CN111968084B - Rapid and accurate identification method for defects of aero-engine blade based on artificial intelligence - Google Patents
Rapid and accurate identification method for defects of aero-engine blade based on artificial intelligence Download PDFInfo
- Publication number
- CN111968084B CN111968084B CN202010791906.5A CN202010791906A CN111968084B CN 111968084 B CN111968084 B CN 111968084B CN 202010791906 A CN202010791906 A CN 202010791906A CN 111968084 B CN111968084 B CN 111968084B
- Authority
- CN
- China
- Prior art keywords
- blade
- defect
- training
- model
- image
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0004—Industrial image inspection
- G06T7/0006—Industrial image inspection using a design-rule based approach
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10072—Tomographic images
- G06T2207/10081—Computed x-ray tomography [CT]
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Physics & Mathematics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Quality & Reliability (AREA)
- Image Analysis (AREA)
Abstract
The invention provides a rapid and accurate identification method for defects of an aircraft engine blade based on artificial intelligence, which is characterized in that a turbine blade detection image database is built by adopting an artificial calibration and data enhancement technology, so that the problem of small sample size is effectively solved, and the generalization capability of a model is favorably improved. Based on an artificial intelligence deep learning method, a turbine blade defect detection and identification network is constructed, the problem of true and false defect identification is solved by adopting technologies such as a maximum entropy principle acceleration strategy and the like, and the training speed and precision of the model are greatly improved. The detection method effectively overcomes the influence of human factors such as experience difference, manual evaluation of eye fatigue, standard understanding and the like, so that the ray detection work of the blade realizes the standardization and intelligentization target, the detection efficiency and accuracy are greatly improved, and the production reliability of the turbine blade is ensured.
Description
Technical Field
The invention belongs to the field of machining, manufacturing and quality inspection of aero-engine blades, and particularly relates to a rapid and accurate identification method for defects of aero-engine blades based on artificial intelligence.
Background
The turbine blade is one of the core components of the aircraft engine, and with the continuous improvement of the performance of the aircraft, the reliability detection requirements of the turbine blade are also increasingly strict. The defect detection of the turbine blade gradually develops from the traditional film detection to the current computerized tomography detection, and meanwhile, the requirements on the objectivity, the accuracy and the reliability of the ray detection are higher and higher.
The existing engine blade is generally formed by adopting non-allowance precision casting due to the complex internal structure. In the process of process forming, the defects of cracks, cold shut, air holes, slag inclusion, looseness and the like exist in the formed blade. Compared with the external defects of the blade, the internal defects of the blade, such as air holes, slag inclusion, looseness and the like, are difficult to find by a conventional method, and the potential hazard is larger. The basic current practice for detecting such defects is to scan the turbine blade by radiation tomography and then manually evaluate the CT images of the blade produced by the scan. However, the method has the influence of human factors such as experience difference, eye fatigue, standard understanding and the like, so that the conditions of missed detection and false detection are easy to occur in the blade defect detection process, and further, flight safety accidents are easy to cause. In recent years, artificial intelligence technology is gradually applied to defect detection and is applied to defect identification with obvious morphological characteristics such as welding seams, holes and the like, however, most of defect identification is still based on traditional image identification technology. The casting process of the turbine blade is complex, the typical defects are various, the typical defects have certain similarity and own characteristics, and the defect detection of the turbine blade is evaluated by means of manual experience for a long time. In recent years, automatic turbine blade defect identification has been studied, but all of them are based on conventional image feature extraction methods such as image segmentation and morphological calculation. The conventional defect detection methods are still high in omission factor and false detection rate, and the delivery quality of the turbine blade cannot be effectively improved. Therefore, an intelligent new method for automatically detecting and identifying the defects of the turbine blade is urgently needed to reduce the missing detection rate and the false detection rate of the defect detection of the turbine blade and improve the detection level of the turbine blade.
Disclosure of Invention
In order to overcome the problems of missed detection, false detection and the like caused by the influence of human factors such as experience difference, eye fatigue, standard understanding and the like due to the evaluation of a blade CT image in a manual mode and avoid the problem of low detection precision in the prior art, the invention provides an aircraft engine blade defect rapid and accurate identification method based on artificial intelligence.
The technical scheme of the invention is as follows:
the method for quickly and accurately identifying the defects of the blade of the aircraft engine based on artificial intelligence is characterized by comprising the following steps of: the method comprises the following steps:
step 1: establishing a turbine blade radiographic image database which comprises a perfect blade image database and a defective blade image database; the defect leaf image database calibrates the type and the position of the defect in each defect leaf digital image;
step 2: establishing and training an aircraft engine blade initial inspection model:
selecting a DenseNet model, removing a Classification layer, connecting two full-connection layers after the 4 th Dense Block of the model, and then connecting a two-Classification output layer, thereby establishing a two-Classification deep convolution network model as an aircraft engine blade initial inspection model;
training an aircraft engine blade initial inspection model by using intact blade images and defective blade images as sample data;
and step 3: establishing and training an aero-engine blade rechecking model:
establishing a defect detection identification model based on a fast RCNN algorithm and an SSD algorithm respectively;
selecting defect detection training data and defect detection test data from a defect blade image database, and respectively training defect detection identification models established based on a fast RCNN algorithm and an SSD algorithm to obtain two sets of aero-engine blade defect review models;
and 4, step 4: detecting whether the aero-engine blade has a defect or not by using the aero-engine blade initial inspection model obtained in the step 2, and further detecting and identifying the type and the position of the defect by using the aero-engine blade rechecking model obtained in the step 3 when the aero-engine blade with the defect is judged; and the final defect type confidence coefficient and position information are obtained by respectively carrying out weighted average on the defect type confidence coefficient and the position information output by the two sets of aero-engine blade defect rechecking models.
Further, in step 1, the digital image in the turbine blade radiographic image database includes a digital image obtained by scanning the blade by a computed tomography technique, a digital image obtained by performing digital processing on an existing blade X-ray scanning film, and a digital image obtained by performing image expansion and data enhancement processing on an existing digital image.
Further, the process of carrying out digital processing on the existing leaf X-ray scanning film comprises the following steps:
adopting a negative film digital scanner to carry out digital processing on the leaf X-ray scanning film, wherein the leaf X-ray scanning film is subjected to DS-level scanning with the blackness range of 0.5-4.5D, the highest scanning resolution is 7980 multiplied by 9690 and 570DPI, the line pair reaches 11LP/mm, and the geometric definition is better than 1% or 2 pixels in the transverse direction and the longitudinal direction;
after digital processing, based on a super-resolution technology, by edge-guided interpolation and based on an image data structure fitting method, noise, gray scale deviation and texture loss in the scanning process are further eliminated.
Further, in the step 1, defect categories are extracted from typical characteristic parameters of defects of the turbine blade in a fuzzy reasoning mode, wherein the typical characteristic parameters of the defects of the turbine blade are selected from 11 items including edge smoothness, end sharpness, trend change rate, relative position, symmetry, included angle, length-to-axis ratio, defects and relative change rate of external gray scale of the defects, and the defect categories are 5 categories including cracks, cold shut, air holes, slag inclusions and looseness;
the typical feature parameters are expressed in the image features as:
C＝{ci|i＝1,2,…,9}
definition of
D＝{dj|j＝1,2,…,5}
Is a defect category set;
establishing a defect label data set: firstly, the typical characteristic parameter ciDefect class d as a rule preconditionjAnd as a rule conclusion, establishing fuzzy reasoning between C and D by taking the rule credibility CF as a reasoning basis:
a decision expression according to the fuzzy generative rule:
determining the defect type of each picture, carrying out manual calibration, and establishing a defect label; wherein, ciAs a rule prerequisite, djIn the conclusion of the rule, ω ≧ 0 is the degree of membership,
CF is the rule confidence.
Further, in the step 2, an image label of the intact leaf image is set to be (1, 0) by adopting one-hot coding, an image label of the defective leaf image is set to be (0, 1), the intact leaf image and the defective leaf image are used as sample data, and the aeroengine leaf initial inspection model is trained by a migration learning method; and in the training process, training the model according to a preset learning rate, wherein when the training is carried out for a preset number of times, the learning rate is attenuated.
Further, the defect detection and identification model established based on the fast RCNN algorithm in the step 3 is characterized in that the initial detection model in the step 2 is used as a main feature extraction network, a VGG network is used as a second main feature extraction network, and two groups of feature graphs obtained by the two main feature extraction networks are connected according to a third dimension; during training, setting the size of a prior frame in an RPN (resilient packet network) of the fast RCNN according to the average height and width of a defect target, alternately training the RPN and a classification regression network in a training step according to the training step of the fast RCNN model, and training to obtain the aircraft engine blade reinspection model based on the fast RCNN algorithm.
Further, when two sets of aero-engine blade recheck models are trained in the step 3, maximum entropy principle algorithm is adopted to accelerate training, the Boltzmann equation H theorem is converted into a mathematical selection path, the gradient is converted into entropy increase and dissipation, the models are trained according to the preset learning rate, and the learning rate is attenuated when the models are trained to the preset times.
Advantageous effects
The invention provides a rapid and accurate identification method for defects of an aircraft engine blade based on artificial intelligence, which is characterized in that a turbine blade detection image database is built by adopting an artificial calibration and data enhancement technology, so that the problem of small sample size is effectively solved, and the generalization capability of a model is favorably improved. Based on an artificial intelligence deep learning method, a turbine blade defect detection and identification network is constructed, the problem of true and false defect identification is solved by adopting technologies such as a maximum entropy principle acceleration strategy, and the training speed and precision of the model are greatly improved. The detection method effectively overcomes the influence of human factors such as experience difference, manual eye fatigue evaluation, standard understanding and the like, so that the ray detection work of the blade realizes the standardization and intelligentized target, the detection efficiency and the accuracy are greatly improved, and the production reliability of the turbine blade is ensured.
Additional aspects and advantages of the invention will be set forth in part in the description which follows and, in part, will be obvious from the description, or may be learned by practice of the invention.
Drawings
The above and/or additional aspects and advantages of the present invention will become apparent and readily appreciated from the following description of the embodiments, taken in conjunction with the accompanying drawings of which:
FIG. 1 is a flow chart of an aircraft engine blade defect rapid and accurate identification method based on artificial intelligence.
FIG. 2 is a flow chart for building a database of turbine blade radiographic images.
FIG. 3 is a flow chart of an exemplary defect label set creation.
Fig. 4 is a schematic diagram of a typical defect.
FIG. 5 is a schematic diagram of a defect detection process based on the Faster R-CNN algorithm model.
FIG. 6 is a schematic diagram of a defect detection flow based on an SSD algorithm model.
Detailed Description
In order to make the objects, technical solutions and advantages of the present application more apparent, the present application is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the present application and are not intended to limit the present application.
In one embodiment, as shown in fig. 1, a method for quickly and accurately identifying a blade defect of an aircraft engine based on artificial intelligence is provided, which includes:
step 1: establishing a turbine blade radiographic image database which comprises a perfect blade image database and a defective blade image database; the defect leaf image database calibrates the type and the position of the defect in each defect leaf digital image.
In step 1, the digital image in the turbine blade radiographic image database includes a digital image obtained by scanning the blade by a Computed Tomography (CT) technique, a digital image obtained by performing digital processing on an existing blade X-ray scanning film, and a digital image obtained by performing image expansion and data enhancement processing on an existing digital image.
For example, for digital photos, the digital photos are directly stored in a database to be used as training samples of the deep learning network. However, for the conventional film, as shown in fig. 2, the digital processing is first performed before the digital processing is stored in the detected image database as a training sample. The Film is digitally processed using a Film digital scanner (Film digital scanner). The DS level (highest level) scanning of the blackness range of 0.5-4.5D is realized on the traditional film, the scanning highest resolution is 7980 multiplied by 9690 and 570DPI, the line pair can reach 11LP/mm, and the geometric definition is better than 1 percent or 2 pixels in the transverse direction and the longitudinal direction.
After digital processing, based on a super-resolution technology, noise, gray scale deviation and texture loss in the scanning process are further eliminated through methods such as edge-oriented interpolation (NEDI) and image data structure Fitting (FS), so that the defect condition of the turbine blade is reflected more accurately by pictures.
Then, about 2000 high-resolution digital pictures are selected, the sizes of the pictures are uniformly fixed to be (4800 ), and each picture is subjected to image processing such as turning, scaling, rotation, brightness increase and decrease, and the quantity of the pictures is expanded by 10 times.
It should be understood that the preset number of the present embodiment is about 2000 pictures, and each picture is respectively subjected to image processing such as flipping, scaling, rotating, brightness increasing and decreasing, and the like, so as to expand by 10 times the picture amount. The number of the sample pictures is related to the effect of the model training, and a person skilled in the art can deduce the required picture amount according to the experimental effect, which is not described herein again.
And manually calibrating the type and position information of the defect in each image as a defect image label for the defective engine blade digital image in the defective blade image database.
The labels are answers in the deep network training process, and as the turbine blade defects are various in types, complex in features and high in detection image dimensionality, as shown in fig. 3, based on typical characteristic parameters of edge smoothness, end sharpness, trend change rate, relative position, symmetry, included angle, length-to-axis ratio, defect and relative change rate of external gray scale of the defect, the main features of 5 defects, namely cracks, cold shut, air holes, slag inclusion and porosity, can be extracted in a fuzzy reasoning mode, as shown in fig. 4.
Specifically, in this embodiment, the defect categories are first classified into 5 categories, i.e., crack, cold shut, air hole, slag inclusion, and porosity.
The typical feature parameters are expressed in the image features as:
C＝{ci|i＝1,2,…,9}
further, define
D＝{dj|j＝1,2,…,5}
Is a defect class set.
Establishing a defect label data set, specifically comprising the following steps: firstly, the typical characteristic parameter ciDefect class d as a rule preconditionjAnd as a rule conclusion, establishing fuzzy reasoning between C and D by taking the rule credibility CF as a reasoning basis:
a decision expression according to the fuzzy generative rule:
and determining the defect type of each picture, manually calibrating and establishing a defect label. Wherein, ciAs a rule prerequisite, djIn order to conclude the rule, the rule is,
is the degree of membership and CF is the rule confidence.
And processing the defective blade image in the database according to the fuzzy reasoning method, calibrating the defective area and the defect type by adopting label making software, and making a label file corresponding to the defective blade image.
Step 2: and establishing an aircraft engine blade initial inspection model.
The method comprises the following steps: the DenseNet model was selected, with the Classification layer removed. The model is used for extracting main features, two full connection layers are connected after 4 th Dense Block of the model, and then a two-classification output layer is connected, so that a two-classification deep convolutional network model is established. And training by using the established deep convolution network model and taking the defective image and intact blade image samples as training data to obtain an aircraft engine blade initial inspection model.
In this embodiment, first, the Classification layers in the conventional DenseNet model are all removed, two full-connection layers are connected after the 4 th Dense Block of the model, and then a two-class output layer is connected to establish a two-class deep convolutional network model.
And (3) establishing a good or defective label file for the samples in the database during training: the one-hot coding is adopted to set the image label with the intact leaf as (1, 0), and the image label with the defective leaf as (0, 1). And setting the Batch-size to be 2 according to the GPU operation speed, and training the detection model by adopting a transfer learning method.
In one embodiment, when the model for the initial blade inspection of the aircraft engine is obtained through training, the method further comprises training the model according to a preset learning rate, and when the model is trained to a preset number of times, sequentially attenuating the learning rate.
For example, the learning rate is adjusted to 0.0001, the model is trained, the iteration is carried out for 120000 times, the learning rate is attenuated at 80000 and 100000 times, and the attenuation rate is 0.5 each time, so that the aircraft engine blade initial inspection model is obtained.
And step 3: establishing and training an aero-engine blade rechecking model:
the method comprises the following steps: firstly, establishing a defect detection identification model based on a Faster RCNN algorithm and an SSD algorithm, randomly extracting 10-20% of images from a defective blade image data set as defect detection test data, using the rest images as defect detection training data, and performing accelerated training by adopting a maximum entropy principle algorithm to obtain two sets of aero-engine blade defect reinspection models based on the Faster RCNN algorithm and the SSD algorithm.
As shown in fig. 5, in this example, a set of aeroengine blade defect detection and identification models is established based on a fast RCNN algorithm framework, the initial inspection model in step 2 is used as a main feature extraction network, and because training data has the characteristics of large information feature quantity and small defect target, a VGG network is used as a second main feature extraction network to connect two obtained feature maps according to a third dimension (channel number).
And then setting the prior frame size in the RPN of the fast RCNN according to the average height and width of the defect target, alternately training the RPN and the classification regression network in the training step according to the training step of the fast RCNN model, and training to obtain the aircraft engine blade defect reinspection model based on the fast RCNN algorithm.
As shown in fig. 6, in the present example, a second set of aeroengine blade defect detection and identification models are established based on the SSD algorithm, the size of the prior frame under the SSD frame is changed according to the defect target average height and width, and the number of layers is increased to extract more features. And setting the Batch-size to be 2 according to the GPU operation speed, training a detection model by adopting a transfer learning method, and training to obtain an aero-engine blade defect recheck model based on the SSD algorithm.
In one embodiment, when the two sets of aero-engine blade defect review models are obtained through training, a maximum entropy principle algorithm acceleration training technology is adopted, and the method specifically comprises the following steps:
transforming the Boltzmann equation H theorem into a mathematical selection path: the cross entropy feedback in the traditional deep learning network is changed into entropy feedback based on the H theorem, namely, a cross entropy feedback function in the traditional algorithm is changed.
Transforming the gradient into entropy increase and dissipation: changing a cross entropy gradient algorithm in a traditional deep learning network into a nonlinear entropy increasing and dissipating function; namely, the traditional method of adjusting the learning network by means of entropy gradient is changed into the method of adjusting the learning network by means of nonlinear entropy increase and dissipation, so that the learning speed is accelerated.
And training the model according to a preset learning rate, and sequentially attenuating the learning rate when the model is trained to a preset number of times. For example, the learning rate is adjusted to 0.0001, the model is trained, the iteration is carried out for 120000 times, the learning rate is attenuated when 80000 and 100000 times are carried out, the attenuation rate is 0.5 each time, and the double-aircraft-engine blade defect detection and identification model is obtained.
And 4, step 4: detecting whether the aero-engine blade has a defect or not by using the aero-engine blade initial inspection model obtained in the step 2, and further detecting and identifying the type and the position of the defect by using the aero-engine blade rechecking model obtained in the step 3 when the aero-engine blade with the defect is judged; and the final defect type confidence coefficient and position information are obtained by respectively carrying out weighted average on the defect type confidence coefficient and the position information in two detection results output by the two sets of aero-engine blade defect rechecking models.
After an aero-engine blade initial inspection model and a double-aero-engine blade defect detection and identification model are obtained through training, the digital image of the aero-engine blade to be detected, which is generated through CT scanning, is sequentially input into the three models, whether the blade has defects or not is detected, and the type and the specific position of the defects are identified.
All modules for identifying the blade defects of the aircraft engine based on artificial intelligence detection can be wholly or partially realized through software, hardware and a combination of the software and the hardware. The modules can be embedded in a hardware form or independent from a processor in the computer device, and can also be stored in a memory in the computer device in a software form, so that the processor can call and execute operations corresponding to the modules.
The memory and the processor are electrically connected, directly or indirectly, to enable transmission or interaction of data. For example, the components may be electrically connected to each other via one or more communication buses or signal lines. The memory stores a computer program that can be executed on the processor, and the processor executes the computer program stored in the memory, thereby implementing the model control in the embodiment of the present invention.
The Memory may be, but is not limited to, a Random Access Memory (RAM), a Read Only Memory (ROM), a Programmable Read-Only Memory (PROM), an Erasable Read-Only Memory (EPROM), an electrically Erasable Read-Only Memory (EEPROM), and the like. The memory is used for storing programs, and the processor executes the programs after receiving the execution instructions.
The processor may be an integrated circuit chip having data processing capabilities. The Processor may be a general-purpose Processor including a Central Processing Unit (CPU), a Network Processor (NP), and the like. The methods, steps, and flow diagrams disclosed in embodiments of the invention may be implemented or performed. A general purpose processor may be a microprocessor or the processor may be any conventional processor or the like.
Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention, and that variations, modifications, substitutions and alterations can be made in the above embodiments by those of ordinary skill in the art without departing from the principle and spirit of the present invention.
Claims (6)
1. The utility model provides an aeroengine blade defect quick accurate identification method based on artificial intelligence which characterized in that: the method comprises the following steps:
step 1: establishing a turbine blade radiographic image database which comprises a perfect blade image database and a defective blade image database; the defect leaf image database calibrates the type and the position of the defect in each defect leaf digital image;
step 2: establishing and training an aircraft engine blade initial inspection model:
selecting a DenseNet model, removing a Classification layer, connecting two full-connection layers after the 4 th Dense Block of the model, and then connecting a two-Classification output layer, thereby establishing a two-Classification deep convolution network model as an aircraft engine blade initial inspection model;
training an aircraft engine blade initial inspection model by using intact blade images and defective blade images as sample data;
and step 3: establishing and training an aero-engine blade rechecking model:
establishing a defect detection identification model based on a fast RCNN algorithm and an SSD algorithm respectively;
selecting defect detection training data and defect detection test data from a defect blade image database, and respectively training defect detection identification models established based on a fast RCNN algorithm and an SSD algorithm to obtain two sets of aero-engine blade defect review models;
and 4, step 4: detecting whether the aero-engine blade has a defect or not by using the aero-engine blade initial inspection model obtained in the step 2, and further detecting and identifying the type and the position of the defect by using the aero-engine blade rechecking model obtained in the step 3 when the aero-engine blade with the defect is judged; and the final defect type confidence coefficient and position information are obtained by respectively carrying out weighted average on the defect type confidence coefficient and the position information output by the two sets of aero-engine blade defect rechecking models.
2. The method for rapidly and accurately identifying the blade defects of the aircraft engine based on the artificial intelligence as claimed in claim 1, wherein the method comprises the following steps: in step 1, the digital image in the turbine blade radiographic image database includes a digital image obtained by scanning the blades through a computed tomography technology, a digital image obtained by performing digital processing on an existing blade X-ray scanning film, and a digital image obtained by performing image expansion and data enhancement processing on an existing digital image.
3. The method for rapidly and accurately identifying the blade defects of the aircraft engine based on the artificial intelligence as claimed in claim 1, wherein the method comprises the following steps: in the step 1, defect categories are extracted from typical characteristic parameters of defects of the turbine blade in a fuzzy reasoning mode, wherein the typical characteristic parameters of the defects of the turbine blade are 9 items in total, namely edge smoothness, end sharpness, trend change rate, relative position, symmetry, included angle, length-axis ratio, defect and relative change rate of external gray scale of the defect, and the defect categories are 5 categories including cracks, cold shut, air holes, slag inclusion and looseness; the typical feature parameters are expressed in the image features as:
C＝{ci|i＝1,2,…,9}
definition of
D＝{dj|j＝1,2,…,5}
Is a defect category set;
establishing a defect label data set: firstly, the typical characteristic parameter ciDefect class d as a rule preconditionjAnd as a rule conclusion, establishing fuzzy reasoning between C and D by taking the rule credibility CF as a reasoning basis:
a decision expression according to the fuzzy generative rule:
determining the defect type of each picture, manually calibrating and establishing a defect label; wherein, ciAs a rule prerequisite, djIn the conclusion of the rule, ω ≧ 0 is the degree of membership,
CF is the rule confidence.
4. The method for rapidly and accurately identifying the blade defects of the aircraft engine based on the artificial intelligence as claimed in claim 1, wherein the method comprises the following steps: in the step 2, setting an image label of a perfect blade image as (1, 0) and an image label of a defective blade image as (0, 1) by adopting one-hot coding, and training an aircraft engine blade initial inspection model by adopting the perfect blade image and the defective blade image as sample data through a transfer learning method; and in the training process, training the model according to a preset learning rate, wherein when the training is carried out for a preset number of times, the learning rate is attenuated.
5. The method for rapidly and accurately identifying the blade defects of the aircraft engine based on the artificial intelligence as claimed in claim 1, wherein the method comprises the following steps: the defect detection and identification model established based on the fast RCNN algorithm in the step 3 is characterized in that the initial detection model in the step 2 is used as a main feature extraction network, a VGG network is used as a second main feature extraction network, and two groups of feature graphs obtained by the two main feature extraction networks are connected according to a third dimension; during training, setting the size of a prior frame in an RPN (resilient packet network) of the fast RCNN according to the average height and width of a defect target, alternately training the RPN and a classification regression network in a training step according to the training step of the fast RCNN model, and training to obtain the aircraft engine blade reinspection model based on the fast RCNN algorithm.
6. The method for rapidly and accurately identifying the blade defects of the aircraft engine based on the artificial intelligence as claimed in claim 1, wherein the method comprises the following steps: and 3, when two sets of aero-engine blade reinspection models are trained in the step 3, accelerating the training by adopting a maximum entropy principle algorithm, wherein the method comprises the steps of converting the Boltzmann equation H theorem into a mathematical selection path, converting the gradient into entropy increase and dissipation, training the models according to a preset learning rate, and attenuating the learning rate when the preset times are trained.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010791906.5A CN111968084B (en) | 2020-08-08 | 2020-08-08 | Rapid and accurate identification method for defects of aero-engine blade based on artificial intelligence |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010791906.5A CN111968084B (en) | 2020-08-08 | 2020-08-08 | Rapid and accurate identification method for defects of aero-engine blade based on artificial intelligence |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111968084A CN111968084A (en) | 2020-11-20 |
| CN111968084B true CN111968084B (en) | 2022-05-20 |
Family
ID=73364795
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010791906.5A Active CN111968084B (en) | 2020-08-08 | 2020-08-08 | Rapid and accurate identification method for defects of aero-engine blade based on artificial intelligence |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111968084B (en) |
Families Citing this family (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112862771A (en) * | 2021-01-29 | 2021-05-28 | 珠海迪沃航空工程有限公司 | Airplane engine blade defect identification and diagnosis system and method based on deep learning |
| CN112907521B (en) * | 2021-02-05 | 2022-11-04 | 上海展湾信息科技有限公司 | SSD and GLCM based welding air hole target detection method |
| CN112991187B (en) * | 2021-04-28 | 2021-07-27 | 四川大学 | Convolution twin-point network blade profile splicing system based on multiple spatial similarities |
| CN113034498B (en) * | 2021-04-28 | 2023-11-28 | 江苏欧密格光电科技股份有限公司 | LED lamp bead defect detection and assessment method, device, computer equipment and medium |
| CN113313695A (en) * | 2021-06-05 | 2021-08-27 | 西北工业大学 | Automatic deep learning defect detection and identification method based on small sample aeroengine blade CT image |
| KR102394093B1 (en) * | 2021-09-13 | 2022-05-04 | 오병후 | Method of reading defects using X-ray images of helicopter blades |
| CN113634876B (en) * | 2021-10-18 | 2021-12-28 | 武汉逸飞激光股份有限公司 | Auxiliary calibration method and device for machine vision, electronic equipment and storage medium |
| CN115096891B (en) * | 2022-05-28 | 2024-05-07 | 国营芜湖机械厂 | Intelligent inspection method for aero-engine blades |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109993094A (en) * | 2019-03-26 | 2019-07-09 | 苏州富莱智能科技有限公司 | Fault in material intelligent checking system and method based on machine vision |
Family Cites Families (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102692429B (en) * | 2011-03-24 | 2013-10-30 | 中国科学院沈阳自动化研究所 | Method for automatic identification and detection of defect in composite material |
| CN103753098B (en) * | 2013-12-31 | 2016-08-31 | 成都青石激光科技有限公司 | Turbine engine blade automatic prosthetic appliance and restorative procedure thereof |
| US10656518B2 (en) * | 2017-12-17 | 2020-05-19 | United Microelectronics Corp. | Automatic inline detection and wafer disposition system and method for automatic inline detection and wafer disposition |
| CN108242054A (en) * | 2018-01-09 | 2018-07-03 | 北京百度网讯科技有限公司 | A kind of steel plate defect detection method, device, equipment and server |
| CN108363382B (en) * | 2018-02-09 | 2019-02-26 | 哈尔滨工业大学（威海） | A kind of complex equipment fault diagnosis method and system |
| US10473593B1 (en) * | 2018-05-04 | 2019-11-12 | United Technologies Corporation | System and method for damage detection by cast shadows |
| CN109522627B (en) * | 2018-11-01 | 2022-12-02 | 西安电子科技大学 | Fan blade icing prediction method based on SCADA (Supervisory control and data acquisition) data |
| CN109871905A (en) * | 2019-03-14 | 2019-06-11 | 同济大学 | A kind of plant leaf identification method based on attention mechanism depth model |
| CN110455822A (en) * | 2019-07-10 | 2019-11-15 | 苏州卓融新能源科技有限公司 | A kind of detection method of pcb board defect |
-
2020
- 2020-08-08 CN CN202010791906.5A patent/CN111968084B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109993094A (en) * | 2019-03-26 | 2019-07-09 | 苏州富莱智能科技有限公司 | Fault in material intelligent checking system and method based on machine vision |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111968084A (en) | 2020-11-20 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111968084B (en) | Rapid and accurate identification method for defects of aero-engine blade based on artificial intelligence | |
| CN107437245B (en) | High-speed railway contact net fault diagnosis method based on deep convolutional neural network | |
| Malarvel et al. | An improved version of Otsu's method for segmentation of weld defects on X-radiography images | |
| CN110148130B (en) | Method and device for detecting part defects | |
| US11501435B2 (en) | Unsupervised content-preserved domain adaptation method for multiple CT lung texture recognition | |
| CN115082464A (en) | Method and system for identifying welding seam data in welding process of dust remover | |
| CN111814867A (en) | Defect detection model training method, defect detection method and related device | |
| CN112465746B (en) | Method for detecting small defects in ray film | |
| CN112102229A (en) | Intelligent industrial CT detection defect identification method based on deep learning | |
| CN114549997B (en) | X-ray image defect detection method and device based on regional feature extraction | |
| CN110796661B (en) | Fungal microscopic image segmentation detection method and system based on convolutional neural network | |
| CN116758075B (en) | Artificial intelligence-based blower motor operation fault detection method | |
| CN111275700A (en) | Terminal defect detection method and system based on deep learning | |
| CN112906689B (en) | Image detection method based on defect detection and segmentation depth convolutional neural network | |
| CN113313107B (en) | Intelligent detection and identification method for multiple types of diseases on cable surface of cable-stayed bridge | |
| TWI786570B (en) | Generating a training set usable for examination of a semiconductor specimen | |
| CN113920096A (en) | Method for detecting metal packaging defects of integrated circuit | |
| JP2022027473A (en) | Generation of training data usable for inspection of semiconductor sample | |
| CN116612111B (en) | High-strength composite material processing quality detection method | |
| CN114419007A (en) | Defect type identification method and system of multi-strategy fusion deep learning network model | |
| CN113077416A (en) | Welding spot welding defect detection method and system based on image processing | |
| CN113313695A (en) | Automatic deep learning defect detection and identification method based on small sample aeroengine blade CT image | |
| CN115587989B (en) | Workpiece CT image defect detection segmentation method and system | |
| CN113627302B (en) | Ascending construction compliance detection method and system | |
| CN115601747A (en) | Method and system for calculating confluency of adherent cells |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |