Engine blade damage identification and reconstruction method based on Mask RCNN algorithm

The Mask RCNN algorithm and motion recovery structure enhance engine blade inspection efficiency by providing high-precision damage identification and reconstruction, addressing the limitations of traditional methods.

CN119068244APending Publication Date: 2024-12-03SICHUAN UNIV
View PDF 4 Cites 0 Cited by

Patent Information

Application Number
CN202411107119.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-13
Publication Date
2024-12-03

AI Technical Summary

Technical Problem

Traditional engine blade inspection methods are time-consuming, labor-intensive, and inefficient for batch inspections, with low recognition accuracy and difficulty in analyzing defects in high-pressure turbine blades.

Method used

An engine blade damage identification and reconstruction method using the Mask RCNN algorithm, combined with motion recovery structure, to improve recognition accuracy and generate low-error three-dimensional models.

Benefits of technology

The method achieves high-precision recognition and reconstruction of blade defects, reducing failure risks and enhancing equipment reliability through accurate damage prediction and detailed three-dimensional modeling.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119068244A_ABST
    Figure CN119068244A_ABST
Patent Text Reader

Abstract

The invention discloses an engine blade damage identification and reconstruction method based on a Mask RCNN algorithm, and belongs to the technical field of engine blade damage detection, and the method comprises the steps: building an engine blade defect data set; constructing an engine blade damage identification model based on a Mask RCNN algorithm, and training and verifying the model; identifying the blade damage by using the verified engine blade damage identification model; and performing three-dimensional reconstruction on the blade damage. According to the method, based on the Mask RCNN algorithm, accurate identification of the blade damage type, size and position is realized, fine three-dimensional reconstruction of the blade and the damage thereof is completed by using the motion recovery structure, and the accuracy and efficiency of engine blade damage identification and reconstruction are improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of engine blade damage detection, and in particular to an engine blade damage recognition and reconstruction method based on a Mask RCNN algorithm. Background Art

[0002] Since the engine works continuously under the harsh conditions of high temperature, high pressure and high load for a long time, the high-pressure turbine, as the core component of the engine, is usually one of the most damaged components and is prone to various defects. The traditional engine blade inspection method mainly includes the borescope inspection method, which uses an endoscope to observe the blade. This method can be used for inspection without disassembling the blade structure, so it can reduce the inspection workload to a certain extent and speed up the inspection process of blade damage. However, it has the disadvantages of being time-consuming and labor-intensive, cumbersome to operate, and difficult to inspect blades in batches. Summary of the invention

[0003] The purpose of the present invention is to provide an engine blade damage identification and reconstruction method based on the Mask RCNN algorithm. In view of the problems of low recognition and difficult analysis in traditional image inspection, the defects are identified and reconstructed respectively by the Mask RCNN algorithm and the motion recovery structure, which improves the accuracy of the recognition result and obtains a low-error three-dimensional model.

[0004] To achieve the above object, the present invention provides an engine blade damage identification and reconstruction method based on Mask RCNN algorithm, the steps comprising:

[0005] S1. Establish an engine blade defect dataset and divide the dataset into a training set, a validation set and a test set;

[0006] S2. Build an engine blade damage recognition model based on the Mask RCNN algorithm, use the training set to train the engine blade damage recognition model, and use the validation set to verify it;

[0007] S3. Using the test set to identify the verified engine blade damage identification model, and obtaining the engine blade damage prediction result;

[0008] S4. Based on the prediction results of engine blade damage, the motion recovery structure is used to reconstruct the engine blade damage.

[0009] Preferably, the step S1 includes:

[0010] S11, acquiring engine blade defect images at different angles and light sources;

[0011] S12, screening engine blade defect images;

[0012] S13, marking defects on the screened engine blade defect images, generating a json file, and integrating them into a data set;

[0013] S14. Divide the dataset into training set, validation set and test set in a ratio of 9:1:1.

[0014] Preferably, the engine blade defect types include four categories: ablation, cracks, coating loss and material loss.

[0015] Preferably, in step S2, the COCO dataset is used for pre-weighting and training parameters are set to train the engine blade damage recognition model, and the training parameters include training cycle, number of trainings, number of trainings, learning rate and learning rate reduction method.

[0016] Preferably, the step S3 includes:

[0017] The test set is input into the verified engine blade damage recognition model. The engine blade damage recognition model extracts features from the input image, draws a prediction box of the blade damage area in the input image, predicts the blade damage area based on the prediction box information, generates prediction results including damage type, damage size and damage location, and outputs and saves each blade damage area prediction box and the corresponding prediction results.

[0018] Preferably, the step S4 includes:

[0019] S41, target interception is performed based on the damage position of the prediction box, a target box is drawn, and the damage type corresponding to the target area is found;

[0020] S42, performing pixel-level segmentation on the target area to obtain the foreground and background of each target area;

[0021] S43, using VisualSFM software to process the image obtained in step S42 to obtain a dense point cloud with real colors;

[0022] S44. Use MeshLab software to process the dense point cloud and generate a completed three-dimensional model of the blade and its damage based on the optimization and reconstruction functions in MeshLab.

[0023] Preferably, VisualSFM is used for image matching and feature point detection, a sparse point cloud is generated using the structure from motion method, and a dense point cloud with true color is generated using the CMVS / PMVS method.

[0024] Preferably, the processing of the dense point cloud in step S44 includes removing noise points, meshing and generating texture.

[0025] Therefore, the present invention adopts the above-mentioned engine blade damage identification and reconstruction method based on Mask RCNN algorithm, which has the following beneficial effects:

[0026] (1) The defects are identified and reconstructed by using the Mask RCNN algorithm and motion recovery structure, respectively, to obtain high-precision recognition results and low-error three-dimensional models. The blade defects are predicted by analyzing the damage recognition model to reduce the risk of failure and improve the reliability and safety of the equipment.

[0027] (2) The calculation is simple, the versatility is high, it is in line with the actual engineering application, it is easy for technical personnel to master and use, it has few limitations, and it is conducive to application and promotion.

[0028] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 is a flow chart of a method according to an embodiment of the present invention;

[0030] Figure 2 This is a real blade coating shedding diagram of an embodiment of the present invention;

[0031] Figure 3 This is an ablation diagram of a blade model according to an embodiment of the present invention;

[0032] Figure 4 This is a display diagram of the recognition result of the actual blade coating shedding according to an embodiment of the present invention;

[0033] Figure 5 This is a display diagram of the recognition result of the coating shedding area of ​​a real blade according to an embodiment of the present invention;

[0034] Figure 6 This is a diagram showing the identification result of the blade model burning according to an embodiment of the present invention;

[0035] Figure 7 This is a diagram showing the identification results of the ablation area of ​​the blade model according to an embodiment of the present invention;

[0036] Figure 8 A three-dimensional model display diagram of a real blade according to an embodiment of the present invention;

[0037] Fig. 9 A three-dimensional model showing the actual blade coating shedding according to an embodiment of the present invention;

[0038] Fig.10 A three-dimensional model display diagram of a blade model according to an embodiment of the present invention;

[0039] Fig.11 This is a three-dimensional model display diagram of the blade model according to an embodiment of the present invention. DETAILED DESCRIPTION

[0040] Example

[0041] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, 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 described embodiments are only part of the embodiments of the present invention, not all of the embodiments.

[0042] Reference Figure 1 , an engine blade damage identification and reconstruction method based on Mask RCNN algorithm, the steps include:

[0043] S1. Establish an engine blade defect dataset. The engine blade defect types include ablation, cracks, coating loss, and material loss.

[0044] The steps to establish an engine blade defect dataset include:

[0045] Acquire engine blade defect images at different angles and lighting conditions to ensure the diversity and comprehensiveness of the data set.

[0046] The defective images of engine blades are screened to remove images with low resolution and uneven lighting.

[0047] The screened engine blade defect images are enhanced, and the Labelme software is used to annotate the engine blade defect images after image data enhancement, including annotating the damage type, damage size and damage location in the engine blade damage images. A json file is generated for each engine blade defect image and integrated into a data set.

[0048] Write code to divide the dataset into training set, validation set, and test set in a ratio of 9:1:1, and each subset contains diverse samples.

[0049] S2. Build an engine blade damage recognition model based on the Mask RCNN algorithm, use the training set to train the engine blade damage recognition model, and use the validation set to verify it.

[0050] The Mask RCNN algorithm is universal across different datasets because its feature extraction module can adapt to a variety of tasks. The model of the present invention is trained using the pre-trained weights of the COCO dataset. The pre-trained weights are universal across different datasets.

[0051] The model training process requires setting training parameters. First, set the total number of training cycles for the model to determine the total number of model training times; second, set the current training generation of the model to determine which training cycle to start training from; third, set the number of images input each time to determine the amount of input data for each batch of training.

[0052] In addition, other training parameters need to be set, including learning rate, optimizer, and learning rate descent method. The learning rate determines the speed at which the model weights are updated, the optimizer is used to adjust the model parameters, and the learning rate descent method sets the learning rate decrease strategy during the training process.

[0053] S3. Use the test set to identify the verified engine blade damage recognition model to obtain the engine blade damage prediction results, which include the predicted type of blade damage, the predicted size of blade damage, and the predicted location of blade damage. The predicted size of blade damage refers to the model's estimate of the size of the damaged area, usually expressed in pixels. The predicted location of blade damage refers to the model's specific coordinate location of the damaged area in the image.

[0054] First, the test set is input into the verified engine blade damage recognition model. The engine blade damage recognition model extracts features from the input image, draws a prediction frame of the blade damage area in the input image, predicts the blade damage area based on the prediction frame information, and generates prediction results including damage type, damage size, and damage location. During the recognition process, the engine blade damage recognition model outputs and saves the prediction frame and corresponding prediction results of each blade damage area as the result data of blade damage recognition.

[0055] S4, reconstructing the engine blade damage using motion recovery structure based on the engine blade damage prediction result, the steps include:

[0056] S41, target interception is performed based on the damage position of the prediction box, a target box is drawn, and the damage type corresponding to the target area is found;

[0057] S42, performing pixel-level segmentation on the target area to obtain the foreground and background of each target area;

[0058] S43, using VisualSFM software to process the image obtained in step S42, VisualSFM is responsible for image matching, SFM (Structure from Motion) processing and conversion from sparse point cloud to dense point cloud. Image matching and feature point detection are performed by VisualSFM, sparse point cloud is generated by using Structure from Motion method, and dense point cloud with real color is generated by CMVS / PMVS method.

[0059] S44. Use MeshLab software to process the dense point cloud, including steps such as removing noise, meshing, and generating textures. Based on the optimization and reconstruction functions in MeshLab, a complete three-dimensional model of the blade and its damage is generated.

[0060] The coating shedding of a real blade and the ablation of a blade model are used as examples for analysis. Figure 2 , Figure 3 As shown. The coating shedding dataset of the real blade and the ablation dataset of the blade model are obtained, and the two datasets are divided into training set, verification set and test set in a ratio of 9:1:1. The training set is used to train the engine blade damage recognition model, and the verification set is used to verify it. The test set is input into the verified engine blade damage recognition model for recognition, and the recognition results of the real blade are shown in Figure 4 , Figure 5 The recognition results of the blade model are shown in Figure 6 , Figure 7 As shown. Finally, based on the recognition results, the blade defects of the two defect types are reconstructed. The reconstruction results of the real blade are shown as follows: Figure 8 , Fig. 9 The reconstruction result of the blade model is shown in Fig.10 , Fig.11 shown.

[0061] Therefore, the present invention uses the above-mentioned engine blade damage identification and reconstruction method based on the Mask RCNN algorithm to construct an engine blade damage identification model based on the Mask RCNN algorithm, which can accurately identify the damage type, size and position of the blade; the use of pre-trained weights improves the efficiency and effect of model training; through the combined use of VisualSFM and MeshLab, high-precision three-dimensional reconstruction is achieved, providing a complete three-dimensional model of the blade and its damage; and detailed blade damage data is provided, providing a reliable basis for subsequent damage analysis and repair plan formulation.

[0062] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.

Claims

1. A method for engine blade damage identification and reconstruction based on Mask RCNN algorithm, characterized in that the steps include: S1. Establish an engine blade defect dataset and divide the dataset into a training set, a validation set and a test set; S2. Build an engine blade damage recognition model based on the Mask RCNN algorithm, use the training set to train the engine blade damage recognition model, and use the validation set to verify it; S3. Using the test set to identify the verified engine blade damage identification model, and obtaining the engine blade damage prediction result; S4. Based on the prediction results of engine blade damage, the motion recovery structure is used to reconstruct the engine blade damage.

2. The engine blade damage identification and reconstruction method based on the Mask RCNN algorithm according to claim 1 is characterized in that: The steps in S1 include: S11, acquiring engine blade defect images at different angles and light sources; S12, screening engine blade defect images; S13, marking defects on the screened engine blade defect images, generating a json file, and integrating them into a data set; S14. Divide the dataset into training set, validation set and test set in a ratio of 9:1:

1.

3. The engine blade damage identification and reconstruction method based on the Mask RCNN algorithm according to claim 2 is characterized in that: The engine blade defect types include ablation, cracks, coating loss and material loss.

4. The engine blade damage identification and reconstruction method based on the Mask RCNN algorithm according to claim 3 is characterized by: In step S2, the engine blade damage recognition model is trained by using the COCO data set for pre-weighting and setting training parameters, wherein the training parameters include training cycle, number of training times, number of training times, learning rate and learning rate reduction method.

5. The engine blade damage identification and reconstruction method based on Mask RCNN algorithm according to claim 4 is characterized in that: The steps in S3 include: The test set is input into the verified engine blade damage recognition model. The engine blade damage recognition model extracts features from the input image, draws a prediction box of the blade damage area in the input image, predicts the blade damage area based on the prediction box information, generates prediction results including damage type, damage size and damage location, and outputs and saves each blade damage area prediction box and the corresponding prediction results.

6. The engine blade damage identification and reconstruction method based on Mask RCNN algorithm according to claim 5 is characterized in that: The step S4 includes: S41, target interception is performed based on the damage position of the prediction box, a target box is drawn, and the damage type corresponding to the target area is found; S42, performing pixel-level segmentation on the target area to obtain the foreground and background of each target area; S43, using VisualSFM software to process the image obtained in step S42 to obtain a dense point cloud with real colors; S44. Use MeshLab software to process the dense point cloud and generate a complete three-dimensional model of the blade and its damage based on the optimization and reconstruction functions in MeshLab.

7. The engine blade damage identification and reconstruction method based on Mask RCNN algorithm according to claim 6 is characterized in that: The step S43 includes: VisualSFM is used for image matching and feature point detection, the motion recovery structure method is used to generate sparse point clouds, and the CMVS / PMVS method is used to generate dense point clouds with real colors.

8. The engine blade damage identification and reconstruction method based on Mask RCNN algorithm according to claim 7 is characterized in that: The processing of the dense point cloud in step S44 includes removing noise points, meshing and generating texture.

Citation Information

Patent Citations

  • Road surface rut shape and structure transverse difference degree evaluation method for unmanned driving

    CN112200779A

  • Wind driven generator blade surface damage lightweight identification method

    CN112581443A

  • Blade damage identification method suitable for small sample condition

    CN115457551A

  • Method and computer program product for automated defect detection during borescoping of an engine

    US20220245785A1