration, thereby improving production efficiency and reducing costs.

Description

Automatic TIG welding arc concentration online monitoring and intelligent early warning method

Technical field

The invention belongs to the field of robot welding technology, and is specifically an online monitoring and early warning method of arc concentration applied to robot automatic TIG welding.

Background technique

In the actual TIG welding process of petrochemical process pipelines, the arc shape has an important impact on the stability of the welding process and welding quality. The welding arc shape represents the heat flow density distribution, and the heat flow density distribution directly affects the penetration of the pipe groove and the welding seam forming quality. Generally, the more concentrated the arc shape is, the more concentrated the heat flow distribution is, and the easier it is for the groove to be penetrated. Therefore, workers generally observe the arc shape through a mask and rely on experience to judge whether the pipe groove is penetrated during the welding process and the forming quality of the back of the pipe. However, workers' judgment accuracy fluctuates greatly and they cannot observe for a long time. Therefore, there are problems such as heavy workload and low quality of monitoring, resulting in waste of materials and labor and reduction of production efficiency.

In response to the above problems, people generally use industrial cameras to take images of the melt pool and predict them through a convolutional neural network prediction model to achieve real-time monitoring of arc morphology. However, in actual situations, due to changes in welding parameters caused by changes in groove size, there is a large difference in the clarity of arc images taken. Therefore, it is difficult for the convolutional neural prediction model to accurately identify the input image, resulting in a greatly reduced prediction accuracy. At the same time, the classification of arc images in the convolutional neural network model mainly relies on manual experience and does not establish objective standards. Therefore, the distinction between image sets will be low and confusing, making it difficult to train a high-accuracy model.

Contents of the invention

The purpose of this invention is to provide an online monitoring and intelligent early warning method for arc concentration in automatic TIG welding using robots, to achieve real-time weld quality prediction during the welding process, and to ensure welding stability and welding quality. The specific technical solutions are as follows:

Step 1: Establish an automatic TIG welding arc exposure adjustment model and a standard arc concentration model;

Further, the automatic TIG welding arc exposure adjustment model includes:

The calculation formula of the exposure adjustment model is as follows:

Among them, E is the exposure parameter value, Xi _and Yi are the original and processed RGB values respectively;

The exposure parameter value is adjusted based on the image exposure parameter value under the welding parameters with a current value of 150A and an arc voltage value of 10V. The specific calculation formula is as follows:

Among them, I is the current value, U is the arc voltage value, α and β are constants, eta is related to the material, and λ is related to the type of protective gas.

Furthermore, the standard arc concentration model includes: calculating the standard arc concentration using the area of the arc area and the size of the tungsten electrode under the welding parameters of a shielding gas flow rate of 10L/min, a current value of 150A, and an arc voltage value of 10V. The specific formula as follows:

Among them, A _G ^* is the standard arc shape, d _min is the minimum size of the tungsten electrode tip, D is the diameter of the tungsten electrode, θ is the angle of the tungsten electrode tip, and S ^* is the area of the arc area.

Step 2: During the TIG welding process of petrochemical pipelines, the welding images are captured by the industrial camera and the welding electrical signals are collected in real time by the welding parameter acquisition device, and the captured welding images are transmitted back to the computer every 1-3 seconds. At the same time, based on the automatic TIG welding arc The exposure adjustment model performs image exposure adjustment.

Step 3: Perform morphological processing on the image after exposure adjustment, and perform edge detection on the morphologically processed image based on the dual threshold determination model to intercept the area of the arc.

Further, the dual threshold determination model includes:

The specific formula for calculating the high threshold and low threshold in edge detection is as follows:

Among them, Th is the high threshold, Tl is the low threshold, and a ₁ , a ₂ , b ₁ , b ₂ , c ₁ and c ₂ are constants.

Step 4: Calculate the area of the arc based on the number of pixels in the arc area, and normalize the arc concentration based on the relationship between the arc concentration and the three factors of current, protective gas flow rate and arc length.

Further, the formula for calculating the normalized arc concentration is as follows:

Among them, δ is the normalized arc concentration, S is the area of the arc area, Q is the protective gas flow rate, I current size, U is the arc voltage, and ε, φ and ρ are constants.

Step 5: Classify arc images based on the ratio of normalized arc concentration and standard arc concentration, and establish and train an arc concentration prediction model for robot automatic TIG welding;

Further, the calculation formula for the ratio of normalized arc concentration and standard arc concentration is as follows:

Arc images are classified based on the size of T. When the T value is between 1 and 2, the arc concentration is defined as good; when the T value is between 2 and 5, the arc concentration is average; when the T value is above 5, the arc concentration is unqualified. The classified arc images were used to create training sets, verification sets and test sets in a ratio of 7:2:1.

Furthermore, AlexNet convolutional neural network is used to establish a tungsten electrode burning state prediction model, which includes 5 convolutional layers, 3 pooling layers and 3 fully connected layers. The last layer uses the Softmax function as the output layer; using ReLU is a nonlinear activation number, and Dropout with 50% of neuron activations as 0 is used between fully connected layers.

Further, the training of the arc concentration prediction model is to load the data set into the model for multiple rounds of training, and use the model with the highest verification accuracy as the final arc concentration prediction model. Finally, load the test set into the model for verification. The prediction accuracy of the model.

Step 6: Develop different treatment plans according to the degree of arc concentration;

Furthermore, the following different treatment plans are formulated for different arc concentration degrees:

(1) When the arc concentration is good, continue welding;

(2) When the arc concentration is normal, the arc can be stopped and the tungsten electrode replaced when the current pipeline welding is completed;

(3) If the arc concentration fails, stop welding immediately and replace the tungsten electrode.

Step 7: Execute different warnings based on the results and processing plans output by the prediction model.

Further, based on the results and processing plans output by the prediction model, the following different early warnings are executed:

(1) When the arc concentration is good, no prompt will appear on the computer interface;

(2) When the arc concentration is normal, the computer interface pops up an early warning prompt to stop the arc and replace the tungsten electrode when the current pipeline welding is completed;

(3) When the arc concentration fails, the computer interface pops up a prompt to stop welding and replace the tungsten electrode and feeds back to the control system to execute the stop welding command.

Compared with the existing technology, the present invention has significant advantages: 1) The present invention can monitor the arc concentration degree online and output the arc status in real time, so that the tungsten electrode can be replaced in time, reducing the waste of production materials and time, and greatly improving the efficiency of the tungsten electrode. Production efficiency and welding quality; 2) This invention normalizes the arc concentration and compares it with the standard arc concentration, thereby eliminating interference factors and achieving accurate judgment of arc concentration; 3) This invention is based on normalization The ratio of the arc concentration degree to the standard arc concentration degree is used to objectively classify images, which avoids the problem of cross confusion between image sets caused by artificial subjective classification of images, and is conducive to training a high-accuracy prediction model; 4) The present invention uses exposure The degree of adjustment model can clearly extract the image of the arc area under different welding parameters, which is beneficial to the neural network prediction model to identify the image, thereby improving the prediction accuracy.

Description of drawings

Figure 1 is a test platform for the automatic TIG welding arc concentration online monitoring and intelligent early warning method of the present invention.

Figure 2 is a flow chart of the automatic TIG welding arc concentration online monitoring and intelligent early warning method of the present invention.

Figure 3 shows the arc status under different welding parameters in the automatic TIG welding arc concentration online monitoring and intelligent early warning method of the present invention.

Figure 4 is a morphology diagram of the tungsten electrode in the automatic TIG welding arc concentration online monitoring and intelligent early warning method of the present invention.

Figure 5 is an original image and a processed image of the arc concentration from good to unqualified in the automatic TIG welding arc concentration online monitoring and intelligent early warning method of the present invention.

Figure 6 is a structural diagram of the convolutional neural network in the pipe-flange example of the embodiment of the present invention.

Figure 7 is the training result of the arc concentration prediction model of the present invention.

Figure 8 is the verification result of the arc concentration prediction model of the present invention.

Detailed ways

The present invention will be further described below in conjunction with the accompanying drawings and examples.

In order to solve the problem that in actual situations, due to the change of welding parameters due to the change of groove size, there is a large difference in the clarity of the arc images taken and no objective standards are established for the classification of the same arc images, the present invention proposes a method for pipe-flange Online monitoring and early warning method for arc concentration in robotic automatic TIG primer welding. This invention establishes a mapping model between image exposure and edge detection threshold size and welding parameters, achieving a clear arc image; and simultaneously establishing an evaluation of arc concentration. The objective standard of degree realizes accurate classification of arc images, thereby realizing a prediction model with high prediction accuracy. The present invention establishes a prediction model of arc concentration through a convolutional neural network, and realizes online monitoring and intelligent early warning of arc concentration, thereby improving production efficiency and reducing costs.

A test platform is actually built before the pipe-flange bottom welding. As shown in Figure 1, the welded pipe base material 1 is made of carbon steel and stainless steel with different diameters. The arc vision sensing system includes an industrial camera 3 and a computer 4. The industrial camera 3 is fixed on the welding gun 2 of the welding robot using a clamp. The industrial camera 3 collects images at a frequency of 30Hz.

Specifically follow the following steps to implement, the specific process is shown in Figure 2:

Step 1: Establish an automatic TIG welding arc exposure adjustment model and a standard arc concentration model.

Further, as shown in Figure 3, the arc shows different brightness under different welding parameters. The smaller the current value, the dimmer the arc. Therefore, it is necessary to increase the exposure to make the arc more obvious. Automatic TIG welding arc exposure adjustment models include:

The calculation formula of the exposure adjustment model is as follows:

Among them, E is the exposure parameter value, Xi _and Yi are the original and processed RGB values respectively;

The exposure parameter value is adjusted based on the image exposure parameter value under the welding parameters with a current value of 150A and an arc voltage value of 10V. The specific calculation formula is as follows:

Among them, I is the current value, U is the arc voltage value, α and β are constants, eta is related to the material, and λ is related to the type of protective gas.

Furthermore, the standard arc concentration model includes: calculating the standard arc concentration using the area of the arc area and the size of the tungsten electrode under the welding parameters of a shielding gas flow rate of 10L/min, a current value of 150A, and an arc voltage value of 10V, in which tungsten The polar morphology is shown in Figure 4, and the specific formula is as follows:

Among them, A _G ^* is the standard arc shape, d _min is the minimum size of the tungsten electrode tip, D is the diameter of the tungsten electrode, θ is the angle of the tungsten electrode tip, and S ^* is the area of the arc area.

Step 2: During the TIG welding process of petrochemical pipelines, the welding images are captured by the industrial camera and the welding electrical signals are collected in real time by the welding parameter acquisition device, and the captured welding images are transmitted back to the computer every 1-3 seconds. At the same time, based on the automatic TIG welding arc The exposure adjustment model performs image exposure adjustment.

Step 3: Perform morphological processing on the image after exposure adjustment, and perform edge detection on the morphologically processed image based on the dual threshold determination model to intercept the area of the arc, as shown in Figure 5.

Further, the arc color image is converted into grayscale.

Further, the brightness and contrast of the arc image are adjusted based on the linear transformation formula. The linear transformation method is as follows:

In the formula, f (i, j) is the pixel value of the original image, and g (i, j) is the pixel value after image processing.

Furthermore, the Laplacian operator with an aperture of 3 is used to sharpen the arc image three times.

Further, the specific formulas for calculating the high threshold and low threshold in edge detection are as follows:

Among them, Th is the high threshold, Tl is the low threshold, and a ₁ , a ₂ , b ₁ , b ₂ , c ₁ and c ₂ are constants.

Further, the arc image is processed using 3×3 expansion morphology with structural elements.

Further, the connected domain with a pixel value of 255 in the image is marked, and the size of each connected domain is calculated. Finally, the largest connected domain is retained to obtain a complete arc edge image.

Step 4: Calculate the area of the arc based on the number of pixels in the arc area, and normalize the arc concentration based on the relationship between the arc concentration and the three factors of current, protective gas flow rate and arc length.

Further, the formula for calculating the normalized arc concentration is as follows:

Among them, δ is the normalized arc concentration, S is the area of the arc area, Q is the protective gas flow rate, I current size, U is the arc voltage, and ε, φ and ρ are constants.

Step 5: Classify arc images based on the ratio of normalized arc concentration and standard arc concentration, and establish and train an arc concentration prediction model for robot automatic TIG welding;

Further, the calculation formula for the ratio of normalized arc concentration and standard arc concentration is as follows:

Arc images are classified based on the size of T. When the T value is between 1 and 2, the arc concentration is defined as good; when the T value is between 2 and 5, the arc concentration is average; when the T value is above 5, the arc concentration is unqualified. The classified arc images were used to create training sets, verification sets and test sets in a ratio of 7:2:1.

Furthermore, the AlexNet convolutional neural network is used to establish the arc concentration prediction model, which includes 5 convolutional layers, 3 pooling layers and 3 fully connected layers. The last layer uses the Softmax function as the output layer; ReLU is used as the output layer. Nonlinear activation number, use Dropout with 50% of neuron activations as 0 between fully connected layers; the convolution kernel and step size of the first convolution layer are set to 11 and 4, and the convolution of the second convolution layer is The convolution kernel and step size are set to 5 and 1. The convolution kernel and step size of the third, fourth and fifth convolutional layers are all set to 3 and 1; the convolution kernel and step size of the three pooling layers are all set to 3. and 2, as shown in Figure 6.

Furthermore, the arc concentration prediction model was trained for a total of 20 rounds, and the validation set with the highest accuracy in the training model was used as the final training model. The accuracy of the training set and validation set in the 20th round is above 97%, as shown in Figures 7 and 8. Therefore, round 20 is selected as the final training model. Finally, the test set is loaded into the model, and the prediction accuracy is above 95%.

Step 6: Develop different treatment plans according to the degree of arc concentration;

Furthermore, the following different treatment plans are formulated for different arc concentration degrees:

(1) When the arc concentration is good, continue welding;

(2) When the arc concentration is normal, the arc can be stopped and the tungsten electrode replaced when the current pipeline welding is completed;

(3) If the arc concentration fails, stop welding immediately and replace the tungsten electrode.

Step 7: Execute different warnings based on the results and processing plans output by the prediction model.

Further, based on the results and processing plans output by the prediction model, the following different early warnings are executed:

(1) When the arc concentration is good, no prompt will appear on the computer interface;

(2) When the arc concentration is normal, the computer interface pops up an early warning prompt to stop the arc and replace the tungsten electrode when the current pipeline welding is completed;

(3) When the arc concentration fails, the computer interface pops up a prompt to stop welding and replace the tungsten electrode and feeds back to the control system to execute the stop welding command.

Claims (2)

1. An automatic TIG welding arc concentration on-line monitoring and intelligent early warning method is characterized by comprising the following steps:

step 1: establishing an automatic TIG welding arc exposure adjustment model and a standard arc aggregation model;

the establishment of the automatic TIG welding arc exposure degree adjusting model comprises the following steps:

the calculation formula of the exposure degree adjustment model is as follows:

，

wherein E is an exposure parameter value, X _i And Yi are the RGB values after the original and the process, respectively;

the exposure parameter value is adjusted by taking the image exposure parameter value under the welding parameters with the current value of 150A and the arc voltage value of 10V as a reference, and the specific calculation formula is as follows:

，

wherein I is a current value, U is an arc voltage value, alpha and beta are constants, eta is related to materials, and lambda is related to the type of protective gas;

the building of the standard arc concentration model comprises the following steps: the standard arc concentration is calculated by adopting the area of an arc area under the welding parameters of 10L/min of shielding gas flow rate, 150A of current value and 10V of arc voltage value and the tungsten electrode size, and the specific formula is as follows:

，

wherein A is _G ^* In the form of standard arc, d _min The minimum size of the tip of the tungsten electrode, D is the diameter of the tungsten electrode, theta is the angle of the end part of the tungsten electrode, S ^* Is the area of the arc region;

step 2: shooting welding images through an industrial camera in the TIG welding process of the petrochemical pipelines, collecting welding electric signals in real time through a welding parameter collecting device, transmitting the shot welding images back to a computer every 1-3s, and adjusting the image exposure based on an automatic TIG welding arc exposure adjusting model;

step 3: carrying out morphological processing on the image after exposure degree adjustment, and simultaneously carrying out edge detection and intercepting an arc area on the image after morphological processing based on a dual-threshold determining model; the dual threshold determination model comprises:

the specific formulas of the high threshold and the low threshold in the edge detection are as follows:

，

wherein Th is a high threshold, tl is a low threshold, a ₁ 、a ₂ 、b ₁ 、b ₂ 、c ₁ And c ₂ Is a constant;

step 4: calculating the area of an electric arc according to the number of pixels in the electric arc area, and normalizing the electric arc aggregation based on three factors of the electric arc aggregation, namely the current, the flow rate of protective gas and the electric arc length;

the normalization processing of the arc concentration degree comprises the following steps:

the normalized arc concentration calculation formula is as follows:

，

wherein delta is normalized electric arc aggregation degree, S is the area of an electric arc region, Q is the flow rate of protective gas, I current is large, U is arc voltage, epsilon, phi and rho are constants;

step 5: classifying arc images based on the ratio of the normalized arc concentration to the standard arc concentration, and establishing and training an arc concentration prediction model of the automatic TIG welding of the robot;

the classifying the arc image based on the normalized arc concentration ratio and the standard arc concentration ratio comprises:

the calculation formula of the ratio of the normalized arc concentration to the standard arc concentration is as follows:

，

classifying the arc images based on the size of T; when the T value is between 1 and 2, the arc aggregation degree is well defined; when the T value is between 2 and 5, the arc aggregation degree is defined as general; when the T value is more than 5, defining that the arc aggregation degree is unqualified; manufacturing a training set, a verification set and a test set according to the proportion of 7:2:1 from the classified arc images;

step 6: different treatment schemes are formulated according to the electric arc concentration degree; comprising the following steps:

(1) When the electric arc aggregation degree is good, continuing welding;

(2) When the arc concentration is common, stopping arc and replacing the tungsten electrode under the condition that the current pipeline welding is completed;

(3) When the arc concentration is unqualified, immediately stopping welding and replacing the tungsten electrode;

step 7: and (3) transmitting the image processed in the step (4) to an arc concentration prediction model, and executing different early warning according to the result output by the prediction model and the processing scheme.

2. The method for on-line monitoring and intelligent early warning of the arc concentration of automatic TIG welding according to claim 1, which is characterized in that: and (3) executing different early warning according to the result and the processing scheme output by the prediction model in the step (7) comprises the following steps:

(1) When the arc concentration is good, no prompt appears on the computer interface;

(2) When the arc concentration is common, the computer interface pops up an early warning prompt of changing tungsten electrode when arc stopping is finished under the current pipeline welding condition;

(3) When the arc concentration is unqualified, the computer interface pops up a prompt for stopping welding and replacing the tungsten electrode, feeds back the prompt to the control system and executes a welding stopping command.