An air outlet leakage and reverse installation detection method based on image detection

Abstract

The application discloses an air outlet leakage and reverse installation prevention detection method based on image detection and belongs to the technical field of optical detection. The application comprises determining the data collection range of image detection in the air outlet area and demarcating the detection area according to the structural parameters of the air outlet. The application divides the conventional and error-prone areas, adopts a differentiated detection strategy and combines three-dimensional point cloud comparison technology to effectively control the unqualified product rate and detection time of the air outlet installation, reduce the labor cost and the repair rate, dynamically optimize the threshold based on historical data, improve the identification accuracy of complex areas, solve the misjudgment of difficult areas, realize the double improvement of detection accuracy and efficiency, and realize the transformation from passive detection to active early warning through a three-level risk signal mechanism, shorten the risk response time, guarantee the product quality stability, reduce the downtime, improve the overall efficiency of the production line, and realize the collaborative optimization of risk control and production continuity.

Description

Technical Field

[0001] This invention relates to the field of optical inspection technology, specifically to an image-based method for detecting leaks and reverse installations at air outlets. Background Technology

[0002] In the production and assembly of equipment such as air conditioners and automotive air conditioners, the air outlet is a core component, and its installation quality directly affects the air delivery efficiency, sealing performance, and service life of the equipment. Traditional inspection methods rely on manual visual inspection, judging the installation status by observing whether the buckles are tight and whether the directional markings are aligned, but there are significant limitations: First, it is inefficient, with each unit taking about 30 seconds to inspect, which is difficult to meet the needs of assembly line production with a cycle time of 10 seconds per unit; second, it lacks precision, as the human eye cannot identify positional deviations or differences in surface fit at the level of 0.1mm, resulting in about 5% of defective products entering the market.

[0003] With the application of image inspection technology, some companies have introduced industrial cameras to achieve automated inspection. However, existing technologies still have many problems: First, they lack versatility, as the inspection parameters are tied to specific models of air outlets. Changing models requires reprogramming, which can take up to two hours, making them unsuitable for multi-variety, small-batch production models. Second, the inspection logic is rigid, failing to distinguish between regular and error-prone areas, resulting in insufficient accuracy in recognizing complex areas such as curved transitions and a high false positive rate. Third, risk management is lagging, only able to determine "qualified / unqualified," unable to predict risk levels based on historical fault data, leading to high rework costs. Therefore, there is an urgent need for an air outlet inspection method that can balance accuracy, efficiency, and versatility.

[0004] To address the aforementioned technical shortcomings, a solution is proposed. Summary of the Invention

[0005] The purpose of this invention is to provide an image detection-based method for detecting leaks and reverse installations at air outlets, in order to solve the problems mentioned above.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for detecting leaks and reverse installations at air outlets based on image detection, comprising the following steps;

[0007] S1. Determine the data acquisition range of the image detection in the air outlet area. Collect several image data of the air outlet installation status using an industrial camera under preset lighting conditions, and mark them as the original image data of the air outlet.

[0008] S2. Obtain historical data and raw image data for joint processing, divide the area into a set of regular installation areas and a set of error-prone installation areas, and use the historical data as a reference source for the threshold of subsequent detection;

[0009] S3. Analyze error-prone installation areas, perform noise reduction and edge enhancement on the original images of error-prone installation areas, extract stable and unstable features, and obtain feature recognition region sets and complex region sets.

[0010] S4. Obtain the scanning model data of the standard installation of the air outlet, compare the feature identification area set and complex area set with the model data one by one, and generate the appropriate area signal and the inappropriate area signal.

[0011] S5. Retrieve comparison data of the unsuitable area, combine historical failure case analysis with actual feature point cloud items, model point cloud items and deviation value items, and generate corresponding risk level signals according to the risk level.

[0012] Furthermore, the process of determining the image detection range and acquiring the raw data in step S1 is as follows:

[0013] Obtain the design drawings of the air outlet to determine the inspection boundary. The boundary range covers the installation reference surface of the air outlet, all fastener positions and direction markings. Use a ring LED light source to provide uniform illumination. Acquire 5 frames of image data of the air outlet installation status continuously by the camera, remove the blurry frames, retain 3 clear images and mark them as the original image data.

[0014] Furthermore, the historical data processing and region division process in step S2 is as follows:

[0015] Retrieve installation history records of air outlets with the same model as the current testing equipment within the past 6 months and mark them as historical datasets. Filter and retain valid records in the historical datasets. Divide the original image data and historical datasets into grids, and calculate the pass rate of the installation areas within the corresponding grids of the original image data and historical datasets. Obtain the pre-stored standard pass rate threshold and compare it with the pass rate of each area. Areas with a pass rate ≥ the pass rate threshold are marked as regular installation areas, forming a set of regular installation areas. Areas with a pass rate < the pass rate threshold are marked as error-prone installation areas, forming a set of error-prone installation areas.

[0016] Furthermore, the analysis process for the set of error-prone installation areas in step S3 is as follows:

[0017] The original images of error-prone installation areas are acquired and preprocessed. A pre-stored Gaussian filter model is retrieved to remove noise, and histogram equalization is used to enhance image contrast and increase the clarity of feature edges. Image edge regions are extracted, and repeatable features are identified as stable features. Regions containing stable features are marked as feature recognition regions. Regions with curved transitions, overlapping areas of multiple features, and areas susceptible to lighting effects are extracted and identified as unstable features. Regions with unstable features are marked as complex regions.

[0018] Furthermore, the process of acquiring and jointly analyzing the model data in step S4 is as follows:

[0019] The scanner acquires scanned model data of the air outlet in its standard installation state. The model data includes the three-dimensional coordinates and feature parameters of each feature point corresponding to the original image data. Each feature in the feature identification region set is compared with the model data to obtain the difference in three-dimensional coordinates and mark it as the position deviation value, and the difference in feature parameters and mark it as the parameter deviation value. The pre-stored position deviation threshold and parameter deviation threshold are retrieved and compared with the position deviation value and parameter deviation value. When the position deviation value is within the range of the position deviation threshold and the parameter deviation value is within the range of the parameter deviation threshold, an adaptation region signal is generated. When the position deviation value is outside the range of the position deviation threshold or the parameter deviation value is outside the range of the parameter deviation threshold, a re-inspection signal is generated.

[0020] Furthermore, when generating the re-inspection signal, a complex region set is obtained, and key feature points that are actually installed and representative are extracted and marked as actual feature point clouds. Key feature points that are representative for subsequent installation in the model data are obtained and marked as model point clouds. The pre-stored fault tolerance difference threshold is retrieved, and the actual feature point clouds and model point clouds are horizontally shifted and stitched together for comparison. When there is a calibration difference between the actual feature point clouds and model point clouds that is greater than the fault tolerance difference threshold, an incompatible region signal is generated. When there is a calibration difference between the actual feature point clouds and model point clouds that is less than the fault tolerance difference threshold, an error marking signal is generated.

[0021] Furthermore, the analysis process for the mismatch signal in step S5 is as follows:

[0022] The comparison data of the mismatched region and historical failure cases in the historical dataset are retrieved, and the failure causes of the historical failure cases are extracted, including actual feature point cloud items, model point cloud items and deviation value items. The mean threshold of actual feature point cloud items, mean threshold of model point cloud items and deviation value items of cases that have not occurred in the historical dataset are obtained. The average difference between the actual feature point cloud items, model point cloud items and deviation value items of the historical failure cases and the mean threshold of actual feature point cloud items, mean threshold of model point cloud items and deviation value items is used as the risk judgment threshold. The comparison data and relevant data of the historical failure cases are brought into the comparison.

[0023] Furthermore, when the actual feature point cloud, model point cloud, and deviation value exceed the threshold values ​​for the average of the actual feature point cloud items, the average of the model point cloud items, and the deviation value item, but do not exceed the risk assessment threshold, a Level 1 risk signal is generated; when the actual feature point cloud, model point cloud, and deviation value exceed the threshold values ​​for the average of the actual feature point cloud items, the average of the model point cloud items, the deviation value item, and the risk assessment threshold, but do not exceed the threshold values ​​for the actual feature point cloud items, the model point cloud items, and the deviation value item, a Level 2 risk signal is generated; when the actual feature point cloud, model point cloud, and deviation value exceed the threshold values ​​for the actual feature point cloud items, the model point cloud items, and the deviation value item, a Level 3 risk signal is generated.

[0024] The beneficial effects of this invention are:

[0025] 1. This invention effectively controls the defect rate and inspection time of air outlet installations by dividing the area into regular / error-prone regions and adopting a differentiated detection strategy, combined with three-dimensional point cloud comparison technology. This significantly reduces labor costs and rework rates. At the same time, the threshold is dynamically optimized based on historical data, which improves the recognition accuracy of complex areas and solves the problem of misjudgment in difficult areas such as curved transitions, thus achieving a dual improvement in detection accuracy and efficiency.

[0026] 2. This invention achieves a shift from passive detection to proactive early warning through a three-level risk signal mechanism. Level 1 risks identify minor deviations in advance, Level 2 risks intercept moderate problems in a timely manner, and Level 3 risks handle serious faults urgently. This shortens the risk response time, not only ensuring the stability of product quality, but also improving the overall efficiency of the production line by reducing downtime and rework rates. This achieves a win-win situation for quality control and production efficiency, and realizes the synergistic optimization of risk management and production continuity. Attached Figure Description

[0027] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0028] Figure 1 This is a schematic flowchart of the method of the present invention. Detailed Implementation

[0029] The technical solutions of 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.

[0030] Example 1: Please refer to Figure 1 As shown, this embodiment is a method for detecting leaks and reverse installations at air outlets based on image detection, including the following steps;

[0031] S1. Determine the data acquisition range for image detection in the air outlet area. The detection area is defined based on the structural parameters of the air outlet, including dimensions and installation baseline. Several image data points of the air outlet's installation state are acquired using an industrial camera under preset lighting conditions, and these are marked as the original image data of the air outlet. The original image data of the air outlet includes RGB three-color channels and depth information, with a resolution of no less than 1920×1080 and a frame rate of 25fps. The process of determining the image detection range and acquiring the original data in step S1 is as follows:

[0032] Obtain the design drawings of the air outlet to determine the inspection boundary. The boundary range covers the installation reference surface of the air outlet, all fastener positions, and directional markings to ensure no blind spots. Use a ring LED light source to provide uniform illumination. Acquire 5 consecutive frames of image data of the air outlet installation status using a camera. Remove blurry frames. The blurry frame standard is adjusted according to the pre-stored standard frame threshold in the historical dataset. All blurry frames below the standard frame threshold are removed. Retain 3 clear images and mark them as the original image data. It should be noted that the industrial camera parameters are configured as follows: lens focal length is set to 16mm, working distance is maintained at 500±50mm; light intensity is controlled at 800-1200 lux; the original image data format is BMP, and the storage path is associated with the device model and the timestamp of the inspection acquisition. The standard frame threshold is used to determine whether the image is blurry. The possible value is 85 points, based on a sharpness scoring algorithm with a maximum score of 100 points. For example, if the sharpness score of a certain image is 82 points, which is lower than 85 points, it is judged as a blurry frame and discarded; if the sharpness score of a certain image is 90 points, which is higher than 85 points, the original image data is retained.

[0033] S2. Obtain historical data and raw image data for joint processing, and divide the area into a set of regular installation areas and a set of error-prone installation areas. The historical data serves as a threshold reference source for subsequent detection to avoid reflections affecting image quality. The historical data processing and area division process in step S2 is as follows:

[0034] Retrieve installation history records of air outlets with the same model as the current testing equipment within the past 6 months and mark them as historical datasets. Historical datasets include testing time, installation area coordinates, pass / fail judgment results, and fault types. Filter and retain valid records in the historical datasets, with a sample size ≥ 500 records. Divide the original image data and historical datasets into grids, and calculate the pass rate of the installation areas within the corresponding grids of the original image data and historical datasets: Pass rate = (Number of pass records / Total number of records) × 100%. Obtain the pre-stored standard pass rate threshold and compare it with the pass rate of each area. It should be noted that the standard pass rate threshold can be set to 95%, used to distinguish between regular and error-prone areas. For example: a certain clip position has 480 pass records out of a total of 500 records, with a pass rate of 96%, its pass rate ≥ 95%, and is classified as a regular installation area; a certain curved transition area has 420 pass records out of a total of 500 records, with a pass rate of 84%, its pass rate < 95%), and is classified as an error-prone installation area.

[0035] Areas with a pass rate greater than or equal to the pass rate threshold are marked as regular installation areas and formed into a set of regular installation areas. The subsequent detection frequency of this area can be reduced to once every 3 frames of images.

[0036] Areas with a pass rate less than the pass rate threshold are marked as error-prone installation areas, forming an error-prone installation area set. Subsequent images must be inspected for each frame, and the pass rate threshold is increased by 20%.

[0037] Historical datasets are automatically updated every 24 hours. Newly added test records are added to the database after manual review and confirmation. If the pass rate of a certain installation area is greater than or equal to the standard pass rate threshold for 30 consecutive days, it will be moved from the error-prone installation area set to the regular installation area set. If the pass rate is less than the standard pass rate threshold for 5 consecutive days, it will be moved from the regular installation area set to the error-prone installation area set to ensure the dynamic adaptability of the area division.

[0038] S3. Analyze error-prone installation areas, perform denoising and edge enhancement processing on the original images of these areas, and extract stable and unstable features to obtain a set of feature-identifiable regions and a set of complex regions. The analysis process of the error-prone installation region set in step S3 is as follows:

[0039] The original image of the error-prone installation area is acquired and preprocessed. A pre-stored Gaussian filter model is retrieved to remove noise. The filter kernel size is 3×3, meaning that the neighboring pixels within a 3×3 range around each pixel participate in the calculation. A corresponding weight matrix is ​​generated based on the Gaussian function. In this matrix, the pixels closer to the center have larger weights, which conforms to the Gaussian distribution law. Then, the gray value of each pixel in the original image is convolved with the weight matrix. That is, the new gray value of each pixel is equal to the product of its own gray value and the corresponding weight, plus the sum of the products of the gray values ​​of neighboring pixels and the corresponding weights. This weighted averaging method smooths the image and reduces the interference of random noise on subsequent feature extraction.

[0040] Histogram equalization enhances image contrast and increases the sharpness of feature edges. The specific process is as follows: First, the grayscale histogram of the original image data is statistically analyzed, that is, the number of pixels corresponding to each grayscale level (0-255) is calculated. Then, the cumulative distribution function (CDF) is calculated, which reflects the proportion of pixels less than or equal to a certain grayscale level in the total number of pixels. Finally, the grayscale values ​​are mapped and transformed according to the CDF, stretching the densely distributed grayscale range in the original image to a wider range, making dark areas darker and bright areas brighter, thereby enhancing the contrast between features and background in the image and highlighting edge details.

[0041] Extract the image edge region and identify repeatable features as stable features, such as circular buckle holes with a diameter of 3±0.5mm and directional lines with a length of 10±1mm. Mark the region containing the stable feature as the feature recognition region set. Extract the region with curved transition, blurred edges in the multi-feature overlapping area, and areas easily affected by lighting in the edge region and determine its unstable features. Mark the region with unstable features as the complex region set.

[0042] S4. Obtain the scanning model data of the standard installation of the air outlet, including three-dimensional coordinates and feature parameters. Compare the feature identification region set and complex region set with the model data one by one to generate suitable region signals and unsuitable region signals. The process of obtaining and jointly analyzing the model data in step S4 is as follows:

[0043] A 3D scanner is used to acquire scanned model data of the air vent in its standard installation state. The model data includes the 3D coordinates and feature parameters of each feature point corresponding to the original image data. The 3D coordinates include X, Y, and Z axes with an accuracy of ±0.01mm. Feature parameters include the diameter of the latching hole and the angle of the engraving line. Each feature in the feature recognition area set is compared with the model data to obtain the difference in 3D coordinates, which is marked as a positional deviation value (e.g., ΔX, ΔY, ΔZ), and the difference in feature parameters, which is marked as a parameter deviation value (e.g., diameter deviation ΔD). Pre-stored positional deviation thresholds and parameter deviation thresholds are retrieved and compared with the positional deviation values ​​and parameter deviation values. It should be noted that... Position deviation threshold: A possible value is ±0.1mm, based on a 3D coordinate accuracy of ±0.01mm. Example: A standard X-coordinate of a buckle hole is 100.00mm, and the actual measured value is 100.08mm. The deviation is 0.08mm ≤ 0.1mm, meeting the threshold requirement. If the actual value is 100.12mm, the deviation is 0.12mm > 0.1mm, exceeding the threshold. Parameter deviation threshold: A possible value is 5%, based on the feature parameter. Example: A standard length of a directional engraving is 10mm, and the actual measured value is 10.4mm. The deviation is 4% ≤ 5%, meeting the requirement. If the actual value is 10.6mm, the deviation is 6% > 5%, exceeding the threshold.

[0044] When the position deviation value is within the position deviation threshold range and the parameter deviation value is within the parameter deviation threshold range, an adaptation area signal is generated, indicating that the installation in the area meets the standard. The qualified information of the area is automatically recorded, and the conveyor belt is triggered to send the air outlet to the next process. At the same time, the historical dataset is updated, and qualified records are accumulated to optimize the area division threshold. Emergency measures: If there are 3 consecutive adaptation signals but the downstream process reports abnormalities, the re-inspection procedure is automatically started.

[0045] When the position deviation value is outside the position deviation threshold range, or the parameter deviation value is outside the parameter deviation threshold range, a re-inspection signal is generated, indicating that there is a deviation in the feature identification area and a second confirmation is required. The industrial camera is then controlled to re-acquire 3 frames of images of the area, using a higher precision algorithm, such as sub-pixel level edge detection, to recalculate the position deviation value and parameter deviation value. If the value still exceeds the threshold, the system automatically switches to the complex area point cloud comparison mode. Emergency measures: If the re-inspection number is ≥3 times and the problem persists, an audible and visual alarm is triggered, prompting the operator to check the camera focal length or light source stability.

[0046] When generating the re-inspection signal, a complex region set is obtained. Key feature points that are actually installed and representative are extracted and marked as actual feature point clouds. Key feature points that are representative for subsequent installation in the model data are obtained and marked as model point clouds. A pre-stored tolerance difference threshold is retrieved, and the actual feature point clouds and model point clouds are horizontally shifted and stitched together for comparison. It should be noted that the tolerance difference threshold can be 0.08mm, used for point cloud comparison in complex regions. For example, if the calibration difference between the actual feature point cloud and the model point cloud is 0.07mm (0.07mm < 0.08mm), it is considered acceptable error; if the difference is 0.09mm (0.09mm > 0.08mm), it is considered incompatible.

[0047] When the calibration difference between the actual feature point cloud and the model point cloud exceeds the tolerance threshold, an incompatibility signal is generated, indicating a significant installation deviation in a complex area. The production line is immediately suspended, and the 3D coordinates and deviation values ​​of the incompatibility area are marked on the display screen. The control actuator, such as a robotic arm, pushes the air outlet to the rework station and generates rework instructions, such as "Snap-hole position offset 0.15mm, it is recommended to move it to the left for adjustment," which are displayed on the monitor of the supervisor. If the same model of air outlet generates 5 incompatibility signals consecutively within 1 hour, the equipment is automatically locked, prompting a check to see if the scanned model data matches the current product model.

[0048] When there is a calibration difference between the actual feature point cloud and the model point cloud, and the difference is less than the tolerance threshold, an error marker signal is generated, indicating that there is a slight but acceptable deviation in the complex area. The deviation value of this area is marked in the inspection report, such as the text "curved surface fit deviation 0.06mm". The data is also synchronized to the production management system for statistical analysis of common error distributions of the air outlet of this model. In subsequent inspections, more attention is paid to this area, such as increasing the number of feature points extracted by 10%. If the error marker signal accounts for ≥30% of the products in the same batch, the operator is prompted to fine-tune the positioning reference of the installation tooling.

[0049] Example 2

[0050] S5. Retrieve the comparison data of the mismatched region, analyze the actual feature point cloud items, model point cloud items, and deviation value items in conjunction with historical failure cases, and generate corresponding risk level signals based on the risk level. The analysis process of the mismatch signal in step S5 is as follows:

[0051] The comparison data of the incompatible region and historical failure cases in the historical dataset are retrieved, and the causes of failure are extracted from the historical failure cases, including actual feature point cloud items, model point cloud items, and deviation values. The mean threshold of actual feature point cloud items, mean threshold of model point cloud items, and threshold of deviation values ​​in historical failure cases without failure are obtained. The average difference between the actual feature point cloud items, model point cloud items, and deviation values ​​of historical failure cases and the mean threshold of actual feature point cloud items, mean threshold of model point cloud items, and deviation values ​​is used as the risk judgment threshold. The comparison data and relevant data of historical failure cases are substituted into the comparison. It should be noted that the mean threshold of actual feature point cloud items: taking the X coordinate of a surface feature point as an example, can be 50.00±0.05mm. Based on the statistics of historical failure-free cases, for example: the actual detected X coordinate is 50.06mm, which exceeds the mean threshold and needs to be further compared with the risk judgment threshold.

[0052] When the actual feature point cloud, model point cloud, and deviation value exceed the threshold values ​​for the average of the actual feature point cloud items, the average of the model point cloud items, and the deviation value item, but do not exceed the risk assessment threshold, a Level 1 risk signal is generated, indicating a minor risk. A yellow warning icon is displayed on the terminal, marking the risk area and deviation value, such as the text "Direction line angle deviation 3°". Production is not suspended, but the risk information is recorded. When there are ≥2 Level 1 risk signals at the same air outlet, it is automatically upgraded to Level 2 risk. Operators summarize the Level 1 risk signals every hour to analyze whether there are common deviations, such as dimensional fluctuations in a batch of parts.

[0053] When the actual feature point cloud, model point cloud, and deviation value exceed the threshold of the average value of the actual feature point cloud item, the average value of the model point cloud item, the threshold of the deviation value item, and the risk judgment threshold, but do not exceed the actual feature point cloud item, model point cloud item, and deviation value item, a level 2 risk signal is generated, indicating that there is a moderate risk, triggering an orange alarm, suspending the flow of the air outlet, and prompting the operator to manually review; after the review is qualified, the operator must manually confirm "risk removed" before production can continue; if the review is unqualified, the process will be transferred to the rework process. If three level 2 risk signals occur consecutively, historical fault cases will be automatically retrieved and solutions for similar faults will be pushed, such as the text "Refer to case 20240512, adjust the buckle installation force";

[0054] When the actual feature point cloud, model point cloud, and deviation value exceed the actual feature point cloud item, model point cloud item, and deviation value item, a level three risk signal is generated, indicating a serious risk, such as missing or reversed installation. A red emergency alarm is activated, the production line power is cut off, and a warning message, including a picture of the fault area and deviation data, is sent to the mobile phone of the management personnel. After manual inspection and replacement of the qualified air outlet, the equipment can only be restarted after the administrator password is entered. If two level three risk signals occur within 24 hours, the quality traceability program is activated to verify the raw material batch and installation process parameters.

[0055] Combining Embodiments 1 and 2, this invention effectively controls the defect rate and inspection time of air outlet installations by dividing them into regular and error-prone areas, adopting differentiated detection strategies and combining them with 3D point cloud comparison technology. This reduces labor costs and rework rates. Based on historical data, thresholds are dynamically optimized to improve the recognition accuracy of complex areas and solve misjudgments in difficult areas, achieving a dual improvement in detection accuracy and efficiency. At the same time, a three-level risk signal mechanism is used to transform from passive detection to proactive early warning, shortening risk response time, ensuring product quality stability, reducing downtime, improving the overall efficiency of the production line, and achieving synergistic optimization of risk control and production continuity.

[0056] The above description is merely an example and illustration of the structure of the present invention. Those skilled in the art can make various modifications or additions to the specific examples described or use similar methods to replace them, as long as they do not deviate from the structure of the invention or exceed the scope defined in the claims, they should all fall within the protection scope of the present invention.

[0057] In the description of this specification, the references to terms such as "an embodiment," "example," and "specific example" indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described can be combined in any suitable manner in one or more embodiments or examples. Related accessories include commonly used mechanical connection components in this field such as couplings, lead screws, gears, and gaskets, but are not limited to these. Specific replacements and adaptations are made according to actual use.

[0058] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to any specific implementation. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A method for detecting leaks and reverse installations at air outlets based on image detection, characterized in that, Includes the following steps; S1. Determine the data acquisition range of the image detection in the air outlet area. Collect several image data of the air outlet installation status using an industrial camera under preset lighting conditions, and mark them as the original image data of the air outlet. S2. Obtain historical data and raw image data for joint processing, divide the area into a set of regular installation areas and a set of error-prone installation areas, and use the historical data as a reference source for the threshold of subsequent detection; S3. Analyze error-prone installation areas, perform noise reduction and edge enhancement on the original images of error-prone installation areas, extract stable and unstable features, and obtain feature recognition region sets and complex region sets. S4. Obtain the scanning model data of the standard installation of the air outlet, including three-dimensional coordinates and feature parameters. Compare the feature identification area set and complex area set with the model data one by one to generate the appropriate area signal and the inappropriate area signal. S5. Retrieve the comparison data of the mismatched area, combine it with the analysis of actual feature point cloud items, model point cloud items and deviation value items in historical failure cases, and generate the corresponding risk level signal according to the risk level. The process of acquiring and jointly analyzing model data in step S4 is as follows: Scanning model data of the standard installation state of the air outlet is acquired using a scanner. The model data includes the three-dimensional coordinates and feature parameters of each feature point corresponding to the original image data. Each feature in the feature identification region set is compared with the model data to obtain the difference in three-dimensional coordinates, which is marked as the position deviation value, and the difference in feature parameters, which is marked as the parameter deviation value. Pre-stored position deviation thresholds and parameter deviation thresholds are retrieved and compared with the position deviation value and parameter deviation value. When the position deviation value is within the range of the position deviation threshold and the parameter deviation value is within the range of the parameter deviation threshold, an adaptation region signal is generated. When the position deviation value is outside the range of the position deviation threshold or the parameter deviation value is outside the range of the parameter deviation threshold, a re-inspection signal is generated. When generating the re-inspection signal, a complex region set is obtained, and key feature points that are actually installed and representative are extracted and marked as actual feature point clouds. Key feature points that are representative for subsequent installation in the model data are obtained and marked as model point clouds. The pre-stored fault tolerance difference threshold is retrieved, and the actual feature point clouds and model point clouds are horizontally shifted and stitched together for comparison. When there is a calibration difference between the actual feature point clouds and model point clouds that is greater than the fault tolerance difference threshold, an incompatible region signal is generated. When there is a calibration difference between the actual feature point clouds and model point clouds that is less than the fault tolerance difference threshold, an error marker signal is generated. The analysis process for the mismatch signal in step S5 is as follows: retrieve the comparison data of the mismatch region and the historical failure cases in the historical dataset, and extract the failure causes of the historical failure cases, including actual feature point cloud items, model point cloud items and deviation value items. Obtain the mean threshold of actual feature point cloud items, mean threshold of model point cloud items and deviation value items of cases that have not experienced failures in the historical dataset. Use the average difference between the actual feature point cloud items, model point cloud items and deviation value items of the historical failure cases and the mean threshold of actual feature point cloud items, mean threshold of model point cloud items and deviation value items as the risk judgment threshold. Introduce the comparison data and relevant data of the historical failure cases into the comparison.

2. The method for detecting leaks and reverse installations at air outlets based on image detection according to claim 1, characterized in that, The process of determining the image detection range and acquiring raw data in step S1 is as follows: Obtain the design drawings of the air outlet to determine the inspection boundary. The boundary range covers the installation reference surface of the air outlet, all fastener positions and direction markings. Use a ring LED light source to provide uniform illumination. Acquire 5 frames of image data of the air outlet installation status continuously by the camera, remove the blurry frames, retain 3 clear images and mark them as the original image data.

3. The method for detecting leaks and reverse installations at air outlets based on image detection according to claim 1, characterized in that, The historical data processing and region division process in step S2 is as follows: Retrieve installation history records of air outlets with the same model as the current testing equipment within the past 6 months and mark them as historical datasets. Filter and retain valid records in the historical datasets. Divide the original image data and historical datasets into grids, and calculate the pass rate of the installation areas within the corresponding grids of the original image data and historical datasets. Obtain the pre-stored standard pass rate threshold and compare it with the pass rate of each area. Areas with a pass rate ≥ the pass rate threshold are marked as regular installation areas, forming a set of regular installation areas. The subsequent testing frequency of these areas can be reduced to once every 3 frames of images. Areas with a pass rate < the pass rate threshold are marked as error-prone installation areas, forming a set of error-prone installation areas. These areas need to be tested every frame of images, and the testing accuracy threshold is increased by 20%.

4. The method for detecting leaks and reverse installations at air outlets based on image detection according to claim 1, characterized in that, The analysis process for the set of error-prone installation areas in step S3 is as follows: The original images of error-prone installation areas are acquired and preprocessed. A pre-stored Gaussian filter model is retrieved to remove noise, and histogram equalization is used to enhance image contrast and increase the clarity of feature edges. Image edge regions are extracted, and repeatable features are identified as stable features, such as circular buckle holes with a diameter of 3±0.5mm and directional lines with a length of 10±1mm. The regions containing stable features are marked as feature recognition regions. Regions with curved transitions, overlapping areas of multiple features, blurred edges, and susceptibility to lighting effects are extracted from the edge regions and identified as unstable features. Regions with unstable features are marked as complex regions.

5. The method for detecting leaks and reverse installations at air outlets based on image detection according to claim 1, characterized in that, A Level 1 risk signal is generated when the actual feature point cloud, model point cloud, and deviation value exceed the threshold values ​​for the average value of the actual feature point cloud item, the average value of the model point cloud item, and the deviation value item, but do not exceed the risk assessment threshold. A Level 2 risk signal is generated when the actual feature point cloud, model point cloud, and deviation value exceed the threshold values ​​for the average value of the actual feature point cloud item, the average value of the model point cloud item, the deviation value item, and the risk assessment threshold, but do not exceed the threshold values ​​for the actual feature point cloud item, the model point cloud item, and the deviation value item. When the actual feature point cloud, model point cloud, and deviation value exceed the actual feature point cloud item, model point cloud item, and deviation value item, a level three risk signal is generated.

Images