AI-based engine blade defect identification and detection method and system

Abstract

The invention provides an AI-based engine blade defect identification and detection method and system, and the method comprises the steps: S1, scanning an engine blade through an eddy current detection probe according to the sequence of the front edge, the blade body, the rear edge and the root of the engine blade, obtaining an eddy current response signal, and transmitting the eddy current response signal to a signal processing system; s2, extracting defect features from the defect eddy current signal of the defective engine blade; extracting a defect-free feature from the defect-free eddy current signal of the defect-free engine blade; s3, establishing and training an engine blade detection model through the extracted defect features and defect-free features; and S4, whether the engine blade to be detected has defects or not is detected through the engine blade model. According to the invention, based on artificial intelligence eddy current detection, the accuracy and efficiency of detection are improved, the probability of missing detection and false detection is reduced, and the technical problem of missing detection or false detection caused by low detection efficiency in the prior art is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of quality inspection of engine blades, and particularly to an AI-based method and system for defect identification and detection of engine blades. Background Art

[0002] The integrity of the blades is crucial for the performance and safety of the engine. Traditional methods for detecting engine blades mainly rely on manual inspection and basic detection equipment, such as ultrasonic testing and eddy current testing. These methods usually require a large amount of manual intervention, have low detection efficiency, and are easily affected by human factors, resulting in missed or false detections. At the same time, due to the complex geometric shape of the engine blades, traditional detection methods are difficult to comprehensively cover all parts of the blades, especially the detection ability for internal defects is limited. The deficiencies of these existing technologies affect the accuracy and reliability of the detection. Summary of the Invention

[0003] Based on the defects in the above background art, the present invention proposes an AI-based method and system for defect identification and detection of engine blades, and the technical solutions adopted are as follows:

[0004] An AI-based method for defect identification and detection of engine blades, the method comprising:

[0005] S1: Scanning the engine blade in the order of the leading edge, blade body, trailing edge, and root of the engine blade through an eddy current detection probe to obtain an eddy current response signal, and transmitting the eddy current response signal to a signal processing system;

[0006] S2: Extracting defect features from the defect eddy current signals of defective engine blades; extracting non-defect features from the non-defect eddy current signals of non-defective engine blades;

[0007] S3: Establishing and training an engine blade detection model through the extracted defect features and non-defect features;

[0008] S4: Detecting whether there are defects in the engine blade to be detected through the engine blade model.

[0009] Preferably, the S1 includes, before scanning the leading edge, blade body, trailing edge, and root of the engine blade, first performing a full-body pre-scan on the engine blade to obtain the main dimensions and shape features of the engine blade, and comparing them with the main dimensions and shape of the designed engine blade to ensure that the overall shape of the engine blade is consistent with the overall shape of the designed engine blade.

[0010] Preferably, the S1 further includes:

[0011] The leading edge, blade body, trailing edge, and root of the engine blade are scanned by an eddy current detection probe to obtain an eddy current response signal, and the eddy current response signal is transmitted to a signal processing system.

[0012] Preferably, the S2 includes:

[0013] Scan defective engine blades to obtain defective eddy current response signals, and transmit the defective eddy current response signals to a signal processing system;

[0014] The signal processing system extracts defect features based on the defective eddy current response signals;

[0015] Take known defect-free engine blades as samples to obtain defect-free features.

[0016] Preferably, the defect features include:

[0017] Frequency features, amplitude features, and phase features containing defects, and the three features are integrated into a defect feature set.

[0018] Preferably, the S3 includes:

[0019] Integrate the defect feature set and the defect-free features to obtain a data set, divide the data set into a training set and a validation set, select a random forest as the engine blade detection model, train the engine blade detection model with the training set, and verify the engine blade detection model with the validation set. When the accuracy of the verification result reaches the set value, it indicates that the training of the engine blade detection model is completed, and the trained engine blade detection model is put into use.

[0020] Preferably, the S4 includes:

[0021] Extract the features to be detected of the engine blade to be detected. The features to be detected include the frequency features, amplitude features, and phase features to be detected, and detect the features to be detected through the engine blade detection model.

[0022] Preferably, the engine blade detection model detects the features to be detected, specifically including:

[0023] Determine the location of the engine blade defect by detecting the amplitude feature to be detected; determine the depth of the engine blade defect by detecting the phase feature to be detected; determine the size and shape of the engine blade defect by detecting the frequency feature to be detected.

[0024] Preferably, the engine blade detection model detects the feature to be detected and generates a detection report on the detection result. Specifically, if no defect is present, a detection report indicating no defect in the engine blade to be detected is generated; if a defect is present, a report on the location, depth, size, and shape of the engine defect is generated.

[0025] An AI-based engine blade defect identification and detection system, the system comprising:

[0026] Signal acquisition system: Scans the engine blade in the order of the leading edge, blade body, trailing edge, and root of the engine blade through an eddy current detection probe, obtains an eddy current response signal, and transmits the eddy current response signal to the signal processing system;

[0027] Feature extraction system: Traverses all engine blades with various defects that have occurred, obtains the defect eddy current response signals of all the engine blades with various defects that have occurred in the same manner as the signal acquisition system, and extracts several defect features from the defect eddy current signals; obtains the defect-free eddy current response signals of defect-free engine blades in the same manner as the signal acquisition system, and extracts several defect-free features from the defect-free eddy current signals;

[0028] Model training system: Establishes an engine blade detection model, uses the defect features and defect-free features as sample data to train the engine blade detection model, and puts the trained engine blade detection model into use;

[0029] Detection decision-making system: Obtains the feature to be detected of the engine blade to be detected in the same manner as the signal acquisition system and the feature extraction system, and uses the engine blade detection model obtained in S3 to detect the feature to be detected, and determines whether there is a defect in the engine blade to be detected according to the detection result.

[0030] Advantages of the present invention: The AI-based engine blade defect identification and detection method and system improve the accuracy and efficiency of detection, and reduce the probability of missed detection and false detection by using machine learning algorithms and automated processes. The system can comprehensively cover all parts of the detected blade and provide accurate positioning and evaluation of internal defects, including the location, depth, size, and geometric shape of the defects. This not only speeds up the detection speed and shortens the time from detection to analysis, but also reduces errors caused by human factors. Overall, this technical solution significantly improves the intelligence level of engine blade detection and provides important support for the safety maintenance of aeroengines. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 An AI-based engine blade defect identification and detection method according to the present invention. DETAILED DESCRIPTION

[0032] The preferred embodiments of the present invention will be described below in conjunction with the accompanying drawings. It should be understood that the preferred embodiments described herein are only for the purpose of illustrating and explaining the present invention, and are not intended to limit the present invention.

[0033] In one embodiment of the present invention, based on the deficiencies in the above-mentioned background art, the present invention proposes a method and system for defect identification and detection of engine blades based on AI, and the technical solutions adopted are as follows:

[0034] A method for defect identification and detection of engine blades based on AI, the method comprising:

[0035] S1: Scanning the engine blade in the order of the leading edge, blade body, trailing edge and root of the engine blade through an eddy current detection probe to obtain an eddy current response signal, and transmitting the eddy current response signal to a signal processing system;

[0036] S2: Extracting defect features from the defect eddy current signals of defective engine blades; extracting non-defect features from the non-defect eddy current signals of non-defective engine blades;

[0037] S3: Establishing and training an engine blade detection model through the extracted defect features and non-defect features;

[0038] S4: Detecting whether there are defects in the engine blade to be detected through the engine blade model.

[0039] The working principle and effect of the above technical solution are as follows: First, use the eddy current detection probe to scan in the order of the leading edge, blade body, trailing edge and root of the engine blade, and transmit the obtained eddy current response signal to the signal processing system, which processes it; then extract the key features that can characterize the defective and non-defective states from the eddy current signals of defective and non-defective engine blades respectively; then use these extracted features to construct and train an engine blade detection model, enabling the model to learn the feature difference patterns in the defective and non-defective states, and continuously adjusting the model parameters during the training process to optimize the recognition ability; finally, input the eddy current signal features of the engine blade to be detected into the trained model, and the model judges whether the measured blade has defects according to the learned feature patterns. Traditional methods for measuring the depth of defects often require cutting or damaging the object, or using complex and expensive tomography techniques. However, this method calculates the defect depth by measuring the phase angle change in combination with material properties and working frequency, which is a non-invasive method and does not require damaging or disassembling the engine blade, and can complete the depth measurement without affecting the normal use of the blade; reducing the detection difficulty.

[0040] In one embodiment of the present invention, S1 includes, before scanning the leading edge, blade body, trailing edge, and root of the engine blade, first performing a full-body pre-scan on the engine blade to obtain the main dimensions and shape features of the engine blade, and comparing them with the main dimensions and shape of the designed engine blade to ensure that the overall shape of the engine blade is consistent with the overall shape of the designed engine blade.

[0041] The working principle and effect of the above technical solution are as follows: By pre-scanning, the actual main dimension and shape feature data of the engine blade are obtained, compared with the main dimensions and shape data of the designed engine blade, and the data matching and difference analysis algorithm is used to determine whether the overall shape of the blade conforms to the design requirements. Pre-scanning the engine blade throughout the body and comparing the shape features can detect in advance whether there are deviations in the overall shape of the blade, avoid interference with subsequent detection results caused by blade manufacturing errors or severe deformation, ensure that the subsequent eddy current scanning detection of the leading edge, blade body, trailing edge, and root is carried out on the correct shape basis, and improve the accuracy of defect identification.

[0042] In one embodiment of the present invention, S1 further includes:

[0043] Scanning the leading edge, blade body, trailing edge, and root of the engine blade through an eddy current detection probe to obtain an eddy current response signal, and transmitting the eddy current response signal to a signal processing system.

[0044] The working principle and effect of the above technical solution are as follows: When the eddy current detection probe approaches the engine blade, an alternating current is passed through the coil in the probe to generate an alternating magnetic field. The magnetic field will generate an induced current, i.e., eddy current, in the engine blade. Due to factors such as the material properties, geometric shapes, and whether there are defects in different parts of the engine blade, the distribution and magnitude of the eddy current will change, and then different eddy current magnetic fields will be generated. This changing eddy current magnetic field will generate an induced electromotive force in the coil of the detection probe, forming an eddy current response signal. By scanning each part of the engine blade with an eddy current detection probe and transmitting the signal to the processing system, the eddy current effect can be used to detect the blade non-contactingly, avoiding damage to the blade; this method has a fast detection speed, is suitable for batch detection, and improves the detection efficiency.

[0045] In one embodiment of the present invention, S2 includes:

[0046] In one embodiment of the present invention, the defect features include:

[0047] Frequency features, amplitude features, and phase features containing defects, and integrating the three features into a defect feature set.

[0048] The working principle and effects of the above technical solution are as follows: In eddy current testing, when there are defects in the engine blade, it will cause changes in the eddy current distribution, which in turn affects the frequency, amplitude, and phase of the eddy current response signal. For the frequency characteristic, the presence of defects changes the eddy current path and distribution, causing the interaction frequency between the eddy current and the probe coil to change. Different types and sizes of defects will cause specific frequency offsets because the defects destroy the uniformity and integrity of the blade material, resulting in changes in the dynamic characteristics of electromagnetic induction. In terms of the amplitude characteristic, the defects will change the intensity of the eddy current. Since the electromagnetic properties such as conductivity and permeability in the defect area are different from those in the defect-free area, the flow of the eddy current in these areas is hindered or enhanced, thus changing the amplitude of the induced eddy current response signal. The larger the defect or the more significant its impact on the electromagnetic properties, the more obvious the amplitude change. In terms of the phase characteristic, the defects will cause a change in the phase relationship between the eddy current magnetic field and the excitation magnetic field. This is because the defects change the propagation path and time delay of the eddy current, resulting in a different phase difference between the induced electromotive force generated by the eddy current and the excitation signal in time. Different types of defects will cause different degrees of phase changes. Integrating the three characteristics unifies the expression form of defect information and reduces the complexity of data storage.

[0049] In one embodiment of the present invention, the S4 includes:

[0050] Extracting the to-be-detected characteristics of the engine blade to be detected, where the to-be-detected characteristics include the to-be-detected frequency characteristic, amplitude characteristic, and phase characteristic, and detecting the to-be-detected characteristics through the engine blade detection model. Moreover, the engine blade detection model determines whether there are defects in the engine blade to be detected according to the feature score, and the feature score is obtained according to the following formula:

[0051]

[0052] where i and j represent the index variables corresponding to the to-be-detected feature values, used to traverse the elements in the set {F, A, P}, F, A, and P respectively represent the frequency feature value, amplitude feature value, and phase feature value, i0 and j0 represent the index variables corresponding to the defect-free feature values, and i1 and j1 respectively represent the standard deviations corresponding to the defect-free feature values.

[0053] When S ≥ the predefined threshold K, it is determined that there are defects;

[0054] When S < the predefined threshold K, it is determined that there are no defects.

[0055] The working principle and effects of the above technical solution are as follows: In the above feature score calculation formula, by subtracting the defect-free feature value from the feature value to be detected and then dividing by the standard deviation of the defect-free feature value, it represents that the deviation between the currently detected feature value and the defect-free feature value is scaled by the standard deviation. This value (feature difference value) indicates how many times this frequency deviation appears within the normal fluctuation range, which is very useful for identifying outliers and defects. Because in the standard normal distribution, the significance of the deviation can be intuitively judged by multiples of the standard deviation. Taking the feature difference values of the frequency feature and the amplitude feature as an example, when considering the product of the feature difference values of frequency and amplitude, it is actually exploring how the change in frequency affects the change in amplitude and what additional information this combined change may reveal. The simultaneous occurrence of significant deviations in two features indicates the existence of a specific defect, and this complex relationship can be more effectively captured through interaction terms. Some physical defects may not only be manifested as anomalies in a single feature but as a result of the combined action of multiple features. For example, a certain material defect causes a change in the signal frequency and simultaneously affects the signal amplitude. Through the product term, the model can capture physical phenomena that cannot be revealed by the change in a single feature. Using historical defect-free data, calculate the feature score distribution during normal operation. Through statistical analysis, calculate the mean and standard deviation of the feature value distribution, and add 2 standard deviations to the mean to obtain the predefined threshold K.

[0056] In one embodiment of the present invention, the engine blade detection model detects the feature to be detected, specifically including:

[0057] By detecting the amplitude feature to be detected, determine the location where the engine blade defect is located; by detecting the phase feature to be detected, determine the depth of the engine blade defect; by detecting the frequency feature to be detected, determine the size and shape of the engine blade defect. And the depth of the engine blade defect is obtained through the following formula:

[0058]

[0059] Where D represents the depth of the engine blade defect, c represents the electromagnetic wave velocity of the engine material, f represents the operating frequency of the alternating current in the eddy current detection probe (the operating frequency is divided into high frequency and low frequency), φ0 represents the phase angle of the eddy current signal measured when the eddy current detection probe scans the defect area, and φ1 represents the phase angle of the eddy current signal measured when the eddy current detection probe scans the defect-free area;

[0060] And the shape of the engine blade defect is judged through the following formula:

[0061]

[0062] Wherein, L represents the length of the defect, f0 represents the resonance frequency of the defect edge, f1 represents the resonance frequency of the defect center, λ represents the skin depth, that is, the effective penetration depth of eddy current in the material, and, Wherein, ω represents the electromagnetic oscillation angular frequency of the eddy current field in the engine blade material, μ represents the magnetic permeability, that is, the magnetic conductivity of the engine blade material, and σ represents the electrical conductivity, that is, the electrical conductivity of the engine blade material;

[0063]

[0064] Wherein, w represents the width of the defect, ΔZ1 represents the impedance change amount when the operating frequency f of the alternating current is in the low-frequency state, and ΔZ2 represents the impedance change amount when the operating frequency f of the alternating current is in the high-frequency state;

[0065] When the aspect ratio is satisfied, the defect is determined to be a crack. When the aspect ratio is satisfied, the defect is determined to be a hole.

[0066] The working principle and effect of the above technical solution are as follows: When the eddy current detection probe scans the engine blade, the defect-free area and the defective area will cause different changes in the phase angle of the eddy current signal. The presence of the defect will interfere with the normal distribution of the eddy current, causing the phase angle to change. By measuring this phase angle change, it is possible to determine whether there is a defect in the blade and the approximate location of the defect. However, the relationship between the defect size and the phase angle is not a simple linear relationship; in the defect size calculation formula for the engine blade, by introducing the ln function, the non-linear relationship can be accurately expressed, thereby improving the accuracy of defect size calculation. By comprehensively analyzing multiple parameters such as the phase angle change, material properties, and operating frequency, rather than relying solely on a single parameter to judge the defect, the misjudgment or inaccuracy caused by single-parameter detection is avoided.

[0067] An embodiment of the present invention, an AI-based engine blade defect identification and detection system, the system includes:

[0068] Signal acquisition system: The engine blade is scanned by an eddy current detection probe in the order of the leading edge, blade body, trailing edge, and root of the engine blade to obtain an eddy current response signal, and the eddy current response signal is transmitted to the signal processing system;

[0069] Feature extraction system: Traverse all engine blades with various defects that have occurred, and obtain the defect eddy current response signals of all engine blades with various defects that have occurred in the same way as the signal acquisition system, and extract several defect features from the defect eddy current signals; obtain the defect-free eddy current response signals of defect-free engine blades in the same way as the signal acquisition system, and extract several defect-free features from the defect-free eddy current signals;

[0070] Model training system: Establish an engine blade detection model, use defect features and defect-free features as sample data to train the engine blade detection model, and put the trained engine blade detection model into use;

[0071] Detection decision-making system: Obtain the features to be detected of the engine blade to be detected in the form of a signal acquisition system and a feature extraction system, and use the engine blade detection model obtained by the model training system to detect the features to be detected, and judge whether there are defects in the engine blade to be detected according to the detection results.

[0072] The working principle and effect of the above technical solution are as follows: First, use an eddy current detection probe to scan in the order of the leading edge, blade body, trailing edge and root of the engine blade, and transmit the obtained eddy current response signal to the signal processing system, and the signal processing system processes it; then extract the key features that can characterize the defective and defect-free states from the eddy current signals of defective and defect-free engine blades respectively; then use these extracted features to construct and train an engine blade detection model, so that the model learns the feature difference patterns in the defective and defect-free states, and continuously adjusts the model parameters during the training process to optimize the recognition ability; finally, input the eddy current signal features of the engine blade to be detected into the trained model, and the model judges whether the measured blade has defects according to the learned feature patterns. Traditional methods for measuring the depth of defects often require cutting or damaging the object, or using complex and expensive tomography techniques. However, this method calculates the defect depth by measuring the phase angle change in combination with material properties and working frequency, which is a non-invasive method that does not require damaging or disassembling the engine blade and can complete the depth measurement without affecting the normal use of the blade; it reduces the detection difficulty.

[0073] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalent technologies, the present invention is also intended to include these modifications and variations.

Claims

1. An AI-based method for identifying and detecting defects in engine blades, characterized in that, The method includes: S1: Scanning the engine blade in the order of the leading edge, blade body, trailing edge, and root of the engine blade by an eddy current detection probe to obtain an eddy current response signal, and transmitting the eddy current response signal to a signal processing system; S2: Extracting defect features from the defect eddy current signals of defective engine blades; extracting defect-free features from the defect-free eddy current signals of defect-free engine blades; S3: Establishing and training an engine blade detection model based on the extracted defect features and defect-free features; S4: Detecting whether there are defects in the engine blade to be detected through the engine blade model.

2. The method for identifying and detecting engine blade defects based on AI according to claim 1, wherein The S1 includes that before scanning the leading edge, blade body, trailing edge, and root of the engine blade, a full-body pre-scan of the engine blade is first performed to obtain the main dimensions and shape features of the engine blade, and comparing them with the main dimensions and shape of the designed engine blade to ensure that the overall shape of the engine blade is consistent with the overall shape of the designed engine blade.

3. The method for identifying and detecting engine blade defects based on AI according to claim 1, wherein, The S1 also includes: Scanning the leading edge, blade body, trailing edge, and root of the engine blade by an eddy current detection probe to obtain an eddy current response signal, and transmitting the eddy current response signal to a signal processing system.

4. The method for identifying and detecting engine blade defects based on AI according to claim 1, characterized in that, The S2 includes: Scanning the defective engine blade to obtain a defect eddy current response signal, and transmitting the defect eddy current response signal to a signal processing system; The signal processing system extracts defect features based on the defect eddy current response signal Taking the known defect-free engine blade as a sample to obtain defect-free features.

5. The method for defect identification and detection of engine blades based on AI according to claim 3, wherein, The defect features include: Frequency features, amplitude features, and phase features containing defects, and integrating the three features into a defect feature set.

6. The method for identifying and detecting defects of engine blades based on AI according to claim 1, wherein, The S3 includes: Integrating the defect feature set and defect-free features to obtain a data set, dividing the data set into a training set and a validation set, selecting a random forest as the engine blade detection model, training the engine blade detection model through the training set, and validating the engine blade detection model through the validation set. When the accuracy of the validation result reaches the set value, it indicates that the training of the engine blade detection model is completed, and the trained engine blade detection model is put into use.

7. A method for identifying and detecting engine blade defects based on AI according to claim 1, characterized in that, The S4 includes: S41: Extracting the features to be detected of the engine blade to be detected, where the features to be detected include the frequency features, amplitude features, and phase features to be detected, and detecting the features to be detected through the engine blade detection model.

8. The method for identifying and detecting engine blade defects based on AI according to claim 6, wherein, The detection of the features to be detected by the engine blade detection model specifically includes: Determining the position where the engine blade defect is located by detecting the amplitude feature to be detected; determining the depth of the engine blade defect by detecting the phase feature to be detected; determining the size and shape of the engine blade defect by detecting the frequency feature to be detected.

9. The method for identifying and detecting engine blade defects based on AI according to claim 6, characterized in that, The detection of the features to be detected by the engine blade detection model generates a detection report for the detection result, specifically including that if no defect exists, a defect-free detection report for the engine blade to be detected is generated; If a defect exists, a report on the location, depth, size, and shape of the engine defect is generated. Moreover, the engine blade detection model determines whether there is a defect in the engine blade to be detected based on the feature score, and the feature score is obtained according to the following formula: Where i and j represent the index variables corresponding to the feature values to be detected, which are used to traverse the elements in the set {F, A, P}, where F, A, and P represent the frequency feature value, amplitude feature value, and phase feature value respectively, i0 and j0 represent the index variables corresponding to the defect-free feature values, and i1 and j1 represent the standard deviations of the defect-free feature values respectively; When S ≥ the predefined threshold K, it is determined that there is a defect; When S < the predefined threshold K, it is determined that there is no defect.

10. An AI-based engine blade defect identification and detection system, characterized in that, The system includes: Signal acquisition system: Scans the engine blade through an eddy current detection probe in the order of the leading edge, blade body, trailing edge, and root of the engine blade, obtains the eddy current response signal, and transmits the eddy current response signal to the signal processing system; Feature extraction system: Traverses all engine blades with various defects that have occurred, obtains the defect eddy current response signals of all the engine blades with various defects that have occurred in the same way as the signal acquisition system, extracts several defect features from the defect eddy current signals; obtains the defect-free eddy current response signals of the defect-free engine blades in the same way as the signal acquisition system, and extracts several defect-free features from the defect-free eddy current signals; Model training system: Establishes an engine blade detection model, uses the defect features and defect-free features as sample data to train the engine blade detection model, and puts the trained engine blade detection model into use; Detection decision system: Obtains the features to be detected of the engine blade to be detected in the same way as the signal acquisition system and the feature extraction system, and uses the engine blade detection model obtained by the model training system to detect the features to be detected, and determines whether there is a defect in the engine blade to be detected according to the detection result.

Images