US-12620079-B2 - YOLOv5-based real-time detection method and device for blade cracks in aeroengine operation and maintenance

US12620079B2US 12620079 B2US12620079 B2US 12620079B2US-12620079-B2

Abstract

Provided are a YOLOv5-based real-time detection method and device for blade cracks in aeroengine operation and maintenance. The method includes: sending a first instruction to obtain images of internal blades of an engine; preprocessing the images of the internal blades of the engine to obtain a test dataset, a training dataset, and a validation dataset; inputting the training dataset into a preset YOLOv5 network model for training, preliminarily evaluating a model effect derived from training by using the validation dataset to adjust the model, testing the model by using a trained weight file and the test dataset, and obtaining an mAP value and a precision-recall curve to finally evaluate the model; obtaining the images of the internal blades of the engine in real time, detecting the internal blades of the engine in real time by using the weight file, and outputting a detection result.

Inventors

Shuangbao LI
Jingyi Yu

Assignees

CIVIL AVIATION UNIVERSITY OF CHINA

Dates

Publication Date: 20260505
Application Date: 20220919
Priority Date: 20210913

Claims (8)

1 . A YOLOv5-based real-time detection method for blade cracks in aeroengine operation and maintenance, applied to an upper computer, and specifically comprising the following steps: sending a first instruction to obtain images of internal blades of an engine; preprocessing the images of the internal blades of the engine to obtain a test dataset, a training dataset, and a validation dataset; inputting the training dataset into a preset YOLOv5 network model for training, preliminarily evaluating a model effect derived from training by using the validation dataset to adjust the model, testing the model by using a trained weight file and the test dataset, and obtaining an mAP value and a precision-recall (P-R) curve to finally evaluate the model; and obtaining the images of the internal blades of the engine in real time, detecting the internal blades of the engine in real time by using the weight file, and outputting a detection result.
2 . The method according to claim 1 , wherein the step of preprocessing the images of the internal blades of the engine to obtain a test dataset, a training dataset, and a validation dataset comprises: performing crack marking, scaling and folding on the images of the internal blades of the engine to expand datasets to obtain the test dataset, the training dataset, and the validation dataset.
3 . The method according to claim 1 , wherein the preset YOLOv5 network model comprises an input end, a Backbone network, a Neck network, and an output end; the input end comprises a data enhancement module and an anchor box selection module, wherein the data enhancement module is configured to splice input data in a manner of scaling, cutting and random arrangement, and the anchor box selection module is configured to calculate and update a size of a crack anchor box marked on the images of the internal blades of the engine; the Backbone network comprises a Focus structure, a Conv+BatchNormalization+LeakyRelu (CBL) structure, a cross stage partial (CSP) structure, and a spatial pyramid pooling (SPP) module; the Focus structure is configured to obtain the crack image at the input end for slicing and parallel convolution operation, and the CBL structure is configured to extract feature information of the crack image subjected to slicing and parallel convolution operation by the Focus structure; the SPP module is configured to equally divide a feature mapping of crack image features and perform pooling operation; the Neck network comprises a feature pyramid network (FPN) structure and a path aggregation network (PAN) structure, wherein the FPN structure and the CBL structure increase a size of a feature map of the crack image, and the PAN structure and the FPN structure perform feature fusion to reduce the size of the feature map of the crack image; and the output end is configured to mark crack information and output confidence to obtain precision and recall.
4 . The method according to claim 3 , wherein the pooling operation is performed by using the following formula: S H * S W = ⌊ h + 2 ⁢ p - f s + 1 ⌋ * ⌊ w + 2 ⁢ p - f s + 1 ⌋ wherein S H is a height of a matrix; S W is a width of the matrix; h is the height of the image; w is the width of the image; p is a filling quantity; f is a filter size; and s is a stride.
5 . The method according to claim 3 , wherein the precision is obtained by using the following formula: Precision = T ⁢ P T ⁢ P + F ⁢ P = T ⁢ P n , wherein n is a total number of recognized images; TP is a number of correctly recognized images; and FP is a number of wrongly recognized images; the recall is obtained by using the following formula: Recall = T ⁢ P T ⁢ P + F ⁢ N = T ⁢ P m , wherein m is a total number images with a target to be recognized; and FN is a number of images with a target but not recognized by a system.
6 . The method according to claim 5 , wherein TP and FP are obtained by using the following steps: obtaining confidence and Intersection over Union (IOU), wherein the IOU is obtained by using the following formula: IOU = area ⁢ ( B P ⋂ B g ⁢ t ) ‘ area ⁢ ( B P ⋃ B g ⁢ t ) , wherein B p is a prediction box; and B gt is a true box; and obtaining TP and FP based on the IOU: if a recognized region has an area greater than an IOU threshold, determining a result as TP, or if the recognized region has an area less than the IOU threshold, determining the result as FP.
7 . The method according to claim 6 , wherein the mAP value is obtained by using the following formula: AP = ∑ k = 1 N ⁢ P ⁡ ( k ) ⁢ Δ ⁢ r ⁡ ( k ) , wherein P is precision; r is recall; and the mAP value is a mean of AP values of category features.
8 . A YOLOv5-based real-time detection device for blade cracks in aeroengine operation and maintenance, comprising: an image acquisition instruction sending module, configured to send a first instruction to obtain images of internal blades of an engine; a sample acquisition module, configured to preprocess the images of the internal blades of the engine to obtain a test dataset, a training dataset, and a validation dataset; a training test module, configured to input the training dataset into a preset YOLOv5 network model for training, preliminarily evaluate a model effect derived from training by using the validation dataset to adjust the model, test the model by using a trained weight file and the test dataset, and obtain an mAP value and a P-R curve to finally evaluate the model; and an output module, configured to obtain the images of the internal blades of the engine in real time, detect the internal blades of the engine in real time by using the weight file, and output a detection result.

Description

CROSS REFERENCE TO RELATED APPLICATION This patent application is a national stage application of International Patent Application No. PCT/CN2022/119657, filed on Sep. 19, 2022, which claims the benefit and priority of Chinese Patent Application No. 202111068098.0, filed with the China National Intellectual Property Administration on Sep. 13, 2021, the disclosure of which is incorporated by reference herein in its entirety as part of the present application. TECHNICAL FIELD The present disclosure relates to the technical field of acro-generator detection, and in particular, a YOLOv5-based real-time detection method and device for blade cracks in aeroengine operation and maintenance. BACKGROUND Normal operation of aeroengine blades can provide continuous flight power for an engine, and the aeroengine blades usually have a long service time. In such an environment, the aeroengine blades are likely to generate fatigue cracks, and the cracks on the internal blades of the engine pose a potential threat to the normal operation of the aeroengine. The cracks that are not treated in a timely manner may further deteriorate, which further leads to the paralysis and failure of the entire engine, thereby posing a serious threat to normal aviation flight. In fact, provided that there are cracks on the internal blades of the engine, no matter how big the cracks are, the cracks may endanger people and pose a serious threat to a machine, or even destroy the machine and cause death, resulting in irreparable losses. For a long time, flight accidents caused by turbine blade fracture are common in flight, so it is very important to regularly detect blade cracks to ensure the safe operation of aeroengines. Existing methods for detecting blade cracks include: conventional methods such as a borescope and penetrant testing method, an X-ray and magnetic particle testing method, eddy current testing, and ultrasonic testing; and image processing methods such as a faster region-based convolutional neural network (R-CNN) two-stage algorithm. The conventional methods mainly have problems such as a limited number of manual marks, poor robustness, many steps in process, time consuming, and labor consuming. Target detection algorithms fall into one-stage algorithms and two-stage algorithms. The one-stage algorithm is to perform positioning prediction after image information is input, and directly output results, which has a fast detection speed, but there are many anchor boxes, so the selection of anchor boxes needs to be optimized. YOLO is a representative algorithm of one-stage target detection, which outputs a position and category confidence of a target box at one time. The two-stage algorithm classifies and regresses the anchor boxes, and performs detection and update for many times, has a slower speed than the one-stage algorithm, and has a structure not flexible enough, but the network fusion is high. In summary, in the prior art, the blade crack detection method relies on manual marking, which is inefficient, and the target detection algorithm cannot meet requirements for both the blade detection speed and blade detection network flexibility. SUMMARY In view of this, an objective of the present disclosure is to provide a YOLOv5-based real-time detection method and device for blade cracks in aeroengine operation and maintenance, so as to increase a speed of detecting blade cracks in the prior art and improve network flexibility of a blade crack detection algorithm. In a first aspect, an embodiment of the present disclosure provides a YOLOv5-based real-time detection method for blade cracks in aeroengine operation and maintenance, which is applied to an upper computer, and specifically includes the following steps: sending a first instruction to obtain images of internal blades of an engine;preprocessing the images of the internal blades of the engine to obtain a test dataset, a training dataset, and a validation dataset;inputting the training dataset into a preset YOLOv5 network model for training, preliminarily evaluating a model effect derived from training by using the validation dataset to adjust the model, testing the model by using a trained weight file and the test dataset, and obtaining an mAP value and a precision-recall (P-R) curve to finally evaluate the model; andobtaining the images of the internal blades of the engine in real time, detecting the internal blades of the engine in real time by using the weight file, and outputting a detection result. Preferably, the step of preprocessing the images of the internal blades of the engine to obtain a test dataset, a training dataset, and a validation dataset includes: performing crack marking, scaling and folding on the images of the internal blades of the engine to expand datasets to obtain the test dataset, the training dataset, and the validation dataset. Preferably, the preset YOLOv5 network model includes an input end, a Backbone network, a Neck network, and an output end