Intelligent pipeline welding quality management and control system and method

Abstract

The application discloses a pipeline welding quality intelligent management and control system and method, the method comprising: synchronously collecting molten pool images, molten pool radiation spectra and welding electric signals during the welding process, extracting and fusing image features, spectrum features and electric signal features, inputting a defect prediction model to obtain a defect probability, and when the defect probability exceeds a preset threshold, adaptively adjusting welding current, voltage and welding speed to realize real-time quality control in the welding process. After the welding is completed, welding process parameters, welding process signals and post-welding detection data are collected to construct a welding process digital twin, multi-source data is registered and analyzed, and a weld quality score and a defect diagnosis result are output. In a high-altitude or low-temperature environment, the environmental parameters are adjusted to adjust the thermal excitation and infrared thermal image acquisition parameters, and non-metal pipeline welding defect identification and quality evaluation are realized.

Description

Technical Field

[0001] This application relates to the field of pipeline welding quality control technology, and more specifically, to an intelligent pipeline welding quality control system and method. Background Technology

[0002] Pipeline welding is a critical process in oil and gas transportation, urban gas pipeline networks, water supply and drainage systems, and chemical pipeline engineering. Welding quality directly affects the structural safety and long-term operational reliability of the pipeline system. In actual engineering projects, the welding process is typically affected by various factors such as material properties, welding process parameters, environmental conditions, and operational stability, making it prone to defects such as porosity, lack of fusion, inclusions, and cracks. If welding defects are not detected or effectively controlled in a timely manner, they may lead to leaks, structural failures, or even safety accidents during service. Therefore, effective monitoring and evaluation of the welding process and weld quality are of great significance.

[0003] In existing technologies, pipeline welding quality control typically relies on welding process specifications, human experience, and post-weld non-destructive testing. For example, after welding, the weld quality is assessed using ultrasonic testing, radiographic testing, or visual inspection. While these methods can detect some welding defects, they are mostly post-weld inspection methods and cannot predict or intervene in defect formation during the welding process in real time. This results in problems such as detection lag, high rework costs, and difficulty in achieving process quality control. Furthermore, single inspection methods provide limited information dimensions and are insufficient for characterizing complex welding process states, making it difficult to comprehensively reflect key characteristics such as energy input, molten pool morphology changes, and arc stability during the welding process.

[0004] In recent years, with the development of sensing and data processing technologies, some studies have begun to explore online monitoring of the welding process using image acquisition, electrical signal monitoring, or spectral analysis. However, existing solutions often focus on single-type signal analysis, lacking a multi-source information fusion mechanism, making it difficult to accurately describe the multi-physics coupling characteristics of the welding process. Furthermore, existing technologies often operate independently in terms of welding process monitoring, post-weld quality assessment, and unified data management, lacking unified data correlation and comprehensive analysis capabilities, thus hindering the formation of a complete welding quality control system.

[0005] Furthermore, during the thermofusion welding of non-metallic pipelines in special environmental conditions such as high altitude or low temperature, factors such as ambient temperature, air pressure, and wind speed can significantly affect the welding thermal process and inspection results. Traditional inspection methods are difficult to adapt effectively to environmental changes, thus reducing inspection accuracy. Therefore, how to construct a pipeline welding quality control technology that can integrate multi-source welding process information, achieve welding defect prediction and parameter control, and combine post-weld inspection and environmental adaptation inspection has become an urgent technical problem to be solved in related fields. Summary of the Invention

[0006] This application provides an intelligent control system and method for pipeline welding quality, which at least solves some of the technical problems existing in the related technologies described above.

[0007] According to a first aspect of the embodiments of this application, a method for intelligent control of pipeline welding quality is provided, comprising: During the welding process, images of the molten pool, radiation spectra of the molten pool, and welding electrical signals are acquired simultaneously. Image features, spectral features, and electrical signal features are extracted respectively. The image features, spectral features, and electrical signal features are then fused and input into a defect prediction model to obtain the defect probability. When the defect probability exceeds a preset threshold, the welding parameter adjustment amount is calculated and the welding parameter adjustment is performed based on the sensitivity of the defect probability to the welding parameters and the deviation between the current defect probability and the preset threshold. After welding is completed, welding process parameters, welding process signals and post-weld inspection data are collected. A digital twin of the welding process is constructed based on the welding process parameters, the welding process signals and the post-weld inspection data. Spatial, temporal and attribute registration is performed on the welding process signals and the post-weld inspection data. The registered fusion features are extracted and the weld quality quantitative score and defect diagnosis results are output. When the welding scenario is non-metallic pipeline hot-melt welding in high-altitude and / or low-temperature environments, environmental parameters are collected, and thermal excitation parameters and infrared thermography acquisition parameters are adjusted according to the environmental parameters. The thermography sequence is preprocessed and features are extracted to identify welding defects and output weld quality grade and inspection report. Welding process data, post-weld evaluation results and environmental detection results are recorded to a unified data platform.

[0008] As an optional solution, the simultaneous acquisition of molten pool images, molten pool radiation spectra, and welding electrical signals includes: The camera, spectral acquisition device, and electrical signal sensor are synchronously triggered by a unified hardware trigger signal; the image features include molten pool morphology features and depth features; the spectral features include characteristic spectral line intensity, spectral line intensity ratio, and plasma temperature; and the electrical signal features include time-domain statistical features and frequency-domain features.

[0009] As an optional approach, the fused features are input into the defect prediction model to obtain the defect probability, including: The image features, spectral features, and electrical signal features are concatenated to form a fused feature vector, which is then input into the multilayer perceptron prediction head. The calculation of the welding parameter adjustment includes: calculating the parameter increment according to the sensitivity of the defect probability to current, voltage, and welding speed, combined with the deviation between the defect probability and the preset threshold, using an incremental proportional-integral-derivative control method.

[0010] As an optional solution, the acquisition of welding process parameters, welding process signals, and post-weld inspection data includes: The process involves collecting welding current, welding voltage, welding speed, and wire feed speed; acquiring images of the molten pool, arc acoustic signals, spectral signals, and temperature field data; and collecting ultrasonic testing data, radiographic testing data, and three-dimensional morphology data of the weld surface. The construction of a digital twin of the welding process includes fusing a finite element thermo-mechanical coupling model with a data-driven model to obtain a digital twin corresponding to the actual welding process.

[0011] As an optional approach, the spatial, temporal, and attribute registration of the welding process signals and post-weld inspection data includes: A digital twin coordinate system is established using the three-dimensional point cloud of the weld surface. Ultrasonic and radiographic inspection data are mapped to the digital twin coordinate system through rigid body transformation. A time correspondence is established using welding timestamps. An attribute association relationship is established between welding process parameters, welding process signals, and defect detection results. The spatial registration results are corrected using an iterative nearest point algorithm.

[0012] As an optional approach, the extraction of registered fusion features and output of weld quality quantitative scoring and defect diagnosis results includes: The molten pool morphology features, spectral features, electrical signal features, ultrasonic detection features, radiographic detection features, and surface morphology features are extracted from the registered multi-source data to form a fused feature vector. The fused feature vector is input into the comprehensive quality assessment model to output a quantitative score of weld quality and defect diagnosis results corresponding to defect type and severity.

[0013] As an optional solution, the acquisition of environmental parameters, and the adjustment of thermal excitation parameters and infrared thermal imaging acquisition parameters based on the environmental parameters, include: Collect ambient temperature, ambient humidity, ambient wind speed, and ambient air pressure; adjust the power of the thermal excitation source based on the ambient temperature and ambient wind speed; select continuous excitation mode, pulse excitation mode, or phase-locked excitation mode based on the depth of the defect to be inspected; and adjust the gain and integration time of the infrared thermal imager based on the ambient temperature.

[0014] As an optional approach, the preprocessing and feature extraction of the thermal image sequence includes: The original thermal image sequence is sequentially filtered for noise reduction, thermal image non-uniformity correction, and contrast enhancement. Peak temperature, heating rate, and cooling rate are extracted in pulse excitation mode, and phase difference and amplitude attenuation are extracted in phase-locked excitation mode. The coupling relationship between emissivity, temperature, and environment is established based on ambient temperature, ambient humidity, and pipe material emissivity to correct the thermal image temperature data.

[0015] As an optional solution, the identification of welding defects and the output of weld quality grade and inspection report include: The preprocessed thermal image feature map is input into the defect segmentation model to obtain the segmentation results corresponding to interface cold welds, overheating, inclusions or porosity; in phase-locked loop excitation mode, the defect depth is calculated based on the phase difference between the defect area and the normal area, the modulation frequency and the material thermal diffusivity; and the weld quality grade is determined according to the defect type, size and depth, generating an inspection report containing environmental parameters, inspection parameters, defect information and quality grade.

[0016] According to a second aspect of the embodiments of this application, a smart control system for pipeline welding quality is also provided, comprising: The multi-source data acquisition module is used to simultaneously acquire images of the molten pool, the radiation spectrum of the molten pool, and welding electrical signals during the welding process; The feature extraction and fusion module is used to process the molten pool image, the molten pool radiation spectrum, and the welding electrical signal respectively, extract image features, spectral features, and electrical signal features, and fuse the image features, spectral features, and electrical signal features; The defect prediction module is used to input the fused features into the defect prediction model to obtain the probability of welding defects. The parameter adaptive adjustment module is used to calculate the welding parameter adjustment amount based on the sensitivity of the defect probability to the welding parameters and the deviation between the current defect probability and the preset threshold when the defect probability exceeds a preset threshold, and to control the welding equipment to perform the welding parameter adjustment. The post-weld data acquisition module is used to acquire welding process parameters, welding process signals, and post-weld inspection data after welding is completed. The digital twin construction and analysis module is used to construct a digital twin of the welding process based on the welding process parameters, the welding process signals and the post-weld inspection data, perform spatial, temporal and attribute registration on the welding process signals and the post-weld inspection data, extract the fusion features after registration, and output the weld quality quantitative score and defect diagnosis results. The environmental adaptability detection module is used to collect environmental parameters when non-metallic pipes are hot-melt welded in high-altitude and / or low-temperature environments. Based on the environmental parameters, the module adjusts the thermal excitation parameters and infrared thermal imaging acquisition parameters, preprocesses the thermal image sequence and extracts features to identify welding defects and output weld quality level and inspection report. The data management module is used to record welding process data, post-weld evaluation results, and environmental monitoring results to a unified data platform.

[0017] Compared with existing technologies, this application simultaneously acquires multi-source information such as molten pool images, radiation spectra, and welding electrical signals during the welding process, and performs fusion analysis of image features, spectral features, and electrical signal features to achieve online prediction of welding defect probability. When the defect probability exceeds a threshold, key process parameters such as welding current, voltage, and welding speed are adaptively adjusted, thereby timely suppressing defect formation during welding and improving the stability and weld quality. Simultaneously, after welding, a digital twin of the welding process is constructed by integrating welding process parameters, welding process signals, and post-weld inspection data. Spatial, temporal, and attribute registration of multi-source data enables comprehensive evaluation of weld quality and defect diagnosis, improving the accuracy and traceability of weld quality assessment. Furthermore, in special environmental conditions such as high altitude or low temperature, by acquiring environmental parameters and adaptively adjusting thermal excitation parameters and infrared thermography acquisition parameters, effective identification of defects in non-metallic pipeline hot-melt welding is achieved. This application thus addresses the needs of in-weld monitoring, post-weld evaluation, and special environmental detection, forming a quality control mechanism that runs through the entire welding process, improving the reliability and engineering applicability of pipeline welding quality management.

[0018] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Furthermore, no embodiment in this disclosure is required to achieve all the effects described above. Attached Figure Description

[0019] The accompanying drawings, which are incorporated in and form a part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure.

[0020] Figure 1 This is a schematic diagram of an intelligent control method for pipeline welding quality provided in an embodiment of this disclosure.

[0021] Figure 2 This is a schematic diagram of the defect probability prediction process provided in an embodiment of the present disclosure.

[0022] Figure 3 This is a schematic diagram of the post-weld comprehensive evaluation process provided in an embodiment of this disclosure.

[0023] Figure 4 This is a schematic diagram of the structure of an intelligent control system for pipeline welding quality provided in an embodiment of this disclosure. Detailed Implementation

[0024] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0025] This implementation method is applicable to online quality control of metal pipeline welding processes, as well as quality inspection scenarios after hot-melt welding of non-metallic pipelines in high-altitude and / or low-temperature environments. Specifically, the site is equipped with a welding power source, welding actuator, multi-source sensor acquisition components, post-weld non-destructive testing components, 3D morphology acquisition components, infrared thermal imaging acquisition components, and a unified data platform. For metal pipeline welding scenarios, the multi-source sensor acquisition components operate synchronously with the welding power source, forming a combined monitoring during welding and post-weld evaluation. For non-metallic pipeline hot-melt welding scenarios, the environmental parameter acquisition components and the infrared thermal imaging acquisition components operate collaboratively, forming special environment detection. The unified data platform establishes a unified weld identifier for the process data, inspection data, and rework information of the same weld, in order to form a continuous data link and a reproducible implementation process.

[0026] First, some nouns or terms that appear in the description of the embodiments of this application shall be interpreted as follows: In this paper, molten pool images refer to a sequence of two-dimensional images acquired around the molten pool region during the welding process; molten pool radiation spectra refer to the wavelength distribution data formed by the welding arc and radiation from the molten pool region; welding electrical signals refer to the time series of welding current and welding voltage; welding process signals refer to the image, acoustic, spectral, electrical, and temperature field data acquired during the welding stage; post-weld inspection data refer to the data obtained after the weld is completed through ultrasonic testing, radiographic testing, and three-dimensional surface morphology scanning; digital twins refer to virtual mapping bodies corresponding to the actual welding process constructed based on welding process parameters, welding process signals, and post-weld inspection data; and fusion features refer to features obtained from multiple sources after spatial, temporal, and attribute registration. Feature vectors are extracted and uniformly expressed from the data; defect probability refers to the probability output of the defect prediction model that the current welding state belongs to a defect state; defect diagnosis result refers to the combined output of defect type and severity; weld quality quantitative score refers to the continuous score given by the comprehensive quality assessment model for the weld state; weld quality grade refers to the grading result determined according to defect type, size and depth during the special environment inspection stage; preset threshold refers to the probability threshold that triggers the adjustment of welding parameters, the value of which can be configured to be obtained based on the statistics of historical welding samples or through process calibration, remain unchanged within the same production batch, and be recalibrated after changing the material grade, pipe diameter or welding method.

[0027] The implementation process of the method described in this application will be described in detail below with reference to specific embodiments. It should be noted that this embodiment is only used to explain this application and is not intended to limit the scope of protection of this application. Conventional adjustments or substitutions of each step by those skilled in the art without departing from the concept of this application should be included in the scope of protection of this application.

[0028] Figure 1 This is a flowchart of an intelligent control method for pipeline welding quality according to an embodiment of the present invention, such as... Figure 1 As shown, the method includes steps S1-S4: In step S1, during the welding process, the molten pool image, molten pool radiation spectrum, and welding electrical signal are acquired simultaneously. Image features, spectral features, and electrical signal features are extracted respectively. The image features, spectral features, and electrical signal features are then fused and input into the defect prediction model to obtain the defect probability.

[0029] Figure 2 A schematic diagram of the defect probability prediction process provided in an embodiment of this disclosure is shown. Figure 2 As shown, in step S201, a unified time reference is first established during the welding stage. Specifically, according to embodiments of this disclosure, the camera, spectral acquisition device, and electrical signal sensor all receive the same hardware trigger signal, which is output by the main controller. The frequency of this hardware trigger signal is consistent with the image acquisition cycle, ensuring that the molten pool image frame, spectral sampling points, and electrical signal sampling window correspond under the same time reference. It is understood that time synchronization here does not merely require the simultaneous activation of the three types of devices, but rather requires that each frame of the molten pool image corresponds to a spectral segment and an electrical signal analysis window. Therefore, the subsequently constructed image features, spectral features, and electrical signal features all have a strict time correspondence, preventing the problem of subsequent models calling data from undefined sources.

[0030] In practical applications, the camera is positioned behind or to the side of the welding torch to cover the leading edge of the molten pool, the main body of the molten pool, and part of the solidified area. Optionally, a narrow-band filter is configured at the front end of the camera, and an auxiliary light source is used if necessary to suppress interference from the arc background on the extraction of the molten pool boundary. The spectral acquisition device is aligned with the coupling area between the molten pool and the arc via an optical fiber to acquire the molten pool radiation spectrum. The electrical signal sensor is connected to the welding power source to acquire the welding current and welding voltage sequences. After the three types of raw data enter the buffer queue, the main controller packages the data according to the same timestamp to form a data packet for a single sampling period. The data packet includes at least the image frame number, image timestamp, spectral timestamp, current window timestamp, voltage window timestamp, and the corresponding welding state number. With this setup, the data retrieved in any subsequent processing step can be traced back to the same sampling period in time.

[0031] In step S202, the molten pool image, molten pool radiation spectrum, and welding electrical signal are acquired simultaneously, and image features, spectral features, and electrical signal features are extracted respectively. Regarding image feature extraction, in one implementation, region cropping, grayscale normalization, and edge enhancement are first performed on the molten pool image. Region cropping uses the center of the molten pool as a reference to crop local image blocks of fixed or adaptive size. Grayscale normalization is used to eliminate numerical shifts caused by brightness drift at different welding times. Edge enhancement strengthens the edges of the molten pool and the wake edges.

[0032] Subsequently, two types of image features are extracted. The first type is molten pool morphology features, including handcrafted features such as molten pool area, aspect ratio, boundary curvature, center offset, and wake length. The second type is depth features, extracted using a convolutional neural network. It's important to understand that depth features do not replace morphological features, but rather provide a high-dimensional complement. The convolutional neural network can be configured as a lightweight residual network, taking a preprocessed single-frame image or a short-time image sequence as input and outputting a fixed-dimensional feature vector. If a single-frame image is used as input, the feature extraction relationship can be represented as:

[0033] in, This indicates the preprocessed image input. This represents the parameters of the image feature extraction network. This represents the image depth feature vector. The network parameters can be configured to be obtained through training with historical welding samples and are fixed during the deployment phase; when the type of welding material or forming method changes to meet the preset update conditions, it is then retrained and updated using offline samples.

[0034] Regarding spectral feature extraction, specifically, baseline correction, smoothing and denoising, and intensity normalization are first performed on the acquired spectral intensity curves. Baseline correction is used to eliminate low-frequency drift caused by background radiation, smoothing and denoising is used to reduce high-frequency random noise, and intensity normalization is used to eliminate the influence of absolute brightness fluctuations. After preprocessing, characteristic spectral line intensities, spectral line intensity ratios, and plasma temperatures are extracted. The characteristic spectral line intensities correspond to several pre-selected spectral line peaks or peak areas, and the spectral line intensity ratio is a dimensionless feature obtained by calculating the ratio of the intensities of two target spectral lines. The plasma temperature is obtained from the temperature estimation relationship constructed from multiple spectral lines, and it participates in subsequent fusion as a scalar feature reflecting changes in thermal state. For different welding material systems, the set of characteristic spectral lines is not the same; therefore, the set of characteristic spectral lines needs to be pre-set based on material composition and welding energy input range during the process introduction stage.

[0035] Regarding electrical signal feature extraction, in some embodiments, a time window corresponding to each frame of the image is used as a reference, and current and voltage sequences are extracted from that time window. Subsequently, time-domain statistical features and frequency-domain features are calculated for the current and voltage sequences, respectively. Time-domain statistical features include mean, variance, skewness, kurtosis, and peak-to-peak value. Taking voltage variance as an example, it can be written as:

[0036] in, Indicates the voltage sampling point. This represents the average voltage within that time window. This indicates the number of sampling points. Frequency domain features are obtained by performing a Fast Fourier Transform on the signal within the time window, including the fundamental amplitude, harmonic amplitude, dominant frequency position, and frequency band energy distribution. It can be understood that time-domain statistical features reflect short-term stability, while frequency domain features reflect periodic fluctuations and oscillation modes; together, they constitute the electrical signal feature vector. To ensure dimensional consistency, different features undergo standardized processing before splicing. Standardized parameters, such as the mean and standard deviation, can be obtained statistically from the training set and are fixed during the online phase; when equipment is replaced or the sampling link is changed, the standardized parameters are updated.

[0037] In S203, after obtaining image features, spectral features, and electrical signal features, feature fusion and defect probability prediction are performed. Specifically, the three types of features are first concatenated in a fixed order to obtain a fused feature vector. This fixed order is, for example, image features first, spectral features in the middle, and electrical signal features last. Once selected, this order remains consistent during the training and deployment phases to avoid semantic drift in the model input. After the fused feature vector is input into the defect prediction model, the defect probability is output. The defect prediction model can be configured as a multilayer perceptron consisting of a feature input layer, several fully connected hidden layers, and an output layer. For example, the input layer receives the fused feature vector; the first hidden layer performs linear transformations and nonlinear activations; the second hidden layer further compresses the high-dimensional features; and the output layer obtains a defect probability between zero and one through a logistic function. The model is denoted as:

[0038] in, Represents the fused feature vector. Indicates the parameters of the prediction model. This represents the defect probability. The defect probability here is the sole trigger for subsequent parameter adjustments; its source is clearly defined, and all subsequent control quantities use it as input. There are no cases where undefined sources are used directly. During model training, the input is the fused feature vector corresponding to historical welding samples, and the supervision label is a marker indicating whether defects such as incomplete fusion, porosity, or cracks have occurred. After training, the model parameters can be exported and deployed in an edge controller or industrial computer.

[0039] In step S2, when the defect probability exceeds a preset threshold, the welding parameter adjustment amount is calculated and the welding parameter adjustment is performed based on the sensitivity of the defect probability to the welding parameters and the deviation between the current defect probability and the preset threshold.

[0040] When the defect probability exceeds a preset threshold, the system enters the self-coordination control stage of welding parameters. The welding parameters referred to here include welding current, welding voltage, and welding speed. Sensitivity refers to the degree of response of the defect probability to changes in each welding parameter, and can be obtained from model gradients, offline perturbation experiments, or statistical calibration. In one example, the sensitivity vector is denoted as:

[0041] in, Indicates welding current. Indicates the welding voltage. This indicates the welding speed. First, calculate the deviation between the current defect probability and a preset threshold.

[0042] in, This represents the preset threshold. Then, the parameter increment is calculated using the incremental proportional-integral-derivative (PID) control method. Its general formula can be written as:

[0043] in, Represents the parameter increment vector. This is the proportionality coefficient. The integral coefficient is... The coefficients are differential coefficients. The proportional coefficient represents the influence of the current deviation on the adjustment amount, the integral coefficient represents the influence of the cumulative deviation on the adjustment amount, and the differential coefficient represents the influence of the deviation change rate on the adjustment amount. The values ​​of these three coefficients can be configured to be obtained through offline process calibration or based on historical stable welding batch statistics. To prevent drastic parameter oscillations, amplitude and rate-of-change limits can be added to the incremental vector. Subsequently, the new current, voltage, and welding speed settings are sent to the welding power source and actuator. If the defect probability in the next sampling period is still higher than the preset threshold, the above incremental calculation and adjustment steps are repeated; if it has fallen below the preset threshold, the current parameters are maintained until a new anomaly is triggered. The current forming state is probabilistically expressed through multi-source features, and the adjustment direction is constrained by sensitivity to ensure that the control quantity remains consistent with the defect trend.

[0044] In step S3, after welding is completed, welding process parameters, welding process signals and post-weld inspection data are collected. A digital twin of the welding process is constructed based on the welding process parameters, the welding process signals and the post-weld inspection data. Spatial, temporal and attribute registration is performed on the welding process signals and the post-weld inspection data. The registered fusion features are extracted and the weld quality quantitative score and defect diagnosis results are output.

[0045] Figure 3 A schematic diagram of the post-weld comprehensive evaluation process provided in an embodiment of this disclosure is shown. For example... Figure 3 As shown, in step S301, welding process parameters, welding process signals, and post-weld inspection data are collected. After welding is completed, the system switches to the post-weld comprehensive evaluation stage. First, welding process parameters, welding process signals, and post-weld inspection data are collected. Welding process parameters include at least welding current, welding voltage, welding speed, and wire feed speed. Welding process signals include at least molten pool images, arc acoustic signals, spectral signals, and temperature field data. Post-weld inspection data includes at least ultrasonic inspection data, radiographic inspection data, and three-dimensional morphology data of the weld surface. The three-dimensional morphology data can be obtained by a laser profilometer or a three-dimensional scanning device, and its output is a weld surface point cloud or mesh. It should be noted that some data generated during the welding stage is already stored in a unified data platform. When these data are retrieved again in the post-weld stage, the unified weld identifier and timestamp are still used as the index key, thus ensuring the continuity of the data chain.

[0046] In step S302, a digital twin of the welding process is subsequently constructed. Specifically, the digital twin consists of a physical model and a data-driven model. The physical model employs a thermo-coupled finite element model to reflect welding heat input, temperature conduction, and stress-strain evolution. The data-driven model establishes a nonlinear mapping between process parameters and geometric shaping amounts, temperature correction amounts, or defect tendency amounts. In one implementation, the heat input distribution is represented using a double ellipsoidal heat source. If the heat source density is denoted as...

[0047] in, Indicates the effective heat input power. , , Indicates the shape parameters of the heat source. The energy distribution coefficient is used to solve for the temperature and stress fields in the finite element model based on the heat source density. The effective heat input power can be calculated from...

[0048] It is confirmed that, among them, Indicates the thermal efficiency coefficient. and These represent the welding voltage and welding current, respectively. The thermal efficiency coefficient can be configured through welding thermal cycle calibration, and its value range is determined by the specific material system and welding method. When the welding material, groove type, or heat source configuration changes, the thermal efficiency coefficient needs to be recalibrated. In this way, the heat input of the digital twin comes from the acquired process parameters, and there is no data dependency related to fracture.

[0049] Beyond the physical model, the data-driven model receives welding current, welding voltage, welding speed, and necessary auxiliary features, outputting weld geometry or temperature corrections. Its training samples come from historical welding experiments, detailing the correspondence between known process parameters and measured geometry. The physical model provides the basic field distribution, while the data-driven model provides deviation corrections; the two are integrated to form the final state of the digital twin. For example, the temperature correction relationship can be written as...

[0050] in, This represents the corrected temperature field in a digital twin. This indicates that the temperature field is calculated using a finite element model. This represents the temperature correction amount output by the data-driven model. The input to the correction amount can be configured as the measured temperature field, process parameters, and historical residual characteristics. Therefore, the digital twin is not a static model, but a dynamic virtual entity that is continuously corrected based on measured data.

[0051] In step S303, after the digital twin is constructed, spatial, temporal, and attribute registration is performed. Specifically, a digital twin coordinate system is first established using the three-dimensional point cloud of the weld surface. Point cloud preprocessing may include outlier removal, smoothing filtering, and datum plane alignment. After the digital twin coordinate system is established, ultrasonic and radiographic inspection data are mapped to this coordinate system. For ultrasonic inspection data, the probe scanning path, sound path information, and defect echo position must first be converted into three-dimensional spatial coordinates, and then mapped to the digital twin coordinate system through rigid body transformation. For radiographic inspection data, projection inversion or calibration mapping must be performed based on imaging geometry, and then converted to the digital twin coordinate system. Rigid body transformation can be expressed as follows:

[0052] in, Indicates the source coordinates. Represents the rotation matrix. Represents the translation vector. The target coordinates are represented. The rotation matrix and translation vector can be obtained through calibration blocks or common feature points. After the initial mapping is completed, the spatial registration result is corrected using the iterative nearest-point algorithm to reduce the coordinate residuals between data from different sources. It should be noted that the coordinate results obtained from spatial registration are directly used in subsequent defect localization and surface topography correlation analysis; therefore, spatial registration must be performed before feature extraction.

[0053] Regarding time registration, a temporal correspondence between welding process signals and post-weld inspection results is established using welding timestamps. Each data packet during the welding phase carries a timestamp, while post-weld inspection data is traced back to the corresponding welding time period through weld location and scanning sequence. For example, if the weld is divided into several segments according to the welding direction, the post-weld defect results of each segment can be associated with image, spectral, and electrical signal segments of the same segment during the welding phase. Thus, time registration does not require post-weld inspection and welding inspection to occur at the same moment, but rather requires them to form a consistent temporal mapping on the same weld segment. Attribute registration further establishes an attribute association model between process parameters, welding process signals, and defect detection results. For example, current fluctuations, abnormal molten pool morphology, and spectral temperature fluctuations in a certain weld segment are associated with the lack of fusion defects exhibited in ultrasonic testing of that segment. Through this attribute association, the subsequent comprehensive evaluation model receives not isolated features, but fused samples with spatial, temporal, and attribute consistency.

[0054] In step S304, after completing the three-dimensional registration, the system extracts the registered fusion features and performs a comprehensive quality assessment. The fusion features include at least six categories: molten pool morphology features, spectral features, electrical signal features, ultrasonic testing features, radiographic testing features, and surface morphology features. Molten pool morphology features may include molten pool area, aspect ratio, and boundary change rate; spectral features may include plasma temperature and spectral line intensity ratio; electrical signal features may include current variance, voltage variance, and harmonic amplitude; ultrasonic testing features may include echo amplitude, defect length, and defect depth; radiographic testing features may include grayscale mean, grayscale contrast, and shape factor; and surface morphology features may include weld reinforcement height, weld width, and surface roughness.

[0055] Before being incorporated into the comprehensive quality assessment model, all features are normalized and dimensionally aligned to form a unified fusion feature vector. Weight coefficients can be defined for certain important features to reflect their contribution to the overall score. The weight coefficient refers to the relative influence assigned to a particular feature class in the comprehensive scoring model; its value can be obtained through statistical learning of training samples or by selecting the parameter set that minimizes the scoring error through cross-validation. The weight coefficients remain fixed before model retraining; offline updates are performed when the sample distribution changes significantly.

[0056] The comprehensive quality assessment model can be configured as a probabilistic neural network or other supervised classification and regression networks. Taking a probabilistic neural network as an example, the input layer receives a fused feature vector, the pattern layer represents the probability distribution response of the training samples, the summation layer aggregates the responses of samples of the same type, and the output layer provides a quantitative score for weld quality, as well as the defect type and severity. Here, the quantitative score for weld quality is a continuous value, while the defect diagnosis result is discrete or graded. Both are obtained simultaneously from the same fused feature vector, so there will be no contradiction between the score and the diagnosis result. If classification and regression are required simultaneously, a shared backbone network with a dual-output head structure can also be used. One output head outputs the quality score, and the other output head outputs the defect type and severity. During model training, the quality score label comes from the results of manual evaluation or standard sample evaluation, and the defect type label comes from the combined results of ultrasonic testing, X-ray testing, and manual review.

[0057] In step S4, when the welding scenario is non-metallic pipe hot-melt welding in a high-altitude and / or low-temperature environment, environmental parameters are collected, and the thermal excitation parameters and infrared thermography acquisition parameters are adjusted according to the environmental parameters. The thermography sequence is preprocessed and features are extracted to identify welding defects and output weld quality level and inspection report. The welding process data, post-weld evaluation results and environmental detection results are recorded to a unified data platform.

[0058] When welding non-metallic pipes in high-altitude and / or low-temperature environments undergoing thermofusion welding, a special environmental monitoring link is activated. First, environmental parameters are collected. These parameters include at least ambient temperature, humidity, wind speed, and air pressure. The environmental parameters are collected by a multi-sensor integrated unit and uploaded to the control unit periodically. The environmental parameters are collected at the very beginning of the process because subsequent adjustments to thermal excitation parameters, infrared thermal imaging parameters, and emissivity correction all rely on these parameters.

[0059] After obtaining the environmental parameters, thermal excitation parameter adjustments are performed. The thermal excitation parameters include at least the thermal excitation source power, excitation frequency, and excitation mode. The excitation mode can be a continuous excitation mode, a pulsed excitation mode, or a phase-locked loop excitation mode. According to embodiments of this disclosure, ambient temperature and ambient wind speed significantly affect surface heat dissipation; therefore, these two are used as the main compensation factors for the thermal excitation source power. The thermal excitation source power can be determined according to the following relationship.

[0060] in, This indicates the adjusted power of the thermal excitation source. Indicates the reference power. Indicates the temperature compensation coefficient. Indicates reference temperature. Indicates ambient temperature. This represents the wind speed compensation coefficient. Indicates ambient wind speed. This represents the wind speed influence function. The physical meaning of the temperature compensation coefficient is the proportionality factor of the change in ambient temperature to the change in thermal excitation demand, and the physical meaning of the wind speed compensation coefficient is the proportionality factor of the change in ambient wind speed to the change in thermal excitation demand.

[0061] Both can be calibrated through environmental simulation tests or obtained through statistical analysis of comparative tests under different temperature and wind speed conditions. Their values ​​can remain fixed within the same combination of equipment, materials, and thermal imaging system; they are updated only when the equipment model or material emissivity changes significantly. The selection of the excitation mode depends on the depth of the defect being inspected. For shallow defects, a continuous excitation mode can be configured; for medium-depth defects, a pulsed excitation mode can be configured; and for deeper defects, a phase-locked loop excitation mode can be configured. Therefore, environmental parameters and defect depth jointly determine the thermal excitation scheme.

[0062] Subsequently, infrared thermal imaging acquisition parameter adjustments and temperature corrections are performed. Infrared thermal imaging acquisition parameters include at least gain and integration time. Gain is used to adjust the thermal imager's response to changes in radiation signals, and integration time is used to adjust the single-frame exposure time. Since radiation intensity decreases at low temperatures, gain should be increased and integration time extended if necessary to maintain thermal image contrast. Regarding the method of parameter determination, a calibration table relating ambient temperature to gain and integration time can optionally be pre-established. During detection, the control unit looks up the target gain and target integration time from the table based on the real-time ambient temperature and sends the parameters to the infrared thermal imager.

[0063] If the ambient temperature fluctuates beyond the update threshold during the detection period, the calibration table is reread and the thermal imaging parameters are updated. Simultaneously, a coupling relationship is established between emissivity, temperature, and the environment to correct the thermal imaging temperature data. Here, material emissivity refers to the material surface's ability to emit infrared radiation; its value can be obtained through standard blackbody comparison tests or material sample calibration. For the same pipe material grade and surface condition, material emissivity can be used as an initial constant; when surface contamination, oxidation, or roughness changes significantly, it is updated based on the field calibration value. This correction ensures that subsequent peak temperature, heating rate, cooling rate, and phase difference characteristics are established on a unified temperature reference.

[0064] After obtaining the thermal image sequence, preprocessing and feature extraction are performed. The preprocessing sequence can be configured as filtering and denoising, thermal image non-uniformity correction, and contrast enhancement. Filtering and denoising can employ median filtering or other nonlinear filtering methods to remove isolated noise points and random perturbations. Thermal image non-uniformity correction is used to eliminate fixed pattern noise caused by differences in detector array response. Contrast enhancement can employ local histogram enhancement to highlight the temperature gradient between the defect area and the background area. After preprocessing, thermal wave features are extracted according to the excitation mode.

[0065] For pulsed excitation mode, peak temperature, heating rate, and cooling rate are extracted from the thermal image sequence. Peak temperature is the temperature at which the thermal image curve reaches its maximum value; heating rate is the first-order rate of change of temperature with respect to time during the heating phase; and cooling rate is the first-order rate of change of temperature with respect to time during the cooling phase. For phase-locked loop (PLL) excitation mode, phase difference and amplitude attenuation are extracted based on the frequency domain representation of the time series. Phase difference is the phase shift of the thermal response of the defect region relative to the normal region; and amplitude attenuation is the quantified result of the attenuation of the thermal wave amplitude with propagation depth. These features are standardized to form a thermal image feature map or thermal image feature vector, providing a unified input for subsequent defect segmentation and depth estimation.

[0066] Regarding defect identification, in one implementation, the defect segmentation model employs a segmentation network that connects an encoder and a decoder. The encoder extracts multi-scale thermal image features layer by layer, while the decoder restores spatial resolution layer by layer, preserving edges and local details through skip connections. The input is a preprocessed thermal image feature map, and the output is a pixel-level segmentation result of the same size as the input. Each pixel in the segmentation result is assigned a label such as interface solder joint defects, overheating, inclusions, porosity, or background.

[0067] The model training data comes from thermal image samples with known defect types and depths. During training, enhancement methods such as rotation, flipping, and noise perturbation can be introduced to improve adaptability to different imaging conditions. The defect segmentation results output by the model can directly provide the defect location and size, and can also be combined with phase difference data from phase-locked loop (PLL) mode for defect depth calculation. Therefore, the sources of defect type, defect size, and defect depth correspond to the segmentation results, spatial scale conversion, and phase difference calculation, respectively, and these three sources are clearly defined and interconnected.

[0068] For depth estimation in phase-locked loop (PLL) mode, specifically, the defect depth is calculated based on the phase difference between the defect region and the normal region, the modulation frequency, and the material's thermal diffusivity. The depth relationship can be written as:

[0069] in, Indicates the depth of the defect. This represents the phase difference between the defective region and the normal region. Indicates the thermal diffusivity of the material. This indicates the modulation frequency. The thermal diffusivity of a material is a parameter representing the ability of heat to propagate and diffuse within the material. It can be obtained through material handbooks, standard sample testing, or field calibration. For the same material grade, it can be considered a fixed value when the temperature range remains relatively constant; however, the thermal diffusivity needs to be updated when the material grade changes or the temperature range changes significantly. It is important to note that defect depth can only be calculated after phase difference extraction has been completed and the current modulation frequency is known.

[0070] After obtaining the defect type, size, and depth, the weld quality grade is determined and an inspection report is generated. The weld quality grade can be determined based on industry standards, internal enterprise control standards, or confirmed project acceptance standards. Specifically, the defect type, size, and depth are input into a grading rule set, which outputs the corresponding weld quality grade. The inspection report includes at least environmental parameters, inspection parameters, defect information, and the quality grade. Environmental parameters include ambient temperature, humidity, wind speed, and air pressure at the time of inspection; inspection parameters include excitation mode, thermal excitation source power, excitation frequency, gain, and integration time; defect information includes defect type, location, size, and depth; and the quality grade is the final grading result. The inspection report is bound to a unified weld identifier and written to a unified data platform, thus establishing a correspondence with the in-weld monitoring results and post-weld comprehensive evaluation results.

[0071] In some embodiments, during actual implementation, the unified data platform establishes a unique weld identifier for the same weld, and uses this weld identifier as the primary key to associate welding process parameters, welding process signals, post-weld inspection data, weld quality quantitative scoring, defect diagnosis results, weld quality grade, inspection report, and rework record.

[0072] Specifically, the welding monitoring results are written after the welding phase ends; the post-weld comprehensive evaluation results are written after the post-weld evaluation ends; the environmental testing results are written after the special environment testing ends; and if rework occurs, the rework time, rework process, and post-rework re-inspection results are added. Since all three types of results are associated through the same weld identifier, the cause, evolution, and handling process of any defect can be directly traced. For subsequent process optimization, preset thresholds, sensitivity estimation models, thermal excitation compensation coefficients, and comprehensive evaluation model parameters can be recalibrated based on historical records in the unified data platform. It should be noted that these parameter updates are all completed offline, and the updated parameters are loaded in the next production cycle, without disrupting the stability of the online implementation process.

[0073] In this embodiment, the three links—in-weld adjustment, post-weld comprehensive evaluation, and special environment testing—are organized according to the data characteristics of different stages, but they are not isolated from each other. The defect probability and parameter adjustment records output during the in-weld stage can be used as input for post-weld defect cause analysis; the defect diagnosis results given by the post-weld comprehensive evaluation can be used as cross-validation basis in the special environment testing report; and the quality level and rework suggestions obtained from the special environment testing can be written back into the unified data platform to become the data source for subsequent welding process calibration.

[0074] Thus, the entire process forms a chain from real-time monitoring, parameter adjustment, post-weld evaluation to special environment detection and quality traceability. A continuous data chain and processing chain are established around the same weld, enabling abnormal states during the welding process to be characterized and adjusted in a timely manner. This allows for comprehensive diagnosis of post-weld defects under a unified coordinate and attribute framework. Furthermore, defect detection of non-metallic pipeline thermofusion welding in high-altitude and / or low-temperature environments can be adaptively configured with parameters according to environmental changes, thereby obtaining consistent quality scores, defect diagnosis results, quality grades, and inspection reports.

[0075] Please see Figure 4 , Figure 4 This is a schematic diagram of a pipe welding quality intelligent control system provided in an embodiment of this application. As shown in the figure, the system includes: The multi-source data acquisition module 401 is used to simultaneously acquire images of the molten pool, the radiation spectrum of the molten pool, and welding electrical signals during the welding process; The feature extraction and fusion module 402 is used to process the molten pool image, the molten pool radiation spectrum and the welding electrical signal respectively, extract image features, spectral features and electrical signal features, and fuse the image features, spectral features and electrical signal features; The defect prediction module 403 is used to input the fused features into the defect prediction model to obtain the welding defect probability. The parameter adaptive adjustment module 404 is used to calculate the welding parameter adjustment amount based on the sensitivity of the defect probability to the welding parameters and the deviation between the current defect probability and the preset threshold when the defect probability exceeds a preset threshold, and to control the welding equipment to perform welding parameter adjustment. The post-weld data acquisition module 405 is used to acquire welding process parameters, welding process signals, and post-weld inspection data after welding is completed. The digital twin construction and analysis module 406 is used to construct a digital twin of the welding process based on the welding process parameters, the welding process signals and the post-weld inspection data, perform spatial, temporal and attribute registration on the welding process signals and the post-weld inspection data, extract the fusion features after registration, and output the weld quality quantitative score and defect diagnosis results. The environmental adaptability detection module 407 is used to collect environmental parameters when non-metallic pipe hot-melt welding is carried out in high-altitude and / or low-temperature environments. Based on the environmental parameters, the thermal excitation parameters and infrared thermal imaging acquisition parameters are adjusted, and the thermal image sequence is preprocessed and feature extracted to identify welding defects and output weld quality level and inspection report. The data management module 408 is used to record welding process data, post-weld evaluation results, and environmental monitoring results to a unified data platform.

[0076] Those skilled in the art will clearly understand that the technical solutions of the embodiments of this application can be implemented by means of software and / or hardware. In this specification, "unit" and "module" refer to software and / or hardware that can independently complete or cooperate with other components to complete a specific function, wherein the hardware may be, for example, a field-programmable gate array (FPGA), an integrated circuit (IC), etc.

[0077] Each processing unit and / or module in the embodiments of this application can be implemented by an analog circuit that implements the functions described in the embodiments of this application, or by software that executes the functions described in the embodiments of this application.

[0078] In the above embodiments, the descriptions of each embodiment have different focuses. Parts not described in detail in a certain embodiment can be referred to in the relevant descriptions of other embodiments. The above descriptions are merely preferred embodiments of this application and explanations of the technical principles used. Those skilled in the art should understand that the scope of the invention involved in this application is not limited to the technical solutions formed by specific combinations of the above technical features, but should also cover other technical solutions formed by arbitrary combinations of the above technical features or their equivalent features without departing from the inventive concept.

Claims

1. A method for intelligent control of pipeline welding quality, characterized in that, include: During the welding process, images of the molten pool, radiation spectra of the molten pool, and welding electrical signals are acquired simultaneously. Image features, spectral features, and electrical signal features are extracted respectively. The image features, spectral features, and electrical signal features are then fused and input into a defect prediction model to obtain the defect probability. When the defect probability exceeds a preset threshold, the welding parameter adjustment amount is calculated and the welding parameter adjustment is performed based on the sensitivity of the defect probability to the welding parameters and the deviation between the current defect probability and the preset threshold. After welding is completed, welding process parameters, welding process signals and post-weld inspection data are collected. A digital twin of the welding process is constructed based on the welding process parameters, the welding process signals and the post-weld inspection data. Spatial, temporal and attribute registration is performed on the welding process signals and the post-weld inspection data. The registered fusion features are extracted and the weld quality quantitative score and defect diagnosis results are output. When the welding scenario is non-metallic pipe hot-melt welding in high-altitude and / or low-temperature environments, environmental parameters are collected, and thermal excitation parameters and infrared thermal imaging acquisition parameters are adjusted according to the environmental parameters. The thermal image sequence is preprocessed and features are extracted to identify welding defects and output weld quality level and inspection report. Welding process data, post-weld evaluation results, and environmental monitoring results are recorded in a unified data platform.

2. The method according to claim 1, characterized in that, The synchronous acquisition of molten pool images, molten pool radiation spectra, and welding electrical signals includes: The camera, spectral acquisition device, and electrical signal sensor are synchronously triggered by a unified hardware trigger signal; the image features include molten pool morphology features and depth features; the spectral features include characteristic spectral line intensity, spectral line intensity ratio, and plasma temperature; and the electrical signal features include time-domain statistical features and frequency-domain features.

3. The method according to claim 1, characterized in that, The fused features are input into the defect prediction model to obtain the defect probability, including: The image features, spectral features, and electrical signal features are concatenated to form a fused feature vector, which is then input into the multilayer perceptron prediction head. The calculation of the welding parameter adjustment includes: calculating the parameter increment according to the sensitivity of the defect probability to current, voltage, and welding speed, combined with the deviation between the defect probability and the preset threshold, using an incremental proportional-integral-derivative control method.

4. The method according to claim 1, characterized in that, The acquisition of welding process parameters, welding process signals, and post-weld inspection data includes: The process involves collecting welding current, welding voltage, welding speed, and wire feed speed; acquiring images of the molten pool, arc acoustic signals, spectral signals, and temperature field data; and collecting ultrasonic testing data, radiographic testing data, and three-dimensional morphology data of the weld surface. The construction of a digital twin of the welding process includes fusing a finite element thermo-mechanical coupling model with a data-driven model to obtain a digital twin corresponding to the actual welding process.

5. The method according to claim 4, characterized in that, The spatial, temporal, and attribute registration of welding process signals and post-weld inspection data includes: A digital twin coordinate system is established using the three-dimensional point cloud of the weld surface. Ultrasonic and radiographic inspection data are mapped to the digital twin coordinate system through rigid body transformation. A time correspondence is established using welding timestamps. An attribute association relationship is established between welding process parameters, welding process signals, and defect detection results. The spatial registration results are corrected using an iterative nearest point algorithm.

6. The method according to claim 1 or 5, characterized in that, The process of extracting and registering the fusion features and outputting a quantitative score for weld quality and defect diagnosis results includes: The molten pool morphology features, spectral features, electrical signal features, ultrasonic detection features, radiographic detection features, and surface morphology features are extracted from the registered multi-source data to form a fused feature vector. The fused feature vector is input into the comprehensive quality assessment model to output a quantitative score of weld quality and defect diagnosis results corresponding to defect type and severity.

7. The method according to claim 1, characterized in that, The acquisition of environmental parameters, and the adjustment of thermal excitation parameters and infrared thermal imaging acquisition parameters based on the environmental parameters, include: Collect ambient temperature, ambient humidity, ambient wind speed, and ambient air pressure; adjust the power of the thermal excitation source based on the ambient temperature and ambient wind speed; select continuous excitation mode, pulse excitation mode, or phase-locked excitation mode based on the depth of the defect to be inspected; and adjust the gain and integration time of the infrared thermal imager based on the ambient temperature.

8. The method according to claim 7, characterized in that, The preprocessing and feature extraction of the thermal image sequence includes: The original thermal image sequence is sequentially filtered for noise reduction, thermal image non-uniformity correction, and contrast enhancement. Peak temperature, heating rate, and cooling rate are extracted in pulse excitation mode, and phase difference and amplitude attenuation are extracted in phase-locked excitation mode. The coupling relationship between emissivity, temperature, and environment is established based on ambient temperature, ambient humidity, and pipe material emissivity to correct the thermal image temperature data.

9. The method according to claim 8, characterized in that, The process of identifying welding defects and outputting weld quality grades and inspection reports includes: The preprocessed thermal image feature map is input into the defect segmentation model to obtain the segmentation results corresponding to interface cold welds, overheating, inclusions or porosity; in phase-locked loop excitation mode, the defect depth is calculated based on the phase difference between the defect area and the normal area, the modulation frequency and the material thermal diffusivity; and the weld quality grade is determined according to the defect type, size and depth, generating an inspection report containing environmental parameters, inspection parameters, defect information and quality grade.

10. An intelligent control system for pipeline welding quality, characterized in that, include: The multi-source data acquisition module is used to simultaneously acquire images of the molten pool, the radiation spectrum of the molten pool, and welding electrical signals during the welding process; The feature extraction and fusion module is used to process the molten pool image, the molten pool radiation spectrum, and the welding electrical signal respectively, extract image features, spectral features, and electrical signal features, and fuse the image features, spectral features, and electrical signal features; The defect prediction module is used to input the fused features into the defect prediction model to obtain the probability of welding defects. The parameter adaptive adjustment module is used to calculate the welding parameter adjustment amount based on the sensitivity of the defect probability to the welding parameters and the deviation between the current defect probability and the preset threshold when the defect probability exceeds a preset threshold, and to control the welding equipment to perform the welding parameter adjustment. The post-weld data acquisition module is used to acquire welding process parameters, welding process signals, and post-weld inspection data after welding is completed. The digital twin construction and analysis module is used to construct a digital twin of the welding process based on the welding process parameters, the welding process signals and the post-weld inspection data, perform spatial, temporal and attribute registration on the welding process signals and the post-weld inspection data, extract the fusion features after registration, and output the weld quality quantitative score and defect diagnosis results. The environmental adaptability detection module is used to collect environmental parameters when non-metallic pipes are hot-melt welded in high-altitude and / or low-temperature environments. Based on the environmental parameters, the module adjusts the thermal excitation parameters and infrared thermal imaging acquisition parameters, preprocesses the thermal image sequence and extracts features to identify welding defects and output weld quality level and inspection report. The data management module is used to record welding process data, post-weld evaluation results, and environmental monitoring results to a unified data platform.

Images