A method and system for evaluating the accuracy of blade root groove test

Abstract

The present invention belongs to the field of data processing technology and discloses a method and system for evaluating the accuracy of blade root notch testing. The method comprises: extracting the strain gauge contour based on a pre-acquired strain gauge image using a pre-trained zero-sampled image region segmentation model based on word embedding, obtaining a processed strain gauge image, and recording the strain gauge attachment angle; using the pre-acquired digital speckle image as the target image and the acquired processed strain gauge image as the source image, obtaining a speckle strain gauge overlap image through affine transformation; calculating the error between non-contact measurement data and contact measurement data based on the speckle strain gauge overlap image and the recorded strain gauge attachment angle; and evaluating the accuracy of the blade root notch test based on the error to obtain an accuracy evaluation result. The technical solution of the present invention can obtain indicators for accuracy evaluation and has good reliability and feasibility.

Description

Technical Field

[0001] The present invention belongs to the field of data processing technology, relates to the field of turbine machinery blade root groove test, and in particular to a method and system for evaluating the accuracy of blade root groove test. Background Art

[0002] Blades are core components of turbines, including steam turbines, gas turbines, and aircraft engines. During turbine operation, the complex working environment can cause blades to exhibit characteristics that differ from those in their designed operating conditions. Specifically, factors such as temperature and pressure field fluctuations and errors in assembly and machining precision can cause blades to experience thermal stress and stress concentration. The blade root, where it meets the groove, often bears extreme loads due to its unique structure and function. Therefore, during blade design and manufacturing, both the root and groove sections must be tested to ensure safe and reliable operation of the turbine.

[0003] Currently, the existing traditional methods for processing data from blade root groove tests to evaluate test accuracy usually require manual selection of corresponding nodes in contact measurement, non-contact measurement, and numerical simulation results for result analysis. The above traditional processing methods can basically meet the requirements for test pieces with low precision requirements and small size.

[0004] However, when dealing with high-precision or large test pieces, the above-mentioned traditional methods will produce errors due to problems such as human selection and image angle, which will ultimately affect the judgment and evaluation of the test results. In addition, if the specimen volume is large and the number of strain gauges used for contact measurement is large, the traditional method will greatly increase the difficulty of node selection and comparison, as well as the time cost of data processing, and its feasibility is poor. Summary of the Invention

[0005] The present invention aims to provide a method and system for evaluating the accuracy of blade root groove testing to address one or more of the aforementioned technical issues. The technical solution provided by the present invention accurately identifies the outlines of contact-measured strain gauges using a zero-sampled image region segmentation model based on word embedding, and then performs image overlap processing through affine transformation. Ultimately, an accuracy evaluation metric (interpretive, i.e., the relative error between non-contact and contact measurement data) is obtained, demonstrating good reliability and feasibility.

[0006] In order to achieve the above object, the present invention adopts the following technical solutions:

[0007] The present invention provides a method for evaluating the accuracy of a blade root groove test, comprising the following steps:

[0008] Acquire strain gauge images, digital speckle images, non-contact measurement data, and contact measurement data during blade root groove tests;

[0009] Based on the acquired strain gauge image, a pre-trained word embedding-based zero-sampling image region segmentation model is used to extract the strain gauge contour, obtain a processed strain gauge image, and record the strain gauge attachment angle;

[0010] The obtained digital speckle image is used as a target image, the obtained processed strain gauge image is used as a source image, and a speckle strain gauge overlapping image is obtained by affine transformation;

[0011] Based on the obtained overlapping images of the speckle strain gauges and the recorded attachment angles of the strain gauges, the errors between the non-contact measurement data and the contact measurement data are calculated; and based on the obtained errors, the accuracy of the blade root groove test is evaluated to obtain an accuracy evaluation result.

[0012] A further improvement of the method of the present invention is that the architecture of the zero-sampling image region segmentation model based on word embedding includes:

[0013] The visible class pre-classification module is used to input the original strain gauge image and perform pre-classification processing within the visible class and between the visible class and the invisible class, thereby obtaining the pixel features of the pre-classified image;

[0014] The word vector semantic description construction module is used to input the names of visible class objects and invisible class objects for word vector processing to obtain the semantic description of the visible class and the semantic description of the invisible class;

[0015] The invisible class pixel generation module is used to input the semantic description of the visible class and the semantic description of the invisible class to perform pixel feature generation processing to obtain the pixel features of the invisible class;

[0016] The secondary classification module is used to input the pixel features of the pre-classified image and the pixel features of the invisible class and perform secondary classification processing between all classes, and output the final image region segmentation image.

[0017] A further improvement of the method of the present invention is that in the zero-sampling image region segmentation model based on word embedding,

[0018] The visible class pre-classification module is composed of a fully supervised model;

[0019] The word vector semantic description building module is composed of two word2vec models;

[0020] The invisible pixel generation module is composed of a generator;

[0021] The secondary classification module is composed of a classifier.

[0022] A further improvement of the method of the present invention is that the training step of the pre-trained word embedding-based zero-sampling image region segmentation model includes:

[0023] 1) Definition and collection of training data, including:

[0024] The categories of all identified objects in the image are denoted as C = S∪U; where S represents the visible class, including the blade root test piece, the wheel groove test piece, and the fastener housing; U represents the invisible class, including the strain gauge and the strain gauge cable;

[0025] The word2vec model is used to construct word vector semantic descriptions. By inputting preset keywords, the semantic descriptions mapped to the visible class and the semantic descriptions of the invisible class are embedded respectively.

[0026] Images of the blade root groove test piece taken at different shooting angles and different lighting conditions are collected as samples to construct a pre-training training set for the visible class. The image size is m×n, the number of samples is M, and the pre-training training set for the visible class is expressed as: Where, is the d of the i-th sample x dimensional pixel features, 1≤i≤M; represents the manually annotated segmented image of the i-th sample; Indicates d a dimensional semantic description, used to embed the semantics of the class into each pixel;

[0027] 2) Pre-training of visible classes, including:

[0028] A fully supervised model is used, pre-trained on a pre-trained training set of the visible class. The pixel features output by the last layer of the fully supervised model are only used to distinguish different categories under the visible class, while all unseen classes are identified as a whole.

[0029] 3) Build the generator, including:

[0030] Construct pixel features based on word vector semantic description Where G is a generator with trainable parameters ω, a is the semantic description of the word vector, and z is a random sample from a fixed multivariate Gaussian distribution;

[0031] The parameters ω of the generator are trained and optimized. The optimized generator obtains the pixel features of the class by inputting the semantic description of the word vector; the input of the generator is Output is pixel features of visible class and invisible class pixel features According to the output value, the unseen training set is constructed as Where, is the d of the i'th sample x dimensional pixel features, 1≤i'≤M'; represents the manually annotated segmented image of the i'th sample; Indicates d a Dimensional semantic description;

[0032] 4) Build a classifier, including:

[0033] Will Embed x S Construct input pixel feature x, combined with D S and As a training set, construct a pixel-level classifier The classifier is trained using a self-training strategy of semi-supervised learning to obtain the pre-trained zero-sampling image region segmentation model based on word embedding.

[0034] A further improvement of the method of the present invention is that the acquired digital speckle image is used as a target image, the acquired processed strain gauge image is used as a source image, and the step of obtaining the speckle strain gauge overlapping image by affine transformation specifically includes:

[0035] (1) Read the source image and obtain the source image size, record the coordinate values ​​(x1, y1) and (x2, y2) of the vertices A and B on the left and right sides of the wheel groove, and the coordinate values ​​(x3, y3) and (x4, y4) of the points C and D at the rounded corners of the wheel groove; obtain the coordinates (u1, v1), (u2, v2), (u3, v3), and (u4, v4) of the four points A, B, C, and D on the target image;

[0036] (2) According to the two sets of coordinate values ​​obtained in step (1), the affine transformation matrix is ​​obtained, which is expressed as Among them, the matrix The diagonal elements of the vector determine the scaling factor, the anti-diagonal elements determine the rotation or shearing, and the vector Decide to translate;

[0037] (3) According to the matrix H, the affine transformation formula of the image is obtained, which is expressed as The affine transformation formula is used to realize the coordinate overlap of the digital speckle image and the processed strain gauge image to obtain the speckle strain gauge overlap image.

[0038] A further improvement of the method of the present invention is that the step of calculating the error between the non-contact measurement data and the contact measurement data based on the obtained overlapping image of the speckle strain gauge and the recorded attachment angle of the strain gauge specifically includes:

[0039] Using the obtained overlapping image of the speckle strain gauge, the angle between all strain gauges and the x-axis is measured, and the angle between the i-th strain gauge and the x-axis is αi , the test strain measurement value of strain gauge No. i is ε i ;

[0040] According to the overlapping image of the speckle strain gauge, all nodes inside and on the edge of the area covered by the current strain gauge are selected, and the number of selected nodes is recorded as k. For each node, the normal strain and shear strain in the x and y directions in the non-contact measurement results are read; the normal strain in the x direction of the j-th node is ε jx , the positive strain in the y direction is ε jy , shear strain is γ jxy , 1≤j≤k;

[0041] According to the obtained strain values ​​of normal strain and shear strain in the x and y directions and the strain gauge attachment angle α i , calculate the linear strain of the node along the strain gauge direction, the calculation expression is,

[0042] ε j =cos 2 α i ·ε jx +sin 2 α i ·ε jy +cosα i sinα i γ jxy ;

[0043] Then the linear strains of all k nodes are summed and averaged to obtain the strain value ε′ at the corresponding position of the i-th strain gauge in the non-contact measurement data. i , the calculation formula is,

[0044] According to the non-contact measurement of strain value ε′ i and the corresponding strain gauge measurement value ε i , calculate the error E i , the calculation expression is,

[0045]

[0046] A further improvement of the method of the present invention is that the step of performing the accuracy assessment of the blade root groove test based on the obtained error and obtaining the accuracy assessment result specifically includes:

[0047] For the error E of the coverage position of strain gauge No. i i , if any The measurement is considered accurate; is the preset accuracy judgment threshold;

[0048] If the error of all strain gauge coverage positions is less than the preset accuracy judgment threshold, the test accuracy is considered high; if there are strain gauge coverage positions with errors E less than the preset value, the test accuracy is considered high. i 'satisfy If yes, the test precision is considered to meet the requirements; otherwise, the test precision is considered to not meet the requirements.

[0049] The present invention provides a system for evaluating the accuracy of a blade root groove test, comprising:

[0050] Data acquisition module, used to obtain strain gauge images, digital speckle images, non-contact measurement data and contact measurement data in blade root groove tests;

[0051] An image segmentation and extraction module is used to extract the strain gauge contour based on the acquired strain gauge image using a pre-trained word embedding-based zero-sampling image region segmentation model, obtain a processed strain gauge image, and record the strain gauge attachment angle;

[0052] an affine transformation module, configured to use the acquired digital speckle image as a target image and the acquired processed strain gauge image as a source image, and obtain a speckle strain gauge overlapping image through affine transformation;

[0053] The error calculation and evaluation module is used to calculate the error between the non-contact measurement data and the contact measurement data based on the obtained overlapping images of the speckle strain gauges and the recorded attachment angles of the strain gauges; and to perform an accuracy evaluation of the blade root groove test based on the obtained error to obtain an accuracy evaluation result.

[0054] A further improvement of the system of the present invention is that the architecture of the zero-sampling image region segmentation model based on word embedding includes:

[0055] The visible class pre-classification module is used to input the original strain gauge image and perform pre-classification processing within the visible class and between the visible class and the invisible class, thereby obtaining the pixel features of the pre-classified image;

[0056] The word vector semantic description construction module is used to input the names of visible class objects and invisible class objects for word vector processing to obtain the semantic description of the visible class and the semantic description of the invisible class;

[0057] The invisible class pixel generation module is used to input the semantic description of the visible class and the semantic description of the invisible class to perform pixel feature generation processing to obtain the pixel features of the invisible class;

[0058] The secondary classification module is used to input the pixel features of the pre-classified image and the pixel features of the invisible class and perform secondary classification processing between all classes, and output the final image region segmentation image.

[0059] A further improvement of the system of the present invention is that in the zero-sampling image region segmentation model based on word embedding,

[0060] The visible class pre-classification module is composed of a fully supervised model;

[0061] The word vector semantic description building module is composed of two word2vec models;

[0062] The invisible pixel generation module is composed of a generator;

[0063] The secondary classification module is composed of a classifier.

[0064] Compared with the prior art, the present invention has the following beneficial effects:

[0065] The technical solution provided by this invention can process systematic data from various blade root groove tests, enabling rapid and accurate test accuracy assessments with high reliability and feasibility. Specifically, the method uses strain gauge images and digital speckle images as its foundation. It accurately identifies the strain gauge contours for contact measurement using a zero-sampled image region segmentation model based on word embedding. It then overlaps the three images using an affine transformation, accurately selecting the nodes corresponding to the different methods. Finally, it calculates the error between contact and non-contact measurements to assess test accuracy.

[0066] In the present invention, a zero-sampling image region segmentation model based on word embedding is adopted, and its training samples do not require manual labeling of strain gauge areas, making contact measurement data easier to process and reducing the time cost of contour extraction.

[0067] In the present invention, affine transformation is adopted to improve the accuracy of selecting corresponding nodes of different measurement methods, and greatly reduce the error introduced by human factors. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art; obviously, the drawings described below are some embodiments of the present invention, and for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0069] Figure 1 1 is a flow chart of a method for evaluating the accuracy of a blade root groove test provided by an embodiment of the present invention;

[0070] Figure 2 is a schematic diagram of image data in an embodiment of the present invention; wherein, Figure 2 (a) is a schematic diagram of the strain gauge image; Figure 2 (b) is a schematic diagram of digital speckle image;

[0071] Figure 3 2 is a schematic diagram of the architecture of a zero-sampling image region segmentation model based on word embedding in an embodiment of the present invention;

[0072] Figure 4 1 is a schematic diagram of an affine transformation process in an embodiment of the present invention;

[0073] Figure 5 Schematic diagram of a system for evaluating the accuracy of a blade root groove test provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0074] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of the present invention.

[0075] It should be noted that the terms "first," "second," and the like in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or precedence. It should be understood that the numbers used in this manner are interchangeable where appropriate so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having," as well as any variations thereof, are intended to cover non-exclusive inclusions. For example, a process, method, system, product, or apparatus comprising a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units that are not explicitly listed or that are inherent to these processes, methods, products, or apparatus.

[0076] The present invention is described in further detail below with reference to the accompanying drawings:

[0077] See also Figure 1 The present invention discloses a method for evaluating the accuracy of a blade root groove test, which specifically includes the following steps:

[0078] Step 1: Acquire strain gauge images, digital speckle patterns, non-contact measurement data, and contact measurement data during the blade root groove test. For further explanation, the strain gauge image may be an image of a blade root groove test piece with strain gauges attached, the digital speckle pattern image may be an image of a blade root groove test piece with sprayed speckles, and electronic speckle pattern measurement may be performed using an electronic speckle pattern measuring instrument. Non-contact measurement data refers to the cloud diagram of electronic speckle pattern measurement results; contact measurement data refers to the strain reading of each strain gauge.

[0079] Step 2: Based on the strain gauge image obtained in step 1, the strain gauge contour is extracted using a pre-trained word embedding-based zero-sampling image region segmentation model to obtain the processed strain gauge image and record the strain gauge attachment angle;

[0080] Step 3: Using the digital speckle image obtained in step 1 as the target image and the processed strain gauge image obtained in step 2 as the source image, an overlapping image of the speckle strain gauge is obtained by affine transformation;

[0081] Step 4: Based on the overlapping images of the speckle strain gauges obtained in step 3 and the strain gauge attachment angles recorded in step 2, the error between the non-contact measurement data and the contact measurement data is calculated; based on the obtained error, the accuracy of the blade root groove test is evaluated to obtain the accuracy evaluation result.

[0082] In the method provided in an embodiment of the present invention, based on strain gauge images and digital speckle images, the strain gauge contours of contact measurements are accurately identified through a zero-sampling image region segmentation model based on word embedding. The three images are then overlapped through affine transformation, thereby accurately selecting the corresponding nodes of different methods. Finally, the error values ​​of contact measurement and non-contact measurement are calculated to evaluate the accuracy of the test, showing good reliability and feasibility.

[0083] See also Figure 2 In a further preferred technical solution of the embodiment of the present invention, the specific implementation steps of step 1 of the above embodiment may include:

[0084] Before the test, a strain gauge was attached to one side of the blade root groove test piece for contact measurement. Surface paint was sprayed on the other side to form speckles, and an electronic speckle pattern meter was used to achieve non-contact measurement. The strain gauge image was obtained by photographing the side where the strain gauge was attached. For example, Figure 2 As shown in (a), it is used to record the position of the strain gauge attachment; the electronic speckle meter needs to record the digital speckle image with a grid generated by the supporting computer, as shown in the example Figure 2 As shown in (b), it is used for subsequent image coordinate overlap and data processing operations;

[0085] During the test, the strain readings of each strain gauge and the measurement cloud map of the electronic speckle pattern meter were recorded for subsequent data processing operations.

[0086] See also Figure 3 In a further preferred technical solution of the embodiment of the present invention, in step 2, the architecture of the zero-sampling image region segmentation model based on word embedding includes:

[0087] The visible class pre-classification module consists of a fully supervised model that inputs the original image for pre-classification within the visible class and between the visible class and the invisible class, and obtains the pixel features x of the pre-classified image. S ; Further explained, in the pre-classified image, different visible classes are displayed using different color blocks, and all invisible classes are displayed using a color block different from the visible class;

[0088] The word vector semantic description construction module consists of two word2vec models, which are used to input the names of visible class objects and invisible class objects for word-to-vector processing to obtain the semantic description of the visible class. And the semantic description of the invisible class

[0089] The invisible class pixel generation module consists of a generator that inputs the semantic description of the visible class And the semantic description of the invisible class Perform pixel feature generation processing to obtain pixel features of invisible classes

[0090] The secondary classification module consists of a classifier that inputs the pixel features x of the pre-classified image. S and pixel features of the invisible class Perform secondary classification processing between all classes and output the final image region segmentation image; further explained, different visible classes in the image are displayed using different color blocks, and different invisible classes are displayed using different color blocks different from the visible classes.

[0091] Specifically, the embodiment of the present invention is explained as follows: by inputting the strain gauge image into a zero-sampling image region segmentation model based on word embedding, its image region segmentation image can be automatically obtained. Different categories of objects in the image can be displayed with different color blocks, thereby extracting the outline of the strain gauge.

[0092] In a further preferred embodiment of the present invention, the training step of the pre-trained word embedding-based zero-sampling image region segmentation model includes:

[0093] (1) Definition and collection of training data, including:

[0094] The categories of all identified objects in the image are denoted as C = S ∪ U. Where S represents the visible class, including the blade root test piece, wheel groove test piece, and fastener housing that are easy to mark; U represents the invisible class, including the strain gauge and strain gauge cable that are not easy to mark.

[0095] The word2vec model is used to construct word vector semantic descriptions. By inputting keywords such as blade root, wheel groove, strain gauge, etc., the semantic descriptions mapped to visible classes are embedded respectively. And the semantic description of the invisible class

[0096] Images of the blade root groove test piece taken at different shooting angles and different lighting conditions are collected as samples to construct a pre-training training set for the visible class. The image size is m×n and the number of samples is M. The training set can be expressed as: in is the d of the i-th sample (1≤k≤M) x dimensional pixel features, represents the manually annotated segmented image of the i-th sample, Indicates d a dimensional semantic description, which can embed the semantics of the class into each pixel. To ensure the pre-training effect, the number of samples should be no less than 100.

[0097] For unseen classes, no sampling is required, only word vector semantic descriptions.

[0098] (2) Pre-training of visible classes; specifically explained as follows:

[0099] A fully supervised image region segmentation model is used, pre-trained on a training set of visible classes. Further explanation: the fully supervised model consists of an input layer, a hidden layer, and a classification layer. The input layer is used to read the pixel features of the image, the hidden layer contains multiple convolutional layers that perform convolution operations on the image and pooling layers that reduce the dimensionality of the image, and the classification layer is used to segment objects of different categories.

[0100] In pre-training, according to the last layer of the model - the classification layer, the pixel feature x is output S , this pixel feature can only be used to distinguish different categories under the visible class, and all invisible classes are identified as a whole.

[0101] (3) Construct the generator G; the specific explanation is:

[0102] Construct pixel features based on word vector semantic description Where G is a generator with trainable parameters ω, a is the semantic description of the word vector, and z is a random sample from a fixed multivariate Gaussian distribution;

[0103] The generator parameter ω is trained and optimized. After the optimization, the pixel features of the class can be obtained by inputting the semantic description of the word vector. To further explain, the input of the generator is: Output is pixel features of visible class and invisible class pixel features Based on this output value, an unseen training set can be constructed in is the d of the i'th sample (1≤k'≤M') x dimensional pixel features, represents the manually annotated segmented image of the i'th sample, Indicates d a Dimensional semantic description.

[0104] (4) Construct a classifier f and output an image; the specific explanation is:

[0105] Will Embed x S Construct input pixel feature x, combined with D S and As a training set, construct a pixel-level classifier The classifier is trained using a self-training strategy of semi-supervised learning. Ultimately, the classifier can output an image with the image region segmented.

[0106] See also Figure 4 In step 3 of the embodiment of the present invention, the specific steps of the affine transformation operation include:

[0107] Read the source image and obtain its size, record the coordinate values ​​(x1, y1) and (x2, y2) of vertices A and B on the left and right sides of the wheel groove, and the coordinate values ​​(x3, y3) and (x4, y4) of points C and D at the rounded corners of the wheel groove;

[0108] Get the coordinates of the four points A, B, C, and D on the target image (u1, v1), (u2, v2), (u3, v3), and (u4, v4). Based on the two sets of coordinate values, calculate the affine transformation matrix. Among them, the matrix The diagonal elements of the vector determine the scaling factor, the anti-diagonal elements determine the rotation or shearing, and the vector Decide to translate;

[0109] According to the matrix H, the affine transformation formula of the image can be obtained

[0110] Using the above formula, the coordinate overlap of the digital speckle image and the processed strain gauge image can be achieved to form a speckle strain gauge overlap image.

[0111] In a further preferred technical solution of the embodiment of the present invention, the step of calculating the error between the non-contact measurement data and the contact measurement data includes: using the above image, measuring the angle between all strain gauges and the x-axis, and recording the angle between the i-th strain gauge and the x-axis as α i At the same time, the test strain measurement value of strain gauge No. i is recorded as ε i According to the overlapping image of the speckle strain gauge, all nodes inside and on the edge of the area covered by the current strain gauge are selected in the post-processing software of the electronic speckle measurement instrument. The number of selected nodes is denoted as k. For each node, the normal strain in the x and y directions and its shear strain in the non-contact measurement results are read. Among them, the normal strain in the x direction of the j-th node (1≤j≤k) is ε jx , the positive strain in the y direction is ε jy , shear strain is γ jxy According to the above three strain values ​​and the strain gauge attachment angle α i , the linear strain of the node along the strain gauge direction can be calculated. The calculation formula is:

[0112] ε j =cos 2 α i ·ε jx +sin 2 α i ·ε jy +cosα i sinα i γ jxy ;

[0113] Then, the linear strains of all k nodes are summed and averaged to obtain the strain value ε′ at the corresponding position of the i-th strain gauge in the non-contact measurement data. i , the calculation formula is,

[0114]

[0115] According to the non-contact measurement of strain value ε′ i and the corresponding strain gauge measurement value ε i , calculate the relative error:

[0116]

[0117] At this point, the entire data processing process of the blade root groove test has been completed.

[0118] In the technical solution of the embodiment of the present invention, the data processing process including image overlap processing, affine transformation and error calculation can all be automatically processed through algorithms. Only the specified pictures and experimental measurement data need to be input to quickly obtain the results, thus realizing an automated processing flow. This has great advantages for various large blade root groove test pieces and contact measurement conditions with a large number of measuring points.

[0119] In a further preferred technical solution of an embodiment of the present invention, the accuracy of the blade root groove test is evaluated based on the obtained error, and the steps of obtaining the accuracy evaluation result specifically include:

[0120] According to the specific requirements of the test, select the accuracy evaluation criteria of the test The relative error E of the coverage position of strain gauge No. i i , if any If the measurement error of all strain gauge positions is less than the judgment standard, the test accuracy is considered high; if there is a relative error E at an individual position i 'satisfy In other cases, the test accuracy is considered to meet the requirements, and the test process needs to be re-inspected and evaluated.

[0121] The following are device embodiments of the present invention, which can be used to perform the method embodiments of the present invention. For details not disclosed in the device embodiments, please refer to the method embodiments of the present invention.

[0122] See also Figure 5 In yet another embodiment of the present invention, a system for evaluating the accuracy of a blade root groove test is provided, comprising:

[0123] Data acquisition module, used to obtain strain gauge images, digital speckle images, non-contact measurement data and contact measurement data in blade root groove tests;

[0124] An image segmentation and extraction module is used to extract the strain gauge contour based on the acquired strain gauge image using a pre-trained word embedding-based zero-sampling image region segmentation model, obtain a processed strain gauge image, and record the strain gauge attachment angle;

[0125] an affine transformation module, configured to use the acquired digital speckle image as a target image and the acquired processed strain gauge image as a source image, and obtain a speckle strain gauge overlapping image through affine transformation;

[0126] The error calculation and evaluation module is used to calculate the error between the non-contact measurement data and the contact measurement data based on the obtained overlapping images of the speckle strain gauges and the recorded attachment angles of the strain gauges; and to perform an accuracy evaluation of the blade root groove test based on the obtained error to obtain an accuracy evaluation result.

[0127] In another embodiment of the present invention, a computer device is provided, comprising a processor and a memory, wherein the memory is used to store a computer program, wherein the computer program includes program instructions, and the processor is used to execute the program instructions stored in the computer storage medium. The processor may be a central processing unit (CPU), or may be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, and is suitable for implementing one or more instructions, specifically for loading and executing one or more instructions in the computer storage medium to implement the corresponding method flow or corresponding function; the processor described in the embodiment of the present invention can be used to operate the method for evaluating the accuracy of the blade root groove test.

[0128] In another embodiment of the present invention, a storage medium is provided, specifically a computer-readable storage medium (Memory). The computer-readable storage medium is a memory device in a computer device, used to store programs and data. It is understood that the computer-readable storage medium herein may include both built-in storage media in the computer device and, of course, extended storage media supported by the computer device. The computer-readable storage medium provides storage space, which stores the terminal's operating system. Furthermore, the storage space also stores one or more instructions suitable for being loaded and executed by a processor. These instructions may be one or more computer programs (including program code). It should be noted that the computer-readable storage medium herein may be a high-speed RAM memory or a non-volatile memory, such as at least one disk storage device. The processor may load and execute the one or more instructions stored in the computer-readable storage medium to implement the corresponding steps of the method for evaluating the accuracy of the blade root groove test in the above-mentioned embodiment.

[0129] Those skilled in the art will appreciate that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0130] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the steps in the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0131] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0132] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0133] 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, ordinary technicians in the field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered by the scope of protection of the claims of the present invention.

Claims

1. A method for evaluating the accuracy of a blade root groove test, characterized in that: The following steps are involved: Acquire strain gauge images, digital speckle images, non-contact measurement data, and contact measurement data during blade root groove tests; Based on the acquired strain gauge image, a pre-trained word embedding-based zero-sampling image region segmentation model is used to extract the strain gauge contour, obtain a processed strain gauge image, and record the strain gauge attachment angle; The obtained digital speckle image is used as a target image, the obtained processed strain gauge image is used as a source image, and a speckle strain gauge overlapping image is obtained by affine transformation; Based on the obtained overlapping images of the speckle strain gauges and the recorded attachment angles of the strain gauges, the errors between the non-contact measurement data and the contact measurement data are calculated; and based on the obtained errors, the accuracy of the blade root groove test is evaluated to obtain an accuracy evaluation result.

2. A method for evaluating the accuracy of a blade root groove test according to claim 1, characterized in that: The architecture of the zero-sampling image region segmentation model based on word embedding includes: The visible class pre-classification module is used to input the original strain gauge image and perform pre-classification processing within the visible class and between the visible class and the invisible class, thereby obtaining the pixel features of the pre-classified image; The word vector semantic description construction module is used to input the names of visible class objects and invisible class objects for word vector processing to obtain the semantic description of the visible class and the semantic description of the invisible class; The invisible class pixel generation module is used to input the semantic description of the visible class and the semantic description of the invisible class to perform pixel feature generation processing to obtain the pixel features of the invisible class; The secondary classification module is used to input the pixel features of the pre-classified image and the pixel features of the invisible class and perform secondary classification processing between all classes, and output the final image region segmentation image.

3. The method for evaluating the accuracy of blade root groove testing according to claim 2, characterized in that: In the zero-sampling image region segmentation model based on word embedding, The visible class pre-classification module is composed of a fully supervised model; The word vector semantic description building module is composed of two word2vec models; The invisible pixel generation module is composed of a generator; The secondary classification module is composed of a classifier.

4. The method for evaluating the accuracy of blade root groove testing according to claim 3, characterized in that: The training steps of the pre-trained word embedding-based zero-sampling image region segmentation model include: 1) Definition and collection of training data, including: The categories of all identified objects in the image are denoted as C = S∪U; where S represents the visible class, including the blade root test piece, the wheel groove test piece, and the fastener housing; U represents the invisible class, including the strain gauge and the strain gauge cable; The word2vec model is used to construct word vector semantic descriptions. By inputting preset keywords, the semantic descriptions mapped to the visible class and the semantic descriptions of the invisible class are embedded respectively. Images of the blade root groove test piece taken at different shooting angles and different lighting conditions are collected as samples to construct a pre-training training set for the visible class. The image size is m×n, the number of samples is M, and the pre-training training set for the visible class is expressed as: Where, is the d of the i-th sample x dimensional pixel features, 1≤i≤M; represents the manually annotated segmented image of the i-th sample; Indicates d a dimensional semantic description, used to embed the semantics of the class into each pixel; 2) Pre-training of visible classes, including: A fully supervised model is used, pre-trained on a pre-trained training set of the visible class. The pixel features output by the last layer of the fully supervised model are only used to distinguish different categories under the visible class, while all unseen classes are identified as a whole. 3) Build the generator, including: Construct pixel features based on word vector semantic description Where G is a generator with trainable parameters ω, a is the semantic description of the word vector, and z is a random sample from a fixed multivariate Gaussian distribution; The parameters ω of the generator are trained and optimized. The optimized generator obtains the pixel features of the class by inputting the semantic description of the word vector; the input of the generator is Output is pixel features of visible class and invisible class pixel features According to the output value, the unseen training set is constructed as Where, is the d of the i'th sample x dimensional pixel features, 1≤i'≤M'; represents the manually annotated segmented image of the i'th sample; Indicates d a Dimensional semantic description; 4) Build a classifier, including: Will Embed x S Construct input pixel feature x, combined with D S and As a training set, construct a pixel-level classifier The classifier is trained using a self-training strategy of semi-supervised learning to obtain the pre-trained zero-sampling image region segmentation model based on word embedding.

5. The method for evaluating the accuracy of blade root groove testing according to claim 1, characterized in that: The step of obtaining the digital speckle image as the target image and the processed strain gauge image as the source image, and obtaining the speckle strain gauge overlapping image by affine transformation specifically includes: (1) Read the source image and obtain the source image size, record the coordinate values ​​of the vertices A and B on the left and right sides of the wheel groove (x1, y1), (x2, y2) and the coordinate values ​​of points C and D at the rounded corners of the wheel groove (x3, y3), (x4, y4); obtain the coordinates of the four points A, B, C, and D on the target image (u1, v1), (u2, v2), (u3, v3), and (u4, v4); (2) According to the two sets of coordinate values ​​obtained in step (1), the affine transformation matrix is ​​obtained, which is expressed as Among them, the matrix The diagonal elements of the vector determine the scaling factor, the anti-diagonal elements determine the rotation or shearing, and the vector Decide to translate; (3) According to the matrix H, the affine transformation formula of the image is obtained, which is expressed as The affine transformation formula is used to realize the coordinate overlap of the digital speckle image and the processed strain gauge image to obtain the speckle strain gauge overlap image.

6. The method for evaluating the accuracy of blade root groove testing according to claim 1, characterized in that: The step of calculating the error between the non-contact measurement data and the contact measurement data based on the obtained speckle strain gauge overlap image and the recorded strain gauge attachment angle specifically includes: Using the obtained overlapping image of the speckle strain gauge, the angle between all strain gauges and the x-axis is measured, and the angle between the i-th strain gauge and the x-axis is α i , the test strain measurement value of strain gauge No. i is ε i ; According to the overlapping image of the speckle strain gauge, all nodes inside and on the edge of the area covered by the current strain gauge are selected, and the number of selected nodes is recorded as k. For each node, the normal strain and shear strain in the x and y directions in the non-contact measurement results are read; the normal strain in the x direction of the j-th node is ε jx , the positive strain in the y direction is ε jy , shear strain is γ jxy , 1≤j≤k; According to the obtained strain values ​​of normal strain and shear strain in the x and y directions and the strain gauge attachment angle α i , calculate the linear strain of the node along the strain gauge direction, the calculation expression is, e j =cos 2 a i ·e jx +sin 2 a i ·e jy +cosα i Sinai i ·c jxy ; Then the linear strains of all k nodes are summed and averaged to obtain the strain value ε of the corresponding position of the i-th strain gauge in the non-contact measurement data. i ′, the calculation formula is, According to the non-contact measurement of strain value ε i ′ and the corresponding strain gauge measurement value ε i , calculate the error E i , the calculation expression is, 7. The method for evaluating the accuracy of blade root groove testing according to claim 6, characterized in that: The step of performing the blade root groove test accuracy assessment based on the obtained error and obtaining the accuracy assessment result specifically includes: For the error E of the coverage position of strain gauge No. i i , if any The measurement is considered accurate; is the preset accuracy judgment threshold; If the error of all strain gauge coverage positions is less than the preset accuracy judgment threshold, the test accuracy is considered high; if there are strain gauge coverage positions with errors E less than the preset value, the test accuracy is considered high. i 'satisfy If yes, the test precision is considered to meet the requirements; otherwise, the test precision is considered to not meet the requirements.

8. A system for evaluating the accuracy of blade root groove testing, characterized in that: include: Data acquisition module, used to obtain strain gauge images, digital speckle images, non-contact measurement data and contact measurement data in blade root groove tests; An image segmentation and extraction module is used to extract the strain gauge contour based on the acquired strain gauge image using a pre-trained word embedding-based zero-sampling image region segmentation model, obtain a processed strain gauge image, and record the strain gauge attachment angle; an affine transformation module, configured to use the acquired digital speckle image as a target image and the acquired processed strain gauge image as a source image, and obtain a speckle strain gauge overlapping image through affine transformation; The error calculation and evaluation module is used to calculate the error between the non-contact measurement data and the contact measurement data based on the obtained overlapping images of the speckle strain gauges and the recorded attachment angles of the strain gauges; and to perform an accuracy evaluation of the blade root groove test based on the obtained error to obtain an accuracy evaluation result.

9. The system for evaluating the accuracy of blade root groove testing according to claim 8, characterized in that: The architecture of the zero-sampling image region segmentation model based on word embedding includes: The visible class pre-classification module is used to input the original strain gauge image and perform pre-classification processing within the visible class and between the visible class and the invisible class, thereby obtaining the pixel features of the pre-classified image; The word vector semantic description construction module is used to input the names of visible class objects and invisible class objects for word vector processing to obtain the semantic description of the visible class and the semantic description of the invisible class; The invisible class pixel generation module is used to input the semantic description of the visible class and the semantic description of the invisible class to perform pixel feature generation processing to obtain the pixel features of the invisible class; The secondary classification module is used to input the pixel features of the pre-classified image and the pixel features of the invisible class and perform secondary classification processing between all classes, and output the final image region segmentation image.

10. The system for evaluating the accuracy of blade root groove testing according to claim 9, characterized in that: In the zero-sampling image region segmentation model based on word embedding, The visible class pre-classification module is composed of a fully supervised model; The word vector semantic description building module is composed of two word2vec models; The invisible pixel generation module is composed of a generator; The secondary classification module is composed of a classifier.

Images