A motor appearance defect recognition method based on machine vision

Abstract

A motor appearance defect recognition method based on machine vision relates to the motor recognition technical field; the motor shell is divided into different components, a first recognition model is constructed according to different components and their appearance images, the first recognition model is used to obtain the local images of each component in different overall images, different defect types of the motor shell and their defect images are obtained, a second recognition model is constructed in combination with the first recognition model, the second recognition model is used to obtain the recognition accuracy of different local images of the same component, and a third recognition model is constructed in combination with the corresponding local image recognition parameter set, and the second recognition model and the third recognition model are used for defect recognition of the motor shell and obtaining the evaluation accuracy; the accuracy of the defect recognition result of the motor shell can be obtained at the same time when the motor shell is recognized, which is beneficial to providing multiple references for relevant personnel.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of motor recognition, and particularly relates to a motor appearance defect recognition method based on machine vision. BACKGROUND

[0002] As a key power equipment in the industrial field, the appearance quality of a motor directly affects performance and reliability. Traditional manual detection is low in efficiency, unstable in precision and susceptible to subjective factors. Therefore, a motor appearance defect recognition method based on machine vision is developed. By using advanced image processing and analysis technology, efficient and accurate defect recognition can be achieved.

[0003] Image recognition of a motor shell is a mature existing technology. However, the existing technology generally recognizes the motor shell as a whole, and ignores that the motor shell also includes different components, failing to recognize different components respectively, resulting in lack of pertinence in defect recognition.

[0004] Although the efficiency of image recognition is significantly improved compared with manual detection, the image itself is affected by the collection direction, so that the final defect recognition result inevitably lacks accuracy. There is no method for evaluating the accuracy of the defect recognition result. In view of the deficiencies of the prior art, the present application provides a motor appearance defect recognition method based on machine vision. SUMMARY

[0005] The present application aims to provide a motor appearance defect recognition method based on machine vision.

[0006] The purpose of the present application can be achieved by the following technical scheme: a motor appearance defect recognition method based on machine vision, comprising the following steps:

[0007] Step S1: dividing the motor shell into different components and obtaining the appearance images of each component, and constructing a first recognition model according to different components and their corresponding appearance images;

[0008] Step S2: obtaining overall images of the motor shell at different positions, and obtaining local images of each component in different overall images by using the first recognition model;

[0009] Step S3: obtaining different defect types of the motor shell and their corresponding defect images, and constructing a second recognition model in combination with the first recognition model, and obtaining the recognition accuracy of different local images of the same component by using the second recognition model;

[0010] Step S4: According to the recognition accuracy of different components in different local images, a third recognition model is constructed combined with the recognition parameter set of the corresponding local image. The motor shell is recognized for defects by using the second recognition model and the third recognition model, and the corresponding evaluation accuracy is obtained.

[0011] Further, the process of dividing the motor shell into different components and obtaining the appearance image of each component includes:

[0012] The components include the base, the end cover, the flange, the heat sink, the mounting foot, the bearing chamber, the ventilation port / protective screen, and the appearance image of each component of the motor shell is obtained, which is a multi-angle image of the corresponding component in the corresponding area of the motor shell.

[0013] Further, the process of constructing the first recognition model according to different components and their corresponding appearance images includes:

[0014] A first recognition set is generated according to the appearance images of different components and their corresponding names, and is divided into a first training set and a first test set;

[0015] A first convolutional neural network is constructed, different appearance images in the first training set are used as input data of the first convolutional neural network, and the corresponding components and their names in the first training set are used as output data of the first convolutional neural network. The first convolutional neural network is trained using the first training set to obtain an initial first convolutional neural network;

[0016] The first test set is used to verify the model of the initial first convolutional neural network, and the initial first convolutional neural network with an output less than or equal to a preset first test error threshold is output as the first recognition model.

[0017] Further, the process of obtaining the overall image of the motor shell at different positions and obtaining the local image of each component in the different overall images by using the first recognition model includes:

[0018] The motor shell with defects is obtained, and the overall image of the motor shell with defects at different positions is obtained, which is an image of the motor shell collected by an image collection device at a single collection position;

[0019] All overall images of different motor shells are input into the first recognition model to obtain the components and their names contained in each overall image, and the image of each component in the corresponding area of each overall image is taken as its local image.

[0020] Further, the process of obtaining different defect types of the motor shell and their corresponding defect images and constructing a second recognition model combined with the first recognition model includes:

[0021] The defect types include pores, shrinkage, notches, sand holes, burrs, scratches, knife marks, oxidation, rust, cracks, and the defect images of different defect types are obtained respectively, the defect image being the multi-angle image of the corresponding defect type in the corresponding area on the motor shell;

[0022] A second recognition set is generated according to the defect images of different defect types and the corresponding names thereof, and is divided into a second training set and a second test set;

[0023] Different defect images in the second training set are taken as the input data of the first recognition model, and the corresponding defect types and names thereof in the second training set are taken as the output data of the first recognition model, the first recognition model is trained by using the second training set to obtain an initial first recognition model;

[0024] The initial first recognition model is verified by using the second test set, and the initial first recognition model with an output less than or equal to a preset second test error threshold is taken as a second recognition model.

[0025] Further, the process of obtaining the recognition accuracy of different local images of the same component by using the second recognition model comprises:

[0026] The local images of the same component in different whole images are respectively input into the second recognition model to obtain the defect types and names thereof contained in each local image, and the names of the defect types recognized in a single local image and the number of each type of defect type are included in the recognized defect set thereof;

[0027] The names of the actual defect types in a single local image and the number of each type of defect type are included in the actual defect set thereof, and the recognition accuracy S of the second recognition model for the single local image is obtained according to the recognized defect set A and the actual defect set B of the single local image t ;

[0028] ;

[0029] is a preset weight value, is the number of defect types in the intersection of the recognized defect set and the actual defect set, is the number of defect types in the union of the recognized defect set and the actual defect set, is the total number of all defects in the recognized defect set, is the total number of all defects in the actual defect set.

[0030] Further, according to the recognition accuracy of different components in different local images, a third recognition model is constructed by combining the recognition parameter set of the corresponding local image.

[0031] The identification parameter set refers to a collection distance and a collection angle between a center point of a single local image and a collection position of a whole image to which the center point belongs, the collection distance refers to a distance between the center point and the collection position, and the collection angle is an included angle between a normal line of the center point and a line connecting the center point and the collection position;

[0032] The identification accuracy of the second identification model for different local images of different components in different whole images is obtained respectively, a third identification set is generated according to the identification defect set, the identification parameter set and the corresponding identification accuracy of different local images, and the third identification set is divided into a third training set and a third test set;

[0033] A second convolutional neural network is constructed, different local images in the third training set and the identification defect set and the identification parameter set thereof are taken as input data of the second convolutional neural network, and the corresponding identification accuracy in the third training set is taken as output data of the second convolutional neural network, the second convolutional neural network is trained by using the third training set to obtain an initial second convolutional neural network;

[0034] The initial second convolutional neural network is subjected to model verification by using the third test set, and the initial second convolutional neural network with an output less than or equal to a preset third test error threshold is taken as the third identification model.

[0035] Further, the process of performing defect identification on the motor shell by using the second identification model and the third identification model and obtaining the corresponding evaluation accuracy comprises:

[0036] In a subsequent application scenario, whole images of the motor shell are obtained at different collection positions by using an image collection device, and the whole images are input into the second identification model to obtain components in each whole image and names of the components, and defect types in a local image corresponding to each component and names of the defect types;

[0037] The number of each type of defect in a single local image is obtained, and an identification defect set thereof is generated, an identification parameter set of a whole image to which the single local image belongs is generated, and the single local image, the identification defect set and the identification parameter set are input into the third identification model to obtain an evaluation accuracy thereof;

[0038] The identification defect set and the evaluation accuracy of each local image in each whole image are obtained respectively, and are fed back to relevant personnel.

[0039] Compared with the prior art, the present application has the following beneficial effects:

[0040] This invention, by constructing a first recognition model, can identify different components of the motor housing and can obtain local images corresponding to each component from the overall image collected. This is beneficial for performing defect identification on different local images to improve the targeting of defect identification. By constructing a second recognition model, it can identify different defect types of the motor housing, which is beneficial for using machine vision to directly identify the type and quantity of defects on the motor housing.

[0041] By comparing the identified defect set of each component with its actual defect set to obtain the corresponding identification accuracy, a method can be provided to evaluate the accuracy of defect identification results. Based on the identified defect set, identification parameter set and corresponding identification accuracy of each local image, a third identification model can be constructed. When identifying defects in the subsequent motor housing, the accuracy of its defect identification results can be obtained simultaneously, which is beneficial to provide relevant personnel with multifaceted references. Attached Figure Description

[0042] Figure 1 This is a schematic diagram of the steps of the present invention. Detailed Implementation

[0043] like Figure 1 As shown, a machine vision-based method for identifying defects in the appearance of motors includes the following steps:

[0044] Step S1: Divide the motor housing into different components and obtain the appearance images of each component. Construct a first recognition model based on the different components and their corresponding appearance images.

[0045] Step S2: Obtain overall images of the motor housing at different locations, and use the first recognition model to obtain local images of each component in different overall images;

[0046] Step S3: Obtain different defect types of the motor housing and their corresponding defect images, and construct a second recognition model by combining the first recognition model. Use the second recognition model to obtain the recognition accuracy of different local images of the same component.

[0047] Step S4: Based on the recognition accuracy of different components in different local images, construct a third recognition model by combining the recognition parameter set of the corresponding local images. Use the second and third recognition models to identify defects in the motor housing and obtain the corresponding evaluation accuracy.

[0048] It should be further explained that, in the specific implementation process, the process of dividing the motor housing into different components and obtaining appearance images of each component includes:

[0049] The motor housing is a critical structural component used to protect internal electromagnetic components (such as the stator and rotor) and provide mounting support. It typically comprises the following different parts:

[0050] The frame, as the main support structure, is used to fix the stator core and windings and bears the main heat dissipation;

[0051] End caps are used to seal both ends of the motor, support the rotor bearings, and protect the winding ends.

[0052] A flange is an interface for mounting motors and equipment, used to ensure coaxiality and load transfer.

[0053] Heat sinks are used to increase the surface area of ​​the chassis and improve the efficiency of air cooling.

[0054] Mounting feet are the fixing points for base-type mounting, used to bear the weight and vibration of the entire machine;

[0055] The bearing housing, located at the center of the end cover, is used to secure the bearing.

[0056] Ventilation openings / protective nets are used for airflow exchange or to prevent the intrusion of foreign objects.

[0057] Taking each component as an object, obtain appearance images of different components of the motor housing. The appearance images refer to multi-angle images of the corresponding regions of the corresponding components in the motor housing, which can reflect the morphological characteristics of the corresponding components.

[0058] It should be further explained that, in the specific implementation process, the process of constructing the first recognition model based on different components and their corresponding appearance images includes:

[0059] A first recognition set is generated based on the appearance images of different components and their corresponding names, and then divided into a first training set and a first test set.

[0060] Construct a first convolutional neural network by using different appearance images from the first training set as input data and the corresponding components and their names from the first training set as output data. Train the first convolutional neural network using the first training set to obtain an initial first convolutional neural network.

[0061] The initial first convolutional neural network is validated using the first test set. The initial first convolutional neural network whose output is less than or equal to the preset first test error threshold is used as the first recognition model. The first recognition model can identify the components contained in the input image and output their names.

[0062] It should be further explained that, in the specific implementation process, the process of acquiring overall images of the motor housing at different locations, and using the first recognition model to acquire local images of each component within different overall images, includes:

[0063] The motor housing that has been identified as defective is acquired, and overall images of the defective motor housing are acquired at different locations. The overall image refers to the image of the motor housing acquired by the image acquisition device at a single acquisition location, which includes all components that can be acquired at the corresponding acquisition location.

[0064] A single overall image of the same motor housing contains different components of the motor housing; all overall images of the same motor housing contain all components of the motor housing; and different overall images of the same motor housing contain the same component of the motor housing.

[0065] All overall images of different motor housings are input into the first recognition model. The first recognition model outputs the components and their names contained in each overall image. The image of the corresponding area of ​​each component in each overall image is taken as its local image. Local images of the same component in different overall images are obtained respectively.

[0066] It should be further explained that, in the specific implementation process, the process of acquiring different defect types of the motor housing and their corresponding defect images, and constructing a second recognition model in conjunction with the first recognition model, includes:

[0067] For motor housings, defects include pores, shrinkage cavities, notches, sand holes, burrs, scratches, knife marks, oxidation, corrosion, cracks, etc.

[0068] Defect images of different defect types are acquired respectively. The defect images refer to multi-angle images of the corresponding area on the motor housing of the corresponding defect type, which can reflect the morphological characteristics of the corresponding defect type.

[0069] A second recognition set is generated based on defect images of different defect types and their corresponding names, and then divided into a second training set and a second test set.

[0070] Different defect images in the second training set are used as input data for the first recognition model, and the corresponding defect types and names in the second training set are used as output data for the first recognition model. The first recognition model is trained using the second training set to obtain an initial first recognition model.

[0071] The initial first recognition model is validated using the second test set. The initial first recognition model that outputs an initial first recognition model that is less than or equal to the preset second test error threshold is used as the second recognition model. The second recognition model can identify the components and defect types contained in the input image and output their names respectively.

[0072] It should be further explained that, in the specific implementation process, the process of using the second recognition model to obtain the recognition accuracy of different local images of the same component includes:

[0073] The local images of the same component in different overall images are input into the second recognition model. The second recognition model outputs the defect types and their names contained in each local image. The names of the defect types identified in a single local image and the number of each type of defect are included in its identification defect set, and the identification defect set corresponding to each local image is obtained respectively.

[0074] The relevant personnel record the actual defect types and their names in each local image, and include the names of the actual defect types and the number of each type of defect in a single local image into its actual defect set, and obtain the actual defect set corresponding to each local image respectively.

[0075] Based on the defect set A and the actual defect set B of a single local image, the recognition accuracy of the second recognition model for that single local image is obtained, denoted as S. t ;

[0076] ;

[0077] The weights are preset, and the sum of the two is 1. To identify the number of defect types in the intersection of the defect set and the actual defect set, To identify the number of defect types in the union of the defect set and the actual defect set, the closer the ratio is to 1, the closer the identified defect set and the actual defect set are in terms of defect types.

[0078] To identify the total number of all defects in the defect set, The ratio of the two is the total number of all defects in the actual defect set. The closer the ratio is to 1, the closer the identified defect set and the actual defect set are in terms of the total number of defects. Therefore, the closer the identification accuracy is to 1, the more accurate the second identification model is in identifying defects in this single local image.

[0079] If A represents {notch 2, pinhole 2, burr 3, scratch 3}, and B represents {pinhole 3, burr 3, scratch 4, tool mark 4}, and the number after each defect type represents its quantity, then its It is 3. It is 5. It is 10. The value is 14. The same method is used to obtain the recognition accuracy of the second recognition model for each local image.

[0080] It should be further explained that, in the specific implementation process, the process of constructing a third recognition model based on the recognition accuracy of different components in different local images, combined with the recognition parameter set of the corresponding local images, includes:

[0081] The recognition accuracy of the second recognition model for local images of different components in different overall images is obtained respectively. Since different local images belong to different overall images, and different overall images have their own acquisition positions when they are acquired.

[0082] The set of recognition parameters refers to the acquisition distance and acquisition angle between the center point of a single local image and the acquisition position of the overall image to which it belongs. The acquisition distance is the distance between the center point and the acquisition position, and the acquisition angle is the angle between the normal of the center point and the line connecting the center point and the acquisition position.

[0083] A third recognition set is generated based on the recognition defect set, recognition parameter set and corresponding recognition accuracy of different local images, and it is divided into a third training set and a third test set.

[0084] Construct a second convolutional neural network by using different local images, their defect sets, and parameter sets from the third training set as input data and the corresponding recognition accuracy from the third training set as output data. Train the second convolutional neural network using the third training set to obtain the initial second convolutional neural network.

[0085] The initial second convolutional neural network is validated using a third test set. The initial second convolutional neural network whose output is less than or equal to a preset third test error threshold is used as the third recognition model. The third recognition model can evaluate the accuracy of the second recognition model for the input image.

[0086] It should be further explained that, in the specific implementation process, the process of using the second and third identification models to identify defects in the motor housing and obtain the corresponding assessment accuracy includes:

[0087] In subsequent application scenarios, the overall image of the motor housing is acquired at different acquisition positions by the image acquisition device and input into the second recognition model. The second recognition model is used to obtain the components and their names in each overall image, as well as the defect types and their names in the local images corresponding to each component.

[0088] The number of various defect types in a single local image is obtained and its identification defect set is generated. Based on the acquisition position of the overall image to which the single local image belongs, its identification parameter set is generated. The single local image, its identification defect set, and its identification parameter set are input into the third identification model to obtain its evaluation accuracy.

[0089] The evaluation accuracy is used to reflect the accuracy of the second recognition model in identifying the defect set of the single local image. The same method is used to obtain the identification defect set and its evaluation accuracy of different local images in each overall image, and feed them back to relevant personnel to achieve defect identification and accuracy evaluation of the motor housing.

[0090] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. A method for identifying appearance defects in motors based on machine vision, characterized in that, Includes the following steps: Step S1: Divide the motor housing into different components and obtain the appearance images of each component. Construct a first recognition model based on the different components and their corresponding appearance images. Step S2: Obtain overall images of the motor housing at different locations, and use the first recognition model to obtain local images of each component in different overall images; Step S3: Obtain different defect types of the motor housing and their corresponding defect images, and construct a second recognition model by combining the first recognition model. Use the second recognition model to obtain the recognition accuracy of different local images of the same component. Step S4: Based on the recognition accuracy of different components in different local images, construct a third recognition model by combining the recognition parameter set of the corresponding local images, and use the second and third recognition models to identify defects in the motor housing and obtain the corresponding evaluation accuracy. The process of constructing a third recognition model includes: The set of recognition parameters refers to the acquisition distance and acquisition angle between the center point of a single local image and the acquisition position of the overall image to which it belongs. The acquisition distance is the distance between the center point and the acquisition position, and the acquisition angle is the angle between the normal of the center point and the line connecting the center point and the acquisition position. The recognition accuracy of the second recognition model for different components in local images of different overall images is obtained respectively. A third recognition set is generated based on the recognition defect set, recognition parameter set and corresponding recognition accuracy of different local images, and it is divided into a third training set and a third test set. Construct a second convolutional neural network by using different local images, their defect sets, and parameter sets from the third training set as input data and the corresponding recognition accuracy from the third training set as output data. Train the second convolutional neural network using the third training set to obtain the initial second convolutional neural network. The initial second convolutional neural network is validated using the third test set. The initial second convolutional neural network whose output is less than or equal to the preset third test error threshold is used as the third recognition model. The process of acquiring defect images and constructing a second recognition model includes: The defect types include porosity, shrinkage cavities, notches, sand holes, burrs, scratches, knife marks, oxidation, corrosion, and cracks. Defect images of different defect types are obtained respectively. The defect images refer to multi-angle images of the corresponding areas on the motor housing for the corresponding defect type. A second recognition set is generated based on defect images of different defect types and their corresponding names, and then divided into a second training set and a second test set. Different defect images in the second training set are used as input data for the first recognition model, and the corresponding defect types and names in the second training set are used as output data for the first recognition model. The first recognition model is trained using the second training set to obtain an initial first recognition model. The initial first recognition model is validated using the second test set, and the initial first recognition model whose output is less than or equal to the preset second test error threshold is used as the second recognition model.

2. The method for identifying motor appearance defects based on machine vision according to claim 1, characterized in that, The process of dividing the components and obtaining appearance images includes: The components include a base, end cover, flange, heat sink, mounting feet, bearing housing, and vent / protective mesh. Appearance images of different components of the motor housing are obtained, and the appearance images refer to multi-angle images of the corresponding components in the corresponding areas of the motor housing.

3. The method for identifying motor appearance defects based on machine vision according to claim 2, characterized in that, The process of constructing the first recognition model includes: A first recognition set is generated based on the appearance images of different components and their corresponding names, and then divided into a first training set and a first test set. Construct a first convolutional neural network by using different appearance images from the first training set as input data and the corresponding components and their names from the first training set as output data. Train the first convolutional neural network using the first training set to obtain an initial first convolutional neural network. The initial first convolutional neural network is validated using the first test set, and the initial first convolutional neural network whose output is less than or equal to the preset first test error threshold is used as the first recognition model.

4. The method for identifying motor appearance defects based on machine vision according to claim 3, characterized in that, The process of acquiring the overall image and then acquiring local images of each component includes: Acquire the motor housing that has been identified as defective, and acquire overall images of the defective motor housing at different locations. The overall image refers to the image of the motor housing acquired by the image acquisition device at a single acquisition location. All overall images of different motor housings are input into the first recognition model to obtain the components and their names contained in each overall image, and the images of the corresponding areas of each component in each overall image are used as its local images.

5. The method for identifying motor appearance defects based on machine vision according to claim 4, characterized in that, The process of obtaining the recognition accuracy of different local images of the same component includes: Local images of the same component in different overall images are input into the second recognition model to obtain the defect types and their names contained in each local image. The names of the defect types identified in a single local image and the number of each type of defect are included in its defect set. The names of the actual defect types and the quantity of each defect type in a single local image are included in its actual defect set. Based on the identified defect set A and the actual defect set B of the single local image, the recognition accuracy S of the second recognition model for that single local image is obtained. t ; ； The preset weight values, To identify the number of defect types in the intersection of the defect set and the actual defect set, To identify the number of defect types in the union of the defect set and the actual defect set, To identify the total number of all defects in the defect set, This represents the total number of all defects in the actual defect set.

6. The method for identifying motor appearance defects based on machine vision according to claim 5, characterized in that, The process of identifying defects in the motor housing and obtaining the accuracy of the assessment includes: In subsequent application scenarios, the overall image of the motor housing is acquired at different acquisition positions by the image acquisition device, and then input into the second recognition model to obtain the components and their names in each overall image, as well as the defect types and their names in the local images corresponding to each component. The number of various defect types in a single local image is obtained and its identification defect set is generated. Based on the acquisition position of the overall image to which the single local image belongs, its identification parameter set is generated. The single local image, its identification defect set, and its identification parameter set are input into the third identification model to obtain its evaluation accuracy. The defect sets and their accuracy assessments for different local images within each overall image are obtained and then fed back to the relevant personnel.

Images