Welding process parameter dynamic detection method and system based on deep learning and Internet of Things

Abstract

The invention provides a welding process parameter dynamic detection method and system based on deep learning and the Internet of Things, and particularly relates to the technical field of welding process on-line monitoring. The method comprises the steps that multi-mode temperature signal data of a welding heat affected zone are obtained and processed through a temperature interference filtering model to generate environment interference resisting temperature data; groove image data are obtained, and an anti-interference groove image is generated through processing of a visual interference filtering model; groove geometrical characteristic parameters are extracted by using a weld joint geometrical characteristic recognition model, and the optimal temperature measurement point position is dynamically determined according to the parameters; and on the basis of a pre-stored material physical property parameter table, heat conduction delay compensation is executed on the anti-interference temperature data corresponding to the optimal temperature measuring point, a compensation temperature value is generated and output to serve as an interlayer temperature monitoring result, and real-time continuous high-precision monitoring of the welding interlayer temperature under the strong interference working condition is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of online monitoring of welding processes, and in particular to a method and system for dynamic detection of welding process parameters based on deep learning and the Internet of Things. Background Art

[0002] As high-parameter thermal power generation units evolve toward 630°C secondary reheat and 650°C ultra-supercritical technology, welded joints must withstand 35MPa high pressure and alternating thermal stress. The accuracy of dynamic interpass temperature control directly impacts weld quality (discreteness must be ≤5%). However, welding sites are subject to multiple interference sources, including arc glare, slag spatter, and smoke. Traditional infrared temperature measurement equipment can have measurement errors as high as ±25°C under dynamic conditions, and manual spot checks cover less than 30% of the equipment. This increases the risk of microcracks caused by interpass temperature excursions (e.g., a ±20°C deviation) by 47%.

[0003] Existing technologies mainly address this issue in two ways: Contact thermocouple temperature measurement: The welding process needs to be interrupted, which destroys the process continuity, and single-point measurement cannot cover the entire heat-affected zone; Single-mode infrared monitoring: susceptible to interference from ambient light, with a false alarm rate of up to 35% when obscured by smoke, and unable to adapt to weld geometry deformation (e.g., changes in groove angle leading to a shift in the optimal temperature measurement point).

[0004] Especially in the narrow groove welding of high-parameter units (groove angle ≤ 15°), the existing technology cannot meet the real-time control requirements of interlayer temperature of ±7°C due to its weak anti-interference and low positioning accuracy (error > ±2mm). In summary, how to achieve real-time, continuous and accurate monitoring of the temperature between welding layers in a strong interference environment is a technical problem that needs to be solved urgently to ensure the welding quality of high-parameter units. Summary of the Invention

[0005] The main purpose of the present invention is to provide a method and system for dynamic detection of welding process parameters based on deep learning and the Internet of Things, so as to at least solve the technical problem of how to achieve real-time, continuous and accurate monitoring of the temperature between welding layers under strong interference environments, thereby realizing real-time, continuous and high-precision monitoring of the temperature between welding layers under strong interference conditions.

[0006] In order to achieve the above objectives, the present invention provides a method and system for dynamic detection of welding process parameters based on deep learning and the Internet of Things.

[0007] In a first aspect, the present invention provides a method for dynamic detection of welding process parameters based on deep learning and the Internet of Things, the method comprising: Acquiring multimodal temperature signal data of a welding heat-affected zone, processing the multimodal temperature signal data through a temperature interference filtering model, and generating temperature data resistant to environmental interference; Acquiring groove image data in the welding heat affected zone, processing the groove image data through a visual interference filtering model, and generating a groove image that is resistant to environmental interference; Extracting groove geometric feature parameters through a weld geometric feature recognition model based on the groove image that is resistant to environmental interference; Dynamically determine the optimal temperature measurement point position of the welding heat affected zone according to the groove geometric characteristic parameters; Based on a pre-stored material property parameter table, performing heat conduction delay compensation on the temperature data corresponding to the optimal temperature measurement point position and the anti-environmental interference temperature value to generate a compensated temperature value; The compensated temperature value is output as the monitoring result of the interlayer temperature.

[0008] Specifically, the acquiring of multimodal temperature signal data of the welding heat affected zone, processing the multimodal temperature signal data through a temperature interference filtering model, and generating temperature data resistant to environmental interference includes: separating the infrared temperature measurement signal and the thermocouple contact signal from the multimodal temperature signal data; When the sudden change value of the infrared temperature measurement signal exceeds a preset threshold, switching to the thermocouple contact signal as the dominant signal source; Narrowband spectrum filtering is applied to the dominant signal source to generate the temperature data that is resistant to environmental interference.

[0009] Specifically, the acquiring of the groove image data in the welding heat affected zone, processing the groove image data by a visual interference filtering model, and generating a groove image resistant to environmental interference includes: Identifying the welding spatter area and the smoke obstruction area in the groove image data by using a smoke segmentation model; Repairing pixel data of the welding spatter area and the smoke occlusion area based on an image restoration algorithm; Contrast enhancement is performed on the repaired groove image data to generate the groove image resistant to environmental interference.

[0010] Specifically, the extracting groove geometric feature parameters based on the environmental interference-resistant groove image through a weld geometric feature recognition model includes: Locating the key points of the groove edge in the groove image resistant to environmental interference through a lightweight convolutional neural network; Calculating the groove angle and groove width according to the key points of the groove edge; The output includes the groove geometric characteristic parameters of the groove angle and groove width.

[0011] Specifically, dynamically determining the optimal temperature measurement point position of the welding heat affected zone according to the groove geometric characteristic parameters includes: Based on the groove angle in the groove geometric characteristic parameters, the lateral offset of the heat-affected zone is solved by a position calculation formula; According to the lateral offset of the heat-affected zone and the real-time coordinates of the welding gun, the servo pan-tilt table is driven to adjust the temperature measuring probe to the target position after the offset; The target position is continuously output as the optimal temperature measurement point position.

[0012] Specifically, the driving servo pan-tilt stage to adjust the temperature measuring probe to the offset target position includes: After the temperature measuring probe reaches the target position, obtaining a welding heat radiation image of the target position; Calculating the difference between the coordinates of the heat radiation center point of the welding heat radiation image and the spatial coordinates of the target position; When the absolute value of the spatial coordinate difference exceeds 0.5 mm, the servo pan / tilt table is driven to move the temperature measuring probe in the opposite direction of the spatial coordinate difference until the coordinates of the heat radiation center point coincide with the target position.

[0013] Specifically, the performing of heat conduction delay compensation on the temperature data against environmental interference corresponding to the optimal temperature measurement point position based on the pre-stored material property parameter table to generate a compensated temperature value includes: Indexing the current temperature value in the temperature data resistant to environmental interference according to the optimal temperature measurement point position; Obtaining the density, specific heat capacity and thermal conductivity of the current welding material based on the pre-stored material physical property parameter table; The time delay compensation is performed on the current temperature value through a one-dimensional heat conduction inverse operation to generate the compensated temperature value.

[0014] In a second aspect, the present invention provides a dynamic detection system for welding process parameters based on deep learning and the Internet of Things. The detection system applies the detection method described in the first aspect, and the detection system includes: An environmental interference resistant temperature generation module is used to obtain multimodal temperature signal data of the welding heat affected zone, process the multimodal temperature signal data through a temperature interference filtering model, and generate environmental interference resistant temperature data; an environmental interference resistant image generation module, configured to obtain groove image data within the welding heat affected zone, process the groove image data through a visual interference filtering model, and generate an environmental interference resistant groove image; A geometric feature extraction module is connected to the anti-environmental interference image generation module, and is used to extract groove geometric feature parameters based on the anti-environmental interference groove image through a weld geometric feature recognition model; A dynamic positioning module is connected to the geometric feature extraction module, and is used to dynamically determine the optimal temperature measurement point position of the welding heat affected zone according to the groove geometric feature parameters; a heat conduction compensation module, connected to the anti-environmental interference temperature generation module and the dynamic positioning module, respectively, and configured to perform heat conduction delay compensation on the anti-environmental interference temperature data corresponding to the optimal temperature measurement point position based on a pre-stored material property parameter table, and generate a compensated temperature value; The monitoring result output module is connected to the heat conduction compensation module, and is used to output the compensated temperature value as the monitoring result of the interlayer temperature.

[0015] Specifically, the anti-environmental interference temperature generation module includes: A signal separation submodule is used to separate the infrared temperature measurement signal and the thermocouple contact signal in the multimodal temperature signal data; A signal switching submodule is connected to the signal separation submodule, and is used to switch the dominant signal source to the thermocouple contact signal when the sudden change value of the infrared temperature measurement signal exceeds a preset threshold; The narrowband filtering submodule is connected to the signal switching submodule, and is used for applying narrowband spectrum filtering to the dominant signal source to generate the temperature data that is resistant to environmental interference.

[0016] Specifically, the anti-environmental interference image generation module includes: The smoke segmentation submodule is used to identify the welding spatter area and smoke obstruction area in the groove image data through the smoke segmentation model; A pixel repair submodule, connected to the smoke segmentation submodule, for repairing pixel data of the welding spatter area and the smoke occlusion area based on an image repair algorithm; The contrast enhancement submodule is connected to the pixel restoration submodule, and is used to perform contrast enhancement on the restored groove image data to generate the groove image resistant to environmental interference.

[0017] The dynamic detection method and system for welding process parameters based on deep learning and the Internet of Things provided in this application are intended to achieve real-time, continuous, and high-precision monitoring of the temperature between welding layers under strong interference conditions. The method first obtains the multimodal temperature signal data of the welding heat-affected zone, processes it using a temperature interference filtering model, and generates temperature data that is resistant to environmental interference; at the same time, the groove image data in the area is obtained, and processed by a visual interference filtering model to obtain an interference-resistant groove image. Based on the interference-resistant groove image, the groove geometric feature parameters are extracted through the weld geometric feature recognition model, and then the optimal temperature measurement point position is dynamically determined. Then, according to the pre-stored material property parameter table, heat conduction delay compensation is performed on the interference-resistant temperature data corresponding to the optimal temperature measurement point position to generate a compensated temperature value. Finally, the compensated temperature value is output as the monitoring result of the interlayer temperature, which effectively solves the technical problem of real-time, continuous, and accurate monitoring of the temperature between welding layers under a strong interference environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The accompanying drawings, which constitute part of this application, are intended to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are intended to explain the present invention and do not constitute an undue limitation of the present invention. In the accompanying drawings: Figure 1 A flow chart of the dynamic detection method of welding process parameters based on deep learning and the Internet of Things provided in this application; Figure 2 This is a connection diagram of the dynamic detection system for welding process parameters based on deep learning and the Internet of Things provided in this application.

[0019] The above drawings illustrate specific embodiments of the present application, which will be described in more detail below. These drawings and the textual description are not intended to limit the scope of the present application in any way, but rather to illustrate the concepts of the present application to those skilled in the art by reference to specific embodiments. DETAILED DESCRIPTION

[0020] To make the purpose, technical solutions, and advantages of this application more clear, the technical solutions in this application will be clearly and completely described below in conjunction with the drawings in this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0021] The terms "first," "second," "third," "fourth," and so forth (if any) in the description and claims of the present invention and in the accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a particular order or sequential sequence. It should be understood that the terms used in this manner are interchangeable where appropriate, such that the embodiments of the present invention described herein can be practiced in sequences other than those illustrated or described herein.

[0022] In the present invention, words such as "exemplary" or "for example" are used to indicate examples, illustrations, or descriptions. Any embodiment or design described in this application as "exemplary" or "for example" should not be construed as being preferred or advantageous over other embodiments or designs. Rather, the use of words such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner.

[0023] This application provides a method and system for dynamic detection of welding process parameters based on deep learning and the Internet of Things. This method first acquires multimodal temperature signal data from the weld heat-affected zone (HAZ) and processes it using a temperature interference filtering model to generate interference-resistant temperature data. Simultaneously, groove image data is acquired and processed using a visual interference filtering model to generate an interference-resistant groove image. The groove geometric characteristic parameters are then extracted based on the interference-resistant image, and the optimal temperature measurement point location is dynamically determined. Finally, the corresponding temperature data is compensated based on a pre-stored material property parameter table, and the compensated temperature value is output as the interlayer temperature monitoring result.

[0024] The following specific embodiments describe in detail the technical solution of the present application and how the technical solution of the present application solves the above-mentioned technical problems. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of the present application will be described below in conjunction with the accompanying drawings.

[0025] Figure 1 The flowchart of the dynamic detection method of welding process parameters based on deep learning and the Internet of Things provided in this application is as follows: Figure 1 As shown in FIG, a dynamic detection method for welding process parameters based on deep learning and the Internet of Things provided in this embodiment includes: S101: Acquire multimodal temperature signal data of a welding heat-affected zone, process the multimodal temperature signal data through a temperature interference filtering model, and generate temperature data that is resistant to environmental interference.

[0026] Specifically, the acquiring of multimodal temperature signal data of the welding heat affected zone, processing the multimodal temperature signal data through a temperature interference filtering model, and generating temperature data resistant to environmental interference includes: separating the infrared temperature measurement signal and the thermocouple contact signal from the multimodal temperature signal data; When the sudden change value of the infrared temperature measurement signal exceeds a preset threshold, switching to the thermocouple contact signal as the dominant signal source; Narrowband spectrum filtering is applied to the dominant signal source to generate the temperature data that is resistant to environmental interference.

[0027] In specific implementation, step S101 includes: 1. Multimodal temperature signal data acquisition 1.1 Use an infrared thermometer and a K-type thermocouple contact sensor to synchronously collect temperature signal data of the welding heat-affected zone, where: The sampling frequency of the infrared thermometer was set to 100 Hz, and the wavelength range was 8–14 μm; The response time of K-type thermocouple is ≤100ms and the temperature measurement range is 0-1300℃.

[0028] 1.2 The original signals collected by the two sensors are combined into multimodal temperature signal data, which is recorded as , where t is the timestamp.

[0029] 2. Signal separation operation Separation by FIR filter bank of digital signal processor : Infrared temperature measurement signal: ; Thermocouple contact signal: .

[0030] 3. Mutation value detection and signal switching 3.1 Calculate the first-order difference of infrared temperature measurement signal: .

[0031] 3.2 When >50℃ / ms (preset threshold): Switch the dominant signal source to the thermocouple contact signal .

[0032] 3.3 Otherwise, keep the infrared temperature measurement signal is the dominant signal source.

[0033] 4. Narrowband spectrum filtering For the dominant signal source Apply a 4th-order Butterworth bandpass filter:

[0034] : Center frequency (set according to the characteristics of the welding arc interference spectrum); :bandwidth.

[0035] 5. Generate temperature data that is resistant to environmental interference Output As temperature data resistant to environmental interference; Data storage format: time series array , time resolution 1ms.

[0036] This step physically separates the infrared and thermocouple signal sources and combines a mutation detection mechanism based on first-order differences to automatically switch to a contact temperature measurement signal with stronger interference resistance under strong arc interference. Furthermore, a Butterworth bandpass filter with a center frequency of 5Hz and a bandwidth of 2Hz is used to effectively suppress environmental electromagnetic noise (especially 50Hz power frequency interference), ultimately outputting clean temperature data with a time resolution of 1ms. Laboratory verification has shown that this solution improves the temperature signal-to-noise ratio by 23dB and maintains a measurement accuracy of ±2°C even when the welding arc is on, laying a data foundation for subsequent precise interlayer temperature monitoring.

[0037] S102: Acquire groove image data in the welding heat affected zone, process the groove image data through a visual interference filtering model, and generate a groove image that is resistant to environmental interference.

[0038] Specifically, the acquiring of the groove image data in the welding heat affected zone, processing the groove image data by a visual interference filtering model, and generating a groove image resistant to environmental interference includes: Identifying the welding spatter area and the smoke obstruction area in the groove image data by using a smoke segmentation model; Repairing pixel data of the welding spatter area and the smoke occlusion area based on an image restoration algorithm; Contrast enhancement is performed on the repaired groove image data to generate the groove image resistant to environmental interference.

[0039] In specific implementation, step S102 includes: 1. Groove image data acquisition An industrial-grade CMOS camera (resolution 1920×1080, frame rate 30fps) was used to photograph the groove area within the weld heat-affected zone; Equipped with a ring-shaped LED fill light (color temperature 5000K, illumination 20000lux) to overcome the interference of welding arc; Output original groove image data , where (x,y) are pixel coordinates.

[0040] 2. Welding spatter area identification Smoke segmentation model processing using U-Net architecture : Encoder: 4 convolutional layers (convolution kernel 3×3, stride 1) + max pooling (pooling window 2×2).

[0041] Decoder: 4 transposed convolutional layers (kernel 2×2, stride 2).

[0042] Output channels: 3 types of segmentation masks (background / splash area / smoke area).

[0043] Split formula: , : feature map of layer i; : Category weight (splash area w=0.6, smoke area w=0.4).

[0044] 3. Pixel data repair Apply the PatchMatch-based image restoration algorithm to the splash area and smoke occlusion area: 3.1 Create the area to be repaired Ω and the known area Φ; 3.2 For each pixel p in Ω, search for the most similar patch in Φ:

[0045] : SSD similarity calculation; : A 5×5 pixel block centered at p.

[0046] 3.3 Use Data fills pixel p 4. Contrast enhancement processing 4.1 Repaired Image Perform CLAHE (Contrast Limited Adaptive Histogram Equalization): Divide the image into 8×8 sub-blocks; Each sub-block histogram clipping threshold: 3.0; Interpolation method: bilinear interpolation.

[0047] 4.2 Output enhanced groove image: .

[0048] Optionally, the smoke segmentation model specifically includes: (1) Encoder structure: Layer 1: Conv3x3 (input channels 3, output channels 64, stride 2) + ReLU; Layer 2: Conv3x3 (64→128, stride 2) + MaxPool2x2; Layer 3: Conv3x3 (128→256, stride 1); Layer 4: Conv3x3 (256→512, stride 2).

[0049] (2) Decoder structure: Layer 1: Deconv2x2 (512→256, stride 2); Layer 2: Deconv2x2 (256→128, stride 2); Layer 3: Deconv2x2 (128→64, stride 1); Layer 4: Conv1x1(64→3, Sigmoid); (3) Training parameters: Loss function: Dice Loss + Cross Entropy; Optimizer: Adam (lr=0.001, β1=0.9, β2=0.999); Input data: 512×512 image normalized to [0,1]; This step uses a U-Net-based smoke segmentation model to accurately identify weld spatter areas (average IoU 0.85) and smoke-occluded areas (average IoU 0.78). Combined with a PatchMatch-based adaptive inpainting algorithm, it effectively restores obscured groove structural details. Finally, CLAHE enhancement technology is used to increase groove edge contrast to over 200%. Field testing has shown that the processed groove images maintain clear and recognizable geometric features even in welding smoke concentrations of up to 15mg / m³, providing reliable image input with an error of less than 0.5° for subsequent groove angle measurement.

[0050] S103: Based on the environmental interference-resistant groove image, groove geometric feature parameters are extracted using a weld geometric feature recognition model.

[0051] Specifically, the extracting groove geometric feature parameters based on the environmental interference-resistant groove image through a weld geometric feature recognition model includes: Locating the key points of the groove edge in the groove image resistant to environmental interference through a lightweight convolutional neural network; Calculating the groove angle and groove width according to the key points of the groove edge; The output includes the groove geometric characteristic parameters of the groove angle and groove width.

[0052] In specific implementation, step S103 includes: 1. Image preprocessing Receive the anti-environmental interference groove image (512×512 pixel RGB format) output by S102. Use bilinear interpolation to scale the image to 128×128 pixels. Normalize each pixel: , : Input pixel value (0-255); 0.5: training set pixel mean (statistical value of 2500 samples); 0.2: Standard deviation of the training set.

[0053] 2. Convolutional Neural Network Key Point Localization Use MobileNetV2 architecture to process normalized images: The input layer receives a 128×128×3 tensor; The 3×3 convolutional layer (stride 2) outputs a 64×64×32 feature map; 3 inverted residual blocks (expansion factor 6) are gradually downsampled to 32×32×32; The output layer generates a 32×32×17 feature map (17 channels corresponding to key points); Forward propagation uses the ReLU activation function: output = max(0, convolution result); 3. Key point coordinate extraction Apply the Sigmoid function to the output feature map to generate a heat map: , K: convolutional neural network output value; Take the maximum value position in each channel heat map ( ) as the key point coordinates, multiply by 4 and map back to the original image size: .

[0054] 4. Calculation of groove geometry parameters 4.1 Width calculation: Take the first key point of the left and right edges and ; , 0.12: Camera calibration factor (mm / pixel).

[0055] 4.2 Angle calculation: Take the left edge reference point and ; . Outputs the groove angle value in degrees.

[0056] 5. Feature parameter output Encapsulate data in JSON format: {"width_mm": W_mm, "angle_deg": θ}.

[0057] Transfer to step S104.

[0058] Model training instructions: The weld geometry recognition model was trained on 2,000 manually annotated weld groove images using Labelme. Training used the Adam optimizer (with parameters β1 = 0.9 and β2 = 0.999), and the loss function used the mean squared error of keypoint coordinates. During deployment, the model was run on an embedded processor using INT8 quantization, achieving a single-frame processing time of 45 milliseconds. This step uses a lightweight convolutional network to precisely locate key groove points, calculate groove width and angle based on coordinates, and provide a precise geometric reference for dynamic temperature measurement point positioning. The measured key point positioning error is 0.3 pixels, the width measurement error is ≤0.2 mm, and the angular resolution is 0.1 degrees. Even in high-interference welding environments, the processing speed remains at 30 frames per second, meeting the millimeter-level control requirements of high-parameter welding processes.

[0059] S104: Dynamically determine the optimal temperature measurement point position of the welding heat affected zone according to the groove geometric characteristic parameters.

[0060] Specifically, dynamically determining the optimal temperature measurement point position of the welding heat affected zone according to the groove geometric characteristic parameters includes: Based on the groove angle in the groove geometric characteristic parameters, the lateral offset of the heat-affected zone is solved by a position calculation formula; According to the lateral offset of the heat-affected zone combined with the real-time coordinates of the welding gun, the servo pan-tilt stage is driven to adjust the temperature measuring probe to the target position after the offset; specifically, the driving of the servo pan-tilt stage to adjust the temperature measuring probe to the target position after the offset specifically includes: after the temperature measuring probe arrives at the target position, obtaining the welding thermal radiation image of the target position; calculating the difference between the thermal radiation center point coordinates of the welding thermal radiation image and the spatial coordinates of the target position; when the absolute value of the spatial coordinate difference exceeds 0.5 mm, driving the servo pan-tilt stage to move the temperature measuring probe in the opposite direction of the spatial coordinate difference until the thermal radiation center point coordinates coincide with the target position.

[0061] The target position is continuously output as the optimal temperature measurement point position.

[0062] In specific implementation, step S104 includes: 1. Calculation of lateral offset Receive the groove geometry parameters (JSON format) output by S103 and parse the groove angle value θ (unit: degree). Calculate the lateral offset of the heat affected zone based on the heat conduction characteristics of the weld heat affected zone: , : lateral offset (mm); : Material deviation coefficient (2.5 for carbon steel and 3.2 for alloy steel); : Groove angle value (from S103).

[0063] 2. Target position calculation and gimbal drive Get the real-time coordinates of the welding gun ( ) (sampled 50 times per second by the welding gun encoder). Calculate the target position coordinates: ; Drive the servo pan / tilt table (repeat positioning accuracy ±0.01mm) to move the temperature probe to the target position ( ).

[0064] 3. Thermal radiation image acquisition and center point positioning After the temperature probe reaches the target position, an infrared thermal imager (resolution 640×480) is used to capture the welding thermal radiation image. The centroid method is used to calculate the coordinates of the thermal radiation center point: , ( ): pixel coordinates; : pixel temperature value (0-65535 grayscale); N: The number of pixels with a temperature greater than 500°C.

[0065] 4. Position deviation calculation and calibration Calculate the spatial coordinate difference between the thermal radiation center point and the target position: when >0.5mm, drive the servo gimbal along the vector ( ) Move the probe in the direction Repeat the image acquisition and deviation calculation until Δd ≤ 0.5 mm.

[0066] 5.0 Best temperature measurement point output When the coordinates of the heat radiation center point and the target position meet When ∣≤0.5mm, lock the current probe position ( ) as the optimal temperature measurement point. The position coordinates are continuously output to step S105 at a frequency of 10 Hz. This step dynamically calculates the lateral offset of the heat-affected zone based on the groove angle and, combined with the welding gun's real-time coordinates, achieves submillimeter positioning of the temperature probe. The probe position is calibrated using the thermal radiation image centroid method to eliminate target point drift errors caused by thermal deformation. Under welding thermal deformation conditions, the final temperature measurement point positioning accuracy reached ±0.3mm (better than the industry standard of ±1mm), with a response delay of less than 20ms, meeting the requirements for continuous interlayer temperature monitoring in high-parameter units.

[0067] S105: Based on a pre-stored material property parameter table, heat conduction delay compensation is performed on the temperature data corresponding to the optimal temperature measurement point position to generate a compensated temperature value, and the compensated temperature value is output as a monitoring result of the interlayer temperature.

[0068] Specifically, the performing of heat conduction delay compensation on the temperature data against environmental interference corresponding to the optimal temperature measurement point position based on the pre-stored material property parameter table to generate a compensated temperature value includes: Indexing the current temperature value in the temperature data resistant to environmental interference according to the optimal temperature measurement point position; Obtaining the density, specific heat capacity and thermal conductivity of the current welding material based on the pre-stored material physical property parameter table; The time delay compensation is performed on the current temperature value through a one-dimensional heat conduction inverse operation to generate the compensated temperature value.

[0069] In specific implementation, step S105 includes: 1. Current temperature value index Receive the optimal temperature measurement point position coordinates output by S104 ( ). Anti-environmental interference temperature data generated from S101 (time series array ), extract the current temperature value corresponding to the location : , : Current timestamp (accuracy 1ms); : Anti-interference temperature data output by S101.

[0070] 2. Obtaining material properties According to the material identification in the welding process parameters (such as "P91"), query the pre-stored material property parameter table (as shown in Table 1): Table 1: 3. One-dimensional heat conduction inverse calculation compensation The true temperature is calculated by inverting the one-dimensional heat conduction equation: L: characteristic thickness of the heat-affected zone (set to 5mm = 0.005m); : Temperature change rate (calculated by central difference method): , Δt: sampling interval (0.1s).

[0071] 4. Compensated temperature value generation Output the compensated temperature value as the final monitoring result: .

[0072] Data format: { "time": t_now, "temp": T_comp, "unit": "℃"}. This step eliminates temperature measurement delays by performing an inverse calculation of heat conduction, dynamically compensating the measured value based on material properties and the temperature change rate. For typical P91 steel welding conditions (heating rate of 100°C / s), the temperature error after compensation is reduced from ±15°C (compared to the traditional method) to ±2.3°C, and the response delay is shortened from 0.8 seconds to 0.05 seconds. This enables real-time and accurate monitoring of interpass temperatures in high-parameter units up to 630°C, providing a reliable basis for closed-loop control of the welding process.

[0073] This embodiment provides a method for dynamic detection of welding process parameters based on deep learning and the Internet of Things, which can realize real-time, continuous, and high-precision monitoring of the temperature between welding layers under strong interference conditions. The method first obtains the multimodal temperature signal data of the welding heat-affected zone, and uses the temperature interference filtering model to process it to generate temperature data that is resistant to environmental interference, thereby eliminating the influence of external interference on temperature measurement. At the same time, the groove image data in the area is obtained and processed by the visual interference filtering model to obtain an interference-resistant groove image. Based on the interference-resistant groove image, the groove geometric feature parameters are extracted using the weld geometric feature recognition model, and the optimal temperature measurement point position of the welding heat-affected zone is dynamically determined based on these parameters. Subsequently, based on the pre-stored material physical property parameter table, heat conduction delay compensation is performed on the interference-resistant temperature data corresponding to the optimal temperature measurement point position to generate a compensated temperature value. Finally, the compensated temperature value is output as the monitoring result of the interlayer temperature, which effectively solves the technical problem of real-time, continuous, and accurate monitoring of the welding interlayer temperature under strong interference environment, thereby improving welding quality and stability.

[0074] Figure 2 The connection diagram of the welding process parameter dynamic detection system based on deep learning and the Internet of Things provided in this application is as follows: Figure 2 As shown in the figure, the welding process parameter dynamic detection system based on deep learning and Internet of Things provided by this embodiment is applied Figure 1 The method for dynamic detection of welding process parameters based on deep learning and the Internet of Things described in the embodiment includes: An environmental interference resistant temperature generation module is used to obtain multimodal temperature signal data of the welding heat affected zone, process the multimodal temperature signal data through a temperature interference filtering model, and generate environmental interference resistant temperature data; an environmental interference resistant image generation module, configured to obtain groove image data within the welding heat affected zone, process the groove image data through a visual interference filtering model, and generate an environmental interference resistant groove image; A geometric feature extraction module is connected to the anti-environmental interference image generation module, and is used to extract groove geometric feature parameters based on the anti-environmental interference groove image through a weld geometric feature recognition model; A dynamic positioning module is connected to the geometric feature extraction module, and is used to dynamically determine the optimal temperature measurement point position of the welding heat affected zone according to the groove geometric feature parameters; a heat conduction compensation module, connected to the anti-environmental interference temperature generation module and the dynamic positioning module, respectively, and configured to perform heat conduction delay compensation on the anti-environmental interference temperature data corresponding to the optimal temperature measurement point position based on a pre-stored material property parameter table, and generate a compensated temperature value; The monitoring result output module is connected to the heat conduction compensation module, and is used to output the compensated temperature value as the monitoring result of the interlayer temperature.

[0075] Specifically, the anti-environmental interference temperature generation module includes: A signal separation submodule is used to separate the infrared temperature measurement signal and the thermocouple contact signal in the multimodal temperature signal data; A signal switching submodule is connected to the signal separation submodule, and is used to switch the dominant signal source to the thermocouple contact signal when the sudden change value of the infrared temperature measurement signal exceeds a preset threshold; The narrowband filtering submodule is connected to the signal switching submodule, and is used for applying narrowband spectrum filtering to the dominant signal source to generate the temperature data that is resistant to environmental interference.

[0076] Specifically, the anti-environmental interference image generation module includes: The smoke segmentation submodule is used to identify the welding spatter area and smoke obstruction area in the groove image data through the smoke segmentation model; A pixel repair submodule, connected to the smoke segmentation submodule, for repairing pixel data of the welding spatter area and the smoke occlusion area based on an image repair algorithm; The contrast enhancement submodule is connected to the pixel restoration submodule, and is used to perform contrast enhancement on the restored groove image data to generate the groove image resistant to environmental interference.

[0077] During implementation, the welding process parameter dynamic detection system based on deep learning and the Internet of Things provided by this embodiment specifically includes: 1. Anti-environmental interference temperature generation module This module includes a signal separation submodule, a signal switching submodule, and a narrowband filtering submodule, and adopts the following physical connections and working methods: 1.1 Signal Separation Submodule: Receives multi-modal temperature signal data input via shielded twisted pair cables and uses a FIR digital filter bank to separate the infrared temperature measurement signal and the thermocouple contact signal. The infrared signal passband is 0.1-10Hz, and the thermocouple signal passband is 0-50Hz.

[0078] 1.2 Signal Switching Submodule: This module is directly connected to the Signal Separation Submodule via a 16-bit parallel data bus. It monitors the rate of change of the infrared temperature measurement signal in real time. When the rate of change exceeds the 50°C / ms threshold, it switches to the thermocouple signal as the dominant signal source.

[0079] 1.3 Narrowband Filter Submodule: Connected to the Signal Switching Submodule via an analog signal line, it uses a fourth-order Butterworth bandpass filter to process the dominant signal source, with a center frequency of 5Hz and a bandwidth of 2Hz, and outputs temperature data that is resistant to environmental interference to the RS485 interface.

[0080] This module eliminates arc light interference and improves the temperature signal-to-noise ratio by 23dB, providing a reliable data basis for subsequent compensation.

[0081] 2. Image generation module that is resistant to environmental interference This module includes smoke segmentation submodule, pixel restoration submodule, and contrast enhancement submodule, which are connected as follows: 2.1 Smoke segmentation submodule: It receives 1920×1080 groove image data through the Camera Link interface, adopts the U-Net convolutional neural network architecture, and outputs binary masks of the welding spatter area and the smoke occlusion area through a four-layer encoder-decoder structure.

[0082] 2.2 Pixel Inpainting Submodule: This module is connected to the Smoke Segmentation Submodule via a PCIe x4 bus. Based on the PatchMatch algorithm, it searches for a 5×5 pixel block in the known image area that is most similar to the area to be inpainted.

[0083] 2.3 Contrast Enhancement Submodule: This module is connected to the Pixel Restoration Submodule via the HDMI video interface. It uses the CLAHE method to divide the image into 8×8 subblocks, constrains the histogram distribution, and then performs bilinear interpolation to output the enhanced image.

[0084] This module eliminates more than 90% of the smoke obstruction effects and increases the groove edge contrast to 200%, ensuring the accuracy of geometric feature recognition.

[0085] 3. Geometric feature extraction module The groove image output by the anti-environmental interference image generation module is received through the Gigabit Ethernet interface.

[0086] It uses the MobileNetV2 lightweight convolutional neural network architecture: it inputs a 128×128 pixel image, processes it through three inverted residual blocks with an expansion factor of 6, and outputs a 17-channel keypoint heat map.

[0087] The key points of the groove edge are analyzed through Sigmoid activation and coordinate mapping, the groove width and angle values ​​are calculated, and the geometric parameters are output in JSON format.

[0088] This module achieves a key point positioning accuracy of 0.3 pixels and a width measurement error of ≤0.2mm, meeting the welding control requirements of high-parameter units.

[0089] 4. Dynamic positioning module The groove angle parameters output by the geometric feature extraction module are received via the CAN bus.

[0090] The lateral offset of the heat-affected zone is calculated based on the angle value. The material coefficient for carbon steel is 2.5, and the coefficient for alloy steel is 3.2.

[0091] Combined with the real-time coordinates of the welding gun encoder, the linear motor servo pan-tilt system is driven to move the temperature measuring probe.

[0092] Thermal radiation images are collected by an infrared thermal imager, and the coordinates of the thermal radiation center point are calculated using the centroid method. When the deviation from the target position exceeds 0.5mm, a closed-loop calibration is triggered.

[0093] This module eliminates positioning drift caused by thermal deformation, with a final positioning accuracy of ±0.3mm and a response delay of <20ms.

[0094] 5. Thermal conduction compensation module The temperature data is obtained by connecting to the anti-environmental interference temperature generation module through the Modbus interface.

[0095] The optimal temperature measurement point position is obtained by connecting to the dynamic positioning module via the RS422 interface.

[0096] Query the ASME material database to obtain the three parameters of density, specific heat capacity and thermal conductivity.

[0097] Perform one-dimensional heat conduction inverse calculation: Based on the characteristic thickness of 5mm and the temperature change rate calculated by the central difference method, compensate for the measurement delay error.

[0098] This module reduces the temperature error from ±15°C to ±2.3°C after compensation, meeting the monitoring requirements of 630°C high-parameter units.

[0099] 6. Monitoring result output module Connect to the thermal conduction compensation module via TCP / IP protocol.

[0100] Encapsulate the compensation temperature value into a JSON data packet: {"time":timestamp, "temp":temperature value, "unit":"℃"}.

[0101] Upload to the factory MES system via the OPC UA protocol and store locally in the SQLite database.

[0102] This module realizes the real-time monitoring data transmission of interlayer temperature in seconds and supports closed-loop control of welding process.

[0103] Those skilled in the art will readily appreciate other embodiments of the present invention after considering the specification and practicing the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of the present invention that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein. The description and examples are to be considered as exemplary only, and the true scope and spirit of the present application are indicated by the following claims.

[0104] It should be understood that the present application is not limited to the exact structure described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present application is limited only by the appended claims.

Claims

1. A dynamic detection method for welding process parameters based on deep learning and the Internet of Things, characterized in that: The method comprises: Acquiring multimodal temperature signal data of a welding heat-affected zone, processing the multimodal temperature signal data through a temperature interference filtering model, and generating temperature data resistant to environmental interference; Acquiring groove image data in the welding heat affected zone, processing the groove image data through a visual interference filtering model, and generating a groove image that is resistant to environmental interference; Extracting groove geometric feature parameters through a weld geometric feature recognition model based on the groove image that is resistant to environmental interference; Dynamically determine the optimal temperature measurement point position of the welding heat affected zone according to the groove geometric characteristic parameters; Based on a pre-stored material property parameter table, performing heat conduction delay compensation on the temperature data corresponding to the optimal temperature measurement point position and the anti-environmental interference temperature value to generate a compensated temperature value; The compensated temperature value is output as the monitoring result of the interlayer temperature.

2. The method for dynamic detection of welding process parameters based on deep learning and the Internet of Things according to claim 1 is characterized in that: The method of obtaining multimodal temperature signal data of the welding heat affected zone and processing the multimodal temperature signal data through a temperature interference filtering model to generate temperature data resistant to environmental interference includes: separating the infrared temperature measurement signal and the thermocouple contact signal from the multimodal temperature signal data; When the sudden change value of the infrared temperature measurement signal exceeds a preset threshold, switching to the thermocouple contact signal as the dominant signal source; Narrowband spectrum filtering is applied to the dominant signal source to generate the temperature data that is resistant to environmental interference.

3. The method for dynamic detection of welding process parameters based on deep learning and the Internet of Things according to claim 1 is characterized in that: The step of acquiring groove image data in the welding heat affected zone and processing the groove image data through a visual interference filtering model to generate a groove image resistant to environmental interference includes: Identifying the welding spatter area and the smoke obstruction area in the groove image data by using a smoke segmentation model; Repairing pixel data of the welding spatter area and the smoke occlusion area based on an image restoration algorithm; Contrast enhancement is performed on the repaired groove image data to generate the groove image resistant to environmental interference.

4. The method for dynamic detection of welding process parameters based on deep learning and the Internet of Things according to claim 1, characterized in that: The method of extracting groove geometric feature parameters based on the groove image resisting environmental interference by using a weld geometric feature recognition model includes: Locating the key points of the groove edge in the groove image resistant to environmental interference through a lightweight convolutional neural network; Calculating the groove angle and groove width according to the key points of the groove edge; The output includes the groove geometric characteristic parameters of the groove angle and groove width.

5. The method for dynamic detection of welding process parameters based on deep learning and the Internet of Things according to claim 4 is characterized in that: The dynamically determining the optimal temperature measurement point position of the welding heat affected zone according to the groove geometric characteristic parameters includes: Based on the groove angle in the groove geometric characteristic parameters, the lateral offset of the heat-affected zone is solved by a position calculation formula; According to the lateral offset of the heat-affected zone and the real-time coordinates of the welding gun, the servo pan-tilt table is driven to adjust the temperature measuring probe to the target position after the offset; The target position is continuously output as the optimal temperature measurement point position.

6. The method for dynamic detection of welding process parameters based on deep learning and the Internet of Things according to claim 5 is characterized in that: The driving servo pan-tilt stage adjusts the temperature measuring probe to the offset target position, comprising: After the temperature measuring probe reaches the target position, obtaining a welding heat radiation image of the target position; Calculating the difference between the coordinates of the heat radiation center point of the welding heat radiation image and the spatial coordinates of the target position; When the absolute value of the spatial coordinate difference exceeds 0.5 mm, the servo pan / tilt table is driven to move the temperature measuring probe in the opposite direction of the spatial coordinate difference until the coordinates of the heat radiation center point coincide with the target position.

7. The method for dynamic detection of welding process parameters based on deep learning and the Internet of Things according to claim 1, characterized in that: The method of performing heat conduction delay compensation on the temperature data against environmental interference corresponding to the optimal temperature measurement point position based on the pre-stored material property parameter table to generate a compensated temperature value includes: Indexing the current temperature value in the temperature data resistant to environmental interference according to the optimal temperature measurement point position; Obtaining the density, specific heat capacity and thermal conductivity of the current welding material based on the pre-stored material physical property parameter table; The time delay compensation is performed on the current temperature value through a one-dimensional heat conduction inverse operation to generate the compensated temperature value.

8. A dynamic detection system for welding process parameters based on deep learning and the Internet of Things, characterized in that: The detection system applies the detection method according to any one of claims 1 to 7, and the detection system comprises: An environmental interference resistant temperature generation module is used to obtain multimodal temperature signal data of the welding heat affected zone, process the multimodal temperature signal data through a temperature interference filtering model, and generate environmental interference resistant temperature data; an environmental interference resistant image generation module, configured to obtain groove image data within the welding heat affected zone, process the groove image data through a visual interference filtering model, and generate an environmental interference resistant groove image; A geometric feature extraction module is connected to the anti-environmental interference image generation module, and is used to extract groove geometric feature parameters based on the anti-environmental interference groove image through a weld geometric feature recognition model; A dynamic positioning module is connected to the geometric feature extraction module, and is used to dynamically determine the optimal temperature measurement point position of the welding heat affected zone according to the groove geometric feature parameters; a heat conduction compensation module, connected to the anti-environmental interference temperature generation module and the dynamic positioning module, respectively, and configured to perform heat conduction delay compensation on the anti-environmental interference temperature data corresponding to the optimal temperature measurement point position based on a pre-stored material property parameter table, and generate a compensated temperature value; The monitoring result output module is connected to the heat conduction compensation module, and is used to output the compensated temperature value as the monitoring result of the interlayer temperature.

9. The detection system according to claim 8, characterized in that: The anti-environmental interference temperature generation module includes: A signal separation submodule is used to separate the infrared temperature measurement signal and the thermocouple contact signal in the multimodal temperature signal data; A signal switching submodule is connected to the signal separation submodule, and is used to switch the dominant signal source to the thermocouple contact signal when the sudden change value of the infrared temperature measurement signal exceeds a preset threshold; The narrowband filtering submodule is connected to the signal switching submodule, and is used for applying narrowband spectrum filtering to the dominant signal source to generate the temperature data that is resistant to environmental interference.

10. The detection system according to claim 8, characterized in that: The anti-environmental interference image generation module includes: The smoke segmentation submodule is used to identify the welding spatter area and smoke obstruction area in the groove image data through the smoke segmentation model; A pixel repair submodule, connected to the smoke segmentation submodule, for repairing pixel data of the welding spatter area and the smoke occlusion area based on an image repair algorithm; The contrast enhancement submodule is connected to the pixel restoration submodule, and is used to perform contrast enhancement on the restored groove image data to generate the groove image resistant to environmental interference.

Images