Microscopic visual detection method for quality defects of nozzle ring belt

Abstract

The invention discloses a microscopic visual detection method for quality defects of a nozzle ring belt, which comprises the following steps of: acquiring a positive focal microscopic image of the nozzle ring belt based on a nozzle ring belt microscopic visual detection device; based on a vision measurement algorithm, analyzing and processing the positive focus microscopic image to obtain a size defect detection and identification result of the nozzle ring belt; through a defect detection algorithm based on a convolutional neural network and a deep learning training framework, analyzing and processing the positive focus microscopic image of the nozzle girdle to obtain a surface defect detection and recognition result of the nozzle girdle; and comprehensively evaluating the processing quality of the nozzle ring belt according to the size defect detection and identification result and the surface defect detection and identification result. Based on the microscopic vision imaging and artificial intelligence technology, vision measurement and defect detection are integrated, so that the detection procedures of size defects and surface defects of the nozzle ring belt are combined into one, the detection tasks of the size defects and the surface defects are synchronously completed, and comprehensive detection and judgment of the quality defects of the nozzle ring belt are achieved.

Description

Technical Field

[0001] The present invention belongs to the field of machine vision and deep learning technology, and in particular relates to a microscopic visual detection method for nozzle ring band quality defects. Background Art

[0002] In the electro-hydraulic servo valve used as the core component of the servo control system, nozzle parts are important components for controlling the pressure, speed, direction, flow rate and jet shape of the oil. Therefore, during the manufacturing process, higher quality control requirements are placed on the geometric dimensions and surface quality of its key parts and important structural features, such as the nozzle jet orifice end face ring band (hereinafter referred to as the "nozzle ring band").

[0003] In the electro-hydraulic servo valve, the nozzle ring is located at the front end of the nozzle body, in a circular ring shape, and the size is mostly in the range of 0.3 to 1.0 mm. According to the technical status control requirements of nozzle parts, the quality defects of the nozzle ring mainly include dimensional defects and surface defects. The former refers to the actual geometric shape and position dimensions of the nozzle ring that exceed the tolerance range of the design value, that is, dimensional deviation, such as outer diameter, inner hole diameter, outer circle roundness, inner hole roundness, and inner and outer circle concentricity; the latter refers to certain undesirable conditions on the surface of the nozzle ring, such as scratches, burrs, bumps, and edge defects. In order to screen out qualified nozzle parts that meet the quality control requirements, it is urgent to realize the rapid detection and accurate identification of the ring quality defects of batch nozzle parts.

[0004] At present, in the machine vision inspection of quality defects of existing mechanical parts, size defects and surface defects are usually inspected separately. For the size defects of the nozzle ring, conventional image measuring equipment with optical or image magnification function is used to realize the visual measurement of its actual geometric and shape and position dimensions, which can achieve high measurement accuracy; and for the surface defects of the nozzle ring, the visual inspection methods used are mainly divided into three categories, namely, inspection methods based on digital image processing, machine learning and deep learning. Among them, the inspection method based on deep learning extracts multiple features of the image through the weight parameters in the convolutional neural network, and then learns the distribution law of the data through the deep learning algorithm to complete the classification task. Compared with the inspection methods based on digital image processing and machine learning, this method does not require manual complex image segmentation, feature recognition and other operations to design feature extractors, and can enable more nonlinear mappings to participate in the calculation, so that its learning ability is more powerful, and it has a series of characteristics such as nonlinearity, self-adaptation, self-organization, self-learning, non-limitation, non-convexity and high fault tolerance. Therefore, since the concept of convolutional neural networks was proposed, deep learning algorithms have been widely used in various fields, such as unmanned driving, speech recognition, big data processing, etc., and the use of deep learning methods to detect surface defects of objects can greatly improve the accuracy and stability of detection results.

[0005] In summary, in the existing defect detection methods based on machine vision, the detection tasks of size defects and surface defects are completed separately on different visual inspection equipment with single functions. On the one hand, it increases the detection steps and makes the quality defect detection process complicated, reducing the detection efficiency; on the other hand, the detection processes of size defects and surface defects of the nozzle annulus are separated from each other and cannot be fully integrated. Summary of the invention

[0006] The technology of the present invention solves the problem: overcomes the shortcomings of the prior art, provides a microscopic visual detection method for nozzle ring band quality defects, integrates visual measurement and defect detection based on microscopic visual imaging and artificial intelligence technology, combines the detection processes of size defects and surface defects of the nozzle ring band into one, and completes the detection tasks of the two simultaneously, thereby realizing comprehensive detection and judgment of nozzle ring band quality defects, thereby improving the efficiency and automation and intelligence level of nozzle ring band quality detection.

[0007] In order to solve the above technical problems, the present invention discloses a microscopic visual detection method for nozzle ring belt quality defects, comprising:

[0008] Construct a nozzle annulus microscopic visual inspection device;

[0009] The nozzle to be tested is clamped onto a nozzle annulus microscopic visual detection device, and a positive-focus microscopic image of the nozzle annulus of the nozzle to be tested is acquired by the nozzle annulus microscopic visual detection device;

[0010] Based on the visual measurement algorithm, the positive focus microscopic image of the nozzle ring is analyzed and processed to obtain the key size parameters; and the size defect detection and identification results of the nozzle ring are obtained based on the comparison results of the key size parameters and the corresponding parameter design values;

[0011] Through the defect detection algorithm based on convolutional neural network and deep learning training framework, the positive focus microscopic image of the nozzle ring belt is analyzed and processed to obtain the surface defect detection and identification results of the nozzle ring belt;

[0012] According to the obtained size defect detection and identification results and surface defect detection and identification results of the nozzle ring band, the processing quality of the nozzle ring band is comprehensively evaluated to obtain a comprehensive evaluation result.

[0013] In the above-mentioned microscopic visual inspection method for nozzle ring band quality defects, the nozzle ring band microscopic visual inspection device includes: a base, an XY-axis motion module, a Z-axis motion module, a special fixture, a microscopic imaging module, a column, a controller, a host computer and a light source module; wherein, the XY-axis motion module and the column are installed on the horizontal surface of the base, and the XY-axis motion module is located in front of the column; the Z-axis motion module is installed on the front surface of the column in the vertical direction; the light source module is installed on the microscopic imaging module, and the microscopic imaging module is installed on the movable pallet of the Z-axis motion module; the special fixture is installed on the movable pallet of the XY-axis motion module; the host computer is connected to the controller through a transmission control cable, and the controller is connected to the XY-axis motion module, the Z-axis motion module, the microscopic imaging module and the light source module through a transmission control cable.

[0014] In the above-mentioned microscopic visual inspection method for nozzle annulus quality defects, the nozzle to be tested is clamped on the nozzle annulus microscopic visual inspection device, and the nozzle annulus microscopic visual inspection device is used to collect a positive focus microscopic image of the nozzle annulus of the nozzle to be tested, including:

[0015] Clamp the nozzle to be tested: Clamp the nozzle to be tested on a special fixture so that the nozzle ring of the nozzle to be tested faces the microscopic imaging module, and make the geometric axis of the nozzle to be tested parallel to the Z-axis direction through mechanical adjustment and alignment. At this time, the nozzle ring is in a position parallel to the XY plane;

[0016] Setting the focus parameters: Adjust the relative posture between the nozzle to be tested and the microscopic imaging module through the XY-axis motion module and the Z-axis motion module, so that the nozzle ring belt completely enters the field of view of the microscopic imaging module, and then set the focus parameters in the host computer;

[0017] Autofocus: The host computer generates control instructions based on the autofocus algorithm and sends the control instructions to the controller; the controller controls the Z-axis motion module to drive the microscopic imaging module to move according to the control instructions, and completes the autofocus of the nozzle ring;

[0018] Image acquisition: After the autofocus is completed, the microscopic imaging module is used to acquire the in-focus microscopic image of the nozzle annulus.

[0019] In the above-mentioned microscopic visual inspection method for quality defects of the nozzle annulus, based on the visual measurement algorithm, the in-focus microscopic image of the nozzle annulus is analyzed and processed to obtain key size parameters; and the size defect detection and identification result of the nozzle annulus is obtained according to the comparison result of the key size parameters and the corresponding parameter design values, including: sending the in-focus microscopic image of the nozzle annulus to the host computer; the host computer calls the size defect detection module to perform image processing on the in-focus microscopic image of the nozzle annulus to obtain the outer edge and inner edge of the nozzle annulus area in the in-focus microscopic image; performing least squares circle fitting on the outer edge and the inner edge respectively to obtain the pixel radius R of the outer edge fitting circlep1 , pixel roundness error δ p1 , pixel center coordinates (u1, v1), and pixel radius R of the inner edge fitting circle p2 , pixel roundness error δ p2 and pixel center coordinates (u2, v2); according to R p1 ,δ p1 、(u1,v1),R p2 ,δ p2 and (u2, v2), combined with the pixel size equivalent K of the calibrated nozzle annulus microscopic visual inspection device, the outer diameter Φ1, inner hole diameter Φ2, outer circle roundness δ1, inner hole roundness δ2 and inner and outer circle concentricity δ of the nozzle annulus are calculated. c ; The calculated outer diameter Φ1, inner diameter Φ2, outer roundness δ1, inner roundness δ2 and inner and outer concentricity δ c The results of nozzle annulus dimensional defect detection and identification are obtained by comparing them with the corresponding parameter design values.

[0020] In the above-mentioned microscopic visual inspection method for nozzle ring quality defects, Φ1=2·K·R p1 ,Φ2=2·K·R p2 ,δ1=K·δ p1 , δ2=K·δ p2 ,

[0021] In the above-mentioned microscopic visual inspection method for the quality defects of the nozzle ring band, the calculated outer diameter Φ1, inner hole diameter Φ2, outer circle roundness δ1, inner hole roundness δ2 and inner and outer circle concentricity δ c The nozzle annulus size defect detection and identification results are obtained by comparing with the corresponding parameter design values, including: the calculated key size parameters of the nozzle annulus are recorded as Q i If Q i Exceeds the design value Q i ′ tolerance range, it is determined that the nozzle ring has a corresponding dimensional defect; where i = 1, 2, 3, 4, 5, Q1 = Φ1, Q2 = Φ2, Q3 = δ1, Q4 = δ2, Q5 = δ c , Q1′=Φ1′, Q2′=Φ′2, Q3′=δ1′, Q4′=δ2′, Q5′=δ c ′, Φ1′ represents the design value of the outer diameter of the nozzle annulus, Φ′2 represents the design value of the inner hole diameter of the nozzle annulus, δ1′ represents the design value of the outer roundness of the nozzle annulus, δ2′ represents the design value of the inner hole roundness of the nozzle annulus, δ c ' represents the design value of the concentricity of the inner and outer circles of the nozzle ring.

[0022] In the above-mentioned microscopic visual inspection method for quality defects of the nozzle ring band, the in-focus microscopic image of the nozzle ring band is analyzed and processed by a defect detection algorithm based on a convolutional neural network and a deep learning training framework to obtain a surface defect detection and identification result of the nozzle ring band, including: sending the in-focus microscopic image of the nozzle ring band to a host computer; the host computer calls a surface defect detection module, and uses a pre-trained deep learning detection model based on a combined neural network to analyze and process the in-focus microscopic image to obtain a surface defect detection and identification result of the nozzle ring band.

[0023] In the above-mentioned microscopic visual inspection method for nozzle ring band quality defects, the dimensional defect detection and identification results include: the nozzle ring band has no dimensional defects and the nozzle ring band has dimensional defects; among them, the types of dimensional defects include: outer circle diameter out of tolerance, inner hole diameter out of tolerance, outer circle roundness out of tolerance, inner hole roundness out of tolerance and inner and outer circle concentricity out of tolerance.

[0024] In the above-mentioned microscopic visual inspection method for nozzle ring band quality defects, the surface defect detection and identification results include: the nozzle ring band has no surface defects and the nozzle ring band has surface defects; among which, the types of surface defects include: scratches, burrs, bumps and edge defects.

[0025] In the above-mentioned microscopic visual inspection method for quality defects of the nozzle ring band, the processing quality of the nozzle ring band is comprehensively evaluated based on the obtained size defect detection and identification results and surface defect detection and identification results of the nozzle ring band, and a comprehensive evaluation result is obtained, including: when the size defect detection and identification result is that the nozzle ring band has no size defects, and the surface defect detection and identification result is that the nozzle ring band has no surface defects, the processing quality of the nozzle ring band is determined to be qualified; when the size defect detection and identification result is that the nozzle ring band has size defects, and / or the surface defect detection and identification result is that the nozzle ring band has surface defects, the processing quality of the nozzle ring band is determined to be unqualified.

[0026] The present invention has the following advantages:

[0027] (1) The present invention discloses a microscopic visual detection method for nozzle ring band quality defects, which integrates visual measurement with defect detection, combines the size defect and surface defect detection of the nozzle ring band, realizes non-contact, automated detection and comprehensive judgment of the size defect and surface defect of the nozzle ring band, and improves the efficiency and automation level of nozzle ring band quality detection.

[0028] (2) The present invention discloses a microscopic visual detection method for nozzle ring band quality defects. In the process of nozzle ring band surface defect detection, a deep learning detection model based on a combined neural network is adopted. The model adopts a structural form based on a convolutional neural network and a deep learning training framework, which combines the advantages of the number of layers and parameters of advanced neural networks with the reliability of a mature training framework. By combining cutting-edge algorithms with mature frameworks, accurate detection and identification of nozzle ring band surface defects can be achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 is a flow chart of a method for microscopic visual inspection of nozzle ring quality defects in an embodiment of the present invention;

[0030] Figure 2 is a schematic structural diagram of a nozzle to be tested in an embodiment of the present invention;

[0031] Figure 3 is a structural schematic diagram of a nozzle annulus microscopic visual inspection device in an embodiment of the present invention;

[0032] Figure 4 is a schematic diagram of a positive focus microscopic image of a nozzle ring zone in an embodiment of the present invention;

[0033] Figure 5 is a schematic diagram of the extraction results of the outer edge and inner edge of the nozzle annulus region in an embodiment of the present invention;

[0034] Figure 6 Schematic diagram of a nozzle ring with excessive concentricity of inner and outer circles in an embodiment of the present invention;

[0035] Figure 7 is a schematic diagram of a nozzle ring band with scratches in an embodiment of the present invention;

[0036] Figure 8 Schematic diagram of a nozzle ring with edge defects in an embodiment of the present invention. DETAILED DESCRIPTION

[0037] In order to make the objectives, technical solutions and advantages of the present invention more clear, the embodiments disclosed in the present invention will be further described in detail below with reference to the accompanying drawings.

[0038] One of the core ideas of the present invention is to provide a microscopic visual detection method for quality defects of a nozzle ring band, which is implemented by relying on a microscopic visual detection device for the nozzle ring band. First, a positive-focus microscopic image of the nozzle ring band is obtained, and then the geometric shape and position dimensions of the nozzle ring band, such as the outer diameter, inner hole diameter, outer circle roundness, inner hole roundness, and inner and outer circle concentricity, are accurately measured through a visual measurement algorithm to identify and judge the size defects of the nozzle ring band; at the same time, accurate detection of adverse conditions such as scratches, burrs, bumps, and edge defects of the nozzle ring band is achieved through a defect detection algorithm based on a convolutional neural network and a deep learning training framework to identify and judge the surface defects of the nozzle ring band.

[0039] In this embodiment, if Figure 1 As shown, the microscopic visual inspection method for the nozzle ring quality defects includes:

[0040] Step 1: construct a nozzle ring microscopic visual inspection device.

[0041] In this embodiment, the nozzle annulus microscopic visual inspection device adopts a forward lighting method, adopts a column-type overall structure, and has three mutually perpendicular linear motion axes: X-axis, Y-axis and Z-axis, which can realize single-axis motion and three-axis linkage; the X-axis and Y-axis are located in the horizontal plane, the Z-axis is in the vertical direction, and conforms to the right-hand rule; the X-axis and the Y-axis are integrated together and realized by the XY-axis motion module; the Z-axis is arranged separately and realized by the Z-axis motion module. Figure 3 As shown, the nozzle annulus microscopic visual inspection device mainly includes: a base 21, an XY axis motion module 22, a Z axis motion module 23, a special fixture 24, a microscopic imaging module 25, a column 26, a controller 27, a host computer 28 and a light source module 29. Among them: the XY axis motion module 22 and the column 26 are installed on the horizontal surface of the base 21, and the XY axis motion module 22 is located in front of the column 26; the Z axis motion module 23 is installed on the front surface of the column 26 along the vertical direction; the XY axis motion module 22 and the Z axis motion module 23 are equipped with high-precision linear guides and linear grating rulers to ensure the movement stability and linear displacement accuracy of each axis. The light source module 29 is installed on the microscopic imaging module 25, and the microscopic imaging module 25 is installed on the moving support plate of the Z axis motion module 23; the microscopic imaging module 25 faces the negative direction of the Z axis, and the imaging optical axis direction of the microscopic imaging module 25 is parallel to the Z axis direction through mechanical adjustment. The Z-axis motion module 23 can drive the microscopic imaging module 25 and the light source module 29 to move along the Z-axis direction. The special fixture 24 is installed on the moving support plate of the XY-axis motion module 22, and the special fixture 24 can be driven to move along the X and Y-axis directions through the XY-axis motion module 22. The nozzle 1 to be tested is clamped in the special fixture 24 so that the geometric axis direction of the nozzle 1 to be tested is parallel to the Z-axis direction. The structure of the nozzle 1 to be tested is as follows: Figure 2As shown; through the X-direction and Y-direction movement of the XY-axis motion module 22 and the Z-direction movement of the Z-axis motion module 23, the nozzle 1 to be tested can be moved to the right below the microscopic imaging module 25, so that the nozzle ring belt 11 completely enters the field of view of the microscopic imaging module 25, and the uniform and stable light emitted by the light source module 29 is emitted from the front of the nozzle 1 to be tested and illuminates the nozzle ring belt 11. After the diffuse reflection light enters the microscopic imaging module 25, it is imaged on the photosensitive surface, and the positive focus microscopic image of the nozzle ring belt 11 can be obtained. The host computer 28 is connected to the controller 27 through a transmission control cable, and the controller 27 is connected to the XY-axis motion module 22, the Z-axis motion module 23, the microscopic imaging module 25 and the light source module 29 through a transmission control cable. The host computer 28 is configured with a host computer software, which can provide a graphical human-machine interface for flexible operation by the operator, and has the functions of system parameter setting, detection program compilation, data acquisition, graphic fitting, surface defect recognition, geometric parameter solution and output display.

[0042] Step 2: Clamp the nozzle 1 to be tested onto the nozzle annulus microscopic visual inspection device, and obtain a focus microscopic image of the nozzle annulus 11 of the nozzle 1 to be tested through the nozzle annulus microscopic visual inspection device.

[0043] In this embodiment, the specific implementation process of step 2 is as follows:

[0044] Clamp the nozzle 1 to be tested: Clamp the nozzle 1 to be tested on the special fixture 24 so that the nozzle ring 11 of the nozzle 1 to be tested faces the microscopic imaging module 25, and make the geometric axis of the nozzle 1 to be tested parallel to the Z-axis direction through mechanical adjustment and alignment. At this time, the nozzle ring 11 is in a position parallel to the XY plane.

[0045] Setting the focus parameters: adjusting the relative position of the nozzle 1 to be tested and the microscopic imaging module 25 through the XY axis motion module 22 and the Z axis motion module 23, so that the nozzle ring belt 11 completely enters the field of view of the microscopic imaging module 25, and then setting the focus parameters such as the light source illumination intensity, focus window, starting position, focus interval and step length in the autofocus process in the host computer 28. Among them, when setting the light source illumination intensity in the autofocus process, there are: adjusting the light source illumination intensity in the autofocus process to the optimal illumination intensity, so that the surface texture and detail features of the nozzle ring belt 11 can be well presented in the field of view of the microscopic imaging module 25, and more details and texture features can be observed in the visual effect, without obvious overexposure or obvious underexposure.

[0046] Autofocus: The host computer 28 generates a control instruction based on the autofocus algorithm and sends the control instruction to the controller 27; the controller 27 controls the Z-axis motion module 23 to drive the microscopic imaging module 25 to move according to the control instruction, and completes the autofocus of the nozzle ring 11. The autofocus algorithm includes: a clarity evaluation function and a peak search algorithm.

[0047] The clarity evaluation function is implemented by the improved Prewitt edge detection operator, which increases the gradient calculation direction of the original Prewitt operator from 2 to 4 to improve the sensitivity and stability of the clarity calculation. For an image region I with a pixel number of M×N, the clarity evaluation function is defined as:

[0048]

[0049] Where D(I) represents the clarity calculation result of image area I, that is, the clarity value; (x, y) represents the coordinates of the pixel in image area I; G x (x,y) and G y (x, y) represents the gradient convolution template of pixel (x, y) in the horizontal and vertical directions respectively; G xy (x,y) and G yx (x, y) represents the gradient convolution template added at the pixel (x, y) in the +45° and -45° directions respectively; M represents the number of rows of pixels in the image area I, and N represents the number of columns of pixels in the image area I;

[0050] The peak search algorithm is implemented through curve fitting method. The clarity value of each image and its position or coordinates form a sample data point set, and then a suitable second-order or third-order function is selected to perform curve fitting on the data point set, so as to approximate the clarity curve and obtain its peak value through analysis.

[0051] Image acquisition: After the automatic focusing is completed, the precise focus position is obtained, and the microscopic image of the nozzle ring zone 11 is acquired at this position through the microscopic imaging module 25, such as Figure 4 shown.

[0052] Step 3, based on the visual measurement algorithm, the focus microscope image of the nozzle ring band 11 is analyzed and processed to obtain key dimensional parameters; and based on the comparison results of the key dimensional parameters with the corresponding parameter design values, the dimensional defect detection and identification results of the nozzle ring band are obtained.

[0053] In this embodiment, the in-focus microscopic image of the nozzle annulus 11 is sent to the host computer 28; the host computer 28 calls the size defect detection module to filter and reduce noise on the in-focus microscopic image of the nozzle annulus 11 (mean filtering, median filtering or Gaussian filtering can be used to suppress noise in the collected in-focus microscopic image), threshold segmentation (a grayscale threshold determined based on the maximum inter-class variance method can be used to segment the denoised image into a foreground area and a background area), disconnected area division (disconnected area division is to further divide the foreground area into several independent sub-areas that are not connected to each other based on whether they are connected or not), and target area selection (the sub-areas can be set to select the target area). The nozzle annulus region is accurately selected from several independent sub-regions that are not connected to each other and is used as the target region based on the area (the number of pixels contained in the sub-region) and shape of the domain, and the target region is used as the target region), hole filling (filling the holes and the like in the target region to make it a closed and complete annular figure), morphological operation (opening operation of first corrosion and then expansion can be used to disconnect small connections and eliminate small protrusions to make the inner and outer edges of the annular figure smooth), and sub-pixel edge extraction (moment estimation method, interpolation method, fitting method, etc. can be used) and other image processing to obtain the outer edge 3 and inner edge 4 of the nozzle annulus region in the positive focus microscopic image, as shown in FIG. Figure 5 As shown; further, the outer edge and the inner edge are respectively fitted with the least squares circle to obtain the pixel radius R of the outer edge fitting circle p1 , pixel roundness error δ p1 , pixel center coordinates (u1, v1), and pixel radius R of the inner edge fitting circle p2 , pixel roundness error δ p2 and pixel center coordinates (u2, v2); according to R p1 ,δ p1 、(u1,v1),R p2 ,δ p2 and (u2, v2), combined with the pixel size equivalent K of the calibrated nozzle annulus microscopic visual inspection device, the outer diameter Φ1, inner hole diameter Φ2, outer circle roundness δ1, inner hole roundness δ2 and inner and outer circle concentricity δ of the nozzle annulus 11 are calculated. c ; The calculated outer diameter Φ1, inner diameter Φ2, outer roundness δ1, inner roundness δ2 and inner and outer concentricity δ of the nozzle ring 11 c The nozzle annulus size defect detection and identification results are obtained by comparing with the corresponding parameter design values. It should be noted that before the microscopic visual inspection of the quality defects of the nozzle annulus, the pixel size equivalent K of the nozzle annulus microscopic visual inspection device is accurately calibrated using a target with good shape and position accuracy and known geometric size information.

[0054] Preferably, the outer diameter Φ1, the inner diameter Φ2, the outer roundness δ1, the inner roundness δ2 and the inner and outer concentricity δ c The calculation formula is as follows:

[0055] Φ1=2·K·R p1

[0056] Φ2=2·K·R p2

[0057] δ1=K·δ p1

[0058] δ2=K·δ p2

[0059]

[0060] Preferably, the outer diameter Φ1, inner diameter Φ2, outer roundness δ1, inner roundness δ2 and inner and outer concentricity δ1 of the nozzle annulus 11 are calculated. c The comparison result with the corresponding parameter design value can determine whether the nozzle ring band 11 has a dimensional defect and what kind of dimensional defect it is. Specifically, the calculated key dimensional parameter of the nozzle ring band 11 can be recorded as Q i If Q i Exceeds the design value Q i ′ tolerance range, it is determined that the nozzle ring 11 has a corresponding dimensional defect; where i = 1, 2, 3, 4, 5, Q1 = Φ1, Q2 = Φ2, Q3 = δ1, Q4 = δ2, Q5 = δ c , Q1′=Φ1′, Q2′=Φ′2, Q3′=δ1′, Q4′=δ2′, Q5′=δ c ′, Φ1′ represents the design value of the outer diameter of the nozzle annulus, Φ′2 represents the design value of the inner hole diameter of the nozzle annulus, δ1′ represents the design value of the outer roundness of the nozzle annulus, δ2′ represents the design value of the inner hole roundness of the nozzle annulus, δ c ' represents the design value of the inner and outer circle concentricity of the nozzle annulus. For example, if the calculated inner and outer circle concentricity δ c Exceeds the design value of inner and outer circle concentricity δ c ′, it is determined that the nozzle ring has a dimensional defect, and the type of dimensional defect is the concentricity of the inner and outer circles is out of tolerance, such as Figure 6 shown.

[0061] Step 4: Analyze and process the positive focus microscopic image of the nozzle ring band 11 through a defect detection algorithm based on a convolutional neural network and a deep learning training framework to obtain a surface defect detection and identification result of the nozzle ring band.

[0062] In this embodiment, the in-focus microscopic image of the nozzle annulus 11 is sent to the host computer 28; the host computer 28 calls the surface defect detection module, and uses a pre-trained deep learning detection model based on a combined neural network to analyze and process the in-focus microscopic image to obtain a surface defect detection and identification result of the nozzle annulus.

[0063] Preferably, the deep learning detection model based on the combined neural network can adopt the structure of "YOLOv7 neural network + PaddlePaddle artificial intelligence learning framework", use the PaddlePaddle framework to import the training set and YOLOv7 neural network, adjust the learning rate, gradient descent rate, neural network hyperparameters and other indicators, continuously optimize the learning process, and import the learning results into the test set to evaluate the training effect.

[0064] Furthermore, before conducting microscopic visual inspection of quality defects of the nozzle ring band, a certain number of sample images of the nozzle ring band can be collected. These sample images should include images with four major types of surface defects such as scratches, burrs, bumps, and edge defects, as well as defect-free images. The marking classification, entry recognition, algorithm modeling, parameter fine-tuning, model training and other processes of various surface defects should be completed, thereby forming a pre-trained deep learning detection model based on a combined neural network. At the same time, a model recovery training window is reserved, and the operator can classify the newly taken images and enter them into the system for supplementary training to continuously improve the detection accuracy of the model.

[0065] Step 5: Based on the obtained size defect detection and recognition results and surface defect detection and recognition results of the nozzle ring band, a comprehensive evaluation is performed on the processing quality of the nozzle ring band 11 to obtain a comprehensive evaluation result.

[0066] In this embodiment, the size defect detection and identification results include: the nozzle ring has no size defects and the nozzle ring has size defects; further, the types of size defects include: outer diameter tolerance, inner hole diameter tolerance, outer circle roundness tolerance, inner hole roundness tolerance and inner and outer circle concentricity tolerance. The surface defect detection and identification results include: the nozzle ring has no surface defects and the nozzle ring has surface defects; further, the types of surface defects include: scratches (such as Figure 7 ), burrs, bumps and edge defects (such as Figure 8As shown), etc. Among them, when the size defect detection and identification result is that the nozzle annulus has no size defect, and the surface defect detection and identification result is that the nozzle annulus has no surface defect, the processing quality of the nozzle annulus is determined to be qualified. When the size defect detection and identification result is that the nozzle annulus has a size defect, and / or the surface defect detection and identification result is that the nozzle annulus has a surface defect, the processing quality of the nozzle annulus is determined to be unqualified; further, if there is a size defect, the type of size defect and the actual measurement result can be further output; if there is a surface defect, the type of surface defect, the position of the annotation box and the confidence level can be further output.

[0067] After the above steps, the comprehensive detection task of the quality defects of the nozzle ring band is finally completed, thereby providing a detection technology support for improving the processing quality and detection efficiency of the nozzle ring band.

[0068] Although the present invention has been disclosed as above in the form of a preferred embodiment, it is not intended to limit the present invention. Any person skilled in the art may make possible changes and modifications to the technical solution of the present invention by using the methods and technical contents disclosed above without departing from the spirit and scope of the present invention. Therefore, any simple modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention shall fall within the protection scope of the technical solution of the present invention.

[0069] The contents not described in detail in the specification of the present invention belong to the common knowledge of the professionals in this field.

Claims

1. A microscopic visual detection method for nozzle ring quality defects, characterized in that: include: Construct a nozzle annulus microscopic visual inspection device; The nozzle (1) to be tested is clamped onto a nozzle annulus microscopic visual detection device, and a positive-focus microscopic image of the nozzle annulus (11) of the nozzle (1) to be tested is acquired by the nozzle annulus microscopic visual detection device; Based on a visual measurement algorithm, a positive focus microscopic image of the nozzle ring band (11) is analyzed and processed to obtain key dimensional parameters; and based on a comparison result between the key dimensional parameters and corresponding parameter design values, a dimensional defect detection and identification result of the nozzle ring band is obtained; By using a defect detection algorithm based on a convolutional neural network and a deep learning training framework, a positive focus microscopic image of the nozzle ring band (11) is analyzed and processed to obtain a surface defect detection and recognition result of the nozzle ring band; Based on the obtained dimensional defect detection and identification results and surface defect detection and identification results of the nozzle ring band, a comprehensive evaluation is performed on the processing quality of the nozzle ring band (11) to obtain a comprehensive evaluation result.

2. The microscopic visual detection method for nozzle ring band quality defects according to claim 1 is characterized in that: A nozzle annulus microscopic visual inspection device comprises: a base (21), an XY axis motion module (22), a Z axis motion module (23), a special fixture (24), a microscopic imaging module (25), a column (26), a controller (27), a host computer (28) and a light source module (29); wherein the XY axis motion module (22) and the column (26) are installed on the horizontal surface of the base (21), and the XY axis motion module (22) is located in front of the column (26); the Z axis motion module (23) is installed on the column in the vertical direction. (26); the light source module (29) is mounted on the microscopic imaging module (25), and the microscopic imaging module (25) is mounted on the movable support plate of the Z-axis motion module (23); the special fixture (24) is mounted on the movable support plate of the XY-axis motion module (22); the host computer (28) is connected to the controller (27) through a transmission control cable, and the controller (27) is connected to the XY-axis motion module (22), the Z-axis motion module (23), the microscopic imaging module (25) and the light source module (29) through a transmission control cable.

3. The microscopic visual detection method for nozzle ring band quality defects according to claim 2 is characterized in that: The nozzle (1) to be tested is clamped onto a nozzle annulus microscopic visual detection device, and a positive-focus microscopic image of the nozzle annulus (11) of the nozzle (1) to be tested is collected by the nozzle annulus microscopic visual detection device, including: Clamping the nozzle (1) to be tested: clamping the nozzle (1) to be tested on a special fixture (24) so ​​that the nozzle ring (11) of the nozzle (1) to be tested faces the microscopic imaging module (25), and through mechanical adjustment and alignment, the geometric axis of the nozzle (1) to be tested is parallel to the Z-axis direction, at which time the nozzle ring (11) is in a position parallel to the XY plane; Setting the focus parameters: adjusting the relative position of the nozzle (1) to be tested and the microscopic imaging module (25) through the XY axis motion module (22) and the Z axis motion module (23) so that the nozzle ring belt (11) completely enters the field of view of the microscopic imaging module (25), and then setting the focus parameters in the host computer (28); Automatic focusing: The host computer (28) generates a control instruction based on the automatic focusing algorithm and sends the control instruction to the controller (27); the controller (27) controls the Z-axis motion module (23) to drive the microscopic imaging module (25) to move according to the control instruction, thereby completing the automatic focusing of the nozzle ring belt (11); Image acquisition: After the automatic focusing is completed, a positive focus microscopic image of the nozzle ring zone (11) is acquired through the microscopic imaging module (25).

4. The microscopic visual detection method for nozzle ring band quality defects according to claim 3 is characterized in that: Based on the visual measurement algorithm, the positive focus microscopic image of the nozzle ring (11) is analyzed and processed to obtain the key size parameters; And according to the comparison result of the key size parameter and the corresponding parameter design value, the size defect detection and identification result of the nozzle ring band is obtained, including: sending the positive focus microscopic image of the nozzle ring band (11) to the host computer (28); The upper computer (28) calls the size defect detection module to perform image processing on the positive focus microscopic image of the nozzle annulus (11) to obtain the outer edge and inner edge of the nozzle annulus region in the positive focus microscopic image; the outer edge and the inner edge are respectively fitted with the least squares circle to obtain the pixel radius R of the outer edge fitting circle. p1 , pixel roundness error δ p1 , pixel center coordinates (u1, v1), and pixel radius R of the inner edge fitting circle p2 , pixel roundness error δ p2 and pixel center coordinates (u2, v2); according to R p1 , δ p1 、(u1,v1),R p2 , δ p2 and (u2, v2), combined with the pixel size equivalent K of the calibrated nozzle annulus microscopic visual detection device, the outer circle diameter Φ1, inner hole diameter Φ2, outer circle roundness δ1, inner hole roundness δ2 and inner and outer circle concentricity δ of the nozzle annulus (11) are calculated. c The calculated outer diameter Φ1, inner diameter Φ2, outer roundness δ1, inner roundness δ2 and inner and outer concentricity δ of the nozzle annulus (11) are c The results of nozzle ring size defect detection and identification are obtained by comparing them with the corresponding parameter design values.

5. The method for microscopic visual inspection of nozzle ring quality defects according to claim 4, characterized in that: Φ1=2·K·R p1 ,Φ2=2·K·R p2 ,δ1=K·δ p1 ,δ2=K·δ p2 , 6. The method for microscopic visual inspection of nozzle ring quality defects according to claim 4, characterized in that: The calculated outer diameter Φ1, inner diameter Φ2, outer roundness δ1, inner roundness δ2 and inner and outer concentricity δ c and comparing them with the corresponding parameter design values ​​to obtain the dimensional defect detection and identification results of the nozzle annulus, including: recording the calculated key dimensional parameters of the nozzle annulus (11) as Q i If Q i Exceeds the design value Q i ′ tolerance range, it is determined that the nozzle ring (11) has a corresponding dimensional defect; wherein i = 1, 2, 3, 4, 5, Q1 = Φ1, Q2 = Φ2, Q3 = δ1, Q4 = δ2, Q5 = δ c , Q1′=Φ1′, Q2′=Φ′2, Q3′=δ1′, Q4′=δ2′, Q5′=δ c ′, Φ1′ represents the design value of the outer diameter of the nozzle annulus, Φ′2 represents the design value of the inner hole diameter of the nozzle annulus, δ1′ represents the design value of the outer roundness of the nozzle annulus, δ2′ represents the design value of the inner hole roundness of the nozzle annulus, δ c ' represents the design value of the concentricity of the inner and outer circles of the nozzle annulus.

7. The method for microscopic visual inspection of nozzle ring quality defects according to claim 4, characterized in that: The method comprises: using a defect detection algorithm based on a convolutional neural network and a deep learning training framework, analyzing and processing the in-focus microscopic image of the nozzle annulus (11) to obtain a surface defect detection and identification result of the nozzle annulus, including: sending the in-focus microscopic image of the nozzle annulus (11) to a host computer (28); the host computer (28) calling a surface defect detection module, using a pre-trained deep learning detection model based on a combined neural network, analyzing and processing the in-focus microscopic image to obtain a surface defect detection and identification result of the nozzle annulus.

8. The method for microscopic visual inspection of nozzle ring quality defects according to claim 7, characterized in that: The results of dimensional defect detection and identification include: no dimensional defects in the nozzle ring and dimensional defects in the nozzle ring; among them, the types of dimensional defects include: outer circle diameter out of tolerance, inner hole diameter out of tolerance, outer circle roundness out of tolerance, inner hole roundness out of tolerance and inner and outer circle concentricity out of tolerance.

9. The method for microscopic visual inspection of nozzle ring quality defects according to claim 8, characterized in that: The surface defect detection and identification results include: the nozzle ring has no surface defects and the nozzle ring has surface defects; among which, the types of surface defects include: scratches, burrs, bumps and edge defects.

10. The method for microscopic visual inspection of nozzle ring quality defects according to claim 9, characterized in that: Based on the obtained size defect detection and identification results and surface defect detection and identification results of the nozzle ring band, a comprehensive evaluation is performed on the processing quality of the nozzle ring band (11), and a comprehensive evaluation result is obtained, including: When the size defect detection and identification result shows that the nozzle ring band has no size defects, and the surface defect detection and identification result shows that the nozzle ring band has no surface defects, the processing quality of the nozzle ring band (11) is determined to be qualified; when the size defect detection and identification result shows that the nozzle ring band has size defects, and / or the surface defect detection and identification result shows that the nozzle ring band has surface defects, the processing quality of the nozzle ring band (11) is determined to be unqualified.

Images