A deep learning-based single-frame structured light gear fault three-dimensional measurement method and system

Abstract

The application provides a single-frame structured light gear fault three-dimensional measurement method and system based on deep learning, and the measurement method steps are as follows: 1) calibrating a projector and a camera, and establishing an internal and external parameter model of the projector-camera; 2) generating a single-frame coded fringe pattern projection to the surface of a measured gear, and collecting a deformed fringe pattern; 3) inputting the single-frame deformed fringe pattern into a wrapped phase recovery model based on a GPD-Net network to wrap the phase; 4) calculating the absolute phase of each pixel point in the image through the wrapped phase; 5) calculating the three-dimensional space coordinates of the corresponding gear fault surface of each pixel point according to the internal and external parameter model and the absolute phase; 6) analyzing the obtained fault three-dimensional point cloud data, and completing the three-dimensional reconstruction and measurement of the gear fault with high precision and speed. The application realizes the real-time online measurement of the three-dimensional morphology of the gear fault through the fringe projection profile technology.

Description

Technical Field

[0001] This invention relates to the field of gear fault diagnosis, and in particular to a single-frame structured light gear fault three-dimensional measurement method and system based on deep learning. Background Technology

[0002] Offshore wind turbines, aero engines, and other critical national equipment require safe and stable operation; any malfunction could result in significant economic losses and casualties. Therefore, early health monitoring of such equipment is crucial. Gears are core transmission components of this equipment and are also the most prone to failure, with common gear faults including pitting, cracking, and spalling. Currently, most researchers use vibration signals for gear fault diagnosis or lifespan prediction. However, these vibration signals include not only gear vibrations but also noise and vibration signals from other components. It is difficult to determine the type of gear fault from these vibration signals, let alone crucial information such as the shape, size, and depth of defects. Therefore, machine vision-based 3D measurement of gear faults has significant engineering value, enabling direct and quantitative assessment of equipment to ensure its safe and efficient operation.

[0003] The invention patent with patent number "CN115272065A" discloses a "dynamic fringe projection three-dimensional measurement method based on fringe pattern super-resolution reconstruction". This application uses the multi-step phase shift method and the phase of multi-frequency heterodyne expansion to perform phase matching to realize the three-dimensional measurement of fringe projection, but this application cannot implement real-time online measurement of fringe projection. Summary of the Invention

[0004] One objective of this invention is to provide a single-frame structured light gear fault 3D measurement method based on deep learning. This invention achieves online measurement of gear faults through a single-frame method based on deep learning.

[0005] The objective of this invention is achieved through the following technical solution, the specific steps of which are as follows:

[0006] 1) System calibration: Adjust the relative positions of the projector, camera and gear under test, calibrate the projector and camera after debugging, and establish the internal and external parameter model of the projector-camera.

[0007] 2) Data acquisition: Generate a single-frame coded fringe pattern and load it into the projector, then project it onto the surface of the gear under test. The camera acquires the deformed fringe pattern at time intervals of △T to obtain a single-frame deformed fringe pattern.

[0008] 3) Calculate the wrapping phase: Input the deformed stripe pattern of the acquired single frame into the wrapping phase recovery model based on the deep learning network GPD-Net to obtain the wrapping phase of each pixel;

[0009] 4) Calculate the absolute phase: Calculate the absolute phase of each pixel in the image based on the obtained wrapper phase of each pixel;

[0010] 5) Calculate coordinates: Based on the internal and external parameter model of the projector-camera and the absolute phase of each pixel, calculate the three-dimensional spatial coordinate values ​​of the gear fault surface corresponding to each pixel.

[0011] 6) Reconstruction and Measurement: Analyze the point cloud data of the obtained three-dimensional spatial coordinate values ​​of the fault to complete the high-precision and rapid three-dimensional reconstruction and measurement of the gear fault.

[0012] Furthermore, the specific steps for system calibration in step 1) are as follows:

[0013] 1-1) Adjust the relative positions of the projector, camera and the gear under test so that the grating projected by the projector covers the fault features of the gear under test, and the fault features of the gear under test are located within the common field of view of the projector and camera.

[0014] 1-2) Disassemble the completed projector and camera, and place them horizontally on a table with the projector's optical axis aligned. Perform off-site calibration to establish the projector-camera internal and external parameter model:

[0015]

[0016] In the formula, M is the product of internal and external parameters, and X... w Y w Z w These are the three-dimensional spatial coordinates of the object being measured.

[0017] After calibration, the projector and camera were reinstalled in their original positions for subsequent measurements.

[0018] Furthermore, the specific steps for data collection in step 2) are as follows:

[0019] 2-1) Generate a single-frame coded fringe pattern and load it into a projector, then project it onto the surface of the gear being measured. The generated single-frame coded fringe pattern I(x,y) is as follows:

[0020]

[0021] In the formula, A and B are background modulations. Get the wrapper phase for each pixel;

[0022] 2-2) The camera acquires the deformed fringe pattern at time intervals of △T to obtain a single frame of deformed fringe pattern.

[0023] Furthermore, the specific steps for calculating the package phase in step 3) are as follows:

[0024] 3-1) The obtained single-frame deformed stripe map I(x,y) is used as input. The single-frame deformed stripe map I(x,y) is first segmented into three feature maps of different scales by the multi-layer convolution C of the deep learning network GPD-Net.

[0025] 3-2) The feature maps at different scales are stretched by the F operation, and the stretched feature maps are used as input features that can be read by the transformer module T.

[0026] 3-3) Use three transformer modules T with different numbers of heads to perform multi-space feature extraction on features at three different scales respectively;

[0027] 3-4) Learnable feature fusion is performed through an A-CRF module with a gated attention mechanism to fuse multi-scale and multi-spatial features in a principled manner;

[0028] 3-5) After the decoding module De outputs M and D, the wrap phase of each pixel is calculated.

[0029]

[0030] Furthermore, the specific method for calculating the absolute phase in step 4) is as follows:

[0031] Calculate the absolute phase of each pixel in the image based on the obtained wrapper phase:

[0032]

[0033] In the formula, k(x,y) is the stripe order.

[0034] Furthermore, the specific steps for calculating the three-dimensional spatial coordinates in step 5) are as follows:

[0035] 5-1) Convert the solved absolute phase into coordinate values ​​u according to the following formula. p :

[0036] u p (x,y)=ψ(x,y) / 2Nπ*Width

[0037] 5-2) Solve for u p Substituting the intrinsic and extrinsic parameter equations of the projector-camera obtained in step 1), and using the three known quantities (u) c v c u p Solve the three equations simultaneously to find the three unknowns (X). W Y W Z W ); where u cv c The pixel coordinates of the camera are known.

[0038] Furthermore, the specific steps for three-dimensional reconstruction and measurement of gear faults in step 6) are as follows:

[0039] 6-1) Select a fault-free section perpendicular to the tooth tip line as a reference plane, project the point cloud data of the three-dimensional spatial coordinates of the tooth surface on the section onto the section to obtain the point set of the tooth profile, and generate a healthy tooth profile curve through polynomial fitting.

[0040] 6-2) Project the point cloud data of the three-dimensional spatial coordinates of the tooth surface of other sections onto the reference plane to obtain its tooth profile curve. The fault depth is calculated by the error between the healthy tooth profile curve and the tooth profile curve to be tested.

[0041] Another objective of this invention is to provide a three-dimensional measurement method for gear faults based on deep learning using a single-frame structured light. This invention achieves online measurement of gear faults by projecting a stripe image onto the surface of the gear under test and acquiring the deformed stripe image.

[0042] The objective of this invention is achieved through the following technical solution: a measuring box, characterized in that it includes a gearbox platform assembly, a data acquisition assembly, and an oil baffle assembly disposed within the measuring box;

[0043] The gearbox platform assembly is used to provide gear fault samples for the measurement system. The gearbox platform assembly includes a gearbox and gears installed inside the gearbox.

[0044] The data acquisition component is used to project a single-frame coded stripe pattern onto the surface of the gear under test and simultaneously acquire stripe images modulated by the surface of the gear under test. The component includes a projector and a CCD industrial camera mounted above the gearbox.

[0045] The oil baffle assembly is used to block the lubricating oil splashed from the gearbox, so that the gearbox cover has a clean area to facilitate the data acquisition component to collect fault images. The oil baffle assembly includes an oil baffle plate integrated on the gearbox cover.

[0046] Furthermore, it also includes a component for controlling the speed increase and decrease of gears during operation, ensuring measurement and control of the gears even when they are rotating at low speeds.

[0047] And an image processing component for implementing the above-mentioned deep learning-based single-frame structured light gear fault 3D measurement method.

[0048] Because of the adoption of the above technical solution, the present invention has the following advantages:

[0049] 1. This application uses stripe projection contour technology to achieve real-time online measurement of the three-dimensional morphology of gear faults.

[0050] 2. This application uses a corresponding deep learning model-based phase algorithm to directly recover the high-precision phase value of the corresponding pixel from the coded stripe image acquired in a single frame, thereby finally obtaining the three-dimensional measurement result of the gear fault.

[0051] Other advantages, objectives, and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination, or may be learned from practice of the invention. The objectives and other advantages of the invention can be realized and obtained from the following description and claims. Attached Figure Description

[0052] The accompanying drawings of this invention are described below.

[0053] Figure 1 This is a flowchart illustrating the three-dimensional measurement method for single-frame structured light gear faults based on deep learning, as described in this invention.

[0054] Figure 2 This is a schematic diagram of the structure of the single-frame structured light gear fault three-dimensional measurement system based on deep learning according to the present invention.

[0055] Figure 3 This is a cross-sectional view of the single-frame structured light gear fault 3D measurement system based on deep learning according to the present invention.

[0056] Figure 4 This is a schematic diagram of the wrap-around phase recovery model based on the GPD-Net network in the single-frame structured light gear fault three-dimensional measurement method based on deep learning of the present invention.

[0057] Figure 5 This is a schematic diagram of the result of the three-dimensional point cloud data of gear fault obtained by the single-frame structured light gear fault three-dimensional measurement method based on deep learning in an embodiment of the present invention.

[0058] Figure 6 This is a schematic diagram of the maximum depth result of gear fault obtained by the single-frame structured light gear fault three-dimensional measurement method based on deep learning in an embodiment of the present invention.

[0059] Figure 7 This is a schematic diagram showing the gear fault calculation results obtained by the single-frame structured light gear fault three-dimensional measurement method based on deep learning in an embodiment of the present invention.

[0060] In the diagram: 1-Measuring box; 2-Gearbox platform assembly; 3-Data acquisition assembly; 4-Oil baffle assembly; 11-Vibration isolation perforated plate; 12-Double-head vibration isolation rubber; 13-Mounting base plate; 21-Gearbox; 22-Gear; 31-Projector; 32-CCD industrial camera; 33-Projector adapter board; 34-Camera adapter board; 41-Oil baffle plate. Detailed Implementation

[0061] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0062] Example 1:

[0063] like Figures 2-3 The single-frame structured light gear fault three-dimensional measurement system based on deep learning is shown, including a measurement box 1, and a gearbox platform assembly 2, a data acquisition assembly 3, and an oil baffle assembly 4 disposed in the measurement box 1.

[0064] The gearbox platform assembly 2 is used to provide gear fault samples for the measurement system. The gearbox platform assembly 2 includes a gearbox 21 and a gear 22 installed in the gearbox 21.

[0065] The data acquisition component 3 is used to project a single-frame coded stripe pattern onto the surface of the gear 22 under test, and simultaneously acquire stripe images modulated by the surface of the gear 22 under test. The component includes a projector 31 and a CCD industrial camera 32 disposed above the gearbox 21.

[0066] The oil baffle assembly 4 is used to block the lubricating oil splashed from the gearbox 21, so that the upper cover of the gearbox 21 has a clean area to facilitate the data acquisition assembly 3 to collect fault images. The oil baffle assembly 4 includes an oil baffle plate 41 integrated on the upper cover of the gearbox 21.

[0067] It also includes a speed control component (not shown in the figure) for controlling the gear 22 during operation to increase or decrease speed, and to ensure that the gear 22 is measured and controlled at low speed.

[0068] And an image processing component (not shown in the figure) for implementing a deep learning-based single-frame structured light gear fault 3D measurement method.

[0069] In this embodiment of the invention, a vibration isolation perforated plate 11 is installed on the inner wall of the top of the measuring box 1. A double-headed vibration isolation rubber 12 is provided between the vibration isolation perforated plate 11 and the inner wall of the top of the measuring box 1. The mounting base plate 13 is installed on the vibration isolation perforated plate 11. The projector 31 and the CCD industrial camera 32 are respectively installed on the mounting base plate 13 through the projector adapter plate 33 and the camera adapter plate 34.

[0070] In this invention, the image processing component is a computer that stores a computer program for a single-frame structured light gear fault 3D measurement method based on deep learning. By running the program, a single-frame structured light 3D reconstruction and quantitative evaluation and analysis of gear faults are completed.

[0071] Example 2:

[0072] like Figure 1 , Figure 4 The following is a method for 3D measurement of gear faults based on deep learning using a single frame of structured light, with the specific steps as follows:

[0073] 1) System Calibration: Adjust the relative positions of the projector, camera, and gear under test; calibrate the projector and camera after adjustment; establish the projector-camera intrinsic and extrinsic parameter model; specific steps are as follows:

[0074] 1-1) Adjust the relative positions of the projector, camera and the gear under test so that the grating projected by the projector covers the fault features of the gear under test, and the fault features of the gear under test are located within the common field of view of the projector and camera.

[0075] 1-2) After the projector and camera have been debugged, remove them from the machine along with the mounting plate. Place the projector horizontally on the table with the optical axis of the projector horizontal, and perform off-machine calibration to establish the internal and external parameter models of the projector-camera system:

[0076]

[0077] In the formula, M is the product of internal and external parameters, and X... w Y w Z w These are the three-dimensional spatial coordinates of the object being measured.

[0078] After calibration, the projector and camera were reinstalled in their original positions for subsequent measurements.

[0079] 2) Data Acquisition: A single-frame coded fringe pattern is generated and loaded into a projector, which then projects it onto the surface of the gear under test. The camera acquires the deformed fringe pattern at time intervals of ΔT to obtain a single-frame deformed fringe pattern. The specific steps are as follows:

[0080] 2-1) Generate a single-frame coded fringe pattern and load it into a projector, then project it onto the surface of the gear being measured. The generated single-frame coded fringe pattern I(x,y) is as follows:

[0081]

[0082] In the formula, A and B are background modulations. Get the wrapper phase for each pixel;

[0083] 2-2) The camera acquires the deformed fringe pattern at time intervals of △T to obtain a single frame of deformed fringe pattern.

[0084] 3) Calculate the wrapping phase: Input the deformed fringe pattern of the acquired single frame into the wrapping phase retrieval model based on the deep learning network GPD-Net to obtain the wrapping phase of each pixel; the specific steps are as follows:

[0085] The specific steps are as follows:

[0086] 3-1) The obtained single-frame deformed stripe map I(x,y) is used as input. The single-frame deformed stripe map I(x,y) is first segmented into three feature maps of different scales by the multi-layer convolution C of the deep learning network GPD-Net.

[0087] 3-2) The feature maps at different scales are stretched by the F operation, and the stretched feature maps are used as input features that can be read by the transformer module T.

[0088] 3-3) Use three transformer modules T with different numbers of heads to perform multi-space feature extraction on features at three different scales respectively;

[0089] 3-4) Learnable feature fusion is performed through an A-CRF module with a gated attention mechanism to fuse multi-scale and multi-spatial features in a principled manner;

[0090] 3-5) After the decoding module De outputs M and D, the wrap phase of each pixel is calculated.

[0091]

[0092] 4) Calculate the absolute phase: Based on the obtained wrapper phase of each pixel, calculate the absolute phase of each pixel in the image:

[0093]

[0094] In the formula, k(x,y) is the stripe order.

[0095] 5) Calculate coordinates: Based on the internal and external parameter model of the projector-camera and the absolute phase of each pixel, calculate the three-dimensional spatial coordinate values ​​of the gear fault surface corresponding to each pixel.

[0096] 5-1) Convert the solved absolute phase into coordinate values ​​u according to the following formula. p :

[0097] u p (x,y)=ψ(x,y) / 2Nπ*Width

[0098] 5-2) Solve for u p Substituting the intrinsic and extrinsic parameter equations of the projector-camera obtained in step 1), and using the three known quantities (u) c v c u p Solve the three equations simultaneously to find the three unknowns (X). W Y W Z W ); where u c v c The pixel coordinates of the camera are known.

[0099] 6) Reconstruction and Measurement: Analyze the point cloud data of the obtained three-dimensional spatial coordinates of the fault to complete high-precision and rapid three-dimensional reconstruction and measurement of the gear fault. The specific steps are as follows:

[0100] 6-1) Select a fault-free section perpendicular to the tooth tip line as a reference plane, project the point cloud data of the three-dimensional spatial coordinates of the tooth surface on the section onto the section to obtain the point set of the tooth profile, and generate a healthy tooth profile curve through polynomial fitting.

[0101] 6-2) Project the point cloud data of the three-dimensional spatial coordinates of the tooth surface of other sections onto the reference plane to obtain its tooth profile curve. The fault depth is calculated by the error between the healthy tooth profile curve and the tooth profile curve to be tested.

[0102] The gear fault was reconstructed and measured according to the above steps, and the results are as follows:

[0103] from Figure 5 As can be seen in (a), the high-precision 3D reconstruction method—the 18-step phase-shifting algorithm—is used as the ground truth for training the network model. The 3D point cloud data output by GPD-Net is close to the ground truth and has high measurement accuracy.

[0104] from Figure 5 As can be seen in (b), by plotting a row of fault depth information of the gear pitting fault point cloud data, it is close to the true value, and the proposed method has high measurement accuracy.

[0105] from Figure 6 As can be seen, the image processing system can calculate the maximum depth information of gear defects.

[0106] from Figure 7 As can be seen, the image processing system can display the calculation results of key fault information such as defect size and depth, and can achieve quantitative assessment of gear faults.

[0107] The image processing system can visualize the depth information of the fault, which can be displayed in the measured three-dimensional point data or in the measured gear image, enabling directional assessment of gear faults.

[0108] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0109] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0110] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0111] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0112] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A single-frame structured light gear fault 3D measurement method based on deep learning, characterized in that, The specific steps are as follows: 1) System calibration: Adjust the relative positions of the projector, camera and gear under test, calibrate the projector and camera after debugging, and establish the internal and external parameter model of the projector-camera. 2) Data Acquisition: A single-frame coded stripe pattern is generated and loaded into a projector, which then projects it onto the surface of the gear being measured. The camera... The stripe pattern after deformation is acquired at time intervals to obtain a single frame of deformed stripe pattern. 3) Calculate the wrapping phase: Input the deformed stripe pattern of the acquired single frame into the wrapping phase recovery model based on the deep learning network GPD-Net to obtain the wrapping phase of each pixel; 4) Calculate the absolute phase: Calculate the absolute phase of each pixel in the image based on the obtained wrapper phase of each pixel; 5) Calculate coordinates: Based on the internal and external parameter model of the projector-camera and the absolute phase of each pixel, calculate the three-dimensional spatial coordinate values ​​of the gear fault surface corresponding to each pixel. 6) Reconstruction and Measurement: Analyze the point cloud data of the obtained three-dimensional spatial coordinate values ​​of the fault to complete high-precision and rapid three-dimensional reconstruction and measurement of gear faults; The specific steps for system calibration in step 1) are as follows: 1-1) Adjust the relative positions of the projector, camera and the gear under test so that the grating projected by the projector covers the fault features of the gear under test, and the fault features of the gear under test are located within the common field of view of the projector and camera. 1-2) After the projector and camera have been debugged, disassemble them and place them horizontally on the table with the projector's optical axis aligned. Perform off-site calibration to establish the internal and external parameter models of the projector-camera system: In the formula, M is the product of internal and external parameters. , , These are the three-dimensional spatial coordinates of the object being measured. After calibration, the projector and camera were reinstalled in their original positions for subsequent measurements. The specific steps for calculating the wrapping phase in step 3) are as follows: 3-1) The obtained single-frame deformed fringe pattern As input, a single-frame deformed fringe pattern First, the input single-frame deformed stripe pattern is processed through multi-layer convolution C of the deep learning network GPD-Net. The feature maps are segmented into three different scales. 3-2) The feature maps at different scales are stretched by the F operation, and the stretched feature maps are used as input features that can be read by the transformer module T. 3-3) Use three transformer modules T with different numbers of heads to perform multi-space feature extraction on features at three different scales respectively; 3-4) Learnable feature fusion is performed through an A-CRF module with a gated attention mechanism to fuse multi-scale and multi-spatial features in a principled manner; 3-5) After the decoding module De outputs M and D, calculate the wrap phase of each pixel. : ； The specific steps for calculating the three-dimensional spatial coordinates in step 5) are as follows: 5-1) Convert the solved absolute phase into coordinate values ​​using the following formula. : 5-2) Solve for u p Substituting the projector-camera intrinsic and extrinsic parameter equations obtained in step 1), and using the three known quantities (u) c v c u p Solve the three equations simultaneously to find the three unknowns (X). W Y W Z W ); where u c v c The pixel coordinates of the camera are known.

2. The method for three-dimensional measurement of gear faults based on deep learning in a single frame, as described in claim 1, is characterized in that... The specific steps for data collection in step 2) are as follows: 2-1) Generate a single-frame coded fringe pattern and load it into a projector, then project it onto the surface of the gear being tested. The generation of the single-frame coded fringe pattern... for: In the formula, A and B are background modulations. The wrapping phase of the pixel; 2-2) Camera with The deformed fringe pattern is acquired at time intervals to obtain a single frame of deformed fringe pattern.

3. The method for three-dimensional measurement of gear faults based on deep learning in a single frame of structured light as described in claim 1, characterized in that, Step 4) The specific method for calculating the absolute phase is as follows: Calculate the absolute phase of each pixel in the image based on the obtained wrapper phase: In the formula, The number of stripes represents the stripe order.

4. The method for three-dimensional measurement of gear faults based on deep learning in a single frame, as described in claim 1, is characterized in that... The specific steps for 3D reconstruction and measurement of gear faults in step 6) are as follows: 6-1) Select a fault-free section perpendicular to the tooth tip line as a reference plane, project the point cloud data of the three-dimensional spatial coordinates of the tooth surface on the section onto the section to obtain the point set of the tooth profile, and generate a healthy tooth profile curve through polynomial fitting. 6-2) Project the point cloud data of the three-dimensional spatial coordinates of the tooth surface of other sections onto the reference plane to obtain its tooth profile curve. The fault depth is calculated by the error between the healthy tooth profile curve and the tooth profile curve to be tested.

5. A single-frame structured light gear fault three-dimensional measurement system based on deep learning, comprising a measurement box (1), characterized in that, Includes a gearbox platform assembly (2), a data acquisition assembly (3), and an oil baffle assembly (4) installed inside the measuring box (1); The gearbox platform assembly (2) is used to provide gear fault samples for the measurement system. The gearbox platform assembly (2) includes a gearbox (21) and a gear (22) installed in the gearbox (21). The data acquisition component (3) is used to project a single-frame coded stripe pattern onto the surface of the gear (22) under test and simultaneously acquire stripe images modulated by the surface of the gear (22) under test. The component includes a projector (31) and a CCD industrial camera (32) set above the gearbox (21). The oil baffle assembly (4) is used to block the lubricating oil splashed from the gearbox (21), so that the upper cover of the gearbox (21) has a clean area to facilitate the data acquisition assembly (3) to collect fault images. The oil baffle assembly (4) includes an oil baffle plate (41) integrated on the upper cover of the gearbox (21). It also includes a speed control component for controlling the gear (22) during operation to ensure that the gear (22) is measured and controlled at low speed. And an image processing component for implementing a deep learning-based single-frame structured light gear fault three-dimensional measurement method as described in any one of claims 1-4.

Images