Oil cylinder and manufacturing device and welding method thereof

Abstract

The invention provides an oil cylinder and a manufacturing device and a welding method thereof, and belongs to the technical field of oil cylinder production. The oil cylinder and the manufacturing device thereof comprise an oil cylinder body; a welding base, which is arranged on the lower side of the oil cylinder body; an oil cylinder fixing seat, which is opened at the upper end of the welding base, and the oil cylinder body is fixedly connected in the oil cylinder fixing seat; a limiting rotating groove, which is opened at the upper end of the welding base; and a booster assembly, wherein the booster assembly is provided with two groups, and each group of the booster assemblies comprises a pressurizing cylinder, a sliding groove, a pressurizing groove, a pneumatic push cylinder, a pneumatic push rod, a welder, a pressurizing push plate, a pressurizing push rod, a connecting air hole and a limiting rotating plate, and can quickly weld oil cylinder bodies with different external shapes, without adjusting the welding route of the welder when welding different oil cylinder bodies.

Description

Technical Field

[0001] The invention belongs to the technical field of oil cylinder production, and in particular relates to an oil cylinder and a manufacturing device and a welding method thereof. Background Art

[0002] The manufacturing of hydraulic cylinders is a complex and precise process. Turning: machining the outer diameter of the cylinder, the frame mouth, boring the stopper, the flange stopper, etc. Drilling and boring: drilling holes at both ends of the cylinder to prepare for the installation of end covers and connectors. Welding: if it is necessary to weld pipe joint seats, flanges and other accessories, professional welding processes must be used to ensure welding quality. Piston and component processing: processing the piston rod to ensure straightness and surface finish. Making the piston, processing the piston's external dimensions to ensure a good fit with the inner wall of the cylinder, and processing the sealing groove. Seal installation.

[0003] After searching, it was found that a welding method for a cylinder structure is disclosed in a Chinese patent with the authorization announcement number "CN110202305B", which belongs to the field of mechanical processing. When welding and forming a cylinder structure, the cylinder body and the end cover can be welded first, and then the cylinder body and the two bottom plates can be spot welded. Then the bottom plate is placed on the tooling plate of the anti-deformation tooling, and the bottom plate and the tooling plate are welded, so that the position of the bottom plate is well supported and fixed. Then, multiple card plates are buckled on the cylinder body, the U-shaped groove presses the cylinder body, the card plates and the bottom plate are welded, and the positions of the cylinder body and the bottom plate are supported and fixed by the card plates and the bottom plate, and then the cylinder body and the bottom plate are welded, and the deformation between the cylinder body and the bottom plate is small. Finally, the card plate is removed, the rib plate is welded between the cylinder body and the bottom plate, the connecting plate is welded between the two bottom plates through the welding hole, and the connecting plate is welded to the cylinder body. When the rib plate and the connecting plate are welded, the bottom plate is still supported by the tooling plate. The dimensional accuracy of the obtained cylinder structure is improved, which can reduce the scrap rate of the cylinder structure.

[0004] When the rib plate and the connecting plate are welded in the above technical solution, the bottom plate is still supported by the tooling plate. The dimensional accuracy of the obtained oil cylinder structure is improved, which can reduce the scrap rate of the oil cylinder structure. However, in the prior art, when welding the oil cylinder, the automatic welding route of the welding head needs to be automatically fixed, and different automatic welding programs need to be loaded when welding oil cylinders of different shapes in one device, which reduces the welding process. Summary of the invention

[0005] The purpose of the present invention is to provide an oil cylinder and its manufacturing device and welding method, aiming to solve the problem in the prior art that when welding the oil cylinder, the automatic welding route of the welding head needs to be automatically fixed, and different automatic welding programs need to be loaded when welding oil cylinders of different shapes in one device, thereby reducing the problem of welding preparation.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] An oil cylinder and a manufacturing device thereof, comprising:

[0008] Cylinder body;

[0009] A welding base, wherein the welding base is arranged on the lower side of the oil cylinder body;

[0010] An oil cylinder fixing seat, wherein the oil cylinder fixing seat is disposed at the upper end of the welding base, and the oil cylinder body is fixedly connected in the oil cylinder fixing seat;

[0011] A limit rotation groove, wherein the limit rotation groove is provided at the upper end of the welding base;

[0012] The boosting assembly is provided with two groups, each group of the boosting assembly includes a pressure cylinder, a sliding groove, a pressure groove, a pneumatic push cylinder, a pneumatic push rod, a welder, a pressure push plate, a pressure push rod, a connecting air hole and a limit rotating plate. The pressure cylinder is arranged on one side of the oil cylinder body, the sliding groove is opened at the lower end of the pressure cylinder, the pressure groove is opened at the upper inner wall of the sliding groove, the pneumatic push cylinder is fixedly connected to the circumferential surface of the pressure cylinder, the pneumatic push rod is slidably connected in the pneumatic push cylinder, the welder is fixedly connected to one side end of the pneumatic push rod, the pressure push plate is slidably connected to the circumferential inner wall of the sliding groove, the limit rotating plate is slidably connected in the limit rotating groove, the pressure push rod is fixedly connected to the upper end of the limit rotating plate and contacts with the lower end of the pressure push plate, and the connecting air hole is opened on the circumferential inner wall of the sliding groove.

[0013] As a preferred solution of the present invention, the upper end of the welding base is fixedly connected to a lifting platform, the upper end of the lifting platform is fixedly connected to two lifting sleeves, a lifting platform is provided on the upper side of the lifting platform, two lifting grooves are provided on the upper end of the lifting platform, the lifting platform is slidably connected to the circumferential surface of the two lifting sleeves through the two lifting grooves, the upper end of the lifting platform is fixedly connected to a driving cylinder, and the output end of the driving cylinder is connected to the upper end of the lifting platform.

[0014] As a preferred solution of the present invention, one side end of the lifting platform is fixedly connected to a support sleeve, the upper ends of the two pressurizing cylinders are fixedly connected to a welding support plate, the upper end of the welding support plate is provided with a support groove, the support sleeve rotates in the support groove, the upper end of the support sleeve is fixedly connected to a drive motor, and the output end of the drive motor is fixedly connected to the upper end of the welding support plate.

[0015] As a preferred solution of the present invention, a pressure relief assembly is provided in each group of the boosting assemblies, and each group of the pressure relief assemblies includes a pressure relief cylinder, a pressure relief port, a pressure relief push plate, a pressure relief push rod and a pressure relief spring, the pressure relief cylinder is fixedly connected to the upper inner wall of the pressurizing groove, the pressure relief port is opened on the circumferential inner wall of the pressure relief cylinder, the pressure relief push plate is slidably connected in the pressure relief cylinder, the pressure relief push rod is fixedly connected to the upper end of the pressure relief push plate and slides in the pressure relief cylinder, and the pressure relief spring is fixedly connected to the upper inner wall of the pressure relief cylinder and the upper end of the pressure relief push rod.

[0016] As a preferred solution of the present invention, an upper opening is formed on the upper inner wall of the pressure relief cylinder.

[0017] As a preferred solution of the present invention, the outer surfaces of the two welders are fixedly connected with a fixing plate, one side end of the two fixing plates is fixedly connected with an arc-shaped push plate, and one side end of the two arc-shaped push plates is provided with a welding hole.

[0018] As a preferred solution of the present invention, one side end of the two fixed plates is fixedly connected to two limiting slide rods, the circumferential surfaces of the two pressurizing cylinders are fixedly connected to two limiting sleeves, and the multiple limiting slide rods slide in the multiple limiting sleeves respectively.

[0019] As a preferred solution of the present invention, a plurality of sealing rings are fixedly connected to the circumferential surfaces of the pressure push plate and the pressure relief push plate.

[0020] As a preferred solution of the present invention, a buffer ring is provided at the upper end of the oil cylinder body, and a sealing sleeve and a protective shell are provided on the outer surface of the oil cylinder body.

[0021] As a preferred solution of the present invention, one side surface of the welding hole is arc-shaped, and the output end of the welder is arranged in the welding hole.

[0022] Furthermore, in order to achieve real-time monitoring and adaptive control, based on the existing oil cylinder and its manufacturing device, a machine learning algorithm is introduced to optimize the welding process and improve production efficiency. The specific process is as follows:

[0023] Step 1: Installation location and selection of sensors and cameras: In order to achieve effective real-time monitoring and adaptive control in the welding device, temperature sensors, pressure sensors and high-resolution cameras need to be installed at key locations;

[0024] 1. Temperature sensor:

[0025] Installation location: The temperature sensor should be installed close to the welding point to ensure that the high temperature generated during the welding process can be accurately measured;

[0026] Sensor type: Use K-type thermocouple with high temperature resistance and fast response speed;

[0027] Temperature range: 0℃-2000℃;

[0028] Accuracy: ±0.5℃;

[0029] Response time: <1 second;

[0030] 2. Pressure sensor:

[0031] Installation location: The pressure sensor is installed in the pipeline between the air pressure push cylinder and the welder to monitor the pressure applied during welding;

[0032] Sensor type: Choose a high-precision, stable pressure sensor (such as a MEMS pressure sensor);

[0033] Pressure range: 0-10MPa;

[0034] Accuracy: ±0.1MPa;

[0035] Response time: <1 ms;

[0036] 3. High-resolution camera:

[0037] Installation location: The high-resolution camera should be installed just above or to the side of the welding point to ensure that the image data of the welding process can be clearly captured;

[0038] Camera type: Use an industrial-grade camera with 5 megapixels and a frame rate of 30fps, which has high resolution and high-speed shooting capabilities;

[0039] Parameter variables:

[0040] Resolution: 5 megapixels;

[0041] Frame rate: 30fps

[0042] Exposure time: automatic adjustment;

[0043] Step 2, data collection and preprocessing: During the welding process of the oil cylinder and its manufacturing device, sensors and cameras are used to collect temperature, pressure and image data in real time, and these data are processed through data preprocessing technology to ensure the quality and consistency of the input data;

[0044] 1. Data acquisition parameter settings

[0045] Temperature collection frequency: 10 times per second, the collected temperature data is T(t);

[0046] Pressure collection frequency: 10 times per second, the collected pressure data P(t);

[0047] Image acquisition frequency: 5 frames of images are collected per second, image data I(t);

[0048] 2. Data Preprocessing

[0049] Temperature and pressure data preprocessing:

[0050] Denoising: Use a sliding average filter to denoise the temperature and pressure data, and the sliding window size is set to 5;

[0051]

[0052]

[0053] Among them, T filtered and P filtered They are the temperature and pressure data after denoising respectively;

[0054] Standardization: Standardize the temperature and pressure data to make their mean 0 and standard deviation 1;

[0055]

[0056]

[0057] Among them, μ T and σ T are the mean and standard deviation of the temperature data, μ P and σ P are the mean and standard deviation of the pressure data, T normalized and P normalized They are the temperature and pressure data after standardization;

[0058] Image data preprocessing:

[0059] Grayscale processing: Convert color images into grayscale images to reduce the amount of data and simplify subsequent processing;

[0060] I gray (x, y) = 0.299·I R (x,y)+0.587·I G (x,y)+0.114·I B (x, y)

[0061] Among them, I R is the red channel value at the pixel point (x, y) in the image; I G is the red channel value at the pixel point (x, y) in the image; I B is the red channel value at the pixel point (x, y) in the image;

[0062] Image denoising: Use Gaussian filtering to denoise the image; the Gaussian filter parameters are set to kernel size 3×3 and standard deviation σ=1.0;

[0063]

[0064] Among them, I denoised is the value of the image pixel (x, y) after denoising;

[0065] Image normalization: normalize the image pixel values ​​to the range of [0, 1];

[0066]

[0067] Among them, I normalized is the value at the image pixel point (x, y) after normalization;

[0068] Step 3: Use the improved convolutional neural network (CNN) to build a welding defect detection model. Improve the traditional CNN and use the deep residual network (ResNet) and attention mechanism (Attention Mechanism) to enhance the performance of the model. The model inputs real-time image data during the welding process and outputs whether there are welding defects and the defect type. The specific steps are as follows:

[0069] 1. Data augmentation: Use techniques such as rotation, scaling, translation, and flipping to augment the dataset:

[0070] Rotation: randomly rotate the image by a certain angle;

[0071] Scale: Randomly scale the image;

[0072] Translation: Randomly translate the image;

[0073] Flip: randomly flip the image horizontally or vertically;

[0074] 2. Data trimming or padding:

[0075] Crop or pad the image to a uniform size of 224×224;

[0076] 3. Convolutional layer and pooling layer

[0077] Convolutional layer: The first layer uses 64 7×7 convolution kernels with a stride of 2 and a padding of 3;

[0078]

[0079] Where X is the input image, W is the convolution kernel, b is the bias, h is the output feature map, i and j are the height and width of the feature map in spatial coordinates, k is the channel index of the output feature map, c is the channel index of the input feature map, M and N are the height and width of the convolution kernel, respectively;

[0080] Calculated output size: 112×112×64;

[0081] Pooling layer: The maximum pooling layer uses a 3×3 pooling window with a stride of 2 and a padding of 1;

[0082] p i,j, k=max m,n h i+m-1，j+n-1，k

[0083] Where p is the pooling output and max is the maximum pooling operation;

[0084] Calculated output size: 56×56×64;

[0085] 4. Residual Block

[0086] Each residual block consists of two convolutional layers and their residual connections:

[0087] y=F(x,{W i})+x

[0088] Among them, F(x, {W i}) represents the learned residual function, x is the input feature map, y is the residual output, W i is the convolution kernel in the residual block, i is the number of convolution layers;

[0089] The specific formula of the residual block is:

[0090] The first convolution layer: 64×3×3, stride 1, padding 1;

[0091] F1(x)=ReLU(W1·x+b1)

[0092] Among them, F1(x) is the output after the first layer of convolution; W1 is the first layer of convolution kernel; b1 is the first layer of convolution bias;

[0093] The second convolution layer: 64×3×3, stride 1, padding 1;

[0094] F2(x)=W2·F1(x)+b2

[0095] Among them, F2(x) is the output after the second layer of convolution; W2 is the second layer of convolution kernel; b2 is the second layer of convolution bias;

[0096] Residual output:

[0097] y = ReLU(F2(x)+x)

[0098] The output size remains the same: 56×56×64;

[0099] 5. Attention Mechanism

[0100] Calculate the output of self-attention:

[0101]

[0102] Where Q is the query matrix, which represents the set of query vectors; K is the key matrix, which represents the set of key vectors; V is the value matrix, which represents the set of value vectors; Q = K = V = X (input feature map); d k is the dimension of the key vector; Attention(Q, K, V) is the output of the attention mechanism;

[0103] Calculation process:

[0104] Q=W Q ·X

[0105] K=W K ·X

[0106] V=W V ·X

[0107] Among them, W Q , W K , W V They are the weight matrices for query, key, and value respectively;

[0108] Calculate the attention score:

[0109]

[0110] Among them, Attention_Scores is the attention score matrix;

[0111] Apply the Softmax function:

[0112] Attention_Weights=softmax(Attention_Scores)

[0113] Among them, Attention_Weights is the attention weight after softmax processing; Softmax is the Softmax activation function;

[0114] Compute the attention output:

[0115] Attention_Output=Attention_Weight·V

[0116] Among them, Attention_Output is the output matrix of the attention mechanism;

[0117] 6. Fully connected layer

[0118] Flatten the feature map into a one-dimensional vector:

[0119] Flatten_Output=flatten(Attention_Output)

[0120] Among them, Flatten_Output is the flattened output vector;

[0121] Output size: 7×7×512=25088;

[0122] After the first fully connected layer, the ReLU activation function is used:

[0123] FC1=ReLU(W fc1 Flatten_Output+b fc1 )

[0124] Among them, FC1 is the output of the first fully connected layer; W fc1 is the weight matrix of the first fully connected layer; b fc1 is the bias vector of the first fully connected layer;

[0125] The second fully connected layer uses the Softmax activation function for multi-classification:

[0126] Output = Softmax(W fc2 FC1+b fc2 )

[0127] Among them, Output is the probability distribution of the final output; W fc2 is the weight matrix of the second fully connected layer; b fc2 is the bias vector of the second fully connected layer;

[0128] Output size: 7 (including 7 categories: normal, crack, pore, undercut, weld accumulation, deviation, and concave);

[0129] 7. Model Training

[0130] Using the cross entropy loss function:

[0131]

[0132] Among them, L is the loss function value; N is the number of samples; C is the number of categories; yi j is the actual label of the jth category of sample i; is the predicted probability of the jth class for sample i;

[0133] Use the Adam optimizer to perform gradient descent optimization and adjust the model weights to minimize the loss function;

[0134] Step 4: During the welding process, deep reinforcement learning (DRL) technology is used to achieve adaptive control. The deep Q network (DQN) algorithm is used. DQN combines Q learning and deep neural networks and is suitable for processing high-dimensional input data from multiple sensors and images during the welding process.

[0135] 1. Environment and state definition

[0136] Environment definition: Oil cylinder manufacturing device, including oil cylinder body, welding base, oil cylinder fixing seat, limit rotation groove, and booster assembly;

[0137] State vector s t :

[0138] s t =[I t , V t , v t , T t , P t , D t ]

[0139] Among them, I t is the current; V t is the voltage; v t is the welding speed; T t is temperature; P t For pressure; D t Welding defects;

[0140] 2. Action definition

[0141] Adjust welding parameters: current I, voltage V, welding speed v;

[0142] Action vector a t :

[0143] a t =[ΔI t , ΔV t , Δv t ]

[0144] 3. Reward Function Design

[0145] Define reward function based on welding quality (defect rate, temperature stability, pressure stability);

[0146] Reward function R t It is expressed as:

[0147] R t =-(α·Rd +β·R T +γ·R P )

[0148] in:

[0149] R t is the reward function; R d is the defect rate (the number of welding defects detected); R T is the absolute difference of temperature from the expected value; R P is the absolute difference of pressure from the expected value; α, β, γ are weight coefficients;

[0150] 4. Q-value function and deep neural network

[0151] The Q-value function Q(s, a) represents the expected cumulative reward for performing action a in state s. A deep neural network is used to approximate the Q-value function: Q(s, a; θ), where θ is the parameter of the neural network.

[0152] Deep Neural Network (DNN) Structure:

[0153] Input layer: state vector s t ;

[0154] Hidden layer: multiple layers of neurons;

[0155] Output layer: Q value Q(s, a; θ);

[0156] 5. Experience replay mechanism and target network

[0157] Store experience data(s t , a t , R t ,s t+1 ) to the experience replay pool, and then sample small batches of data from the experience replay pool for training to break data correlation;

[0158] Set up a target network Q(s, a; θ-);

[0159] Among them, θ- is the parameter of the target network, which is updated to the current network parameter every fixed number of steps;

[0160] 6. Loss Function and Gradient Update

[0161] Define the loss function L(θ):

[0162] L(θ)=E(s,a,R,s′)[(R+γmax a′ Q(s′, a′; θ - )-Q(s,a;θ)) 2 ]

[0163] Among them, γ is the discount factor;

[0164] Minimize the loss function by gradient descent and update the neural network parameters:

[0165]

[0166] Among them, α is the learning rate;

[0167] Repeat the above steps and continuously optimize the DQN model until the welding parameters reach the optimal value;

[0168] Compared with the prior art, the present invention has the following beneficial effects:

[0169] 1. Through this device, the driving assembly drives the two pressurized cylinders to rotate, so that the two welders can quickly weld the welding parts of the cylinder body, and the high air pressure in the pressurized groove provides thrust to keep the welding head of the welder in contact with the outer surface of the cylinder body. When the welding head of the welder contacts the edges or irregular positions of the outer surface of the cylinder body, the protrusion on the outer surface of the cylinder body pushes the welder back, so that the device can quickly weld the cylinder bodies with different external shapes, without adjusting the welding route of the welder when welding different cylinder bodies.

[0170] 2. Through this device, the pressure relief component controls the maximum air pressure in the pressurized tank, and then controls the thrust on the welder, to avoid the welder being squeezed and damaged when the welder encounters the protrusion on the surface of the cylinder body and the air pressure is too high to make the welder slide back.

[0171] 3. Through this device, before the pressurized push plate slides to a position lower than the connecting air hole, the air in the pressurized groove is depressurized once, resulting in the air pressure in the pressurized groove being lower than one atmosphere. The negative pressure generates suction to pull the air pressure push rod back to its original position, completing the resetting of the welder until the next welding.

[0172] 4. Through this device, temperature sensors, pressure sensors and high-resolution cameras are installed at key positions of the welding device, and the present invention realizes comprehensive real-time monitoring of the welding process. These sensors can collect temperature, pressure and image data in the welding process in real time, thereby providing comprehensive process monitoring information. Through this device, a welding defect detection model constructed using an improved convolutional neural network (CNN) can accurately identify welding defects and determine the defect type by analyzing real-time image data during the welding process. This solves the problem of low efficiency and low accuracy of traditional manual defect detection. Through this device, a deep reinforcement learning (DRL) algorithm is introduced to construct an adaptive control model. The model can output the optimal welding parameter adjustment strategy based on real-time monitoring data (including temperature, pressure, and welding defect detection results), thereby realizing adaptive control of the welding process. BRIEF DESCRIPTION OF THE DRAWINGS

[0173] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0174] Figure 1 It is a structural stereogram of the present invention;

[0175] Figure 2 It is an exploded cross-sectional view of the first structure in the present invention;

[0176] Figure 3 It is an exploded cross-sectional view of the second structure in the present invention;

[0177] Figure 4 It is an exploded cross-sectional view of the third structure in the present invention;

[0178] Figure 5 It is an exploded cross-sectional view of the fourth structure in the present invention;

[0179] Figure 6 For the present invention Figure 3 The enlarged view of point A in the middle;

[0180] Figure 7 For the present invention Figure 3 The enlarged view of point B in the middle;

[0181] Figure 8 For the present invention Figure 3 Enlarged view of point C in the middle;

[0182] Fig. 9 For the present invention Figure 5 The enlarged view of point D in the middle;

[0183] Fig.10 This is a flow chart of the intelligent control algorithm of the present invention.

[0184] In the figure: 1. Cylinder body; 2. Welding support plate; 3. Pressurizing cylinder; 4. Sliding groove; 5. Pressurizing groove; 6. Air pressure push cylinder; 7. Air pressure push rod; 8. Welder; 9. Pressurizing push plate; 10. Pressurizing push rod; 11. Connecting air hole; 12. Welding base; 13. Limiting rotating groove; 14. Limiting rotating plate; 16. Pressure relief cylinder; 17. Upper opening; 18. Pressure relief port; 19. Pressure relief push plate; 20. Pressure relief push rod; 21. Pressure relief spring; 22. Fixed plate; 23. Arc push plate; 24. Welding hole; 25. Limiting slide bar; 26. Limiting sleeve; 27. Lifting platform; 28. Lifting platform; 29. ​​Lifting sleeve; 30. Lifting groove; 31. Driving cylinder; 32. Support sleeve; 33. Support groove; 34. Driving motor; 35. Cylinder fixing seat. DETAILED DESCRIPTION

[0185] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0186] Example 1

[0187] See also Figure 1-Figure 9 , the present invention provides the following technical solutions:

[0188] An oil cylinder and a manufacturing device thereof, comprising:

[0189] Cylinder body 1;

[0190] A welding base 12, which is arranged on the lower side of the oil cylinder body 1;

[0191] The oil cylinder fixing seat 35 is provided at the upper end of the welding base 12, and the oil cylinder body 1 is fixedly connected in the oil cylinder fixing seat 35;

[0192] A limit rotation groove 13, which is provided at the upper end of the welding base 12;

[0193] The boosting assembly is provided with two groups, each group of boosting assemblies includes a pressurizing cylinder 3, a sliding groove 4, a pressurizing groove 5, a pneumatic push cylinder 6, a pneumatic push rod 7, a welder 8, a pressurizing push plate 9, a pressurizing push rod 10, a connecting air hole 11 and a limit rotating plate 14. The pressurizing cylinder 3 is arranged on one side of the oil cylinder body 1, the sliding groove 4 is opened at the lower end of the pressurizing cylinder 3, the pressurizing groove 5 is opened on the upper inner wall of the sliding groove 4, the pneumatic push cylinder 6 is fixedly connected to the circumferential surface of the pressurizing cylinder 3, the pneumatic push rod 7 is slidably connected in the pneumatic push cylinder 6, the welder 8 is fixedly connected to one side end of the pneumatic push rod 7, the pressurizing push plate 9 is slidably connected to the circumferential inner wall of the sliding groove 4, the limit rotating plate 14 is slidably connected in the limit rotating groove 13, the pressurizing push rod 10 is fixedly connected to the upper end of the limit rotating plate 14 and contacts with the lower end of the pressurizing push plate 9, and the connecting air hole 11 is opened on the circumferential inner wall of the sliding groove 4.

[0194] In a specific embodiment of the present invention, the cylinder body 1 is a device base, and the cylinder fixing seat 35 is arranged at the upper end of the cylinder body 1 for fixing the cylinder body 1. The welder 8 is a welding device. The two welders 8 are connected to the booster assembly to perform high-speed welding on the outer surface of the cylinder body 1. The pressurizing push rod 10 is limited by the limit rotating plate 14, so that the pressurizing push rod 10 can only slide in the limit rotating groove 13 following the limit rotating plate 14 but cannot move vertically. When it is necessary to weld the cylinder body 1, the driving assembly drives the pressurizing cylinder 3 to move downward. As the pressurizing cylinder 3 moves, the pressurizing push rod 10 is inserted into the sliding groove 4 and contacts the pressurizing push plate 9. At this time, the air pressure in the sliding groove 4 and the pressurizing groove 5 is a normal atmospheric pressure, and the pressurizing push plate 9 is pushed to move upward. When the pressurizing push plate 9 slides to the upper end of the connecting air hole 11, the sliding groove 4 and the pressurizing groove 5 are in a sealed state, the lower end is blocked by the pressurizing push plate 9, and the side end is blocked by the air pressure push rod 7 The upper end is blocked by the pressure relief push plate 19, and as the space in the sliding groove 4 and the pressurizing groove 5 of the pressurizing push plate 9 decreases, the air pressure in the pressurizing groove 5 gradually increases, and the high pressure pushes the air pressure push rod 7 to slide to the other side of the air pressure push cylinder 6, thereby pushing the welder 8 to move forward until the output head of the welder 8 is against the outer surface of the cylinder body 1. When the welder 8 moves to the welding position, the driving assembly drives the two pressurizing cylinders 3 to rotate, so that the two welders 8 can quickly weld the welding parts of the cylinder body 1, and the high air pressure in the pressurizing groove 5 provides thrust to make the welding head of the welder 8 always contact with the outer surface of the cylinder body 1. When the welding head of the welder 8 contacts the edges or irregular positions of the outer surface of the cylinder body 1, the protrusion on the outer surface of the cylinder body 1 pushes the welder 8 back, so that the device can quickly weld the cylinder bodies 1 with different external shapes, and there is no need to adjust the welding route of the welder 8 when welding different cylinder bodies 1.

[0195] For details, please refer to Figure 1-Figure 9 The upper end of the welding base 12 is fixedly connected with a lifting platform 27, and the upper end of the lifting platform 27 is fixedly connected with two lifting sleeve rods 29. A lifting platform 28 is provided on the upper side of the lifting platform 27, and two lifting grooves 30 are provided on the upper end of the lifting platform 28. The lifting platform 28 is slidably connected to the circumferential surface of the two lifting sleeve rods 29 through the two lifting grooves 30. The upper end of the lifting platform 28 is fixedly connected with a driving cylinder 31, and the output end of the driving cylinder 31 is connected to the upper end of the lifting platform 27.

[0196] In this embodiment: the lifting sleeve rod 29 at the upper end of the lifting platform 27 is used to limit the movement of the lifting platform 28 so that the lifting platform 28 will not deviate when sliding. The telescopic end of the driving cylinder 31 passes through the lifting platform 28 and is fixed to the lifting platform 27. When the cylinder body 1 needs to be welded, the driving cylinder 31 drives the lifting platform 28 to move downward to lower the welder 8 to the welding position, and increase the air pressure in the pressurized tank 5 to push the welding head of the welder 8 to move to the welding position.

[0197] For details, please refer to Figure 1-Figure 9 A support sleeve 32 is fixedly connected to one side end of the lifting platform 28, and the upper ends of the two pressurizing cylinders 3 are fixedly connected to the welding support plate 2. A support groove 33 is opened at the upper end of the welding support plate 2. The support sleeve 32 rotates in the support groove 33. The upper end of the support sleeve 32 is fixedly connected to a drive motor 34, and the output end of the drive motor 34 is fixedly connected to the upper end of the welding support plate 2.

[0198] In this embodiment: the support sleeve 32 is cylindrical and inserted into the support groove 33 to support the welding support plate 2, and the welding support plate 2 can rotate at the lower end of the lifting platform 28. The two pressurizing cylinders 3 are fixed at the lower end of the welding support plate 2. The output end of the drive motor 34 passes through the support sleeve 32 and is fixed to the upper end of the welding support plate 2. When the lifting platform 28 moves down to the welding position, the drive motor 34 drives the welding support plate 2 to rotate, thereby enabling the two welders 8 to perform rotational welding on the cylinder body 1.

[0199] For details, please refer to Figure 1-Figure 9 A pressure relief assembly is provided in each group of boosting components, and each group of pressure relief components includes a pressure relief cylinder 16, a pressure relief port 18, a pressure relief push plate 19, a pressure relief push rod 20 and a pressure relief spring 21. The pressure relief cylinder 16 is fixedly connected to the upper inner wall of the pressurizing groove 5, the pressure relief port 18 is opened on the circumferential inner wall of the pressure relief cylinder 16, the pressure relief push plate 19 is slidably connected in the pressure relief cylinder 16, the pressure relief push rod 20 is fixedly connected to the upper end of the pressure relief push plate 19 and slides in the pressure relief cylinder 16, and the pressure relief spring 21 is fixedly connected to the upper inner wall of the pressure relief cylinder 16 and the upper end of the pressure relief push rod 20.

[0200] In this embodiment: the pressure relief push plate 19 is arranged in the pressure relief cylinder 16 to seal the upper end of the pressurizing groove 5. When the air pressure in the pressurizing groove 5 becomes larger and larger as the pressurizing cylinder 3 moves downward, the air pressure thrust in the pressurizing groove 5 overcomes the resistance of the pressure relief spring 21 to push the pressure relief push plate 19 upward. The sliding groove 4 and the pressurizing groove 5 divide the space in the pressurizing cylinder 3 into two equal parts. When the pressurizing push plate 9 slides to the upper inner wall close to the sliding groove 4 so that the air pressure in the pressurizing groove 5 is greater than 160 kPa, the pressure relief push plate 19 is pushed to the upper end of the pressure relief port 18 by the air pressure in the oil cylinder body 1. At this time, the air in the pressurizing groove 5 is discharged from the pressure relief port 18, thereby reducing the pressure in the pressurizing groove 5. After the air pressure drops below 160 kPa, the pressure relief spring 21 pushes the pressure relief push plate 19 upward. The plate 19 pushes back the lower end of the pressure relief port 18 to re-seal the space in the pressurized groove 5, and the maximum air pressure in the pressurized groove 5 is controlled by the pressure relief component, thereby controlling the thrust on the welder 8 to prevent the welder 8 from being squeezed and damaged due to the air pressure being too high to make the welder 8 slide back after the welder 8 encounters the protrusion on the surface of the cylinder body 1. When the welding is completed and the pressurized cylinder 3 is driven to move upward, the pressurized push plate 9 slides downward in the sliding groove 4. Before the pressurized push plate 9 slides to a position lower than the connecting air hole 11, the air in the pressurized groove 5 has been depressurized once, resulting in the air pressure in the pressurized groove 5 being lower than one atmosphere. The negative pressure generates suction to pull the air pressure push rod 7 back to its original position, completing the resetting of the welder 8 until the next welding is performed.

[0201] For details, please refer to Figure 1-Figure 9 An upper opening 17 is provided on the upper inner wall of the pressure relief cylinder 16 .

[0202] In this embodiment, the upper opening 17 is opened above the pressure relief cylinder 16. When the pressure relief push plate 19 slides upward, the air pressure at the upper end of the pressure relief push rod 20 will not change due to the opening of the upper opening 17, thereby avoiding affecting the pressure relief in the pressurized groove 5.

[0203] For details, please refer to Figure 1-Figure 9 The outer surfaces of the two welders 8 are fixedly connected with a fixing plate 22, one side end of the two fixing plates 22 is fixedly connected with an arc-shaped push plate 23, and one side end of the two arc-shaped push plates 23 is provided with a welding hole 24.

[0204] In this embodiment: the fixing plates 22 are arranged on both sides of the welder 8 for fixing the arc-shaped push plate 23, the welding holes 24 are opened at the welding heads of the welder 8, the welding heads weld the surface of the cylinder body 1 through the welding holes 24, and the arc-shaped push plate 23 contacts the outer surface of the cylinder body 1 instead of the welder 8, avoiding direct contact between the welding head of the welder 8 and the cylinder body 1, causing the welder 8 to be squeezed and damaged.

[0205] For details, please refer to Figure 1-Figure 9Two limiting slide bars 25 are fixedly connected to one side end of the two fixing plates 22, and two limiting sleeves 26 are fixedly connected to the circumferential surfaces of the two pressurizing cylinders 3. The plurality of limiting slide bars 25 slide in the plurality of limiting sleeves 26 respectively.

[0206] In this embodiment: the limiting slide rod 25 is arranged on both sides of the welder 8 and slides in the limiting sleeve 26 to limit the moving direction of the welder 8, thereby preventing the pneumatic push rod 7 from pushing the welder 8 to move forward and backward for welding and causing the position to shift, thereby ensuring the fixation of the welding point.

[0207] For details, please refer to Figure 1-Figure 9 The circumferential surfaces of the pressure push plate 9 and the pressure relief push plate 19 are fixedly connected with a plurality of sealing rings.

[0208] In this embodiment: the circumferential surfaces of the pressure relief push plate 19 and the pressure push plate 9 are both provided with rubber sealing rings, which are used to seal the air in the sliding groove 4 and the pressure groove 5 to prevent the high-pressure air in the sliding groove 4 and the pressure groove 5 from leaking out when the pressure push plate 9 and the pressure relief push plate 19 slide.

[0209] For details, please refer to Figure 1-Figure 9 A buffer ring is provided at the upper end of the oil cylinder body 1, and a sealing sleeve and a protective shell are provided on the outer surface of the oil cylinder body 1.

[0210] In this embodiment: a buffer ring is provided at the upper end of the cylinder body 1 for buffering when the piston inside the cylinder body 1 hits the end of the cylinder body 1. The outer surface of the cylinder body 1 is also provided with components such as a pipe joint seat, all of which are welded after driving the welder 8 to move to different heights of the cylinder body 1.

[0211] For details, please refer to Figure 1-Figure 9 One side surface of the welding hole 24 is arc-shaped, and the output end of the welder 8 is arranged in the welding hole 24.

[0212] In this embodiment: the contact surface between the arc push plate 23 and the cylinder body 1 is arc-shaped, so that the protrusion on the surface of the cylinder body 1 will not jam the arc push plate 23 when the arc push plate 23 rotates, and the protruding part pushes the arc surface of the arc push plate 23 to push the arc push plate 23 back. The output head of the welder 8 is arranged in the welding hole 24, and a certain distance is set from the arc surface of the welding hole 24, so that the welding head is closer to the surface of the cylinder body 1 for effective welding, and does not directly contact the cylinder body 1 to avoid scratching the surface of the weld.

[0213] It should be noted that the welding device 8, driving cylinder 31 and driving motor 34 used in the present device are all prior art and will not be described in detail here.

[0214] The working principle and use process of the present invention are as follows: when it is necessary to weld the cylinder body 1, the driving cylinder 31 drives the lifting platform 28 to move downward to lower the welder 8 to the welding position, and increases the air pressure in the pressurizing groove 5 to push the welding head of the welder 8 to move to the welding position. After the lifting platform 28 moves down to the welding position, the driving motor 34 drives the welding support plate 2 to rotate, thereby causing the two welders 8 to perform rotational welding on the cylinder body 1. When it is necessary to weld the cylinder body 1, the driving assembly drives the pressurizing cylinder 3 to move downward. As the pressurizing cylinder 3 moves, the pressurizing push rod 10 is inserted into the sliding groove 4 and contacts the pressurizing push plate 9. At this time, the air pressure in the sliding groove 4 and the pressurizing groove 5 is a normal atmospheric pressure, and the pressurizing push plate 9 is pushed to move upward. When the pressurizing push plate 9 slides to the upper end of the connecting air hole 11, the sliding groove 4 and the pressurizing groove 5 are in a sealed state, the lower end is blocked by the pressurizing push plate 9, and the side end is blocked by the air pressure push rod 7 The upper end is blocked by the pressure relief push plate 19, and as the space in the sliding groove 4 and the pressurizing groove 5 of the pressurizing push plate 9 is reduced, the air pressure in the pressurizing groove 5 gradually increases, and the high pressure pushes the air pressure push rod 7 to slide to the other side of the air pressure push cylinder 6, thereby pushing the welder 8 to move forward until the output head of the welder 8 is against the outer surface of the cylinder body 1. When the welder 8 moves to the welding position, the driving assembly drives the two pressurizing cylinders 3 to rotate, so that the two welders 8 can quickly weld the welding parts of the cylinder body 1, and the high air pressure in the pressurizing groove 5 provides thrust to make the welding head of the welder 8 always contact with the outer surface of the cylinder body 1. When the welding head of the welder 8 contacts the edges or irregular positions of the outer surface of the cylinder body 1, the protrusion on the outer surface of the cylinder body 1 pushes the welder 8 back, so that the device can quickly weld the cylinder bodies 1 with different external shapes, and there is no need to adjust the welding route of the welder 8 when welding different cylinder bodies 1.

[0215] In order to achieve real-time monitoring and adaptive control, based on the existing oil cylinder and its manufacturing device, a machine learning algorithm is introduced to optimize the welding process and improve production efficiency. The specific process is as follows:

[0216] Step 1: Installation location and selection of sensors and cameras: In order to achieve effective real-time monitoring and adaptive control in the welding device, temperature sensors, pressure sensors and high-resolution cameras need to be installed at key locations;

[0217] 1. Temperature sensor:

[0218] Installation location: The temperature sensor should be installed close to the welding point to ensure that the high temperature generated during the welding process can be accurately measured;

[0219] Sensor type: Use K-type thermocouple with high temperature resistance and fast response speed;

[0220] Temperature range: 0℃-2000℃;

[0221] Accuracy: ±0.5℃;

[0222] Response time: <1 second;

[0223] 2. Pressure sensor:

[0224] Installation location: The pressure sensor is installed in the pipeline between the air pressure push cylinder and the welder to monitor the pressure applied during welding;

[0225] Sensor type: Choose a high-precision, stable pressure sensor (such as a MEMS pressure sensor);

[0226] Pressure range: 0-10MPa;

[0227] Accuracy: ±0.1MPa;

[0228] Response time: <1 ms;

[0229] 3. High-resolution camera:

[0230] Installation location: The high-resolution camera should be installed just above or to the side of the welding point to ensure that the image data of the welding process can be clearly captured;

[0231] Camera type: Use an industrial-grade camera with 5 megapixels and a frame rate of 30fps, which has high resolution and high-speed shooting capabilities;

[0232] Parameter variables:

[0233] Resolution: 5 megapixels;

[0234] Frame rate: 30fps

[0235] Exposure time: automatic adjustment;

[0236] Step 2, data collection and preprocessing: During the welding process of the oil cylinder and its manufacturing device, sensors and cameras are used to collect temperature, pressure and image data in real time, and these data are processed through data preprocessing technology to ensure the quality and consistency of the input data;

[0237] 1. Data acquisition parameter settings

[0238] Temperature collection frequency: 10 times per second, the collected temperature data is T(t);

[0239] Pressure collection frequency: 10 times per second, the collected pressure data P(t);

[0240] Image acquisition frequency: 5 frames of images are collected per second, image data I(t);

[0241] 2. Data Preprocessing

[0242] Temperature and pressure data preprocessing:

[0243] Denoising: Use a sliding average filter to denoise the temperature and pressure data, and the sliding window size is set to 5;

[0244]

[0245]

[0246] Among them, T filtered and P filtered They are the temperature and pressure data after denoising respectively;

[0247] Standardization: Standardize the temperature and pressure data to make their mean 0 and standard deviation 1;

[0248]

[0249]

[0250] Among them, μ T and σ T are the mean and standard deviation of the temperature data, μ P and σ P are the mean and standard deviation of the pressure data, T normalized and P normalized They are the temperature and pressure data after standardization;

[0251] Image data preprocessing:

[0252] Grayscale processing: Convert color images into grayscale images to reduce the amount of data and simplify subsequent processing;

[0253] I gray (x, y) = 0.299·I R (x,y)+0.587·I G (x,y)+0.114·I B (x, y)

[0254] Image denoising: Use Gaussian filtering to denoise the image; the Gaussian filter parameters are set to kernel size 3×3 and standard deviation σ=1.0;

[0255]

[0256] Image normalization: normalize the image pixel values ​​to the range of [0, 1];

[0257]

[0258] Step 3: Use the improved convolutional neural network (CNN) to build a welding defect detection model. Improve the traditional CNN and use the deep residual network (ResNet) and attention mechanism (Attention Mechanism) to enhance the performance of the model. The model inputs real-time image data during the welding process and outputs whether there are welding defects and the defect type. The specific steps are as follows:

[0259] 1. Data augmentation: Use techniques such as rotation, scaling, translation, and flipping to augment the dataset:

[0260] Rotation: randomly rotate the image by a certain angle;

[0261] Scale: Randomly scale the image;

[0262] Translation: Randomly translate the image;

[0263] Flip: randomly flip the image horizontally or vertically;

[0264] 2. Data trimming or padding:

[0265] Crop or pad the image to a uniform size of 224×224;

[0266] 3. Convolutional layer and pooling layer

[0267] Convolutional layer: The first layer uses 64 7×7 convolution kernels with a stride of 2 and a padding of 3;

[0268]

[0269] Where X is the input image, W is the convolution kernel, b is the bias, h is the output feature map, i and j are the height and width of the feature map in spatial coordinates, k is the channel index of the output feature map, c is the channel index of the input feature map, M and N are the height and width of the convolution kernel, respectively;

[0270] Calculated output size: 112×112×64;

[0271] Pooling layer: The maximum pooling layer uses a 3×3 pooling window with a stride of 2 and a padding of 1;

[0272] P i,j,k =max m,n h i+m-1，j+n-1，k

[0273] Where p is the pooling output and max is the maximum pooling operation;

[0274] Calculated output size: 56×56×64;

[0275] 4. Residual Block

[0276] Each residual block consists of two convolutional layers and their residual connections:

[0277] y=F(x,{W i})+x

[0278] Among them, F(x, {W i}) represents the learned residual function, x is the input feature map, y is the residual output, W i is the convolution kernel in the residual block, i is the number of convolution layers;

[0279] The specific formula of the residual block is:

[0280] The first convolution layer: 64×3×3, stride 1, padding 1;

[0281] F1(x)=ReLU(W1·x+b1)

[0282] The second convolution layer: 64×3×3, stride 1, padding 1;

[0283] F2(x)=W2·F1(x)+b2

[0284] Residual output:

[0285] y = ReLU(F2(x)+x)

[0286] The output size remains the same: 56×56×64;

[0287] 5. Attention Mechanism

[0288] Calculate the output of self-attention:

[0289]

[0290] Where Q is the query matrix, which represents the set of query vectors; K is the key matrix, which represents the set of key vectors; V is the value matrix, which represents the set of value vectors; Q = K = V = X (input feature map); d k is the dimension of the key vector; Attention(Q, K, V) is the output of the attention mechanism;

[0291] Calculation process:

[0292] Q=W Q ·X

[0293] K=W K ·X

[0294] V=W V ·X

[0295] Among them, W Q , W K , W VThey are the weight matrices for query, key, and value respectively;

[0296] Calculate the attention score:

[0297]

[0298] Among them, Attention_Scores is the attention score matrix;

[0299] Apply the Softmax function:

[0300] Attention_Weights=softmax(Attention_Scores)

[0301] Among them, Attention_Weights is the attention weight after softmax processing;

[0302] Compute the attention output:

[0303] Attention_Output=Attention_Weight·V

[0304] Among them, Attention_Output is the output matrix of the attention mechanism;

[0305] 6. Fully connected layer

[0306] Flatten the feature map into a one-dimensional vector:

[0307] Flatten_Output=flatten(Attention_Output)

[0308] Among them, Flatten_Output is the flattened output vector;

[0309] Output size: 7×7×512=25088;

[0310] After the first fully connected layer, the ReLU activation function is used:

[0311] FC1=ReLU(W fc1 Flatten_Output+b fc1 )

[0312] Among them, FC1 is the output of the first fully connected layer; W fc1 is the weight matrix of the first fully connected layer; b fc1 is the bias vector of the first fully connected layer;

[0313] The second fully connected layer uses the Softmax activation function for multi-classification:

[0314] Output = Softmax(W fc2 FC1+b fc2 )

[0315] Among them, W fc2 is the weight matrix of the second fully connected layer; b fc2 is the bias vector of the second fully connected layer; Softmax is the Softmax activation function;

[0316] Output size: 7 (including 7 categories: normal, crack, pore, undercut, weld accumulation, deviation, and concave);

[0317] 7. Model Training

[0318] Using the cross entropy loss function:

[0319]

[0320] Among them, L is the loss function value; N is the number of samples; C is the number of categories; yi j is the actual label of the jth category of sample i; is the predicted probability of the jth class for sample i;

[0321] Use the Adam optimizer to perform gradient descent optimization and adjust the model weights to minimize the loss function;

[0322] Step 4: During the welding process, deep reinforcement learning (DRL) technology is used to achieve adaptive control. The deep Q network (DQN) algorithm is used. DQN combines Q learning and deep neural networks and is suitable for processing high-dimensional input data from multiple sensors and images during the welding process.

[0323] 1. Environment and state definition

[0324] Environment definition: Oil cylinder manufacturing device, including oil cylinder body, welding base, oil cylinder fixing seat, limit rotation groove, and booster assembly;

[0325] State vector s t :

[0326] s t =[I t , V t , v t , T t , P t , D t ]

[0327] Among them, I t is the current; V t is the voltage; v t is the welding speed; T t is temperature; Pt For pressure; D t Welding defects;

[0328] 2. Action definition

[0329] Adjust welding parameters: current I, voltage V, welding speed v;

[0330] Action vector a t :

[0331] a t =[ΔI t , ΔV t , Δv t ]

[0332] 3. Reward Function Design

[0333] Define reward function based on welding quality (defect rate, temperature stability, pressure stability);

[0334] Reward function R t It is expressed as:

[0335] R t =-(α·R d +β·R T +γ·R P )

[0336] in:

[0337] R d is the defect rate (the number of welding defects detected); R T is the absolute difference of temperature from the expected value; R P is the absolute difference of pressure from the expected value; α, β, γ are weight coefficients;

[0338] 4. Q-value function and deep neural network

[0339] The Q-value function Q(s, a) represents the expected cumulative reward for performing action a in state s. A deep neural network is used to approximate the Q-value function: Q(s, a; θ); where θ is the parameter of the neural network.

[0340] Deep Neural Network (DNN) Structure:

[0341] Input layer: state vector s t ;

[0342] Hidden layer: multiple layers of neurons;

[0343] Output layer: Q value Q(s, a; θ);

[0344] 5. Experience replay mechanism and target network

[0345] Store experience data(s t , a t , R t ,s t+1 ) to the experience replay pool, and then sample small batches of data from the experience replay pool for training to break data correlation;

[0346] Set a target network Q(s, a; θ - );

[0347] 6. Loss Function and Gradient Update

[0348] Define the loss function L(θ):

[0349] L(θ)=E(s,a,R,s′)[(R+γmax a′ Q(s′, a′; θ - )-Q(s,a;θ)) 2 ]

[0350] Among them, γ is the discount factor;

[0351] Minimize the loss function by gradient descent and update the neural network parameters:

[0352]

[0353] Among them, α is the learning rate;

[0354] Repeat the above steps and continuously optimize the DQN model until the welding parameters reach the optimal value;

[0355] Numerical case analysis:

[0356] Step 1, data collection and preprocessing:

[0357] Temperature data (raw data for the past 5 seconds): 1450, 1480, 1500, 1520, 1490 (unit: degrees Celsius);

[0358] Pressure data (raw data for the past 5 seconds): 4.8, 5.0, 5.2, 5.1, 5.0 (unit: MPa);

[0359] Image data (original image at a certain moment): Assume that the image resolution is 100×100 pixels, the original color image data I R , I G , I B (Red, green and blue channels respectively);

[0360] 1. Denoising:

[0361]

[0362]

[0363] Among them, T filtered and P filtered They are the temperature and pressure data after denoising respectively;

[0364] 2. Standardization processing:

[0365]

[0366]

[0367] Among them, T normalized and P normalized They are the temperature and pressure data after standardization;

[0368] 3. Grayscale processing:

[0369] The original color of a pixel is I R (x, y) = 100, I G (x, y) = 150, I B (x, y) = 200.

[0370] I gray (x, y)=0.299·100+0.587·150+0.114·200=140

[0371] 4. Image denoising:

[0372]

[0373] Assume the result is 135.

[0374] 5. Image Standardization:

[0375] Assume that the minimum pixel value of the image is 100 and the maximum pixel value is 200.

[0376]

[0377] Step 2, welding defect detection algorithm:

[0378] Input image size: 224×224×3;

[0379] Input image value range: [0, 1];

[0380] Convolutional Layer 1:

[0381] Number of convolution kernels: 64;

[0382] Convolution kernel size: 7×7;

[0383] Step length: 2;

[0384] padding: 3;

[0385] Compute the convolution:

[0386]

[0387] Where X is the input image, W is the convolution kernel, b is the bias, h is the output feature map, i and j are the height and width of the feature map in spatial coordinates, k is the channel index of the output feature map, c is the channel index of the input feature map, M and N are the height and width of the convolution kernel, respectively; the output feature map size is 112×112×64;

[0388] Max Pooling Layer 1:

[0389] Pooling window size: 3×3;

[0390] Step length: 2;

[0391] padding: 1;

[0392] Pooling operation:

[0393] p i，j，k =max m，n h i+m-1，j+n-1，k

[0394] Where p is the pooling output and max is the maximum pooling operation;

[0395] Output feature map size: 56×56×64;

[0396] Residual Block 1:

[0397] The first convolution layer:

[0398] Convolution kernel size: 3×3;

[0399] Step length: 1;

[0400] padding: 1;

[0401] Calculate the first layer of convolution:

[0402] F1(x) = ReLU(W1 x + b1) Output feature layer size: 56 × 56 × 64;

[0403] The second convolution layer:

[0404] Convolution kernel size: 3×3;

[0405] Step length: 1;

[0406] padding: 1;

[0407] Calculate the second layer of convolution:

[0408] F2(x)=W2·F1(x)+b2 Output feature layer size: 56×56×64;

[0409] Residual Connection:

[0410] y = ReLU(F2(x)+x)

[0411] Output feature size: 56×56×64;

[0412] Attention Mechanism:

[0413] Assume that the input feature map size is 7×7×512;

[0414] Compute the query matrix Q, key matrix K, and value matrix V:

[0415] Q=W Q ·X

[0416] K=W K ·X

[0417] V=W V ·X

[0418] W Q , W Q , W V All are weight matrices of size 512×512;

[0419] Calculate the attention score:

[0420]

[0421] Among them, Attention_Scores is the attention score matrix;

[0422] Assumption d k is 512;

[0423] Apply the Softmax function and calculate the attention output:

[0424] Attention_Weights=softmax(Attention_Scores)

[0425] Attention_Output=Attention_Weight·V

[0426] Among them, Attention_Weights is the attention weight after softmax processing; Attention_Output is the output matrix of the attention mechanism;

[0427] Output feature map size: 7×7×512;

[0428] Fully connected layer:

[0429] Flatten the feature map into a one-dimensional vector Flatten_Output:

[0430] Flatten_Output=flatten(Attention_Output)

[0431] Flattened vector size: 25088;

[0432] The first fully connected layer:

[0433] Input: 25088;

[0434] Output: 512;

[0435] FC1=ReLU(W fc1 Flatten_Output+b fc1 )

[0436] Among them, FC1 is the output of the first fully connected layer; W fc1 is the weight matrix of the first fully connected layer; b fc1 is the bias vector of the first fully connected layer;

[0437] Output vector size: 512;

[0438] The second fully connected layer:

[0439] Input: 512;

[0440] Output: 7;

[0441] Output = Softmax(W fc2 FC1+b fc2 )

[0442] Among them, W fc2 is the weight matrix of the second fully connected layer; b fc2 is the bias vector of the second fully connected layer; Softmax is the Softmax activation function;

[0443] Output vector size: 7 (7 types of welding defects);

[0444] Step 3, Deep Reinforcement Learning Model:

[0445] Initial welding parameters: I0=100A, V0=20V, v0=5mm / s;

[0446] Current Status t =[100,20,5,1500,5,0](no defects, temperature 1500℃, pressure 5Mpa);

[0447] Select action a according to the epsilon-greedy strategy t :

[0448] a t =[ΔI t =5A, ΔV t =-1V,Δv t =0.5mm / s]

[0449] Action: Adjust welding parameters:

[0450] I t+1 =105A

[0451] V t+1 =19V

[0452] v t+1 =5.5mmm / s

[0453] Reward calculation:

[0454] Defect rate R d =0.01;

[0455] Temperature deviation R T =|1500-1500|=0;

[0456] Pressure deviation R P =| 5-5| = 0;

[0457] Reward value: R t =-(α·0.01+β·0+γ·0)=-0.01;

[0458] Storage experience(s) t , a t , R t ,s t+1 );

[0459] Update Q value:

[0460]

[0461] The neural network parameters θ are updated by the gradient descent method until the welding parameters reach the optimal value.

[0462] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein by equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A manufacturing device for an oil cylinder, characterized in that: include: Cylinder body (1); A welding base (12), wherein the welding base (12) is arranged on the lower side of the oil cylinder body (1); An oil cylinder fixing seat (35), wherein the oil cylinder fixing seat (35) is disposed at the upper end of the welding base (12), and the oil cylinder body (1) is fixedly connected to the oil cylinder fixing seat (35); A limit rotation groove (13), wherein the limit rotation groove (13) is provided at the upper end of the welding base (12); A booster assembly, wherein the booster assembly is provided with two groups, each group of the booster assembly comprises a pressurizing cylinder (3), a sliding groove (4), a pressurizing groove (5), a pneumatic push cylinder (6), a pneumatic push rod (7), a welder (8), a pressurizing push plate (9), a pressurizing push rod (10), a connecting air hole (11) and a limit rotating plate (14), wherein the pressurizing cylinder (3) is provided on one side of the oil cylinder body (1), the sliding groove (4) is provided at the lower end of the pressurizing cylinder (3), the pressurizing groove (5) is provided on the upper inner wall of the sliding groove (4), and the pneumatic push cylinder (6) is fixed Connected to the circumferential surface of the pressurizing cylinder (3), the pneumatic push rod (7) is slidably connected to the inside of the pneumatic push cylinder (6), the welder (8) is fixedly connected to one side end of the pneumatic push rod (7), the pressurizing push plate (9) is slidably connected to the circumferential inner wall of the sliding groove (4), the limit rotating plate (14) is slidably connected to the inside of the limit rotating groove (13), the pressurizing push rod (10) is fixedly connected to the upper end of the limit rotating plate (14) and contacts with the lower end of the pressurizing push plate (9), and the connecting air hole (11) is opened on the circumferential inner wall of the sliding groove (4); The oil cylinder body (1) is a device base, the oil cylinder fixing seat (35) is arranged at the upper end of the oil cylinder body (1) for fixing the oil cylinder body (1), the welder (8) is a welding device, and the two welders (8) are connected to the booster assembly to perform high-speed welding on the outer surface of the oil cylinder body (1), and the pressurizing push rod (10) is limited by the limit rotating plate (14), so that the pressurizing push rod (10) can only slide in the limit rotating groove (13) following the limit rotating plate (14) but cannot move vertically. When the oil cylinder body (1) needs to be welded When connected, the driving assembly drives the pressurizing cylinder (3) to move downward. As the pressurizing cylinder (3) moves, the pressurizing push rod (10) is inserted into the sliding groove (4) and contacts with the pressurizing push plate (9). At this time, the air pressure in the sliding groove (4) and the pressurizing groove (5) is a normal atmospheric pressure. The pressurizing push plate (9) is pushed to move upward. When the pressurizing push plate (9) slides to the upper end of the connecting air hole (11), the sliding groove (4) and the pressurizing groove (5) are in a sealed state. The lower end is blocked by the pressurizing push plate (9) and the side end is blocked by the air pressure push rod (7). The upper end is blocked by the pressure relief push plate (19), and as the pressure push plate (9) slides, the space in the sliding groove (4) and the pressure groove (5) decreases, and the air pressure in the pressure groove (5) gradually increases. The high pressure pushes the air pressure push rod (7) to slide to the other side of the air pressure push cylinder (6), thereby pushing the welder (8) forward until the output head of the welder (8) abuts against the outer surface of the cylinder body (1). When the welder (8) moves to the welding position, the driving assembly drives the two pressure cylinders (3) to rotate, so that the two welders (8) are pressed against the cylinder body (1). The welding part of the main body (1) is welded quickly, and the high gas pressure in the pressurized groove (5) provides a thrust to keep the welding head of the welder (8) in contact with the outer surface of the oil cylinder body (1). When the welding head of the welder (8) contacts the corners or irregular positions of the outer surface of the oil cylinder body (1), the protrusion on the outer surface of the oil cylinder body (1) pushes the welder (8) back, so that the device can weld oil cylinder bodies (1) with different external shapes quickly, without adjusting the welding route of the welder (8) when welding different oil cylinder bodies (1); The upper end of the welding base (12) is fixedly connected to a lifting platform (27), the upper end of the lifting platform (27) is fixedly connected to two lifting sleeve rods (29), a lifting platform (28) is provided on the upper side of the lifting platform (27), two lifting grooves (30) are provided on the upper end of the lifting platform (28), the lifting platform (28) is slidably connected to the circumferential surface of the two lifting sleeve rods (29) through the two lifting grooves (30), the upper end of the lifting platform (28) is fixedly connected to a driving cylinder (31), and the output end of the driving cylinder (31) is connected to the upper end of the lifting platform (27); The lifting sleeve rod (29) at the upper end of the lifting platform (27) is used to limit the movement of the lifting platform (28) so that the lifting platform (28) will not deviate when sliding. The telescopic end of the driving cylinder (31) passes through the lifting platform (28) and is fixed to the lifting platform (27). When the oil cylinder body (1) needs to be welded, the driving cylinder (31) drives the lifting platform (28) to move downward to lower the welder (8) to the welding position, and increases the air pressure in the pressurizing tank (5) to push the welding head of the welder (8) to move to the welding position. A support sleeve (32) is fixedly connected to one side end of the lifting platform (28); the upper ends of the two pressurizing cylinders (3) are fixedly connected to a welding support plate (2); a support groove (33) is provided at the upper end of the welding support plate (2); the support sleeve (32) rotates in the support groove (33); the upper end of the support sleeve (32) is fixedly connected to a drive motor (34); the output end of the drive motor (34) is fixedly connected to the upper end of the welding support plate (2); The support sleeve (32) is cylindrical and inserted into the support groove (33) to support the welding support plate (2), and the welding support plate (2) can rotate at the lower end of the lifting platform (28). The two pressurizing cylinders (3) are fixed at the lower end of the welding support plate (2). The output end of the driving motor (34) passes through the support sleeve (32) and is fixed to the upper end of the welding support plate (2). When the lifting platform (28) moves down to the welding position, the driving motor (34) drives the welding support plate (2) to rotate, thereby enabling the two welding devices (8) to perform rotational welding on the oil cylinder body (1); Each group of the boosting components is provided with a pressure relief component, and each group of the pressure relief components includes a pressure relief cylinder (16), a pressure relief port (18), a pressure relief push plate (19), a pressure relief push rod (20) and a pressure relief spring (21); the pressure relief cylinder (16) is fixedly connected to the upper inner wall of the pressurizing groove (5); the pressure relief port (18) is opened on the circumferential inner wall of the pressure relief cylinder (16); the pressure relief push plate (19) is slidably connected to the pressure relief cylinder (16); the pressure relief push rod (20) is fixedly connected to the upper end of the pressure relief push plate (19) and slides in the pressure relief cylinder (16); the pressure relief spring (21) is fixedly connected to the upper inner wall of the pressure relief cylinder (16) and the upper end of the pressure relief push rod (20); The pressure relief push plate (19) is arranged in the pressure relief cylinder (16) and is used to block the upper end of the pressurizing groove (5). When the air pressure in the pressurizing groove (5) increases as the pressurizing cylinder (3) moves downward, the air pressure in the pressurizing groove (5) overcomes the resistance of the pressure relief spring (21) and pushes the pressure relief push plate (19) upward. The sliding groove (4) and the pressurizing groove (5) divide the space in the pressurizing cylinder (3) into two equal parts. When the pressurizing push plate (9) slides to the upper inner wall close to the sliding groove (4) so ​​that the air pressure in the pressurizing groove (5) is greater than 160 kPa, the pressure relief push plate (19) is pushed to the upper end of the pressure relief port (18) by the air pressure in the oil cylinder body (1). At this time, the air in the pressurizing groove (5) is discharged from the pressure relief port (18), thereby reducing the pressure in the pressurizing groove (5). After the air pressure drops below 160 kPa, the pressure relief spring (21) pushes the pressure relief push plate (19) upward. The plate (19) pushes back the lower end of the pressure relief port (18) to reseal the space in the pressurized groove (5). The maximum air pressure in the pressurized groove (5) is controlled by the pressure relief component, thereby controlling the thrust on the welder (8) to prevent the welder (8) from being squeezed and damaged due to the excessive air pressure that prevents the welder (8) from sliding back after the welder (8) encounters the protruding portion on the surface of the oil cylinder body (1). When the welding is completed and the pressurized cylinder (3) is driven to move upward, the pressurized push plate (9) slides downward in the sliding groove (4). Before the pressurized push plate (9) slides to a position lower than the connecting air hole (11), the air in the pressurized groove (5) has been depressurized once, resulting in the air pressure in the pressurized groove (5) being lower than one atmospheric pressure. The negative pressure generates suction to pull the air pressure push rod (7) back to its original position, completing the reset of the welder (8) until the next welding is performed.

2. The manufacturing device of the oil cylinder according to claim 1, characterized in that: An upper opening (17) is formed on the upper inner wall of the pressure relief cylinder (16).

3. The manufacturing device of the oil cylinder according to claim 2, characterized in that: The outer surfaces of the two welders (8) are fixedly connected with a fixing plate (22), one side end of the two fixing plates (22) is fixedly connected with an arc-shaped push plate (23), and one side end of the two arc-shaped push plates (23) is provided with a welding hole (24).

4. The manufacturing device of the oil cylinder according to claim 3, characterized in that: Two limiting slide bars (25) are fixedly connected to one side end of the two fixing plates (22), two limiting sleeves (26) are fixedly connected to the circumferential surfaces of the two pressurizing cylinders (3), and the plurality of limiting slide bars (25) slide in the plurality of limiting sleeves (26) respectively.

5. The manufacturing device of the oil cylinder according to claim 4, characterized in that: The circumferential surfaces of the pressure push plate (9) and the pressure relief push plate (19) are both fixedly connected with a plurality of sealing rings.

6. The manufacturing device of the oil cylinder according to claim 5, characterized in that: The upper end of the oil cylinder body (1) is provided with a buffer ring, and the outer surface of the oil cylinder body (1) is provided with a sealing sleeve and a protective shell.

7. The oil cylinder and the manufacturing device thereof according to claim 6, characterized in that: One side surface of the welding hole (24) is arc-shaped, and the output end of the welder (8) is arranged in the welding hole (24).

8. The welding method for manufacturing a cylinder according to claim 7, characterized in that: The specific process is: Step 1, installation location and selection of sensors and cameras: In the welding device, temperature sensors, pressure sensors and high-resolution cameras are installed at key locations; 1.

1. Temperature sensor: Installation location: The temperature sensor should be installed close to the welding point to ensure that the high temperature generated during the welding process can be accurately measured; Sensor type: Use K-type thermocouple with high temperature resistance and fast response speed; Temperature range: 0℃-2000℃; Accuracy: ±0.5℃; Response time: <1 second; 1.

2. Pressure sensor: Installation location: The pressure sensor is installed in the pipeline between the air pressure push cylinder and the welder to monitor the pressure applied during welding; Sensor type: Use high-precision and stable MEMS pressure sensor; Pressure range: 0-10MPa; Accuracy: ±0.1MPa; Response time: <1 ms; 1.

3. High-resolution camera: Installation location: The high-resolution camera should be installed just above or to the side of the welding point to ensure that the image data of the welding process can be clearly captured; Camera type: Use an industrial-grade camera with 5 megapixels and a frame rate of 30fps, which has high resolution and high-speed shooting capabilities; Parameter variables: Resolution: 5 megapixels; Frame rate: 30fps Exposure time: automatic adjustment; Step 2, data collection and preprocessing: During the welding process of the oil cylinder and its manufacturing device, sensors and cameras are used to collect temperature, pressure and image data in real time, and these data are processed through data preprocessing technology to ensure the quality and consistency of the input data; 2.

1. Data acquisition parameter settings Temperature collection frequency: 10 times per second, the collected temperature data is T(t); Pressure collection frequency: 10 times per second, the collected pressure data P(t); Image acquisition frequency: 5 frames of images are collected per second, image data I(t); Data Preprocessing Temperature and pressure data preprocessing: Denoising: Use a sliding average filter to denoise the temperature and pressure data, and the sliding window size is set to 5; Among them, T filtered and P filtered They are the temperature and pressure data after denoising respectively; Standardization: Standardize the temperature and pressure data to make their mean 0 and standard deviation 1; Among them, μ T and σ T are the mean and standard deviation of the temperature data, μ P and σ P are the mean and standard deviation of the pressure data, T normalized and P normalized They are the temperature and pressure data after standardization; Image data preprocessing: Grayscale processing: Convert color images into grayscale images to reduce the amount of data and simplify subsequent processing; I gray (x,y)=0.299·I R (x,y)+0.587·I G (x,y)+0.114·I B (x,y) Among them, I R is the red channel value at the pixel point (x, y) in the image; I G is the red channel value at the pixel point (x, y) in the image; I B is the red channel value at the pixel point (x, y) in the image; Image denoising: Use Gaussian filtering to denoise the image; the Gaussian filter parameters are set to kernel size 3×3 and standard deviation σ=1.0; Among them, I denoised is the value of the image pixel (x, y) after denoising; Image normalization: normalize image pixel values ​​to the range of [0,1]; Among them, I normalized is the value of the image pixel (x, y) after normalization; Step 3: Use the improved convolutional neural network (CNN) to build a welding defect detection model; use the deep residual network (ResNet) and attention mechanism (Attention Mechanism) to enhance the performance of the model; the model implements the input of real-time image data during the welding process and outputs whether there is a welding defect and the defect type; the specific steps are as follows: 3.

1. Data augmentation: Use techniques such as rotation, scaling, translation, and flipping to expand the dataset: Rotation: randomly rotate the image by a certain angle; Scale: Randomly scale the image; Translation: Randomly translate the image; Flip: randomly flip the image horizontally or vertically; 3.

2. Data trimming or padding: Crop or pad the image to a uniform size of 224×224; 3.

3. Convolutional and Pooling Layers Convolutional layer: The first layer uses 64 7×7 convolution kernels with a stride of 2 and a padding of 3; Where X is the input image, W is the convolution kernel, b is the bias, h is the output feature map, the subscripts i and j are the height and width of the feature map in spatial coordinates, k is the channel index of the output feature map, c is the channel index of the input feature map, and M and N are the height and width of the convolution kernel, respectively. Calculated output size: 112×112×64; Pooling layer: The maximum pooling layer uses a 3×3 pooling window with a stride of 2 and a padding of 1; p i,j,k =max m,n h i+m-1,j+n-1,k Where p is the pooling output and max is the maximum pooling operation; Calculated output size: 56×56×64; 3.

4. Residual Block Each residual block consists of two convolutional layers and their residual connections: y=F(x,{W i })+x Among them, F(x,{W i }) represents the learned residual function, x is the input feature map, y is the residual output, W i is the convolution kernel in the residual block, i is the number of convolution layers; The specific formula of the residual block is: The first convolution layer: 64×3×3, stride 1, padding 1; F1(x)=ReLU(W1·x+b1) Among them, F1(x) is the output after the first layer of convolution; W1 is the first layer of convolution kernel; b1 is the first layer of convolution bias; The second convolution layer: 64×3×3, stride 1, padding 1; F2(x)=W2·F1(x)+b2 Among them, F2(x) is the output after the second layer of convolution; W2 is the second layer of convolution kernel; b2 is the second layer of convolution bias; Residual output: y = ReLU(F2(x)+x) The output size remains the same: 56×56×64; 3.

5. Attention Mechanism Calculate the output of self-attention: Where Q is the query matrix, which represents the set of query vectors; K is the key matrix, which represents the set of key vectors; V is the value matrix, which represents the set of value vectors; Q = K = V = X (input feature map); d k is the dimension of the key vector; Attention(Q,K,V) is the output of the attention mechanism; Calculation process: Q=W Q ·X K=W K ·X V=W V ·X Among them, W Q , W K , W V They are the weight matrices for query, key, and value respectively; Calculate the attention score: Among them, Attention_Scores is the attention score matrix; Apply the Softmax function: Attention_Weights=softmax(Attention_Scores) Among them, Attention_Weights is the attention weight after softmax processing; Softmax is the Softmax activation function; Compute the attention output: Attention_Output=Attention_Weight·V Among them, Attention_Output is the output matrix of the attention mechanism; 3.

6. Fully Connected Layer Flatten the feature map into a one-dimensional vector: Flatten_Output=flatten(Attention_Output) Among them, Flatten_Output is the flattened output vector; Output size: 7×7×512=25088; After the first fully connected layer, the ReLU activation function is used: FC1=ReLU(W fc1 ·Flatten_Output+b fc1 ) Among them, FC1 is the output of the first fully connected layer; W fc1 is the weight matrix of the first fully connected layer; b fc1 is the bias vector of the first fully connected layer; The second fully connected layer uses the Softmax activation function for multi-classification: Output=Softmax(W fc2 ·FC1+b fc2 ) Among them, Output is the probability distribution of the final output; W fc2 is the weight matrix of the second fully connected layer; b fc2 is the bias vector of the second fully connected layer; Output size: 7, including 7 categories: normal, crack, pore, undercut, weld accumulation, deviation, and concave; Model Training Using the cross entropy loss function: Among them, L is the loss function value; N is the number of samples; C is the number of categories; y ij is the actual label of the jth category of sample i; is the predicted probability of the jth class for sample i; Use the Adam optimizer for gradient descent optimization to adjust the model weights to minimize the loss function; Step 4: During the welding process, deep reinforcement learning (DRL) technology is used to achieve adaptive control. The deep Q network (DQN) algorithm is used. DQN combines Q learning and deep neural networks and is suitable for processing high-dimensional input data from multiple sensors and images during the welding process. 4.

1. Environment and state definition Environment definition: Oil cylinder manufacturing device, including oil cylinder body, welding base, oil cylinder fixing seat, limit rotation groove, and booster assembly; State vector s t : s t =[I t ,V t ,v t ,T t ,P t ,D t ] Among them, I t is the current; V t is the voltage; v t is the welding speed; T t is temperature; P t For pressure; D t Welding defects; 4.

2. Action Definition Adjust welding parameters: current I, voltage V, welding speed v; Action vector a t : a t =[ΔI t ,ΔV t ,Δv t ] 4.

3. Reward Function Design Define a reward function based on welding defect rate, temperature stability, and pressure stability; Reward function R t Expressed as: R t =-(α·R d +β·R T +γ·R P ) in: R t is the reward function; R d is the defect rate, which is the number of welding defects detected; R T is the absolute difference of temperature from the expected value; R P is the absolute difference of pressure from the expected value; α, β, γ are weight coefficients; 4.

4. Q-value function and deep neural network The Q-value function Q(s,a) represents the expected cumulative reward for performing action a in state s. A deep neural network is used to approximate the Q-value function: Q(s,a;θ), where θ is the parameter of the neural network. Deep Neural Network (DNN) Structure: Input layer: state vector s t ; Hidden layer: multiple layers of neurons; Output layer: Q value Q(s,a;θ); 4.

5. Experience Replay Mechanism and Target Network Store experience data(s t ,a t ,R t ,s t+1 ) to the experience replay pool, and then sample small batches of data from the experience replay pool for training to break data correlation; Set a target network Q(s,a;θ - ); Among them, θ - The parameters of the target network are updated to the current network parameters every fixed number of steps; 4.

6. Loss Function and Gradient Update Define the loss function L(θ): L(θ)=E(s,a,R,s′)[(R+γmax a′ Q(s′,a′;θ - )-Q(s,a;θ)) 2 ] Among them, γ is the discount factor; Minimize the loss function by gradient descent and update the neural network parameters: Among them, α is the learning rate; Repeat the above steps and continuously optimize the DQN model until the welding parameters reach the optimal value.

Images