An error correction method for the replacement state of a wind tunnel test model

Abstract

A method for correcting errors in wind tunnel test model replacement states belongs to the field of aerodynamic wind tunnel testing technology. The method includes model training, which involves deep learning training of a model image recognition and analysis system through model image data acquisition and classification. After training, model images are collected at the wind tunnel test site and input into the model image recognition and analysis system for automatic component recognition, with the component images and component label information within the images output. A test state preparation step is also included. After the model state is changed, image acquisition is performed, and the test model component information after image acquisition is compared with the pre-set training model component information. Based on the shape, position, and angle information of the test model components after the model state is changed, the test model replacement state is comprehensively determined, and the corresponding results are output and fed back to the work machine interface. This reduces the risk of test state replacement errors caused by human factors and improves test efficiency and accuracy.

Description

Technical Field

[0001] The invention relates to an error correction method for a wind tunnel test model replacement state, and belongs to the technical field of aviation aerodynamic wind tunnel tests. Background Art

[0002] In traditional wind tunnel testing, the change and confirmation of model test conditions primarily relies on the on-duty personnel's operations and confirmation. After a model state test is completed and the test conditions are changed, the on-duty personnel must visit the site to confirm the model state before closing the tunnel for testing. This method of testing and confirming test conditions is not only time-consuming and labor-intensive, but also carries the risk of human error leading to incorrect model state changes. If model state change errors are not promptly identified, not only will the accuracy of the test results be affected, delaying the test schedule, but also impacting the design and development of the model.

[0003] Therefore, it is urgent to propose a method for correcting the change status of wind tunnel test models to solve the above technical problems. Summary of the Invention

[0004] The present invention aims to address the problem that existing methods for testing and confirming test conditions are not only time-consuming and labor-intensive but also carry the risk of model state changes being incorrectly altered due to human error. If these errors are not promptly identified, not only can the accuracy of test results be affected, test progress can be delayed, and the design and development of the model can be impacted. Based on the wind tunnel test process, the present invention proposes a method for correcting model state changes in wind tunnel tests. This method captures images of the model state after the model state is changed during a wind tunnel test, autonomously determines the accuracy of the change, and provides an on-site result indicating whether the model state change is accurate. Based on this result, on-duty personnel can then conduct tunnel closing tests or adjust the model state accordingly. This method enables test condition detection using artificial intelligence. A brief overview of the present invention is provided below to provide a basic understanding of certain aspects of the present invention. It should be understood that this overview is not an exhaustive overview of the present invention. It is not intended to identify key or important aspects of the present invention, nor is it intended to limit the scope of the present invention.

[0005] The technical solution of the present invention:

[0006] A method for correcting an error in a wind tunnel test model replacement state, comprising:

[0007] S1, model training, obtains training model component information, performs deep learning training on the model image recognition and analysis system through model image data collection and classification, and after training, collects model images at the wind tunnel test site, inputs them into the model image recognition and analysis system for automatic component recognition, and outputs component images and component label information in the images;

[0008] S2, test state preparation. When preparing for the wind tunnel test, the test model components are numbered in ascending order according to their angles, using the naming rule of 0, 1, 2...9. These are marked on the surface of the test model components. Then, a model state table is obtained based on the names of all components and the angle information of each test model component.

[0009] S3, test status detection. During the wind tunnel test, image acquisition is performed after the model status is changed. The test model component information after image acquisition is compared with the previously set training model component information. According to the shape, position and angle information of the test model components after the model status is changed, the test model replacement status is comprehensively judged, and the corresponding result output is fed back to the working machine interface.

[0010] Preferably, the model training in step 1 specifically includes:

[0011] S1.1. Prepare the model. Sort the training model components in ascending order of angle using the naming rule 0, 1, 2, ..., 9. Mark the rudder surfaces of the training model components with numbers according to the naming rule.

[0012] S1.2, install the states of the trained model components according to the test order, ensuring that none of the marked numbers are obscured;

[0013] S1.3, fix the training model of the required collected images to the image acquisition platform;

[0014] S1.4, turn on the camera, check whether the camera is turned on, if yes, jump to S1.5, if not, output a warning message, check the camera connection, and jump to S1.5 after detecting that the camera is turned on;

[0015] S1.5, use a camera to collect training model components from multiple angles;

[0016] S1.6, replace the training model components according to the test sequence, and continue to use the camera to collect multi-angle images of the training model components until all test states are replaced;

[0017] S1.7, remove the current training model component, replace it with the next training model component, and repeat steps S1.1 to S1.6 until all training model components are collected and stored;

[0018] S1.8, model rudder classification and labeling: open the collected images with the labeling software Wizard Assistant, classify and name each training model component according to its name, and label all images with components;

[0019] S1.9, extracting the rudder surface image, extracting component information from the labeled image according to different categories and saving it;

[0020] S1.10, rudder digital classification, together with the rudder image, serves as the input file of the model image recognition and analysis system, and is input into the deep learning model system for model training.

[0021] Preferably, the test status detection in step 3 specifically includes:

[0022] S3.1. Input all test states of the model. On the test site work plane, enter the names of all test model components and their angle information. Names include flaps, ailerons, V-tail, rudder, elevator, and horizontal stabilizer. Change the model state to generate a model state table.

[0023] S3.2, importing the model state table into the model image recognition and analysis system;

[0024] S3.3, input the test vehicle number;

[0025] S3.4, turn on the on-site camera and determine whether the camera is detected. If so, jump to S3.6. If not, repeat S3.5;

[0026] S3.5, output warning, warning message is: the camera is not detected, please check the connection, after checking the connection, repeat S3.4;

[0027] S3.6, controlling the on-site camera to collect data, inputting the collected model information into the training model, performing model rudder surface recognition and model rudder surface labeling digital recognition, respectively, to obtain model rudder surface recognition results and model rudder surface labeling digital recognition results;

[0028] S3.7, compare the digital markings on the rudder surface with the surface numbers of the test model components, compare the model rudder surface recognition results and the model rudder surface marking digital recognition results obtained in S3.6 with the model rudder surface names and rudder surface angles obtained in the test vehicle number, and obtain a conclusion on whether the test model status detected by the camera is consistent with the test vehicle number. If they are consistent, output the status of each rudder surface and prompt that the status is correct. If they are inconsistent, output the status of each rudder surface and prompt that the rudder surface status is incorrect.

[0029] The present invention has the following beneficial effects:

[0030] 1. This invention replaces the traditional wind tunnel test process that requires confirmation by on-duty personnel with an artificial intelligence pattern recognition method to identify and classify model component information, thus avoiding the possibility of test condition changes caused by human factors.

[0031] 2. The present invention can determine the model status through pattern recognition during wind tunnel testing, significantly improving wind tunnel testing efficiency and the accuracy of test condition changes, thereby improving the overall test accuracy.

[0032] 3. The present invention is highly robust to the background and lighting of the model environment and is applicable to a wide range of usage scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 It is a logic block diagram of model training in the first specific embodiment of the present invention;

[0034] Figure 2 It is a logic block diagram of the test state detection of the first specific embodiment of the present invention. Implementation Method

[0035] To make the objectives, technical solutions, and advantages of the present invention more clearly apparent, the present invention is described below using specific embodiments shown in the accompanying drawings. However, it should be understood that these descriptions are merely illustrative and are not intended to limit the scope of the present invention. In addition, in the following description, descriptions of well-known structures and technologies are omitted to avoid unnecessary confusion of the concepts of the present invention.

[0036] The connections mentioned in the present invention are divided into fixed connections and detachable connections. The fixed connection refers to a non-detachable connection, including but not limited to conventional fixed connection methods such as hem connection, rivet connection, adhesive connection, and welding connection. The detachable connection refers to but not limited to conventional detachable connection methods such as threaded connection, snap connection, pin connection, and hinge connection. When the specific connection method is not clearly specified, it is assumed that at least one connection method can always be found among the existing connection methods to achieve the function. Those skilled in the art can choose according to their needs. For example, a welded connection is selected for a fixed connection, and a hinge connection is selected for a detachable connection.

[0037] Specific implementation method 1: Combination Figure 1-Figure 2 This embodiment is described. This embodiment is a method for correcting the state of a wind tunnel test model replacement. It is implemented based on a test condition detection system. The test condition detection system used mainly realizes the functions of image acquisition control and test state judgment feedback at the wind tunnel test site. At the test site, the on-duty analyst uses a public machine to control the image acquisition system to control four cameras to acquire model images. The collected model images are subjected to image component information recognition, and the comparison results are compared with the correct test state. The test condition detection system includes four cameras, four camera bracket working machines, a graphics workstation and other hardware and deep learning model software for image recognition and analysis. The image recognition and analysis model uses the more mature YOLO framework for image recognition and segmentation, and uses the handwriting recognition and judgment method in deep learning to confirm the component status. The test site camera is connected to the column at the entrance of the second throat of the wind tunnel through the camera bracket. The camera power cord and data cable are connected to the wind tunnel body through the wiring fixture, and are connected to the power supply and public machine through the common wiring port. The specific steps include:

[0038] S1, model training, obtains training model component information, performs deep learning training on the model image recognition and analysis system through model image data collection and classification, and after training, collects model images at the wind tunnel test site, inputs them into the model image recognition and analysis system for automatic component recognition, and outputs component images and component label information in the images;

[0039] S1.1. Prepare the model. Sort the training model components in ascending order of angle using the naming rule 0, 1, 2, ..., 9. Mark the rudder surfaces of the training model components with numbers according to the naming rule.

[0040] S1.2, install the states of the trained model components according to the test order, ensuring that none of the marked numbers are obscured;

[0041] S1.3, fix the training model of the required collected images to the image acquisition platform;

[0042] S1.4, turn on the camera, check whether the camera is turned on, if yes, jump to S1.5, if not, output a warning message, check the camera connection, and jump to S1.5 after detecting that the camera is turned on;

[0043] S1.5, use a camera to collect training model components from multiple angles;

[0044] S1.6, replace the training model components according to the test sequence, and continue to use the camera to collect multi-angle images of the training model components until all test states are replaced;

[0045] S1.7, remove the current training model component, replace it with the next training model component, and repeat steps S1.1 to S1.6 until all training model components are collected and stored;

[0046] S1.8, model rudder classification and labeling: open the collected images with the labeling software Wizard Assistant, classify and name each training model component according to its name, and label all images with components;

[0047] S1.9, extracting the rudder surface image, extracting component information from the labeled image according to different categories and saving it;

[0048] S1.10, rudder digital classification, together with the rudder image, serves as the input file of the model image recognition and analysis system, and is input into the deep learning model system for model training.

[0049] S2, test state preparation. When preparing for the wind tunnel test, the test model components are numbered in ascending order according to their angles, using the naming rule of 0, 1, 2...9. These are marked on the surface of the test model components. Then, a model state table is obtained based on the names of all components and the angle information of each test model component.

[0050] S3, test status detection. During the wind tunnel test, image acquisition is performed after the model status is changed. The test model component information after image acquisition is compared with the previously set training model component information. According to the shape, position and angle information of the test model components after the model status is changed, the test model replacement status is comprehensively judged, and the corresponding result output is fed back to the working machine interface.

[0051] S3.1. Input all test states of the model. On the test site work plane, enter the names of all test model components and their angle information. Names include flaps, ailerons, V-tail, rudder, elevator, and horizontal stabilizer. Change the model state to generate a model state table.

[0052] S3.2, importing the model state table into the model image recognition and analysis system;

[0053] S3.3, input the test vehicle number;

[0054] S3.4, turn on the on-site camera and determine whether the camera is detected. If so, jump to S3.6. If not, repeat S3.5;

[0055] S3.5, output warning, warning message is: the camera is not detected, please check the connection, after checking the connection, repeat S3.4;

[0056] S3.6, controlling the on-site camera to collect data, inputting the collected model information into the training model, performing model rudder surface recognition and model rudder surface labeling digital recognition, respectively, to obtain model rudder surface recognition results and model rudder surface labeling digital recognition results;

[0057] S3.7, compare the digital markings on the rudder surface with the surface numbers of the test model components, compare the model rudder surface recognition results and the model rudder surface marking digital recognition results obtained in S3.6 with the model rudder surface names and rudder surface angles obtained in the test vehicle number, and obtain a conclusion on whether the test model status detected by the camera is consistent with the test vehicle number. If they are consistent, output the status of each rudder surface and prompt that the status is correct. If they are inconsistent, output the status of each rudder surface and prompt that the rudder surface status is incorrect.

[0058] The test condition detection system categorizes aircraft model configurations into missile-type and aircraft-type models. The missile-type model is suitable for conventional layouts, while the aircraft-type model is suitable for identifying various control surfaces. This approach is not applicable to components obscured by all four cameras in this implementation. This implementation reduces the risk of test state changes due to human error, eliminating the need for on-the-spot analysis and confirmation, improving test efficiency and accuracy. Furthermore, it is robust to the model's environmental background and lighting conditions, allowing for a wider range of applications.

[0059] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention are intended to be within the scope of protection of the present invention.

Claims

1. A method for correcting the error of a wind tunnel test model replacement state, characterized in that: include: S1, model training, obtains training model component information, performs deep learning training on the model image recognition and analysis system through model image data collection and classification, and after training, collects model images at the wind tunnel test site, inputs them into the model image recognition and analysis system for automatic component recognition, and outputs component images and component label information in the images; S2, test state preparation. When preparing for the wind tunnel test, the test model components are numbered in ascending order according to their angles, using the naming rule of 0, 1, 2, ..., 9. These are marked on the surface of the test model components. Then, based on the names of all components and the angle information of each test model component, a model state table is obtained. S3, test state detection: During the wind tunnel test, image acquisition is performed after the model state is changed. The test model component information after image acquisition is compared with the previously set training model component information. Based on the shape, position and angle information of the test model components after the model state is changed, the test model change state is comprehensively judged, and the corresponding results are output and fed back to the working machine interface; The model training in step 1 specifically includes: S1.

1. Prepare the model. Sort the training model components in ascending order of angle using the naming rule 0, 1, 2, ..., 9. Mark the rudder surfaces of the training model components with numbers according to the naming rule. S1.2, install the states of the trained model components according to the test order, ensuring that none of the marked numbers are obscured; S1.3, fix the training model of the required collected images to the image acquisition platform; S1.4, turn on the camera, check whether the camera is turned on, if yes, jump to S1.5, if not, output a warning message, check the camera connection, and jump to S1.5 after detecting that the camera is turned on; S1.5, use a camera to collect training model components from multiple angles; S1.6, replace the training model components according to the test sequence, and continue to use the camera to collect multi-angle images of the training model components until all test states are replaced; S1.7, remove the current training model component, replace it with the next training model component, and repeat steps S1.1 to S1.6 until all training model components are collected and stored; S1.8, model rudder classification and labeling: open the collected images with the labeling software Wizard Assistant, classify and name each training model component according to its name, and label all images with components; S1.9, extracting the rudder surface image, extracting component information from the labeled image according to different categories and saving it; S1.10, rudder digital classification, which together with the rudder image serves as the input file of the model image recognition and analysis system and is input into the deep learning model system for model training; The test status detection of step 3 specifically includes: S3.

1. Input all test states of the model. On the test site work plane, enter the names of all test model components and their angle information. Names include flaps, ailerons, V-tail, rudder, elevator, and horizontal stabilizer. Change the model state to generate a model state table. S3.2, importing the model state table into the model image recognition and analysis system; S3.3, input the test vehicle number; S3.4, turn on the on-site camera and determine whether the camera is detected. If so, jump to S3.

6. If not, repeat S3.5; S3.5, output warning, warning message is: the camera is not detected, please check the connection, after checking the connection, repeat S3.4; S3.6, controlling the on-site camera to collect data, inputting the collected model information into the training model, performing model rudder surface recognition and model rudder surface labeling digital recognition, respectively, to obtain model rudder surface recognition results and model rudder surface labeling digital recognition results; S3.7, compare the digital markings on the rudder surface with the surface numbers of the test model components, compare the model rudder surface recognition results and the model rudder surface marking digital recognition results obtained in S3.6 with the model rudder surface names and rudder surface angles obtained in the test vehicle number, and obtain a conclusion on whether the test model status detected by the camera is consistent with the test vehicle number. If they are consistent, output the status of each rudder surface and prompt that the status is correct. If they are inconsistent, output the status of each rudder surface and prompt that the rudder surface status is incorrect.

Images