Sound barrier stand column weld quality on-line detection and automatic welding method and system

Abstract

The present application relates to the technical field of sound barrier steel structure manufacturing, and particularly relates to a sound barrier stand column welding seam quality online detection and automatic welding method and system. The method comprises the following steps: obtaining a bevel feature and generating a welding track parameter through line laser scanning; starting a multi-source sensing system when multi-layer multi-pass welding is performed; collecting a welding seam appearance, a thermal field distribution, an arc sound print and an internal ultrasonic signal; extracting each channel feature quantity after time and space alignment; distributing a weight according to a sensitivity and dynamically adjusting a fusion weight in combination with a real-time signal-to-noise ratio; calculating a quality evaluation index after fusing the normalized features and the weight of each channel; and determining three levels of a current welding pass, i.e., qualified, repairable or unrepairable. The method performs parameter adaptive fine tuning, online remelting and reinspection iteration, emergency shutdown and defect marking grading closed-loop control according to different quality levels. The present application realizes closed-loop management of online detection and quality control in the welding process, and improves the consistency and reliability of the sound barrier stand column welding quality.

Description

Technical Field

[0001] This invention relates to the field of sound barrier steel structure manufacturing technology, and in particular to a method and system for online inspection and automatic welding of sound barrier column weld quality. Background Technology

[0002] Noise barriers are crucial infrastructure for reducing traffic noise. Their columns are assembled from H-beams and base plates through welding, and the quality of the welds directly determines the strength and stability of the connection points. According to relevant construction quality acceptance standards, the welds of noise barrier columns must reach a quality level of at least Grade II, and must not have defects such as incomplete fusion, incomplete penetration, porosity, cracks, or burn-through. Because H-beams are long and heavy, traditional manual welding is labor-intensive, inefficient, and struggles to consistently meet weld quality requirements. Uneven heating during welding can easily cause deformation, leading to dynamic changes in the bevel gap. Regarding quality inspection, current technologies mostly employ post-weld sampling inspections, primarily relying on piezoelectric ultrasonic testing or radiographic testing. However, piezoelectric ultrasonic testing requires the use of coupling agents, and residues can contaminate the column surface, affecting subsequent anti-corrosion treatment. Furthermore, once problems are discovered during post-weld inspection, they cannot be remedied online, requiring rework or scrapping.

[0003] Regarding automated welding, existing patents disclose robotic automated equipment for sound barrier steel columns, achieving automated welding through trusses, positioning mechanisms, and multiple robots. However, these patents do not address real-time quality detection and closed-loop feedback control during the welding process. In terms of multi-sensor fusion detection, existing research has proposed solutions using infrared cameras and microphones to identify welding quality issues in real time. However, these solutions are mostly general-purpose and lack specific design for the unique geometry of T-shaped fillet welds in sound barrier columns. Furthermore, electromagnetic ultrasonic testing technology, which requires no coupling agent and allows for non-contact inspection, has been used for online weld inspection. However, existing solutions are mostly general-purpose designs, lacking specific adaptation for fillet welds in sound barrier columns and failing to form a complete closed loop with adaptive adjustment of welding parameters. Summary of the Invention

[0004] To overcome the above shortcomings, this invention provides an online detection and automatic welding method for the weld quality of sound barrier columns, aiming to improve the problems of lack of sensing ability and lagging quality control in the existing technology.

[0005] In a first aspect, the present invention provides the following technical solution: an online detection and automatic welding method for the weld quality of sound barrier columns, comprising:

[0006] Acquire the bevel feature data of the fillet weld, and generate the walking trajectory of the welding robot and the initial values ​​of the welding parameters based on the bevel feature data;

[0007] Multi-layer, multi-pass welding is performed according to the walking trajectory and initial values ​​of welding parameters, and multi-source heterogeneous sensing data including visual morphology, thermal field distribution, acoustic signature features and internal defect ultrasound are acquired synchronously through a follow-up detection system.

[0008] Based on the multi-source heterogeneous sensing data, a spatiotemporal synchronization mapping relationship is established for each sensing channel under a unified time reference and weld space coordinate reference.

[0009] Feature extraction is performed based on the spatiotemporally aligned multi-source heterogeneous sensor data. Weight coefficients are dynamically assigned according to the detection sensitivity of each sensor channel for different defect types. A multi-dimensional defect judgment model is constructed to evaluate the quality level of the current weld.

[0010] Based on the quality level, hierarchical closed-loop control is executed. If the current quality level meets the conditions for continuing welding, the welding parameters are adaptively adjusted and welding continues; otherwise, the welding process is interrupted and the defect location is marked.

[0011] Furthermore, the multi-layer, multi-pass welding specifically includes:

[0012] Clamp the sound barrier column into the rotary positioning fixture and adjust the fillet weld to the flat weld position;

[0013] During the welding process, the interpass temperature is monitored, and the welding speed is adjusted according to temperature fluctuations to keep the cooling rate of the molten pool within the preset process window.

[0014] After each weld bead is completed, the welding parameters are updated based on the bevel condition of the next layer, and welding continues until all weld bead bead is completed.

[0015] Furthermore, the multi-source heterogeneous sensing data includes weld surface contour point cloud data obtained by line laser triangulation, temperature field data of the molten pool and its heat-affected zone obtained by infrared radiation thermometry, arc acoustic signature data obtained by acoustic sensing, and ultrasonic echo data inside the weld obtained by an electromagnetic ultrasonic transducer without the need for a coupling agent.

[0016] Furthermore, before collecting the ultrasonic echo data inside the weld, a magnetostrictive coating is sprayed onto the weld surface, and the coverage integrity of the sprayed area is checked; if the check result does not meet the preset quality requirements, the spraying and check are repeated until the preset quality requirements are met.

[0017] Furthermore, the step of achieving alignment under a unified time reference and weld space coordinate reference includes:

[0018] The same clock source is applied to all sensing channels to achieve time synchronization, and coordinate transformation is performed based on the spatial pose of each sensor relative to the welding torch, so that the data collected by each channel is associated with a unified weld space coordinate reference.

[0019] Furthermore, the feature extraction specifically includes:

[0020] Geometric contour features, temperature field features, acoustic spectrum features, and ultrasonic echo features were extracted from visual morphology data, thermal field distribution data, acoustic signature feature data, and internal defect ultrasonic data, respectively.

[0021] Furthermore, the step of dynamically allocating weight coefficients includes:

[0022] Based on the preset detection sensitivity of each sensor channel for different defect types, basic weights are assigned, and the real-time signal-to-noise ratio of each sensor channel is evaluated simultaneously. The weights of each channel are adjusted by reducing the weight of the corresponding channel when the signal-to-noise ratio is lower than a preset threshold, and a fusion judgment vector is generated.

[0023] Furthermore, the step of assessing the quality grade of the current weld bead includes:

[0024] The quality evaluation index is calculated by comparing the fusion results of each sensor channel with the preset judgment criteria.

[0025] The weld quality is marked as qualified, repairable defect, or unrepairable defect based on the threshold range of the quality evaluation index.

[0026] Furthermore, the hierarchical closed-loop control specifically includes:

[0027] When the quality level is repairable, the correction amount of welding parameters is calculated according to the defect type and severity, and online remelting is performed by adjusting the welding current and walking speed.

[0028] After the remelting is completed, the follow-up detection system is called to re-inspect and verify whether the defect has been eliminated. If the defect has not been completely eliminated, the remelting is repeated.

[0029] When the quality level is an unrepairable defect, the welding process is interrupted and the defect location is marked.

[0030] Secondly, the present invention provides the following technical solution: an online inspection and automatic welding system for the weld quality of sound barrier columns, used to implement the above-mentioned online inspection and automatic welding method for weld quality, the system comprising:

[0031] The path planning module is used to acquire fillet weld groove feature data and generate the walking trajectory of the welding robot and the initial values ​​of welding parameters based on the groove feature data.

[0032] The welding and acquisition module is used to perform multi-layer and multi-pass welding according to the walking trajectory and initial values ​​of welding parameters, and to synchronously acquire multi-source heterogeneous sensor data including visual morphology, thermal field distribution, acoustic features and internal defect ultrasound through the follow-up detection system.

[0033] The spatiotemporal mapping module is used to establish a spatiotemporal synchronization mapping relationship between each sensing channel under a unified time reference and weld space coordinate reference based on the multi-source heterogeneous sensing data.

[0034] The fusion judgment module is used to extract features based on the spatiotemporally aligned multi-source heterogeneous sensor data, dynamically allocate weight coefficients according to the detection sensitivity of each sensor channel for different defect types, construct a multi-dimensional defect judgment model, and evaluate the quality level of the current weld.

[0035] The closed-loop control module is used to perform graded closed-loop control based on the quality level. If the current quality level meets the conditions for continuing welding, the welding parameters are adaptively adjusted and welding continues; otherwise, the welding process is interrupted and the defect location is marked.

[0036] The present invention has the following beneficial effects:

[0037] 1. In this invention, a multi-source sensing system of vision, infrared, acoustic signature, and electromagnetic ultrasound is simultaneously activated during the welding process to perform comprehensive online detection of weld surface morphology, thermal field distribution, arc stability, and internal defects. The electromagnetic ultrasound detection is enhanced by a magnetostrictive coating, which can obtain internal defect information without the need for a coupling agent. This fundamentally solves the problem of surface contamination caused by the use of coupling agents in traditional piezoelectric ultrasonic detection. At the same time, it realizes the simultaneous execution of welding process and quality inspection, avoiding the drawback of not being able to remedy problems online when they are found during post-weld sampling inspection.

[0038] 2. In this invention, by establishing a spatiotemporal mapping relationship between each sensing channel under a unified time reference and weld space coordinate reference, the detection information of different physical dimensions is aligned under a unified spatiotemporal framework, eliminating data misalignment caused by differences in sensor distribution spacing and acquisition timing. At the same time, basic weights are assigned according to the detection sensitivity of each sensing channel for different defect types, and the fusion weights are dynamically adjusted in combination with the real-time signal-to-noise ratio. This allows multi-source features to adaptively adjust their contribution level according to the real-time data quality of each channel during the fusion process, thereby obtaining a quality evaluation index that objectively reflects the overall quality of the weld and improving the accuracy and reliability of defect identification.

[0039] 3. In this invention, the weld quality is divided into three levels: qualified, repairable defects, and unrepairable defects according to the quality evaluation index. For each level, a hierarchical closed-loop control is implemented, which includes adaptive parameter fine-tuning, online remelting and re-inspection iteration, emergency shutdown and defect marking. The remelting process increases heat input by increasing current and decreasing speed to eliminate defects. Welding can only continue after the remelting is confirmed to be qualified. This forms a complete closed loop from quality inspection to process control, realizing in-process intervention in welding quality rather than post-process remediation, and reducing rework costs. Attached Figure Description

[0040] Figure 1 This is a flowchart of an online inspection and automatic welding method for the weld quality of a sound barrier column proposed in this invention;

[0041] Figure 2 This is a flowchart of the multi-source sensor data fusion judgment process based on dynamic weight allocation proposed in this invention;

[0042] Figure 3 This is a flowchart of the hierarchical closed-loop control based on quality level proposed in this invention;

[0043] Figure 4 This is a diagram of an online inspection and automatic welding system for the weld quality of a sound barrier column proposed in this invention. Detailed Implementation

[0044] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0045] Example 1

[0046] In the first embodiment of the present invention, the present invention provides an online inspection and automatic welding method for the weld quality of sound barrier columns, such as... Figure 1 As shown, it includes the following steps:

[0047] S100: Obtain the bevel feature data of the fillet weld, and generate the walking trajectory of the welding robot and the initial values ​​of the welding parameters based on the bevel feature data;

[0048] Specifically, a line laser vision sensor installed at the end effector of the welding robot pre-scans the fillet welds of the sound barrier columns. The line laser vision sensor acquires raw point cloud data of the weld area at a sampling frequency of 100Hz to 200Hz. By performing median filtering and background noise removal on the point cloud data, key geometric features of the bevel, including bevel width, root gap, and bevel angle, are extracted. The bevel feature points in the sensor coordinate system are then transformed to the robot base coordinate system through hand-eye calibration and robot kinematic transformation to obtain the spatial position of the bevel feature points in the robot base coordinate system.

[0049] Based on the extracted bevel feature data, the system calculates the filling cross-sectional area of ​​each section. For a typical fillet weld, the filler cross-sectional area is... The calculation method is as follows:

[0050] ;

[0051] in, This represents the theoretical filler cross-sectional area of ​​the current weld bead, in units of... The value ranges from 20 to 80. ; This is the weld reinforcement coefficient, with a value range of [value range missing]. ; and These are the lengths of the two right-angled sides of the fillet weld, in units of... The value ranges from 6 to 15. ; Root gap, unit: The value ranges from 0 to 3. ; The thickness of the sheet material is given in units of 1. The value ranges from 10 to 20. .

[0052] Using the above formula, the system dynamically calculates the required filling cross-sectional area for each section based on the bevel geometry parameters, providing a data basis for the precise matching of welding parameters and avoiding incomplete penetration due to insufficient filling or excessive reinforcement due to excessive filling.

[0053] Based on the calculated filling cross-sectional area The system generates initial values ​​for welding parameters. The welding parameters must be set to balance the amount of wire deposited and the amount of groove filling. The welding travel speed is calculated using the following formula:

[0054] ;

[0055] in, The walking speed of the welding robot, in units of The value ranges from 5 to 12. ; The diameter of the welding wire is expressed in units of 1 / 200 mm. In this embodiment, the value is taken as 1.2. ; The wire feeding speed is expressed in units of... The value ranges from 6 to 12. ; For the deposition efficiency, the value range is: .

[0056] By balancing the above calculations of the welding wire deposition amount and the groove filling amount, it is ensured that the molten pool is fully filled with metal during the welding process without producing excessive excess height.

[0057] When generating initial values ​​for welding parameters, the system synchronously matches the welding current. With welding voltage In this embodiment, the matching relationship between current and voltage is determined by... Confirmed. For sound barrier column steel plates with a thickness of 10 mm to 20 mm, the initial welding current... The initial welding voltage is set between 220 amps and 320 amps. The voltage is set between 24 volts and 32 volts. These parameters can be varied within a reasonable range depending on the actual working conditions. The calculated spatial path point sequence is bound to the corresponding welding parameters to generate a complete welding command stream, which is then downloaded to the robot controller, completing the path planning and parameter initialization before welding.

[0058] By acquiring the geometric features of the bevel using a line laser vision sensor and calculating the theoretical filling cross-sectional area accordingly, and then matching welding parameters based on the principle of balancing wire deposition and bevel filling, precise planning of the initial travel trajectory and welding parameters for the fillet weld of the sound barrier column is achieved. This provides a reliable data benchmark for online detection and adaptive control, and improves the adaptability and consistency of the multi-layer, multi-pass welding process.

[0059] S200: Perform multi-layer and multi-pass welding according to the walking trajectory and initial values ​​of welding parameters, and synchronously acquire multi-source heterogeneous sensing data including visual morphology, thermal field distribution, acoustic features and internal defect ultrasound through a follow-up detection system.

[0060] Furthermore, the multi-layer, multi-pass welding specifically includes:

[0061] Clamp the sound barrier column into the rotary positioning fixture and adjust the fillet weld to the flat weld position;

[0062] During the welding process, the interpass temperature is monitored, and the welding speed is adjusted according to temperature fluctuations to keep the cooling rate of the molten pool within the preset process window.

[0063] After each weld bead is completed, the welding parameters are updated based on the bevel condition of the next layer, and welding continues until all weld bead bead is completed.

[0064] Specifically, the robot drives the welding torch and the follow-up detection system to perform multi-layer, multi-pass welding operations according to the planned path. The sound barrier columns are fixed on the rotary positioning fixture. The fillet weld is adjusted to the flat welding position by the axial rotation of the positioner, that is, the center line of the weld is horizontal and the two plates are at a 45-degree angle to the horizontal plane, so as to use gravity to promote the transfer of molten droplets and the formation of weld beads.

[0065] During welding, the interpass temperature is monitored in real time using an infrared radiation temperature sensor. To ensure the mechanical properties of the welded joint, the welding speed is adjusted in real time based on temperature fluctuations, constraining the cooling rate of the molten pool within a preset process window. The cooling rate is calculated using the following empirical formula:

[0066] ;

[0067] in, The weld cooling rate is set between 10℃ / s and 30℃ / s. The thermal efficiency coefficient is 0.8 for gas shielded arc welding. and These are the real-time welding voltage and current, respectively. For the thickness of the board, and the thickness of the board. The value range is consistent, that is, 10. Up to 20 .

[0068] This formula is an empirical approximation, where To determine the overall thermal efficiency coefficient, its specific value is calibrated through process experiments. The cooling rate is empirically inversely proportional to the linear energy and plate thickness. When the interlayer temperature exceeds 250℃, the system adjusts its response based on the current... , , and target cooling rate Rewrite the formula as Calculate the target walking speed Based on this, the welding speed is increased first; when the speed adjustment is limited by the process window, the interpass dwell time is extended to ensure that the initial temperature before the next welding is in the range of 150℃ to 200℃.

[0069] Furthermore, the multi-source heterogeneous sensing data includes weld surface contour point cloud data obtained by line laser triangulation, temperature field data of the molten pool and its heat-affected zone obtained by infrared radiation thermometry, arc acoustic signature data obtained by acoustic sensing, and ultrasonic echo data inside the weld obtained by an electromagnetic ultrasonic transducer without the need for a coupling agent.

[0070] Furthermore, before collecting the ultrasonic echo data inside the weld, a magnetostrictive coating is sprayed onto the weld surface, and the coverage integrity of the sprayed area is checked; if the check result does not meet the preset quality requirements, the spraying and check are repeated until the preset quality requirements are met.

[0071] Specifically, the servo detection system is activated synchronously during the welding process, acquiring multi-source heterogeneous sensor data. The line laser vision sensor obtains three-dimensional point cloud data of the weld surface through the principle of laser triangulation, with a sampling depth resolution set to 0.05. It is used to describe the weld surface reinforcement, back width, and forming characteristics.

[0072] The infrared radiation thermometry system acquires temperature field data of the molten pool and its heat-affected zone (HAZ) through a non-contact temperature array. The sensor has a temperature measurement range of 300℃ to 2000℃ and a sampling frequency of 50Hz. Based on the temperature distribution curve perpendicular to the weld direction in the temperature field, the system uses the location corresponding to the peak temperature as the center, the width corresponding to a preset temperature threshold as the HAZ width, and the rate of temperature change within this range as the temperature gradient to analyze the width of the HAZ and the temperature gradient distribution.

[0073] The acoustic sensing system collects acoustic signature data of the electric arc during the welding process using a condenser microphone, with the audio sampling rate set to 44.1 kHz or 48 kHz. By analyzing the root mean square value of the arc sound pressure level in the time domain and the dominant frequency energy distribution in the frequency domain, characteristic values ​​reflecting the stability of the welding process are extracted.

[0074] Internal defect detection utilizes an electromagnetic ultrasonic transducer. Before acquiring ultrasonic echo data, an automatic spraying device applies a 10mm thick layer of coating to the surface of the weld to be inspected. Up to 30 A magnetostrictive coating of iron-cobalt alloy is applied. After spraying, the visual verification module uses an image contrast recognition algorithm to detect the integrity of the coating coverage. Image processing is used to calculate the ratio of the area covered by the coating to the total area to be inspected. If the coverage rate is lower than the preset quality requirement of 98%, the system controls the spraying device to repeat the spraying and verification. After passing the verification, the electromagnetic ultrasonic transducer excites ultrasonic shear waves without the need for a coupling agent, acquiring echo data from inside the weld with frequencies between 1MHz and 5MHz.

[0075] After each layer of weld is welded and inspected, the system updates the geometric space data of the next layer of bevel in real time based on the current weld surface morphology obtained by the line laser sensor, and adjusts the welding trajectory compensation amount and parameter initial values ​​accordingly, until all multi-layer and multi-pass stacking of the fillet weld of the sound barrier column is completed.

[0076] The fillet weld is adjusted to a flat welding position using a rotary positioning tool to ensure weld bead formation stability; the welding speed is adjusted according to interpass temperature fluctuations to constrain the molten pool cooling rate and avoid joint defects; a multi-source sensing system is simultaneously activated during welding to comprehensively monitor surface morphology, thermal field distribution, process stability, and internal quality online, with electromagnetic ultrasound enhancing the signal through a magnetostrictive coating, enabling the acquisition of internal defect information without the need for a coupling agent; after each weld bead is completed, the bevel data for the next layer is updated based on the surface morphology, and welding parameters are adjusted to achieve adaptive stacking layer by layer.

[0077] S300: Based on the multi-source heterogeneous sensing data, establish a spatiotemporal synchronization mapping relationship between each sensing channel under a unified time reference and weld space coordinate reference.

[0078] Furthermore, the step of achieving alignment under a unified time reference and weld space coordinate reference includes:

[0079] The same clock source is applied to all sensing channels to achieve time synchronization, and coordinate transformation is performed based on the spatial pose of each sensor relative to the welding torch, so that the data collected by each channel is associated with a unified weld space coordinate reference.

[0080] Specifically, the system performs spatiotemporal alignment processing on the collected multi-source heterogeneous sensor data to ensure that information from different physical dimensions is fused within a unified coordinate framework.

[0081] For time synchronization, the system uses the synchronization trigger pulse generated by the main control unit as the reference clock source. The acquisition frequencies of each sensor channel differ significantly: the line laser scanning frequency is 100Hz, the infrared temperature acquisition frequency is 50Hz, the arc acoustic signature sampling rate is 48kHz, and the ultrasonic echo sampling rate is 10MHz. The system records the timestamp of each frame of data relative to the trigger pulse, and uses a high-precision crystal oscillator to control the relative time error between channels to within 1 millisecond.

[0082] For spatial coordinate alignment, the system performs spatial mapping based on the installation offset of each sensor relative to the center of the welding torch nozzle. The line laser sensor is mounted 30 degrees in front of the welding torch. The infrared sensor is installed 15 cm behind the welding torch. At this location, the electromagnetic ultrasonic transducer is installed 80 degrees behind the welding torch. Because the sensors are physically spaced along the welding direction, different sensors detect different points on the physical location of the weld at the same time. The system establishes a weld space coordinate system with the starting welding point as the origin, and rearranges the time-stamped data from each channel according to their corresponding weld space positions to form a spatially aligned multidimensional data matrix.

[0083] The spatial grid points of this multidimensional data matrix contain the surface morphology, temperature gradient, acoustic spectrum, and internal ultrasonic echo intensity at that location, providing standardized data input for feature extraction and defect identification.

[0084] By synchronously triggering pulses to assign a unified time reference to each sensor data, and combining the sensor installation offset, the time series signal is mapped to the weld space coordinate system, so that the detection information of different physical dimensions is aligned in a unified spatiotemporal framework, thereby eliminating data misalignment caused by differences in sensor distribution spacing and acquisition timing, and providing a standardized data foundation with spatiotemporal consistency for the collaborative analysis of multi-source information.

[0085] S400: Based on the spatiotemporally aligned multi-source heterogeneous sensor data, feature extraction is performed, and weight coefficients are dynamically allocated according to the detection sensitivity of each sensor channel for different defect types to construct a multi-dimensional defect judgment model and evaluate the quality level of the current weld.

[0086] Furthermore, the feature extraction specifically includes:

[0087] Geometric contour features, temperature field features, acoustic spectrum features, and ultrasonic echo features were extracted from visual morphology data, thermal field distribution data, acoustic signature feature data, and internal defect ultrasonic data, respectively.

[0088] Specifically, the system performs multi-dimensional analysis on spatiotemporally aligned multi-source heterogeneous sensor data, and achieves quantitative assessment of weld quality through feature extraction and data fusion. The system extracts characterization parameters reflecting welding quality from four types of sensor channels.

[0089] For visual topography data, the system extracts the geometric contour features of the weld seam using a point cloud algorithm. The system samples the 3D point cloud data in equally spaced slices along the weld seam length, performs linear fitting on the point cloud within each slice to obtain the weld seam surface contour, and calculates the weld reinforcement height, weld width, and weld toe angle based on the contour. The weld reinforcement height is the vertical distance between the highest point of the weld surface and the base metal reference plane, and its determination reference range is 1. Up to 3 The weld width is the lateral distance between the inflection points on both sides of the weld contour, and its judgment criterion range is 8. Up to 16 The weld toe angle is the angle between the tangent direction of the weld profile and the base material reference plane, and its determination criterion range is... to When the above geometric features exceed the judgment criteria range, it indicates that there are surface forming defects such as undercut or excessive weld height.

[0090] For the thermal field distribution data, the system extracts the molten pool width and its longitudinal temperature gradient. The molten pool width is extracted from the maximum lateral distance of the melting point isotherms above 1450°C in the infrared thermal image; this feature is typically distributed within 8... Up to 14 Within the specified range. The longitudinal temperature gradient is calculated by extracting the temperature drop rate along the welding direction and is used to assess whether the heat input is uniform and whether there are localized overheating areas. When the weld pool width exceeds the upper limit of the reference range, it indicates that excessive heat input may lead to burn-through; when the weld pool width is below the lower limit of the reference range, it indicates that insufficient heat input may lead to incomplete fusion.

[0091] For acoustic signature data, the system converts the time-domain acoustic signal into a frequency-domain power spectrum using a Fast Fourier Transform (FFT), and extracts the energy proportion of the characteristic frequency band from 2kHz to 8kHz as the acoustic signature feature. This frequency band corresponds to the short-circuit transition frequency during the arc combustion process, and its energy fluctuations can reflect the stability of the droplet transition. The energy proportion is calculated as the ratio of the power spectrum integral value within the characteristic frequency band to the power spectrum integral value across the entire frequency band; a higher ratio indicates better arc stability.

[0092] For ultrasonic data of internal defects, the system extracts the peak amplitude and time of flight of the ultrasonic echo. The system uses the bottom echo amplitude as a reference benchmark and uses the ratio of the echo amplitude to the bottom echo amplitude as the defect discrimination criterion. When this ratio exceeds a preset threshold, it is determined that an internal defect exists.

[0093] In this embodiment, when the echo amplitude exceeds 30% of the bottom echo amplitude, an internal defect is determined to exist at that location. The system calculates the embedment depth of the defect in the weld cross-section based on the echo's time of flight. Different defect types correspond to different echo characteristic patterns: porosity defects have relatively low echo amplitudes and exhibit isolated spikes, with amplitude ratios typically between 30% and 50%; slag inclusion defects have echo amplitudes between porosity and incomplete penetration, with rapid attenuation at the tail, and amplitude ratios typically between 50% and 70%; incomplete penetration defects have high echo amplitudes and a large time interval with the bottom echo, with amplitude ratios typically above 70%. The aforementioned amplitude ratio thresholds and defect classification ranges are obtained through calibration using standard samples with pre-existing defects and can be adaptively adjusted according to the actual plate thickness and welding process.

[0094] Furthermore, such as Figure 2 As shown, the step of dynamically allocating weight coefficients includes:

[0095] Based on the preset detection sensitivity of each sensor channel for different defect types, a basic weight is assigned, and the real-time signal-to-noise ratio of each sensor channel is evaluated simultaneously. The weight of each channel is adjusted by reducing the weight of the corresponding channel when the signal-to-noise ratio is lower than a preset threshold, and a fusion judgment vector is generated.

[0096] Specifically, basic weights are assigned to each sensor based on its physical detection sensitivity to different defect types. The visual sensing channel has high detection sensitivity for surface forming defects, namely undercut and abnormal weld height, and its basic weight is set to 0.30. The infrared sensing channel has high detection sensitivity for abnormal heat input, namely burn-through and lack of fusion, and its basic weight is set to 0.25. The acoustic signature sensing channel has high detection sensitivity for process stability, namely arc fluctuation and spatter abnormalities, and its basic weight is set to 0.20. The electromagnetic ultrasonic sensing channel has the highest detection sensitivity for internal defects, namely porosity, slag inclusion, and incomplete penetration, and its basic weight is set to 0.40. The sum of the basic weights of all channels is 1.15, which is greater than 1. After normalization, the sum of the weights of all channels equals 1.

[0097] The signal-to-noise ratio (SNR) of each sensing channel is then calculated in real time. The SNR is determined by the ratio of the characteristic signal power to the ambient noise power, calculated by taking the logarithm of the ratio to base 10 and then multiplying by 10, with the unit being decibels (dB). The system's set SNR threshold is 12 dB. When the SNR of a channel falls below 12 dB, it indicates that the sensor is affected by environmental factors such as arc overexposure, electromagnetic interference, or splash obstruction, resulting in reduced data reliability. The system then reduces the weight of this channel using a penalty factor. The penalty factor decreases linearly as the SNR decreases.

[0098] When the signal-to-noise ratio (SNR) is 12 dB, the penalty factor is 1.0. When the SNR drops to 6 dB, the penalty factor is 0.2. When the SNR is between 6 dB and 12 dB, the penalty factor is calculated using linear interpolation. When the SNR is higher than 12 dB, the penalty factor remains at 1.0. The system multiplies the base weight of each channel by the penalty factor, then divides by the sum of the products of all channels for normalization, ensuring that the final sum of the fused weights equals 1.

[0099] Furthermore, the step of assessing the quality grade of the current weld bead includes:

[0100] The quality evaluation index is calculated by comparing the fusion results of each sensor channel with the preset judgment criteria.

[0101] The weld quality is marked as qualified, repairable defect, or unrepairable defect based on the threshold range of the quality evaluation index.

[0102] Specifically, the system normalizes the original feature values ​​of each channel. The normalization method involves linearly mapping the feature values ​​of each channel to a preset reference range. The above reference range is determined based on the statistical distribution of characteristic quantities corresponding to qualified weld beads in process experiments, and can be adaptively adjusted according to the actual plate thickness and welding process. For weld reinforcement, a range of 1... Up to 3 For reference only, values ​​less than 1 Mapped to 0, greater than 3 Mapped to 1; for the molten pool width, mapped to 8. Up to 14 For reference only, less than 8 Mapped to 0, greater than 14 The mapping is 1; for the acoustic energy ratio, the reference range is 0.3 to 0.8, less than 0.3 is mapped to 0, and greater than 0.8 is mapped to 1; for the ultrasonic echo amplitude ratio, the reference range is 0 to 1, and the amplitude ratio is directly used as the normalized score. The higher the normalized score of each channel, the better the weld quality detected by that channel.

[0103] The quality evaluation index is obtained by multiplying the normalized feature scores of each channel by their corresponding final fusion weights and then summing the results. Its value range is The system is based on a quality evaluation index. The current threshold range divides the quality of the weld into three levels.

[0104] when When the value is ≥0.85, the system marks the weld as qualified, indicating that the weld is well formed and there are no obvious internal or external defects;

[0105] when When this happens, the system marks the weld as a repairable defect, indicating that there are defects such as local undercut and surface porosity that can be eliminated through online remelting;

[0106] when When the value is less than 0.60, the system marks the weld as an unrepairable defect, indicating the presence of severe incomplete penetration, through-cracks, or continuous internal defects, requiring manual intervention or scrapping.

[0107] By extracting feature parameters reflecting weld geometry, thermal input state, process stability, and internal integrity from four types of sensor data—visual, thermal field, acoustic, and ultrasonic—basic weights are assigned based on the physical detection sensitivity of each sensor channel for different defect types. The fusion weights are then dynamically adjusted according to the real-time signal-to-noise ratio, enabling the contribution of multi-source features to adaptively adjust based on the real-time data quality of each channel during the fusion process. The normalized feature scores of each channel are fused with their corresponding weights to obtain a quantified quality evaluation index. Based on the threshold range of the evaluation index, the weld quality is divided into different levels, achieving an objective quantitative assessment of welding quality levels.

[0108] S500: Based on the quality level, perform graded closed-loop control. If the current quality level meets the conditions for continuing welding, then adaptively adjust the welding parameters and continue welding; otherwise, interrupt the welding process and mark the defect location.

[0109] Furthermore, such as Figure 3 As shown, the hierarchical closed-loop control specifically includes:

[0110] When the quality level is repairable, the correction amount of welding parameters is calculated according to the defect type and severity, and online remelting is performed by adjusting the welding current and walking speed.

[0111] After the remelting is completed, the follow-up detection system is called to re-inspect and verify whether the defect has been eliminated. If the defect has not been completely eliminated, the remelting is repeated.

[0112] When the quality level is an unrepairable defect, the welding process is interrupted and the defect location is marked.

[0113] Specifically, corresponding control operations are performed based on the assessed quality level. In a qualified state, when... When the value is ≥0.85, the system determines that the current weld formation quality is good. At this time, the control module enters adaptive adjustment mode, incrementally correcting the welding current and travel speed based on the real-time deviation of the weld pool width, so that the weld pool width approaches the target value. The current correction is controlled within ±10 amperes, and the speed correction is controlled within ±0.5 amperes. Within the specified range. The specific value of the correction amount is determined proportionally based on the deviation of the molten pool width. The proportionality coefficient is calibrated through process testing to ensure that the correction amount does not exceed the above-mentioned safe adjustment range when the deviation is within 1 mm.

[0114] In a repairable defect state, when the quality evaluation index When the system detects minor defects such as porosity, undercut, or lack of fusion in the weld, it immediately initiates an online remelting procedure. The control module calculates the compensation amount for welding parameters based on the severity of the defect and performs in-situ or overlay secondary welding. During remelting, the welding current is increased to 1.1 to 1.25 times the initial current to increase heat input and promote metal remelting and filling at the defect site; the walking speed is reduced to 0.7 to 0.85 times the initial speed to extend the arc duration and ensure sufficient molten pool filling. After the remelting operation is completed, the robot, driven by a follow-up detection system, re-inspects the defect location. The system re-executes the data acquisition and evaluation process. If the quality evaluation index after re-inspection improves to above 0.85, the defect is considered eliminated and normal welding resumes; if the re-inspection result is still below 0.85, the system repeats the remelting command, with a maximum repetition limit of 2 times. If the limit is exceeded, the weld is marked as an unrepairable defect and the system is shut down.

[0115] In an irreparable defect state, when When the value is less than 0.60, the system determines that the current weld has a crack or severe incomplete penetration. The system immediately triggers an emergency stop command, interrupting the welding current and wire feeding process. At the same time, the system uses the weld space coordinate reference to accurately locate and record the three-dimensional coordinates of the defect's starting position. The defect coordinate information is written to the system database log and fed back to the workshop control panel. The robot indicator light switches to alarm mode, and the process can only be manually reset after manual intervention to clean the defect or replace the workpiece.

[0116] Through the above-mentioned hierarchical control, the transformation from quality inspection to real-time control of welding quality has been realized. The welding process is kept stable when it is qualified, defects are eliminated through remelting when it is repairable, and the continued stacking of defective welds is terminated in time when it is irreparable, so as to prevent the defective welds from being covered by further stacking and thus increasing the cost of rework.

[0117] Example 2

[0118] In the existing welding process of sound barrier columns, uneven heating of long and large H-beam steel components easily leads to thermal deformation, causing dynamic changes in the bevel gap. Traditional fixed-parameter welding methods are difficult to adapt to, resulting in defects such as incomplete fusion, burn-through, and incomplete penetration between layers. Furthermore, current weld quality inspection methods mostly rely on post-weld sampling, depending on piezoelectric ultrasonic or radiographic testing. This disconnect between inspection and the welding process means that problems cannot be remedied online once they are discovered. Additionally, piezoelectric ultrasonic testing requires the use of coupling agents, and residues contaminate the column surface, affecting subsequent anti-corrosion spraying processes. To solve these problems, this invention provides an online inspection and automatic welding system for sound barrier column weld quality, the structure of which is as follows: Figure 4 As shown. The system includes a path planning module, a welding and acquisition module, a spatiotemporal mapping module, a fusion judgment module, and a closed-loop control module. Its specific implementation process is as follows:

[0119] Path planning module

[0120] This module obtains raw point cloud data by scanning the fillet weld groove with a line laser vision sensor. After filtering and background removal, it extracts the groove width, root gap and groove angle. Based on the groove geometric parameters, it calculates the filling cross-sectional area of ​​each section and generates the walking trajectory of the welding robot and the initial values ​​of welding current, voltage and walking speed according to the principle of balancing welding wire deposition and groove filling.

[0121] Welding and data acquisition module

[0122] The module clamps the sound barrier column into a rotary positioning fixture and adjusts the fillet weld to the flat welding position. Then, it performs multi-layer and multi-pass welding according to the planned path. During the welding process, the follow-up detection system is started simultaneously. The system collects the weld surface contour point cloud, molten pool temperature field, arc sound pattern and internal ultrasonic echo data through a line laser vision sensor, infrared radiation temperature sensor, acoustic sensing system and electromagnetic ultrasonic transducer, respectively. Before ultrasonic acquisition, a magnetostrictive coating is sprayed onto the weld surface and the integrity of the coverage is checked.

[0123] Spacetime mapping module

[0124] This module uses the synchronous trigger pulse generated by the main control unit as the reference clock source to record a timestamp for each frame of data from each sensor channel, keeping the relative time error within 1 millisecond. At the same time, based on the installation offset of each sensor relative to the center of the welding torch nozzle, a weld space coordinate system is established with the starting welding point as the origin. The data from each channel with timestamps are rearranged according to their corresponding weld space positions to form a spatially aligned multidimensional data matrix.

[0125] Fusion Judgment Module

[0126] This module extracts features such as weld reinforcement height, weld pool width, acoustic energy ratio, and ultrasonic echo amplitude ratio from four types of data: visual, thermal field, acoustic fingerprint, and ultrasonic. It assigns basic weights based on the detection sensitivity of each sensor channel for different defect types, and adjusts the weights by reducing the corresponding channel weights through a penalty factor when the real-time signal-to-noise ratio is below 12 dB. The normalized feature scores of each channel are fused with the corresponding weights to obtain a quality evaluation index, and the weld quality is divided into three levels: qualified, repairable defect, or unrepairable defect based on the threshold range of the index.

[0127] Closed-loop control module

[0128] This module performs graded control based on the assessed quality level: under qualified conditions, the welding current and travel speed are incrementally corrected according to the deviation of the weld pool width to maintain process stability; under repairable defect conditions, the current is increased to 1.1 to 1.25 times the initial value and the speed is reduced to 0.7 to 0.85 times to perform remelting, and a re-inspection is performed after remelting. If it is not qualified, remelting is repeated, with a maximum of 2 times. If the limit is exceeded, the machine is stopped; under unrepairable defect conditions, welding is immediately interrupted and the defect location coordinates are recorded.

[0129] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for online inspection and automatic welding of weld quality of sound barrier columns, characterized in that, include: Acquire the bevel feature data of the fillet weld, and generate the walking trajectory of the welding robot and the initial values ​​of the welding parameters based on the bevel feature data; Multi-layer, multi-pass welding is performed according to the walking trajectory and initial values ​​of welding parameters, and multi-source heterogeneous sensing data including visual morphology, thermal field distribution, acoustic signature features and internal defect ultrasound are acquired synchronously through a follow-up detection system. Based on the multi-source heterogeneous sensing data, a spatiotemporal synchronization mapping relationship is established for each sensing channel under a unified time reference and weld space coordinate reference. Feature extraction is performed based on the spatiotemporally aligned multi-source heterogeneous sensor data. Weight coefficients are dynamically assigned according to the detection sensitivity of each sensor channel for different defect types. A multi-dimensional defect judgment model is constructed to evaluate the quality level of the current weld. Based on the quality level, hierarchical closed-loop control is executed. If the current quality level meets the conditions for continuing welding, the welding parameters are adaptively adjusted and welding continues; otherwise, the welding process is interrupted and the defect location is marked.

2. The method for online inspection and automatic welding of weld quality of sound barrier column according to claim 1, characterized in that, The multi-layer, multi-pass welding specifically includes: Clamp the sound barrier column into the rotary positioning fixture and adjust the fillet weld to the flat weld position; During the welding process, the interpass temperature is monitored, and the welding speed is adjusted according to temperature fluctuations to keep the cooling rate of the molten pool within the preset process window. After each weld bead is completed, the welding parameters are updated based on the bevel condition of the next layer, and welding continues until all weld bead bead is completed.

3. The method for online inspection and automatic welding of weld quality of sound barrier column according to claim 1, characterized in that, The multi-source heterogeneous sensing data includes weld surface contour point cloud data obtained by line laser triangulation, temperature field data of the molten pool and its heat-affected zone obtained by infrared radiation thermometry, arc acoustic signature data obtained by acoustic sensing, and ultrasonic echo data inside the weld obtained by an electromagnetic ultrasonic transducer without the need for a coupling agent.

4. The method for online inspection and automatic welding of weld quality of sound barrier column according to claim 3, characterized in that, Before collecting the ultrasonic echo data inside the weld, a magnetostrictive coating is sprayed onto the weld surface, and the coverage integrity of the sprayed area is checked. If the check result does not meet the preset quality requirements, the spraying and check are repeated until the preset quality requirements are met.

5. The method for online inspection and automatic welding of weld quality of sound barrier column according to claim 1, characterized in that, The steps for achieving alignment under a unified time reference and weld space coordinate reference include: The same clock source is applied to all sensing channels to achieve time synchronization, and coordinate transformation is performed based on the spatial pose of each sensor relative to the welding torch, so that the data collected by each channel is associated with a unified weld space coordinate reference.

6. The method for online inspection and automatic welding of weld quality of sound barrier column according to claim 1, characterized in that, The feature extraction specifically includes: Geometric contour features, temperature field features, acoustic spectrum features, and ultrasonic echo features were extracted from visual morphology data, thermal field distribution data, acoustic signature feature data, and internal defect ultrasonic data, respectively.

7. The method for online inspection and automatic welding of weld quality of sound barrier column according to claim 1, characterized in that, The steps for dynamically allocating weight coefficients include: Based on the preset detection sensitivity of each sensor channel for different defect types, a basic weight is assigned, and the real-time signal-to-noise ratio of each sensor channel is evaluated simultaneously. The weight of each channel is adjusted by reducing the weight of the corresponding channel when the signal-to-noise ratio is lower than a preset threshold, and a fusion judgment vector is generated.

8. The method for online inspection and automatic welding of weld quality of sound barrier column according to claim 1, characterized in that, The steps for assessing the quality grade of the current weld bead include: The quality evaluation index is calculated by comparing the fusion results of each sensor channel with the preset judgment criteria. The weld quality is marked as qualified, repairable defect, or unrepairable defect based on the threshold range of the quality evaluation index.

9. The method for online inspection and automatic welding of weld quality of sound barrier column according to claim 1, characterized in that, The hierarchical closed-loop control specifically includes: When the quality level is repairable, the correction amount of welding parameters is calculated according to the defect type and severity, and online remelting is performed by adjusting the welding current and walking speed. After the remelting is completed, the follow-up detection system is called to re-inspect and verify whether the defect has been eliminated. If the defect has not been completely eliminated, the remelting is repeated. When the quality level is an unrepairable defect, the welding process is interrupted and the defect location is marked.

10. An online inspection and automatic welding system for the weld quality of sound barrier columns, characterized in that, The system for the online weld quality inspection and automatic welding method according to any one of claims 1-9, the system comprising: The path planning module is used to acquire fillet weld groove feature data and generate the walking trajectory of the welding robot and the initial values ​​of welding parameters based on the groove feature data. The welding and acquisition module is used to perform multi-layer and multi-pass welding according to the walking trajectory and the initial value of welding parameters, and to synchronously acquire multi-source heterogeneous sensor data including visual morphology, thermal field distribution, acoustic features and internal defect ultrasound through the follow-up detection system. The spatiotemporal mapping module is used to establish a spatiotemporal synchronization mapping relationship between each sensing channel under a unified time reference and weld space coordinate reference based on the multi-source heterogeneous sensing data. The fusion judgment module is used to extract features based on the spatiotemporally aligned multi-source heterogeneous sensor data, dynamically allocate weight coefficients according to the detection sensitivity of each sensor channel for different defect types, construct a multi-dimensional defect judgment model, and evaluate the quality level of the current weld. The closed-loop control module is used to perform graded closed-loop control based on the quality level. If the current quality level meets the conditions for continuing welding, the welding parameters are adaptively adjusted and welding continues; otherwise, the welding process is interrupted and the defect location is marked.

Images