All-welded heat exchanger industrial detection method and system based on machine vision

Abstract

The invention relates to the technical field of machine vision industrial detection, and discloses an all-welded heat exchanger industrial detection method and system based on machine vision, and the method comprises the steps: obtaining a welding line region image of a to-be-detected heat exchanger, and extracting a boundary contour of a welding line region through an image preprocessing algorithm; calculating a welding line center line according to the boundary contour of the welding line area, and dividing the welding line center line into a plurality of detection units with equal length; in each detection unit, linear light scanning data of the surface of the welding seam and ultrasonic echo signals of a welding seam area are collected; generating a three-dimensional shape model of the surface of the welding seam according to the linear light scanning data, and extracting acoustic characteristic parameters in the welding seam; preprocessing the three-dimensional shape model and the acoustic characteristic parameters, and inputting the preprocessed three-dimensional shape model and acoustic characteristic parameters into a pre-trained defect evaluation model to obtain a comprehensive defect index of the welding seam area; and according to a comparison result of the comprehensive defect index and a specified threshold group, judging whether obvious defects exist in the welding seam area, and completing comprehensive evaluation of the welding seam quality.

Description

Technical Field

[0001] The present application relates to the technical field of machine vision industrial inspection, and in particular to a machine vision-based industrial inspection method and system for fully welded heat exchangers. Background Art

[0002] In modern industrial equipment, fully welded plate heat exchangers are widely used due to their efficient heat transfer performance and compact structure. However, their complex weld structure and small gaps between plates make quality inspection a major challenge.

[0003] Traditional manual inspection methods rely primarily on the senses of experienced inspectors and simple tools. These inspectors use powerful flashlights, magnifying glasses, and even endoscopes (for hard-to-see locations) to observe the weld surface, checking for surface cracks, undercuts, weld beads / spatter pores (visible on the surface), lack of fusion / insufficient penetration (surface signs), significant depressions / insufficient height, and severe surface contamination. While these methods offer low initial investment costs, they exhibit low efficiency and a high rate of missed detections in mass production, making inspection consistency difficult to ensure.

[0004] Existing machine vision solutions mostly focus on two-dimensional image analysis. Although they can identify some surface defects, the surface reflection characteristics of the weld area are complex, and the internal structure is invisible. Therefore, there are certain limitations in the quantitative evaluation of three-dimensional weld morphological defects (such as edge depth and weld bead height) and internal leakage.

[0005] Therefore, there is an urgent need to improve the coverage and accuracy of defect identification to meet the needs of industrial production for high-quality inspection. Summary of the Invention

[0006] In order to improve the quality inspection efficiency and accuracy of all-welded heat exchangers, the present application provides an industrial inspection method and system for all-welded heat exchangers based on machine vision.

[0007] In a first aspect, the present application provides a machine vision-based industrial inspection method for all-welded heat exchangers, which employs the following technical solutions: A machine vision-based industrial inspection method for fully welded heat exchangers, comprising: S1. Obtain an image of the weld area of ​​the heat exchanger to be inspected, and extract the boundary contour of the weld area using an image preprocessing algorithm; S2. Calculate the weld centerline according to the boundary contour of the weld area, and divide the weld centerline into a plurality of detection units of equal length; S3. In each detection unit, linear optical scanning data of the weld surface and ultrasonic echo signals of the weld area are collected respectively; S4, performing three-dimensional reconstruction based on the linear light scanning data to generate a three-dimensional topography model of the weld surface; S5. Perform spectrum analysis on the ultrasonic echo signal to extract acoustic characteristic parameters inside the weld; S6. Preprocessing the three-dimensional morphology model and the acoustic characteristic parameters and inputting them into a pre-trained defect assessment model to obtain a comprehensive defect index of the weld area; S7. Based on the comparison result of the comprehensive defect index and the specified threshold value group, determine whether there are obvious defects in the weld area.

[0008] By adopting the above technical solution, the weld surface is first reconstructed in three dimensions using linear light scanning data to obtain three-dimensional morphological information of the weld surface, so that surface defects such as edge depth and weld bead height can be quantified; at the same time, the acoustic characteristic parameters inside the weld are extracted through spectral analysis of the ultrasonic echo signal, which can be used to evaluate potential defects such as cracks or voids inside the weld; the three-dimensional morphological model and acoustic characteristic parameters are then combined and input into a pre-trained defect assessment model to form a comprehensive defect index, completing a comprehensive assessment of the weld quality. This not only covers the multi-dimensional detection needs of surface and internal defects, but also significantly improves detection efficiency and accuracy.

[0009] Optionally, the method further comprises the following steps: Obtain comprehensive defect indices of multiple recent weld areas to form a comprehensive defect index sequence; Fit the comprehensive defect index sequence into a comprehensive defect change curve, and calculate the growth trend of the comprehensive defect change curve; The ultrasonic emission frequency is adjusted according to the growth trend value. The larger the growth trend value is, the lower the ultrasonic emission frequency is; the smaller the growth trend value is, the higher the ultrasonic emission frequency is.

[0010] By employing this technical solution, the changing trend of the comprehensive defect index reflects the distribution and evolution of weld defects. Dynamically adjusting the ultrasonic transmission frequency based on this changing trend can optimize the balance between acoustic penetration depth and resolution.

[0011] Optionally, the following steps are also included: SZ1: For a test unit with obvious defects, divide it into at least three equal parts along the direction of the weld center to obtain the first subdivision unit, the second subdivision unit, and the third subdivision unit with a mutual overlap of less than a specified percentage; SZ2. Obtain the comprehensive defect index corresponding to each subdivision unit and select the subdivision unit corresponding to the highest comprehensive defect index; Determine whether the subdivision unit is centered. If not, reselect the detection unit with the subdivision unit as the center and return to step SZ1. Otherwise, go to step SZ3. SZ3, obtaining the comprehensive defect index and unit area corresponding to the first non-overlapping unit, the first overlapping subdivision unit, the second non-overlapping unit, the second overlapping subdivision unit, and the third non-overlapping unit contained in the current detection unit; SZ4: Perform a pairwise, non-repetitive comparison of the comprehensive defect indexes of non-overlapping units and overlapping units, and accumulate abnormal values ​​based on the comparison results; SZ5: If the two cells with the top two outlier values ​​are located at the two ends of the detection cell at the beginning of this round, these two cells are set as the center to form the key exploration detection cell and then proceed to step SZ3; otherwise, these two cells are used as the two ends of the next round of detection cells and then proceed to step SZ3; SZ6. When the area of ​​the inspection unit is less than or equal to the specified area threshold, it is marked as a potential defect and the loop ends.

[0012] By adopting the above technical solution, through the gradual subdivision of the detection units and the cumulative calculation of outliers, the position of potential defects can be located relatively more accurately. This method is particularly suitable for the identification of local defects in complex weld structures, avoiding the problem of missed detection caused by the traditional detection method due to the large detection range.

[0013] Optionally, the process of accumulating abnormal values ​​according to the comparison situation includes: Configuring an outlier cumulative coefficient and a cumulative base value of the current round, where the outlier value is the product of the cumulative coefficient and the cumulative base value; the outlier cumulative coefficient is inversely proportional to the unit area, and the cumulative base value is related to whether it is an overlapping area; In the pairwise non-repeated comparison, the one with the larger comprehensive defect index is accumulated as an outlier, while the one with the smaller comprehensive defect index is not accumulated as an outlier.

[0014] By adopting the above technical solution, the calculation method of outliers combines the influence of unit area and detection rounds, making the distribution of outliers more consistent with the spatial characteristics of actual defects and the dynamic changes of the detection process, which can effectively improve the accuracy of potential defect identification.

[0015] Optionally, the following steps are also included: For weld areas marked as potentially defective, a focused heat source is used to apply thermal stimulation perpendicular to the weld centerline. Before and after the thermal excitation is applied, a time series of infrared thermal images is acquired; Extract infrared spatiotemporal feature parameters of infrared thermal image time series; The infrared spatiotemporal characteristic parameters are pre-processed into an infrared parameter matrix and then input into a pre-trained infrared defect assessment model; The authenticity of the potential defect is determined according to the output of the infrared defect assessment model.

[0016] By adopting the above technical solution, the thermal excitation stimulated by the focused heat source can make the potential defects in the weld area appear in the infrared thermal image, thereby achieving further verification of the authenticity of the defects, which is suitable for the confirmation of minor defects or hidden defects.

[0017] Optionally, the following steps are also included: Obtain the spatial position order of all potential defects; Based on the spatial position sequence, the intermediate position between any two adjacent potential defect positions in space is calculated to form an intermediate position sequence; Calculate the degree of dispersion of the intermediate position sequence; If the degree of dispersion is less than the preset discrete threshold, the average value of all the middle positions in the middle position sequence is calculated as the re-inspection recommended point.

[0018] By adopting the above technical solutions, statistical analysis of the spatial location of potential defects can determine the key areas for re-inspection, thereby reducing unnecessary repeated inspection workload and improving overall inspection efficiency.

[0019] Optionally, the ultrasonic emission frequency uses different first frequencies and second frequencies for detection respectively: First, a preset first frequency is used to collect a first ultrasonic echo signal in the weld area; Then, a second ultrasonic echo signal from the weld area is collected using a preset second frequency; extracting first acoustic sub-feature parameter sets based on the first ultrasonic echo signals respectively; extracting second acoustic sub-feature parameter sets based on the second ultrasonic echo signals respectively; In the step of calculating the comprehensive defect index of the weld area, both the first acoustic sub-feature parameter set and the second acoustic sub-feature parameter set are input into the pre-trained defect assessment model.

[0020] By employing this technical solution, ultrasonic signals of different frequencies respond differently to internal weld defects. Therefore, multi-frequency testing can provide more comprehensive acoustic signature information. By inputting these characteristic parameter sets into the defect assessment model, the reliability of the comprehensive defect index can be further improved.

[0021] Optionally, the following steps are also included: If it is determined based on the first acoustic sub-feature parameter set and / or the second acoustic sub-feature parameter set that there are suspected features of a specific type of defect, and the confidence level is lower than a preset threshold, a third ultrasonic echo signal is collected from the weld area using a preset third frequency; extracting a third acoustic sub-feature parameter set based on the third ultrasonic echo signal; In the step of calculating the comprehensive defect index of the weld area, the third acoustic sub-feature parameter set is also input into the pre-trained defect assessment model.

[0022] By adopting the above technical solution, when there is uncertainty in the initial detection results, the third frequency ultrasonic signal is introduced for supplementary detection, which can further improve the confidence and accuracy of defect identification.

[0023] In a second aspect, the present application provides a machine vision-based industrial inspection system for all-welded heat exchangers, which adopts the following technical solutions: A machine vision-based industrial inspection system for all-welded heat exchangers includes a processor that executes a program of any one of the above-mentioned machine vision-based industrial inspection methods for all-welded heat exchangers.

[0024] Optionally, it also includes an image acquisition module, an ultrasonic detection module, an infrared thermal imaging module, a linear light scanning module, a defect assessment model, and an infrared defect assessment model. The modules work together through signal transmission and data processing to complete comprehensive detection of the weld area.

[0025] In summary, this application includes at least one of the following beneficial technical effects: This application uses linear light scanning data to perform three-dimensional reconstruction of the weld surface to obtain three-dimensional morphological information of the weld surface, so as to quantify surface defects such as edge depth and weld bead height; at the same time, the acoustic characteristic parameters inside the weld are extracted through spectral analysis of the ultrasonic echo signal, which can be used to evaluate potential defects such as cracks or voids inside the weld; the three-dimensional morphological model and the acoustic characteristic parameters are then combined and input into a pre-trained defect assessment model to form a comprehensive defect index, completing a comprehensive assessment of the weld quality. This not only covers the multi-dimensional detection needs of surface and internal defects, but also significantly improves the detection efficiency and accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 This is a module connection diagram of the fully welded heat exchanger industrial detection system of this application.

[0027] Figure 2 It is a schematic diagram of the processing of the detection unit in the industrial detection method of the fully welded heat exchanger of this application. DETAILED DESCRIPTION

[0028] Embodiments of the present application are described in detail below, examples of which are illustrated in the accompanying drawings.

[0029] Throughout this specification, reference to the terms "certain embodiments," "one embodiment," "some embodiments," "illustrative embodiments," "examples," "specific examples," or "some examples" means that a specific feature, structure, material, or characteristic described in conjunction with the embodiment or example is included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.

[0030] The embodiment of the present application discloses a method and system for industrial detection of fully welded heat exchangers based on machine vision. Figure 1 For detailed description. Figure 1 As shown in Figure 1, the system includes a processor, an image acquisition module, a linear light scanning module, an ultrasonic detection module, an infrared thermal excitation module, and a defect assessment model. These modules work together through physical connections and data transmission to complete a comprehensive inspection of the weld area of ​​a fully welded plate heat exchanger.

[0031] During actual operation, the image acquisition module first captures an image of the weld seam of the core or plate bundle of a fully welded plate heat exchanger to be inspected. The image acquisition module typically consists of an industrial camera and a light source assembly. The industrial camera is mounted above the weld seam and perpendicular to the weld surface, while the light source assemblies are placed on both sides of the weld to ensure uniform illumination.

[0032] The image acquisition module transmits the captured raw image data to the processor, which uses a built-in image preprocessing algorithm to extract the weld area's boundary contours. This contour extraction process uses grayscale threshold segmentation and edge detection techniques, analyzing the pixel differences between the weld and background areas to ultimately generate precise weld boundary information. Based on this information, the processor further calculates the weld centerline and divides it into multiple equal-length detection units. The length of each unit can be pre-defined based on the actual weld size and detection accuracy requirements.

[0033] Next, the linear light scanning module and the ultrasonic detection module respectively collect data on the weld area within each detection unit.

[0034] The linear light scanning module utilizes a camera + line laser + motion mechanism / galvanometer configuration. The motion mechanism can be equipped with a stable linear motor. A single laser line sweep acquires the height information of a cross section. Combined with a motion platform or galvanometer, it can quickly scan long welds, making it ideal for automated online inspection. The module captures the light signal reflected from the weld surface and transmits the data to a processor, which uses the received linear light scanning data for 3D reconstruction, generating a 3D topographic model of the weld surface. This 3D topographic model is constructed using the principle of triangulation. By analyzing the intensity and angle of reflected light at different points on the weld surface, the weld surface's height variations and geometric characteristics (such as the standard deviation of height differences, maximum indentation depth, local curvature entropy, and waviness index) are calculated.

[0035] At the same time, the ultrasonic detection module transmits ultrasonic signals through the ultrasonic probe and receives echo signals from the weld area. The echo signals undergo analog-to-digital conversion and are then transmitted to the processor. The processor performs spectral analysis on the received ultrasonic echo signals to extract the acoustic characteristic parameters within the weld. These acoustic characteristic parameters include the echo signal's frequency content, amplitude distribution, and phase variation, which are used to assess potential defects within the weld.

[0036] After completing data collection, the processor inputs the generated three-dimensional morphology model of the weld surface and the acoustic characteristic parameters inside the weld into a pre-trained defect assessment model; this defect assessment model is pre-trained based on the existing dual-stream convolutional neural network (Dual-Stream DNN) architecture and embedded in the processor. It performs a comprehensive analysis of the input data and outputs a comprehensive defect index for the weld area.

[0037] In the process of obtaining the comprehensive defect index, it is necessary to perform feature preprocessing on geometric features such as height difference and curvature change in the three-dimensional morphology model and acoustic features such as frequency offset and amplitude anomaly in the acoustic feature parameters to achieve geometric feature normalization (which can be based on the plate thickness) and acoustic feature normalization (which can be based on the defect-free area).

[0038] Afterwards, the defect level of the weld area is determined based on the size of the comprehensive defect index. If the comprehensive defect index is less than the first threshold, it is determined that the weld area has no obvious defects. If the comprehensive defect index is between the first threshold and the second threshold, it is determined that there is a slight defect in the weld area; If the comprehensive defect index is between the second threshold and the third threshold, it is determined that there is a moderate defect in the weld area; If the comprehensive defect index is greater than the third threshold, it is determined that there are serious defects in the weld area.

[0039] In terms of multi-frequency detection, the ultrasonic detection module uses different first and second frequencies for separate detection. The processor controls the ultrasonic probe to first use a preset first frequency to collect the first ultrasonic echo signal from the weld area, and then use a preset second frequency to collect the second ultrasonic echo signal from the weld area. The echo signals of the two frequencies are subjected to spectral analysis to extract the first acoustic sub-feature parameter set and the second acoustic sub-feature parameter set, and these parameter sets are input into the defect assessment model to further improve the reliability of the comprehensive defect index. If there is uncertainty in the preliminary detection results and the confidence level is lower than the preset threshold, the processor controls the ultrasonic probe to use a preset third frequency to collect the third ultrasonic echo signal from the weld area and extract the third acoustic sub-feature parameter set as supplementary data.

[0040] Among them, the first frequency, the second frequency and the third frequency can be selected within the low frequency range of 20-100kHz, which can penetrate thick plates, and the high frequency can be selected within the range of 4.6-5.2MHz, which can catch surface micro cracks.

[0041] In order to further optimize the detection effect, the processor can also be used to perform statistical analysis on the comprehensive defect index of multiple continuous weld areas to form a comprehensive defect index sequence. The comprehensive defect index sequence is a comprehensive defect change curve generated by a fitting algorithm. The processor dynamically adjusts the ultrasonic emission frequency according to the growth trend of the curve.

[0042] The consensus in the welding field is that it is better to accept visible shallow defects than to ignore hidden deep defects. However, the core or plate bundle targeted in the application is an important component of the fully welded plate heat exchanger. Because the heat exchanger may often operate in an environment with corrosive media, surface pores caused by shallow defects may become the starting point of stress corrosion cracking. Moreover, such welds are usually thin (4-12mm). In this case, shallow undercut is more dangerous than deep slag inclusions because it directly reduces the pressure-bearing wall thickness. If such defects can be detected with high coverage during the manufacturing stage, the scrap cost can be far lower than after-sales failure.

[0043] Therefore, in this embodiment, the weight of shallow defects on the comprehensive defect index is designed to be higher than that of deep defects. When the growth trend value is large, it means that shallow defects dominate and surface defects are possible and obvious. At this time, the ultrasonic emission frequency can be reduced to enhance the detection capability of deep defects and improve the rationality of the comprehensive defect index; when the growth trend value is small, it means that shallow defects are not obvious. At this time, it is necessary to increase the ultrasonic emission frequency to further improve the resolution of shallow defects; for deep defects, low-frequency ultrasonic waves can be used for subsequent inspection through random sampling.

[0044] Furthermore, to further pinpoint the defect location, inspection cells containing at least minor defects are divided along the weld centerline into three sub-cells with less than 50% overlap. The comprehensive defect index is recalculated for each sub-cell. The processor determines whether to reselect an inspection cell based on the location of the sub-cell with the highest comprehensive defect index. The processor then compares the comprehensive defect indices of non-overlapping and overlapping cells to calculate the cumulative outlier value, pinpointing the location of potential defects.

[0045] In this embodiment, if Figure 2 As shown, taking a long rectangular inspection unit as an example, it can be divided into three equal parts: the first subdivision unit K1, the second subdivision unit K2, and the third subdivision unit K3, with a pairwise overlap of 45%. Then, the comprehensive defect index corresponding to each of the three subdivision units is obtained, such as KC1, KC2, and KC3.

[0046] If KC1 or KC3 is the largest, a new rectangular detection unit is selected with the first subdivision unit K1 or the third subdivision unit K3 as the center and then divided into three equal parts; If KC2 is the largest, it indicates that the probability of defects being concentrated in the middle of the inspection unit is relatively high. Based on the overlap, the current inspection unit can be further divided from left to right into the first non-overlapping unit Q1, the first overlapping subdivision unit A1, the second non-overlapping unit Q2, the second overlapping subdivision unit A2, and the third non-overlapping unit Q3. The first-round comprehensive defect index of each overlapping or non-overlapping unit can be obtained. Taking the first non-overlapping unit Q1 and the first overlapping subdivision unit A1 as an example, QC1 represents the comprehensive defect index of the first non-overlapping unit, AC1 represents the comprehensive defect index of the first overlapping unit, SQ1 represents the unit area of ​​the first non-overlapping unit, SA1 represents the unit area of ​​the first overlapping unit, and the same applies to the others.

[0047] Perform a pairwise, non-repetitive comparison of the comprehensive defect indexes of non-overlapping units and overlapping units. If QC1 is less than AC1, AC1 accumulates an outlier value once. If QC2 is greater than AC2, QC2 accumulates an outlier value once. After obtaining the final outlier value accumulation result, sort the outlier value sequence in descending order and obtain the top two units. If the first two units are Q1 and Q3 (rarely in actual situations), two new non-overlapping detection units are obtained with Q1 and Q3 as the centers respectively; If the first two units are A1 and A2 (which is often the case in practice), A1 and A2 are used as the two ends of the next round of inspection units, and then the next round of subdivision begins. If the area of ​​the inspection unit is less than or equal to the specified area threshold, it is marked as a potential defect and the cycle ends. The specified area threshold can be determined based on the historical defect size.

[0048] In the process of accumulating abnormal values ​​according to the comparison results, first configure the abnormal value accumulation coefficient and the cumulative base value of the current round. Abnormal value = accumulation coefficient * cumulative base value; Because in the training process of the defect assessment model, samples with smaller areas are easier to obtain and more in number, and if defect features exist, they are easier to capture. However, welding defects are mostly continuous defects with a small unit area, but the comprehensive defect index can show that the defect features are more obvious. Therefore, the outlier cumulative coefficient is designed to be inversely proportional to the unit area. Since the overlapping area is located within K2, the cumulative basic value corresponding to the overlapping area is larger; in the non-repetitive comparison between the two, the one with a larger comprehensive defect index accumulates an outlier value once, and the one with a smaller comprehensive defect index does not accumulate an outlier value.

[0049] For easier understanding, let's take part of the first round of comparison as an example: If QC1 is less than AC1, and SQ1 (55%) is greater than SA1 (45%), then the accumulated abnormal value of A1 is 0.8*1=0.8, and Q1 is not accumulated; If QC2 is greater than AC2, and SQ2 (55%) is greater than SA2, then the accumulated abnormal value of Q2 this time is 1*0.9=0.9, and A2 is not accumulated.

[0050] In addition, before the start of this round of abnormal value accumulation, the variance of the comprehensive defect index series of non-overlapping units and overlapping units can be calculated. If the variance value is less than the specified variance value, the overlap degree can be reduced until the variance value is greater than or equal to the specified variance value.

[0051] The infrared thermal excitation module includes a heat source generator and an infrared thermal imager. For weld areas marked for potential defects, the heat source generator uses a laser light source or a high-power density heat gun, perpendicular to the weld centerline, to apply short, high-energy-density pulsed thermal excitation directly above the potential defect area. A medium-wave or long-wave infrared thermal imager continuously captures infrared thermal image sequences before and after the application of the thermal excitation. The energy density and pulse width ranges for the pulsed thermal excitation are determined by the material type and thickness range of the heat exchanger plate being inspected, as well as the expected maximum defect depth. These ranges are determined through thermal conductivity simulation and experimental calibration to ensure that the excitation energy effectively reaches the expected defect depth and produces a detectable temperature rise, while avoiding surface damage.

[0052] It can start from 1 second before the thermal excitation is applied and end 5 seconds after the thermal excitation ends. During this process, an infrared thermal imager is used to continuously collect a time series of infrared thermal images of the potential defect micro-area and the surrounding appropriately extended area (the extended range at least covers the outside of the micro-area boundary) at a sampling frequency of not less than 50 Hz.

[0053] Then, the infrared thermal image time series is processed to extract the infrared spatiotemporal feature parameter set that characterizes the abnormal thermal response of the potential defect micro-area, including but not limited to: The maximum temperature rise in the micro-region (relative to the baseline temperature before excitation); Curve of variation of micro-area average temperature with time t; The maximum heating rate during the micro-area average temperature rise stage; The phase delay of the average temperature peak of the micro-region relative to the average temperature peak of the defect-free reference region (selected near the micro-region); The average temperature difference between the micro-area and its adjacent defect-free reference ring (the inner diameter of the ring is equal to the diameter of the circumscribed circle of the micro-area, and the outer diameter is set according to thermal diffusion) at time t; At a specific moment (such as the peak moment or the end of sampling), the pixel area in the micro area whose temperature exceeds the set threshold; The average rate of decrease of the average temperature of a micro-region during a specific period of time during the cooling stage.

[0054] Afterwards, the extracted infrared spatiotemporal feature parameter set is input into a pre-trained infrared defect assessment model; the infrared defect assessment model can output the confidence level of the defect existence. When the confidence level is greater than 90%, it can be confirmed that a defect actually exists in the micro-area and the infrared assessment result is recorded. Otherwise, further verification is required; the infrared defect assessment model is trained based on the existing lightweight convolutional LSTM network.

[0055] Finally, the processor performs a statistical analysis on the spatial position sequence of all potential defects. Because according to the above precautions, for each inspection unit, considering the quality inspection cost, only one real defect micro-area is obtained in the end, and welding defects are often continuous. Therefore, in this scheme, the middle position between any two adjacent defect positions in space can be calculated to form an middle position sequence. Then, according to the degree of discreteness of the middle position sequence, it is determined whether it is necessary to set a re-inspection recommended point; if the degree of discreteness is less than the preset discrete threshold, the processor calculates the average value of all middle positions in the middle position sequence as the re-inspection recommended point, thereby reducing unnecessary repeated inspection workload and improving overall inspection efficiency.

[0056] In summary, this invention, through the integration of a multi-dimensional collaborative inspection method, achieves efficient and accurate inspection of the weld area of ​​fully welded plate heat exchangers. The coordinated operation of these modules ensures an automated and intelligent inspection process, significantly improving inspection efficiency and accuracy and providing strong technical support for industrial production.

[0057] Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limitations on the present application. Ordinary technicians in this field can change, modify, replace and modify the above embodiments within the scope of the present application.

Claims

1. A machine vision-based industrial inspection method for fully welded heat exchangers, characterized in that: include: S1. Obtain an image of the weld area of ​​the heat exchanger to be inspected, and extract the boundary contour of the weld area using an image preprocessing algorithm; S2. Calculate the weld centerline according to the boundary contour of the weld area, and divide the weld centerline into a plurality of detection units of equal length; S3. In each detection unit, linear optical scanning data of the weld surface and ultrasonic echo signals of the weld area are collected respectively; S4, performing three-dimensional reconstruction based on the linear light scanning data to generate a three-dimensional topography model of the weld surface; S5. Perform spectrum analysis on the ultrasonic echo signal to extract acoustic characteristic parameters inside the weld; S6. Preprocessing the three-dimensional morphology model and the acoustic characteristic parameters and inputting them into a pre-trained defect assessment model to obtain a comprehensive defect index of the weld area; S7. Based on the comparison result of the comprehensive defect index and the specified threshold value group, determine whether there are obvious defects in the weld area.

2. The machine vision-based industrial inspection method for all-welded heat exchangers according to claim 1 is characterized in that: The method further comprises the steps of: Obtain comprehensive defect indices of multiple recent weld areas to form a comprehensive defect index sequence; Fit the comprehensive defect index sequence into a comprehensive defect change curve, and calculate the growth trend of the comprehensive defect change curve; The ultrasonic emission frequency is adjusted according to the growth trend value. The larger the growth trend value is, the lower the ultrasonic emission frequency is; the smaller the growth trend value is, the higher the ultrasonic emission frequency is.

3. The machine vision-based industrial inspection method for all-welded heat exchangers according to claim 1, characterized in that: The following steps are also included: SZ1: For a test unit with obvious defects, divide it into at least three equal parts along the direction of the weld center to obtain the first subdivision unit, the second subdivision unit, and the third subdivision unit with a mutual overlap of less than a specified percentage; SZ2. Obtain the comprehensive defect index corresponding to each subdivision unit and select the subdivision unit corresponding to the highest comprehensive defect index; Determine whether the subdivision unit is centered. If not, reselect the detection unit with the subdivision unit as the center and return to step SZ1. Otherwise, go to step SZ3. SZ3, obtaining the comprehensive defect index and unit area corresponding to the first non-overlapping unit, the first overlapping subdivision unit, the second non-overlapping unit, the second overlapping subdivision unit, and the third non-overlapping unit contained in the current detection unit; SZ4: Perform a pairwise, non-repetitive comparison of the comprehensive defect indexes of non-overlapping units and overlapping units, and accumulate abnormal values ​​based on the comparison results; SZ5: If the two cells with the top two outlier values ​​are located at the two ends of the detection cell at the beginning of this round, these two cells are set as the center to form the key exploration detection cell and then proceed to step SZ3; otherwise, these two cells are used as the two ends of the next round of detection cells and then proceed to step SZ3; SZ6. When the area of ​​the inspection unit is less than or equal to the specified area threshold, it is marked as a potential defect and the loop ends.

4. The machine vision-based industrial inspection method for all-welded heat exchangers according to claim 3 is characterized in that: The process of accumulating abnormal values ​​according to the comparison situation includes: Configuring an outlier cumulative coefficient and a cumulative base value of the current round, where the outlier value is the product of the cumulative coefficient and the cumulative base value; the outlier cumulative coefficient is inversely proportional to the unit area, and the cumulative base value is related to whether it is an overlapping area; In the pairwise non-repeated comparison, the one with the larger comprehensive defect index is accumulated as an outlier, while the one with the smaller comprehensive defect index is not accumulated as an outlier.

5. The machine vision-based industrial inspection method for all-welded heat exchangers according to claim 3 is characterized in that: The following steps are also included: For weld areas marked as potentially defective, a focused heat source is used to apply thermal stimulation perpendicular to the weld centerline. Before and after the thermal excitation is applied, a time series of infrared thermal images is acquired; Extract infrared spatiotemporal feature parameters of infrared thermal image time series; The infrared spatiotemporal characteristic parameters are pre-processed into an infrared parameter matrix and then input into a pre-trained infrared defect assessment model; The authenticity of the potential defect is determined according to the output of the infrared defect assessment model.

6. The machine vision-based industrial inspection method for all-welded heat exchangers according to claim 5, characterized in that: The following steps are also included: Obtain the spatial position order of all potential defects; Based on the spatial position sequence, the intermediate position between any two adjacent potential defect positions in space is calculated to form an intermediate position sequence; Calculate the degree of dispersion of the intermediate position sequence; If the degree of dispersion is less than the preset discrete threshold, the average value of all the middle positions in the middle position sequence is calculated as the re-inspection recommended point.

7. The machine vision-based industrial inspection method for all-welded heat exchangers according to claim 1, characterized in that: Ultrasonic emission frequencies use different first and second frequencies for detection: First, a preset first frequency is used to collect a first ultrasonic echo signal in the weld area; Then, a second ultrasonic echo signal from the weld area is collected using a preset second frequency; extracting first acoustic sub-feature parameter sets based on the first ultrasonic echo signals respectively; extracting second acoustic sub-feature parameter sets based on the second ultrasonic echo signals respectively; In the step of calculating the comprehensive defect index of the weld area, both the first acoustic sub-feature parameter set and the second acoustic sub-feature parameter set are input into the pre-trained defect assessment model.

8. The machine vision-based industrial inspection method for all-welded heat exchangers according to claim 7, characterized in that: The following steps are also included: If it is determined based on the first acoustic sub-feature parameter set and / or the second acoustic sub-feature parameter set that there are suspected features of a specific type of defect, and the confidence level is lower than a preset threshold, a third ultrasonic echo signal is collected from the weld area using a preset third frequency; extracting a third acoustic sub-feature parameter set based on the third ultrasonic echo signal; In the step of calculating the comprehensive defect index of the weld area, the third acoustic sub-feature parameter set is also input into the pre-trained defect assessment model.

9. A fully welded heat exchanger industrial inspection system based on machine vision, characterized in that: The method comprises a processor, wherein the processor executes the steps of the all-welded heat exchanger industrial detection method based on machine vision according to any one of claims 1 to 8.

10. The machine vision-based all-welded heat exchanger industrial inspection system according to claim 9, characterized in that: It also includes an image acquisition module, an ultrasonic detection module, an infrared thermal imaging module, a linear light scanning module, a defect assessment model, and an infrared defect assessment model. The modules work together through signal transmission and data processing to complete comprehensive inspection of the weld area.

Images