A K-TIG multi-position welding penetration control system and method

Abstract

The present invention relates to a K-TIG multi-position welding penetration control system and method. The control system comprises a welding system, a magnetic field device, and a control system. The magnetic field device comprises an iron core and a coil, the iron core being sleeved on a welding torch, the coil being wound around the iron core, and a DC power supply being externally connected at both ends. The DC power supply regulates the output current to change the magnetic field strength generated by the magnetic field device at the welding torch tip. Consequently, during multi-position tilted welding, the magnetic field interacts with the current flowing through the molten pool to generate a Lorentz force opposite to gravity, suppressing the tendency of the molten pool to flow downward during tilted welding, facilitating the formation of a stable keyhole and penetration state, and improving weld formation. The control system employs a TF-CNN neural network to identify penetration of arc sound signals during welding. This signal is then mapped to corresponding current changes according to fuzzy control rules, and the welding power supply current is adjusted to control the penetration state by adjusting the heat input during welding in real time.

Description

Technical Field

[0001] The present invention relates to the technical field of K-TIG welding, and in particular to a K-TIG multi-position welding penetration control system and method. Background Art

[0002] Keyhole effect tungsten inert gas welding (K-TIG welding) is an innovation of traditional TIG welding. K-TIG welding is a highly efficient deep penetration welding technology. When sufficient heat is input during the welding process to melt the welded parts, a keyhole is formed in the weld. The formation of the keyhole can form a narrow channel between the surface and the interior of the welded parts, which can directly transfer energy to the weld joint and improve the efficiency of energy transfer. As long as the workpiece is in a state of full penetration, the keyhole will remain stable throughout the welding process. The state of full penetration during K-TIG welding has a huge impact on the weld quality. By ensuring that the keyhole is always in a stable state of full penetration during the welding process, better weld quality can be obtained.

[0003] Existing K-TIG welding systems usually include a welding power supply, a K-TIG welding gun, an argon gas cylinder, a cooling water tank and a robot, among which the welding power supply is used to set the welding current of the welding gun, and the robot is used to set the welding speed of the K-TIG welding gun. For flat welding, that is, when the welding angle is 0°, the existing K-TIG welding system can form a stable keyhole and achieve a stable penetration state during the welding process by changing welding parameters such as welding current and welding speed. However, when the welding angle changes, for example, when welding large circular storage tanks or welding vertical workpieces, due to the influence of gravity, during the K-TIG welding process, the molten pool flows downward in the direction of gravity, making it difficult to form a stable keyhole and penetration state, resulting in affected weld formation. The weld after welding will have biting, which will also affect the mechanical properties of the weld joint. Summary of the Invention

[0004] In response to the problems existing in the prior art, the purpose of the present invention is to provide a K-TIG multi-position welding penetration control system and method, which can suppress the molten pool from dripping during multi-angle welding, facilitate the formation of a stable keyhole and penetration state, and improve the weld formation effect.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions:

[0006] A K-TIG multi-position welding penetration control system includes a welding system;

[0007] The welding system includes a magnetic field device, a K-TIG welding gun coaxially mounted with the magnetic field device, and a welding power source connected to the K-TIG welding gun;

[0008] The magnetic field device includes an iron core and a coil. The iron core is sleeved on the K-TIG welding gun, and the coil is wound on the iron core. A DC power supply is connected to both ends of the coil. The output current is adjusted by the DC power supply to change the magnetic field strength of the magnetic field generated by the magnetic field device at the tip of the K-TIG welding gun. The magnetic field then interacts with the current flowing through the molten pool to generate a Lorentz force opposite to the direction of gravity, thereby suppressing the tendency of the molten pool to flow down during K-TIG tilt welding.

[0009] Furthermore, it also includes a sound signal acquisition system and a control system. The sound signal acquisition system is used to collect arc sound signals in the welding process in real time and perform analog-to-digital conversion to transmit them to the control system. The control system is used to preprocess and identify the arc sound signals, determine the real-time penetration state of the current welding process, and then adjust the current of the welding power supply according to the real-time penetration state to affect the heat input on the surface of the weldment to control the weld to always be in a stable normal penetration state, thereby ensuring good weld formation and weld performance.

[0010] Furthermore, the sound signal acquisition system includes a microphone, a power supply and a data acquisition card. The microphone is installed on a conical surface at a certain angle to the plane of the welding workpiece. The power supply is used to power the microphone and amplify the collected sound signal and transmit it to the data acquisition card. The data acquisition card is connected to the control system.

[0011] Furthermore, the control system includes an industrial computer and a PLC. The industrial computer is connected to the data acquisition card and the PLC respectively, and is used to run the trained TF-CNN neural network to identify the penetration state in the welding process in real time and adjust the current of the welding system through the fuzzy control strategy and PLC control based on the identification results. Among them, the TF-CNN network model is composed of an input layer, a first self-attention mechanism layer, a first pooling layer, a second self-attention mechanism layer, a second pooling layer, an attention mechanism module, a fully connected layer and an output layer connected in sequence.

[0012] A K-TIG multi-position welding penetration control method comprises the following steps:

[0013] The iron core of the magnetic field device is placed on the K-TIG welding gun, and the coil is wound on the iron core. The two ends of the coil are connected to an external DC power supply. Before welding, the output current of the DC power supply is adjusted according to the different inclination angles of the welding workpiece, and the magnetic field strength generated by the magnetic field device at the tip of the K-TIG welding gun is changed. Then, the magnetic field interacts with the current flowing through the molten pool to generate a Lorentz force in the direction opposite to gravity, thereby suppressing the tendency of the molten pool to drip during K-TIG tilt welding.

[0014] Further, the welding process includes the following steps,

[0015] S1, collects data, adjusts welding parameters in the welding system, collects arc sound signals of K-TIG welding at different penetration states in multiple positions, and forms an arc sound signal database;

[0016] S2, preprocessing, preprocessing the collected arc sound signals during the K-TIG multi-position welding process;

[0017] S3, training model, using the pre-processed arc sound signal database as input, training the improved TF-CNN neural network model, and establishing a TF-CNN neural network suitable for penetration recognition in K-TIG multi-position welding;

[0018] S4, real-time control process, collects arc sound signals in the K-TIG multi-position welding process in real time, and transmits the arc sound signals to the control system, repeats the preprocessing process of S2, and then uses the TF-CNN neural network model suitable for K-TIG multi-position welding penetration identification established in S3 to identify the penetration state in the welding process in real time and perform closed-loop control of the welding current through fuzzy control method.

[0019] Furthermore, the adjusted welding parameters include welding current, magnetic field strength at the welding gun tip, welding speed and welding table angle, among which the adjustment range of welding current is 420A-600A, the adjustment range of welding speed is 3mm / s-4mm / s, and the adjustment range of welding table angle is 0-90°. Before welding, according to the different inclination angles of the welding workpiece, the output current of the DC power supply is adjusted to change the adjustment range of the magnetic field strength generated by the magnetic field device at the tip of the K-TIG welding gun to 0-50mT. Finally, the arc sound signals in three penetration states of normal penetration, over-penetration and under-penetration at different welding angles between flat welding and horizontal welding are obtained to form an arc sound signal database.

[0020] Furthermore, the arc sound signal is preprocessed, including first removing the zero drift, intercepting the sound signal in the 10s time period before arcing, calculating the zero drift DC component by averaging, and removing the drift from the overall signal; then using the improved spectral subtraction method to reduce the noise of the collected arc sound signal, defining the sound signal before arcing as environmental noise, and the sound signal collected after arcing as the coupling of the environmental noise before arcing and the pure arc sound. The improved spectral subtraction method is to subtract the noise power spectrum from the noisy signal power spectrum to obtain a relatively pure sound signal; then using the Hanning window to frame the noise-reduced sound signal, with 50ms as a frame, and there is a certain overlap between adjacent frames, and finally obtaining multiple segments of arc sound signals in the multi-position K-TIG welding process with a length of 50ms.

[0021] Furthermore, the arc sound signal segments in the preprocessed multi-position K-TIG welding process were divided into three penetration states: incomplete penetration, normal penetration, and over-penetration, and labeled. They were then distributed as training and validation sets in a ratio of 8:2. The improved TF-CNN network model was trained to establish a TF-CNN network model suitable for penetration recognition of K-TIG welding arc sound signals.

[0022] Furthermore, the fuzzy control method includes taking the penetration state quantity of the current welding position as the input quantity, wherein the over-penetration state is -1, the normal penetration state is 0, and the non-penetration state is 1, and setting the penetration state identified at the current moment as s(t), the penetration states identified at the previous three moments are s(t-3), s(t-2), and s(t-1), and the value range is [-1, 1], which are divided into 3 fuzzy sets, namely NS (negative small), ZO (zero), and PS (positive small), calculating the state sum of the three moments before the current moment Σs(t)=s(t-3)+s(t-2)+s(t-1), and the value range is [-3, 3], which is divided into 7 fuzzy sets, namely NB (negative large), NM (negative medium), NS (negative small), ZO (zero), PS (positive small), PM (positive medium), and PB (positive large), and setting the value range of the output quantity I(t) to [-14,14], divided into 7 fuzzy sets, namely NB (negative large), NM (negative medium), NS (negative small), ZO (zero), PS (positive small), PM (positive medium), PB (positive large), s(t) and Σs(t) are fuzzified, and Gaussian membership function is used as the membership function. Fuzzy reasoning is performed based on the fuzzified penetration state membership vectors Σs(t) and s(t) as input to obtain the current membership vector I(t). The maximum membership method is used to defuzzify I(t) and obtain the welding power supply current control quantity ΔI(t) as the output. The welding power supply current control quantity ΔI(t) output by the fuzzy control rule is then sent to the PLC. The PLC is used to control the switch of the electromagnetic relay and trigger the current adjustment switch of the welding power supply multiple times. The current in the welding process is adjusted in real time to ensure that the weld quality meets the requirements.

[0023] In general, the present invention has the following advantages:

[0024] The present invention provides a K-TIG welding gun with a magnetic field device. During welding, the coil of the magnetic field device can generate an axial magnetic field at the tip of the K-TIG welding gun. Changing the output current of the DC power supply can change the magnetic field strength of the magnetic field generated by the magnetic field device at the tip of the K-TIG welding gun. The interaction between the magnetic field and the current flowing through the molten pool generates a Lorentz force opposite to the direction of gravity, thereby suppressing the tendency of the molten pool to flow downward during K-TIG welding at an inclined angle, which is beneficial for improving the weld formation effect and weld quality. At the same time, during the real-time welding process, an improved TF-CNN neural network is used to identify the penetration state. Then, a fuzzy control strategy is used to control the current of the welding power supply in the welding system to control the heat input applied to the weldment surface during the real-time welding process, thereby forming a stable keyhole, maintaining a stable penetration state during the welding process, and ensuring that the weld quality during the welding process meets the requirements. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 Schematic diagram of the structure of a K-TIG multi-position welding penetration control system according to an embodiment of the present invention.

[0026] Figure 2 This is a schematic diagram of the installation location structure of the sound sensing system.

[0027] Figure 3 This is the installation method and schematic diagram of the magnetic field device.

[0028] Figure 4 This is a flow chart of the K-TIG multi-position welding penetration control method of the present invention.

[0029] Figure 5 This is a time-frequency diagram of the arc sound signal after preprocessing.

[0030] Figure 6 This is a structural diagram of the TF-CNN neural network used in an embodiment of the present invention.

[0031] In the picture:

[0032] 1-Robot control cabinet, 11-KUKA robot;

[0033] 21-Industrial Computer, 22-PLC;

[0034] 31-Argon gas cylinder, 32-K-TIG welding gun, 33-Welding power supply, 34-Electromagnetic relay;

[0035] 41-microphone, 42-power supply, 43-data acquisition card;

[0036] 5-Magnetic field device, 51-DC power supply. DETAILED DESCRIPTION

[0037] The present invention will be described in further detail below.

[0038] like Figure 1 As shown, a K-TIG multi-position welding penetration control system includes: a welding system, a sound sensing system and a control system.

[0039] The welding system includes a HTIG-1000 welding power supply 33, a K-TIG welding gun 32, an argon gas cylinder 31, a CW-5200 cooling water tank, a magnetic field device 5, an adjustable welding table, and a KUKA robot 11. The welding power supply 33 is used to set the welding current during welding, the KUKA robot 11 is used to set the welding speed, the welding table is used to set the welding angle, and the magnetic field device 5 is used to set parameters such as the magnetic field strength at the tip of the tungsten needle. The magnetic field device 5 receives signals from the control system and adjusts these parameters in real time. The welding power supply 33 has an adjustable current range of 50A-1000A, and the welding table has an adjustable angle of 0-90°.

[0040] An iron core is placed inside the magnetic field device 5, and multiple turns of enameled wire are wrapped around the outside to form a coil. During installation, the coil is kept coaxial with the K-TIG welding gun 32. External DC power supplies 51 are connected to both ends of the coil to generate an axial magnetic field at the tip of the K-TIG welding gun 32. By changing the output current of the DC power supply 51, the magnetic field intensity generated by the magnetic field device 5 at the tip of the K-TIG welding gun 32 can be changed. The important function of this device is to generate a Lorentz force opposite to the direction of gravity through the interaction between the magnetic field and the current flowing through the molten pool, thereby suppressing the tendency of the molten pool to flow downward during K-TIG welding at an inclined angle, and improving the weld formation and weld quality of K-TIG welding at an inclined angle. The installation method of the magnetic field device 5 is shown in the schematic diagram. Figure 3 shown.

[0041] The sound sensing system includes a microphone 41 (model MPA201), a power supply 42 (model MC141), and a data acquisition card 43 (model NI-USB6221). These collect arc sound signals during welding and convert them into digital signals for transmission to the control system. Microphone 41 has a range of 20Hz-20kHz and a noise floor of less than 16dBA. Microphone 41 is mounted on a conical surface with an angle of θ = 75° to the plane of the workpiece, 250mm from the tip of the tungsten needle. Figure 2 The power supply 42 uses a BNC interface to power the microphone 41 and amplifies the collected sound signal with a gain of 10. The amplified signal is transmitted to the data acquisition card 43, and the sampling frequency is set to 42 kHz according to the Quest sampling theorem.

[0042] The control system, consisting of an industrial computer 21 and a programmable logic controller (PLC) 22, processes and identifies arc sound signals transmitted by the sound sensor system. Based on the identification results, it sends control signals to the welding system to adjust welding parameters and ensure welding quality. The industrial computer 21 runs a trained TF-CNN neural network to identify the penetration level during welding in real time. The identification results are then used to control the current of the welding power source 33 in the welding system using a fuzzy control strategy and the PLC 22.

[0043] This example uses butt welding of 12mm thick 304 stainless steel without a groove. The weld gap is 0-1mm, the welding angle is 0°-90°, and the current range is 420-600A. The magnetic field strength applied by magnetic field device 5 to the tip of the tungsten needle varies from 0-50mT depending on the welding angle. The weld width W on the back of the workpiece after welding is used as the criterion for determining penetration. W < 1mm indicates incomplete penetration, 1mm ≤ W < 2.5mm indicates normal penetration, and W ≥ 2.5mm indicates overpenetration. Multiple welding tests were conducted to obtain arc sound signals for three penetration states: overpenetration, incomplete penetration, and normal penetration.

[0044] like Figure 4 As shown, this example also provides a K-TIG multi-position welding penetration control method, including:

[0045] Step 1: Change the welding parameters and collect K-TIG welding arc sound signals at multiple positions with different penetration states.

[0046] Step 2: Pre-process the collected arc sound signal, which includes three steps: zero drift removal, noise reduction, and frame division. Specifically, it includes:

[0047] Step 21: First, remove the zero drift. Intercept the acoustic signal in the 10s period before arcing, calculate the DC component of the zero drift by averaging, and remove the drift from the overall signal.

[0048] Step 22: Then, the improved spectral subtraction method is used to reduce the noise of the collected arc sound signal. First, the sound signal before arcing is defined as environmental noise. The noise is mainly composed of the fan sound of the robot control cabinet 1 and the fan sound of the welding power supply 33 radiator. The sound signal collected after arcing can be regarded as the coupling of the environmental noise before arcing and the pure arc sound. The improved spectral subtraction method is used to reduce the noise of the arc sound signal. The method is to subtract the noise power spectrum from the noisy signal power spectrum to obtain a relatively pure sound signal.

[0049] Step 23: Use the Hanning window to divide the noise-reduced sound signal into frames, with 50ms as a frame and a certain overlap between adjacent frames. Finally, multiple segments of 50ms long arc sound signals in the K-TIG welding process are obtained, such as Figure 5shown.

[0050] Step 3: The preprocessed arc sound signal segments are divided into three penetration states: incomplete penetration, normal penetration, and excessive penetration, and labeled. They are then divided into training and validation sets in a ratio of 8:2. The improved TF-CNN network model is trained to establish a TF-CNN network model suitable for K-TIG welding arc sound signal penetration recognition.

[0051] The improved TF-CNN network model described above consists of the input layer, the first self-attention mechanism layer, the first pooling layer, the second self-attention mechanism layer, the second pooling layer, the attention mechanism module, the fully connected layer and the output layer. Figure 6 The figure shows a schematic diagram of the TF-CNN network model structure.

[0052] Traditional neural network models struggle to determine which input information is important and which is not. Therefore, when using traditional network models to classify penetration, manual feature extraction is required on the preprocessed arc sound signal to eliminate unimportant information and retain the important information. To simplify the cumbersome feature extraction process and reduce subjectivity in penetration identification, this method addresses this issue by employing a TF-CNN network model with an attention mechanism. The improved TF-CNN neural network replaces the convolutional layers in traditional neural networks with a self-attention layer. Experiments have shown that replacing convolutional layers with self-attention layers in the field of penetration identification using arc sound signals can achieve better feature extraction. During the TF-CNN training process, the introduction of the self-attention layer enables the network to calculate the similarity between the requested pixel and all input image pixels in the training set, and then performs softmax normalization. This ensures that the features in the arc sound signal that identify penetration are given greater weight during training, effectively completing the feature extraction process that would otherwise require prior work within the neural network. After training, the recognition accuracy of the established TF-CNN network model can reach over 90%.

[0053] Step 4: Conduct a K-TIG multi-position welding test, collect and process the arc sound signal during the K-TIG welding process in real time, and use the established TF-CNN network model to identify the penetration state in real time.

[0054] Step 5: Using the penetration state identified in real time by the TF-CNN neural network model as input, the current of the welding power source 33 is adjusted using PLC 22 according to the fuzzy control rules to ensure that the K-TIG welding process is in a normal penetration state. This includes the following steps:

[0055] Step 51: Take the penetration state at the current welding position as the input, where the overpenetration state is -1, the normal penetration state is 0, and the incomplete penetration state is 1. Let the identified penetration state at the current moment be s(t). The identified penetration states at the previous three moments are s(t-3), s(t-2), and s(t-1), with values ​​ranging from -1 to 1, and are divided into three fuzzy sets: NS (negative small), ZO (zero), and PS (positive small). Calculate the state sum of the three moments before the current moment, Σs(t) = s(t-3) + s(t-2) + s(t-1), with values ​​ranging from -3 to 3, and are divided into seven fuzzy sets: NB (negative large), NM (negative medium), NS (negative small), ZO (zero), PS (positive small), PM (positive medium), and PB (positive large). The output value I(t) is set to a range of [-14, 14] and divided into seven fuzzy sets: NB (negative large), NM (negative medium), NS (negative small), ZO (zero), PS (positive small), PM (positive medium), and PB (positive large). s(t) and Σs(t) are fuzzified, and the Gaussian membership function is used as the membership function. Fuzzy reasoning is performed using the fuzzified penetration state membership vectors Σs(t) and s(t) as inputs to obtain the current membership vector I(t). The maximum membership method is used to defuzzify I(t) and obtain the welding power supply current control variable ΔI(t) as the output. The established fuzzy rule table is shown in Table 1.

[0056] Table 1: ΔI(t) fuzzy control rules

[0057]

[0058] Step 52: The current control variable ΔI(t) output by the fuzzy control rule is sent to PLC 22. PLC 22 controls the on / off of electromagnetic relay 34, repeatedly triggering the current adjustment switch of welding power supply 33 to adjust the current in real time during the welding process. Welding power supply 33 supports current control steps of ±1A, ±10A, and ±100A. In this example, the control step size used is ±1A.

[0059] The above embodiments are preferred implementation modes of the present invention, but the implementation modes of the present invention are not limited to the above embodiments. Any other changes, modifications, substitutions, combinations, and simplifications that do not deviate from the spirit and principles of the present invention should be considered as equivalent replacement methods and are included in the scope of protection of the present invention.

Claims

1. A K-TIG multi-position welding penetration control system, characterized by: Including welding system; The welding system includes a magnetic field device, a K-TIG welding gun coaxially mounted with the magnetic field device, and a welding power source connected to the K-TIG welding gun; The magnetic field device includes an iron core and a coil. The iron core is mounted on the K-TIG welding gun, and the coil is wound around the iron core. A DC power supply is connected to both ends of the coil. The output current is adjusted by the DC power supply to change the magnetic field strength generated by the magnetic field device at the tip of the K-TIG welding gun. The interaction between the magnetic field and the current flowing through the molten pool generates a Lorentz force in the opposite direction of gravity, which suppresses the tendency of the molten pool to flow downward during K-TIG tilt welding. It also includes a sound signal acquisition system and a control system. The sound signal acquisition system is used to collect arc sound signals in real time during the welding process and perform analog-to-digital conversion to transmit them to the control system. The control system is used to pre-process and identify the arc sound signals, determine the real-time penetration state of the current welding process, and then adjust the current of the welding power supply according to the real-time penetration state to affect the heat input on the surface of the weldment to control the weld to always be in a stable and normal penetration state, thereby ensuring good weld formation and weld performance; The sound signal acquisition system includes a microphone, a power supply, and a data acquisition card. The microphone is installed on a cone at a certain angle to the plane of the welding workpiece. The power supply is used to power the microphone and amplify the collected sound signal before transmitting it to the data acquisition card. The data acquisition card is connected to the control system. The control system includes an industrial computer and a PLC. The industrial computer is connected to the data acquisition card and PLC respectively, and is used to run the trained TF-CNN neural network to identify the penetration state in the welding process in real time and adjust the current of the welding system through fuzzy control strategy and PLC control based on the identification results. Among them, the TF-CNN network model consists of an input layer, a first self-attention mechanism layer, a first pooling layer, a second self-attention mechanism layer, a second pooling layer, an attention mechanism module, a fully connected layer and an output layer connected in sequence.

2. The method of controlling the penetration control system of K-TIG multi-position welding according to claim 1, characterized in that: The following steps are included: The iron core of the magnetic field device is placed on the K-TIG welding gun, and the coil is wound on the iron core. The two ends of the coil are connected to an external DC power supply. Before welding, the output current of the DC power supply is adjusted according to the different inclination angles of the welding workpiece, and the magnetic field strength generated by the magnetic field device at the tip of the K-TIG welding gun is changed. Then, the magnetic field interacts with the current flowing through the molten pool to generate a Lorentz force in the direction opposite to gravity, thereby suppressing the tendency of the molten pool to drip during K-TIG tilt welding.

3. The method according to claim 2, wherein: The welding process includes the following steps, S1, collects data, adjusts welding parameters in the welding system, collects arc sound signals of K-TIG welding at different penetration states in multiple positions, and forms an arc sound signal database; S2, preprocessing, preprocessing the collected arc sound signals during the K-TIG multi-position welding process; S3, training model, using the pre-processed arc sound signal database as input, training the improved TF-CNN neural network model, and establishing a TF-CNN neural network suitable for penetration recognition in K-TIG multi-position welding; S4, real-time control process, collects arc sound signals in the K-TIG multi-position welding process in real time, and transmits the arc sound signals to the control system, repeats the preprocessing process of S2, and then uses the TF-CNN neural network model suitable for K-TIG multi-position welding penetration identification established in S3 to identify the penetration state in the welding process in real time and perform closed-loop control of the welding current through fuzzy control method.

4. The method according to claim 3, wherein: The adjusted welding parameters include welding current, magnetic field strength at the welding gun tip, welding speed and welding table angle. The adjustment range of welding current is 420A-600A, the adjustment range of welding speed is 3mm / s-4mm / s, and the adjustment range of welding table angle is 0-90°. Before welding, according to the different inclination angles of the welding workpiece, the output current of the DC power supply is adjusted to change the adjustment range of the magnetic field strength generated by the magnetic field device at the tip of the K-TIG welding gun to 0-50mT. Finally, the arc sound signals in three penetration states of normal penetration, over-penetration and under-penetration at different welding angles between flat welding and horizontal welding are obtained to form an arc sound signal database.

5. The method according to claim 3, wherein: The preprocessing of arc sound signals includes first removing zero drift, intercepting the sound signal in the 10s period before arcing, calculating the zero drift DC component by averaging, and removing the drift from the overall signal; The collected arc sound signal is then subjected to noise reduction processing using an improved spectral subtraction method. The sound signal before arcing is defined as ambient noise, and the sound signal collected after arcing can be regarded as the coupling of the ambient noise before arcing and the pure arc sound. The improved spectral subtraction method subtracts the noise power spectrum from the noisy signal power spectrum to obtain a relatively pure sound signal. The noise-reduced sound signal is then framed using a Hanning window, with 50ms as a frame and a certain overlap between adjacent frames. Finally, multiple segments of 50ms in length are obtained during the multi-position K-TIG welding process.

6. The method according to claim 5, characterized in that: The arc sound signal segments in the preprocessed multi-position K-TIG welding process are divided into three penetration states: incomplete penetration, normal penetration, and over-penetration, and labeled. Then, they are divided into training and validation sets in a ratio of 8:

2. The improved TF-CNN network model is trained to establish a TF-CNN network model suitable for penetration recognition of K-TIG welding arc sound signals.

7. The method according to claim 6, characterized in that: The fuzzy control method includes taking the penetration state of the current welding position as the input, where the over-penetration state is -1, the normal penetration state is 0, and the incomplete penetration state is 1, and setting the penetration state identified at the current moment as s(t), the penetration states identified at the previous three moments are s(t-3), s(t-2), and s(t-1), and the value range is [-1, 1], which are divided into 3 fuzzy sets, namely NS (negative small), ZO (zero), and PS (positive small), calculating the state sum of the three moments before the current moment Σs(t)=s(t-3)+s(t-2)+s(t-1), and the value range is [-3, 3], which is divided into 7 fuzzy sets, namely NB (negative large), NM (negative medium), NS (negative small), ZO (zero), PS (positive small), PM (positive medium), and PB (positive large), and setting the value range of the output quantity I(t) to [- 14,14], divided into 7 fuzzy sets, namely NB (negative large), NM (negative medium), NS (negative small), ZO (zero), PS (positive small), PM (positive medium), PB (positive large), s(t) and Σs(t) were fuzzified, and Gaussian membership function was used as the membership function. Fuzzy reasoning was performed based on the fuzzified penetration state membership vectors Σs(t) and s(t) as inputs to obtain the current membership vector I(t). The maximum membership method was used to defuzzify I(t) and obtain the welding power supply current control quantity ΔI(t) as the output. The welding power supply current control quantity ΔI(t) output by the fuzzy control rule was then sent to the PLC. The PLC was used to control the switch of the electromagnetic relay and trigger the current adjustment switch of the welding power supply multiple times. The current in the welding process was adjusted in real time to ensure that the weld quality met the requirements.

Images