Single-probe blade crack online detection method based on blade tip vibration envelope profile diagram

Abstract

The invention relates to a single-probe blade crack online detection method based on a blade tip vibration envelope profile diagram, and belongs to the field of rotating machinery blade monitoring, and the method comprises the following steps: collecting a pulse voltage waveform when a blade rotates to pass through a capacitive sensor, and recording the time when equipment rotates by each circle; calculating the vibration characteristics of the blade tip at the moment of scanning the capacitance sensor to trigger the pulse voltage waveform in the blade rotation process based on the pulse voltage waveform; the vibration characteristics comprise blade tip vibration displacement and instantaneous vibration speed; carrying out noise reduction processing on the solved blade tip vibration displacement and instantaneous vibration speed; based on the blade tip vibration displacement and the instantaneous vibration speed, constructing an envelope profile diagram of resonance area displacement response; and inputting the envelope profile diagram fused with the vibration characteristics into a neural network model for crack diagnosis. According to the method, the blade tip vibration displacement, the instantaneous vibration speed and the dynamic response characteristics of the resonance area are fused in the form of the envelope profile diagram, and blade respiration crack online diagnosis of the single sensor probe is achieved.

Description

Technical Field

[0001] The present invention relates to the field of rotating machinery blade monitoring, and in particular to a single-probe blade crack online detection method based on a blade tip vibration envelope contour diagram, which is suitable for real-time diagnosis of early fatigue cracks in blades of high-speed rotating machinery such as aircraft engines and gas turbines. Background Art

[0002] During the operation of large rotor machinery, monitoring blade condition and promptly diagnosing fatigue crack failures are crucial to ensuring safe equipment operation. Traditional non-contact blade tip vibration displacement measurement methods analyze the blade tip vibration displacement signal to identify vibration parameters for blade condition monitoring. However, classic blade tip vibration displacement measurement methods, such as blade tip timing (BTT), fundamentally suffer from undersampling, meaning that the measurement data available during equipment operation is limited. This poses a significant challenge to determining blade vibration parameters and extracting crack-related features from the signal. Furthermore, increasing the number of sensors due to signal undersampling increases costs and can easily damage the original equipment structure. Therefore, an online blade crack diagnosis solution that can use a small number of probes for measurement, acquire multiple feature data for crack diagnosis, and exhibit strong noise robustness is urgently needed to address these issues. Summary of the Invention

[0003] In order to solve the problems existing in the above-mentioned prior art, the present invention provides a single-probe blade crack online detection method based on the blade tip vibration envelope contour map, which collects the pulse voltage waveform through a single sensor probe and constructs an envelope contour map, and inputs the envelope contour map into the neural network model to perform blade breathing crack detection.

[0004] In order to achieve the above object, the present invention provides the following technical solutions:

[0005] A single-probe blade crack online detection method based on a blade tip vibration envelope profile diagram comprises the following steps:

[0006] S100, collecting a pulse voltage waveform when the blade rotates through the capacitive sensor, and recording the time it takes for the device to complete each rotation;

[0007] S200, calculating the vibration characteristics of the blade tip when the blade sweeps across the capacitance sensor triggering pulse voltage waveform during the blade rotation process based on the pulse voltage waveform;

[0008] The vibration characteristics include blade tip vibration displacement and instantaneous vibration velocity;

[0009] S300, performing noise reduction processing on the solved blade tip vibration displacement and instantaneous vibration velocity;

[0010] S400, constructing an envelope contour diagram of the displacement response of the resonance region based on the blade tip vibration displacement and the instantaneous vibration velocity;

[0011] S500: Inputting the envelope contour image of the fused vibration characteristics into a neural network model for crack diagnosis.

[0012] Furthermore, in step S400, the following steps are used to construct an envelope contour diagram of the displacement response of the resonance region:

[0013] S401, selecting a resonance region time window;

[0014] S402, intercepting the instantaneous vibration velocity within the time window, and converting the instantaneous vibration velocity into a set of blade tip vibration displacements within the resonance region;

[0015] S403, the first blade tip vibration displacement d solved from the measurement results in the time window B The second blade tip vibration displacement obtained after conversion is Perform a symmetric transformation along the x-axis and plot the signal d in the same coordinate system B (t),–d B (t), and

[0016] S404, finding the outermost point set of the area covered by all scattered points consisting of the four groups of first blade tip vibration displacement and second blade tip vibration displacement;

[0017] S405, using a smooth curve to fit the upper and lower boundary points of the outermost point set, the area enclosed by the smooth curve is the envelope contour of the vibration signal of the resonance area;

[0018] S406 , filling the area enclosed by the envelope contour image and converting it into a standard binary envelope contour image with pixels of H×W, where H is the height of the image and W is the width of the image.

[0019] Furthermore, in step S402, the instantaneous vibration velocity v captured in the time window is converted into B Converted into a set of blade tip vibration displacements in the resonance region

[0020]

[0021] Among them, D AM is the peak-to-peak value of the blade tip vibration displacement measurement signal within the time window, V AM is the peak-to-peak value of the velocity signal in the time window, v B is the instantaneous vibration speed.

[0022] Furthermore, in step S404, the Alpha Shapes non-convex hull algorithm is used to identify the signal d B (t),–d B (t), and The outermost point of the scattered point set.

[0023] Furthermore, in step S401, the process of selecting the resonance region time window is:

[0024] After the blade resonates, record the time T corresponding to the peak value of the blade tip vibration displacement peak ;

[0025] T peak The starting time T is when the vibration displacement of the blade tip near the blade resonance zone begins to rise. start , the displacement signal decays after passing the resonance peak T peak -T start The final time T end , the middle section is the time window containing the blade resonance area; the time window satisfies T peak -T start =T end -T peak .

[0026] Furthermore, in step S500, the neural network model is an AlexNet-EAFU model, which includes an EAFU module, an AlexNet main part, a 1×1 convolution kernel, a convolution layer, a pooling layer and a fully connected layer; the ends of the EAFU module and the AlexNet main part are respectively connected to the 1×1 convolution kernel data, and the 1×1 convolution kernel data is input into the convolution layer, the pooling layer, and the fully connected layer in sequence; wherein, the EAFU module includes a convolution layer, a pooling layer, a hole convolution layer, and a spatial attention module, and the AlexNet main part includes a convolution layer and a pooling layer.

[0027] Furthermore, in step S500, the specific steps of determining whether the blade has cracks using the neural network model are as follows:

[0028] S501, constructing a training data set;

[0029] For the blade to be tested, a finite element dynamics model is established based on the set parameter information, and numerical simulation is performed to calculate multiple groups of vibration signals of crack-free blades and multiple groups of vibration signals of cracked blades. In step S400, multiple binary envelope contour images are constructed as training data sets;

[0030] S502, training the training data set samples using an iterative optimization method in a neural network model to obtain parameter weight values ​​of an AlexNet-EAFU model;

[0031] S503, using the trained AlexNet-EAFU model to extract the image feature vector of the newly measured leaf envelope contour image;

[0032] S504 , using a sigmoid function to determine the classification corresponding to the image features of the newly measured blade envelope contour image, and to determine whether the blade belongs to a cracked fault class or a crack-free intact class.

[0033] Furthermore, the process of determining whether the blade is a cracked blade in step S504 is as follows:

[0034] The image of the leaf envelope contour is input into the AlexNet-EAFU model and divided into two branches;

[0035] First branch: After the leaf envelope image is input into the EAFU module, the edges of the input envelope image are first extracted and converted into a 227×227 edge map; the edge map is compressed into features after passing through a 7×7 convolution layer, a 3×3 maximum pooling layer, a 3×3 dilated convolution layer with a dilation rate of 2, a 3×3 convolution layer, a 3×3 dilated convolution layer with a dilation rate of 4, and a 3×3 convolution layer. feature Then the edge branch features are output through the spatial attention module

[0036] Second branch: The image of the leaf envelope is input into the main part of AlexNet, and the main part of AlexNet compresses the input image into a 27×27×256 feature map.

[0037] The output features of the two branches are fused through a 1×1 convolution kernel:

[0038]

[0039] Output new feature F out ;

[0040] New feature F out Continue processing through successive convolutional layers and pooling layers, and finally compress it into a feature vector z through a fully connected layer; use the sigmoid function to classify the feature vector z:

[0041]

[0042] When P>0.5, the envelope contour of the blade is determined to belong to the crack class, and when P<0.5, the envelope contour of the blade is determined to belong to the intact class.

[0043] Furthermore, in step S300, the calculated signal sequence d B (t) and v B (t) Use the smooth function command to perform noise reduction.

[0044] Furthermore, in step S200, the steps for solving the blade tip vibration displacement and instantaneous vibration velocity are as follows:

[0045] Step S201: Calculating the bending angle of the blade tip caused by vibration by fitting polynomial coefficients;

[0046] The pulse voltage u(t) collected by the capacitance sensor and the blade tip angle The relationship is as follows:

[0047]

[0048] Among them, U m is the peak value of the pulse voltage, λ is a constant related to the capacitance sensor, and t is the time;

[0049] The corner of the leaf tip in, is the initial angle of the blade, is the angle at which the blade sweeps across the sensing area of ​​the capacitive sensor, which is proportional to the rotational speed ω. is the bending angle of the blade tip caused by vibration;

[0050] The blade tip bending angle within the excitation pulse voltage period is approximated to a quadratic curve:

[0051]

[0052] Where a, b, and c are the coefficients of the blade tip bending angle expression obtained by curve fitting;

[0053] Step S202: Calculate the tip vibration displacement based on the bending angle of the blade tip vibration;

[0054] Where R is the distance from the center of the blade to the tip of the blade;

[0055] Step S203: According to the blade tip vibration displacement d B Solve for the instantaneous vibration velocity v of the blade tip B ;

[0056] v B =d′ B =R(2at+b).

[0057] Beneficial effects of the present invention:

[0058] Compared to traditional blade crack identification methods, the single-probe online blade crack detection method based on blade tip vibration envelope contours, proposed by this invention, uses a single sensor probe to collect pulse voltage waveforms. This waveform, in the form of envelope contours, is then used to perform crack diagnosis through image recognition technology that fuses multi-source feature information, including blade tip vibration displacement, instantaneous vibration velocity, and the dynamic response characteristics of the resonance region. This integrated approach enriches diagnostic information sources while ensuring identification accuracy.

[0059] By integrating diverse feature information, this method reduces the number of required measurement sensors to a single one, significantly reducing the complexity and cost of sensor installation. Furthermore, the capacitive sensor generates a pulse voltage as the blade sweeps across the sensing area, without direct contact with the blade. This simplified design significantly enhances the engineering applicability of non-contact blade tip vibration measurement.

[0060] After intercepting the instantaneous vibration velocity within the time window, the present invention can convert it into a set of blade tip vibration displacements within the resonance zone, which is equivalent to obtaining two sets of blade tip vibration displacements with one sensor probe. While reducing the number of sensors, the sampling quantity can still be met.

[0061] The present invention inputs the envelope contour map into the AlexNet-EAFU model for crack diagnosis. The EAFU module designed in the AlexNet-EAFU model enhances the boundary perception ability of the network model by combining the edge information of the envelope contour map with the attention mechanism, thereby improving the accuracy of the model in envelope contour image classification.

[0062] In addition, the present invention has undergone noise reduction processing and has verified noise resistance. It can still maintain crack recognition capability under strong noise interference, has strong noise robustness, and has potential engineering application value. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 This is a flow chart of the single-probe blade crack online detection method based on the blade tip vibration envelope contour diagram of the present invention;

[0064] Figure 2 Schematic diagram of a single-probe blade vibration sampling system in the present invention;

[0065] Figure 3 Schematic diagram of noise reduction before and after with a signal-to-noise ratio of 5dB in the present invention;

[0066] Figure 4 Schematic diagram of the process of converting instantaneous vibration velocity into blade tip vibration displacement in the present invention;

[0067] Figure 5Schematic diagram of converting instantaneous vibration velocity into blade tip vibration displacement in the present invention;

[0068] Figure 6 Schematic diagram of the envelope contour diagram of the present invention;

[0069] Figure 7 This is the network architecture of the AlexNet-EAFU model in the present invention;

[0070] Figure 8 Schematic diagram of the structural framework of the EAFU module in the AlexNet-EAFU model framework of the present invention. DETAILED DESCRIPTION

[0071] In order to enable those skilled in the art to better understand the technical solution of the present application, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0072] The terms "up", "down", "left", "right", "front", and "back" in this application are based on the positional relationships shown in the accompanying drawings. The corresponding positional relationships may vary depending on the drawings, and should not be construed as limiting the scope of protection.

[0073] In the present invention, the terms "installed," "connected," "connected," "connected," "fixed," etc. should be understood in a broad sense. For example, they may refer to fixed connection, detachable connection, integral connection, mechanical connection, electrical connection, or mutual communication. They may be directly connected or indirectly connected through an intermediate medium. They may refer to internal communication between two components or interaction between two components. Those skilled in the art will understand the specific meanings of the above terms in the present invention based on specific circumstances.

[0074] This embodiment records a single-probe blade crack online detection method based on the blade tip vibration envelope contour map, which uses a single-probe sensor to collect pulse waveforms, construct an envelope contour map, and input the envelope contour map that integrates vibration characteristics into a neural network model for crack diagnosis, thereby realizing online detection of blade cracks.

[0075] like Figures 1 to 8 As shown, the blade crack online detection method includes the following steps:

[0076] Step S100: collecting the pulse voltage waveform when the blade rotates through the capacitive sensor, and recording the time it takes for the device to complete each rotation;

[0077] When high-speed rotating machinery, such as aircraft engines and gas turbines, is in operation, the blades of the rotating mechanism rotate continuously. A single capacitive sensor is fixed above the blade. In actual use, the capacitive sensor is embedded in the outer casing of the blade. This sensor collects the pulse voltage as the blade sweeps past it and records the resulting pulse voltage waveform. Simultaneously, a speed sensor attached to the rotor records the start time of each blade rotation, allowing the blade speed to be calculated.

[0078] In this embodiment, the capacitance sensor inputs the collected pulse voltage into the acquisition card through the signal conditioning module, and then the acquisition card stores it in the computer.

[0079] Step S200: Calculating the vibration characteristics of the blade tip at the moment when the blade sweeps across the capacitance sensor triggering pulse voltage waveform during the blade rotation process based on the pulse voltage waveform.

[0080] In this embodiment, the vibration characteristics of the blade tip mainly include the blade tip vibration displacement and instantaneous vibration velocity, and the solution steps are as follows:

[0081] Step S201: Calculate the bending angle of the blade tip caused by vibration by fitting polynomial coefficients.

[0082] First, the pulse voltage u(t) collected by the capacitance sensor and the blade tip angle are known. The following relationship exists:

[0083]

[0084] Among them, U m is the peak value of the pulse voltage, λ is a constant related to the capacitance sensor, and t is the time.

[0085] The corner of the leaf tip It consists of three parts:

[0086]

[0087] in, is the initial angle of the blade (see Figure 2 ), is the angle at which the blade sweeps across the sensing area of ​​the capacitive sensor, which is proportional to the rotational speed ω. is the bending angle of the blade tip caused by vibration.

[0088] Since the period of the excitation pulse voltage is very short when the blade sweeps through the sensing area of ​​the capacitive sensor, the blade tip bending angle during this period can be approximately expressed as a quadratic curve:

[0089]

[0090] According to the measured pulse voltage signal, the coefficients a, b and c of the blade tip bending angle expression can be obtained by curve fitting, thereby calculating the blade tip bending angle caused by blade vibration.

[0091] Step S202: Calculating the blade tip vibration displacement based on the bending angle of the blade tip vibration.

[0092] according to Figure 2 The blade tip vibration displacement d is obtained by the single-probe blade vibration sampling system shown in FIG. B :

[0093]

[0094] Where R is the distance from the center of the blade to the tip of the blade.

[0095] Step S203: According to the blade tip vibration displacement d B Solve for the instantaneous vibration velocity v of the blade tip B .

[0096] Since the instantaneous vibration velocity is the derivative of the blade tip vibration displacement, we can further find that the instantaneous vibration velocity of the blade tip satisfies:

[0097] v B =d′ B =R(2at+b)

[0098] Step S300: performing noise reduction processing on the blade tip vibration displacement and instantaneous vibration velocity obtained above.

[0099] The calculated signal sequence d B (t) and v B (t) Use the smooth function in MATLAB software to perform noise reduction. The selected signal contains the resonant signal with the most obvious oscillation trend. The core principle of the smooth function is to fit the data of several adjacent points before and after each data point to smooth the entire time series measurement data, thereby eliminating its noise.

[0100] like Figure 3 (a) shows a set of measurement signals with a signal-to-noise ratio (SNR) of 5dB. After being processed by the smoothing function, the noise-eliminated signal is as follows: Figure 3 As shown in (b), it can be seen that the signal after noise elimination is basically consistent with the real signal, eliminating the influence of noise on the measurement.

[0101] Step S400: constructing an envelope contour diagram of the displacement response of the resonance region based on the blade tip vibration displacement and the instantaneous vibration velocity.

[0102] An envelope profile is a smooth curve that encompasses all instantaneous extreme values ​​(maxima and minima) of a signal. It describes the overall profile of the signal amplitude over time. For example, a study titled "A novel dynamic characteristic for detecting breathing cracks in blades based on vibration response envelope analysis" shows that when a crack develops in a blade, the damping of the blade's vibration system increases and nonlinear stiffness is introduced into the vibration response signal. These factors cause the shape of the envelope profile of the cracked blade's vibration displacement response to change compared to that of an intact blade. Therefore, the shape of the envelope profile corresponds to whether a blade has a crack fault, and blade crack detection can be achieved by identifying the shape of the envelope profile.

[0103] like Figures 4 to 6 As shown in Figure 2, constructing the envelope contour of the displacement response in the resonance region includes the following steps:

[0104] Step S401: Selecting a resonance region time window.

[0105] When the equipment is started, the blades begin to rotate at a uniform acceleration condition, increasing from a low speed to a high working speed at a constant acceleration. During the blade acceleration period, the blades are excited to resonate.

[0106] After the blade resonates, the tip vibration displacement increases significantly according to the capacitance sensor until the peak value of the tip vibration displacement is reached. The time T corresponding to the peak value of the tip vibration displacement is recorded. peak .

[0107] The time window is T peak The starting time T is when the vibration displacement of the blade tip near the blade resonance zone begins to rise. start , the displacement signal decays after passing the resonance peak T peak -T start The final time T end The middle section is the time window containing the blade resonance area. The time window of all test blades satisfies T peak -T start =T end -T peak .

[0108] Step S402: intercepting the instantaneous vibration velocity within the time window, and converting the instantaneous vibration velocity into a set of blade tip vibration displacements within the resonance region.

[0109] like Figure 4 As shown, the instantaneous vibration velocity v intercepted in the time window is BMultiply by a coefficient to solve the converted blade tip vibration displacement

[0110]

[0111] Among them, D AM is the peak-to-peak value of the blade tip vibration displacement measurement signal within the time window, V AM is the peak-to-peak value of the velocity signal in the time window, v B is the instantaneous vibration speed.

[0112] After conversion, the instantaneous vibration velocity v B Transform to obtain a set of transformed blade tip vibration displacements This set of blade tip vibration displacements is equivalent to installing another capacitance sensor probe at a position where the actual measurement capacitance sensor probe is 90 degrees apart, and obtaining the blade tip vibration displacement by the newly installed probe.

[0113] During the operation of the equipment, the rotating blades are usually subjected to periodic harmonic excitation. Therefore, the vibration displacement of the blades at each instant can be expressed in the form of a trigonometric function, that is, d B (t) = D m sin(αt), where D m is the vibration amplitude, α is the circular frequency of vibration. And because velocity is the derivative of displacement, v B (t) = V m cos(αt). According to the transformation relationship between trigonometric functions, Therefore, the vibration displacement of the blade tip at any moment is proportional to the instantaneous vibration velocity at the same moment in amplitude, which is equivalent to a 90-degree rotation in phase.

[0114] After this conversion, one capacitance sensor can only obtain one set of blade tip vibration displacement, but now it can be converted into two sets of blade tip vibration displacement after conversion.

[0115] Step S403: The first blade tip vibration displacement d obtained from the measurement results within the time window is converted to B The second blade tip vibration displacement obtained after conversion is Perform a symmetric transformation along the x-axis and plot the signal d in the same coordinate system B (t),–d B (t), and

[0116] Step S404: Calculate the outermost point set of the area covered by all scattered points consisting of the four groups of first blade tip vibration displacements and second blade tip vibration displacements.

[0117] Use Alpha Shapes non-convex hull algorithm to identify signal dB (t),–d B (t), and The outermost point of the scattered point set.

[0118] Step S405: Use a smooth curve to fit the upper and lower boundary points of the outermost point set. The area surrounded by the smooth curve is the envelope contour of the vibration signal in the resonance area. Figure 5 shown.

[0119] Step S406: Figure 6 The area enclosed by the filled envelope contour image is converted into a standard binary envelope contour image with pixels of H×W, where H is the height of the image and W is the width of the image.

[0120] The generated binary envelope contour image is used for classification and recognition using a neural network model. The image's height H and width W must be consistent with the input size of the classification and recognition model. For example, if the neural network model is the AlexNet-EAFU model, the input image must be 227×227 pixels. Therefore, the envelope contour image is converted to a 227×227 pixel image.

[0121] Step S500: Inputting the envelope contour image of the fused vibration features into the neural network model for crack diagnosis.

[0122] The neural network model of this embodiment is a convolutional neural network model. Preferably, the convolutional neural network model is an AlexNet-EAFU (Edge-Attentive Fusion Unit) model. This embodiment is described using the AlexNet-EAFU model as an example.

[0123] like Figure 7 and Figure 8 As shown in Figure 2, the AlexNet-EAFU model consists of the EAFU module, the AlexNet main body, 1×1 convolution kernels, convolution layers, pooling layers, and fully connected layers. The EAFU module and the AlexNet main body are each connected to the 1×1 convolution kernel data, which is then fed into the convolution layer, pooling layer, and fully connected layer in sequence. The EAFU module includes convolution layers, pooling layers, atrous convolution layers, and a spatial attention module, while the AlexNet main body includes convolution layers and pooling layers.

[0124] The vibration signal of the test blade is converted into a standard binary envelope contour map and then input into the AlexNet-EAFU model. It is classified through various convolutional layers, pooling layers, and EAFU modules until the fully connected layer performs the final classification to determine whether the blade contains cracks. The details are as follows:

[0125] Step S501: construct a training data set.

[0126] For the blade to be tested, a finite element dynamic model was established using the parameter information set in Table 1, and numerical simulation was performed. According to steps S100 to S400, 1500 sets of vibration signals of crack-free blades and 7500 sets of vibration signals of cracked blades were calculated. In addition, according to the method described in step S400, a total of 9000 binary envelope contour images were constructed as a training data set.

[0127] Step S502: The training data set samples are trained using an iterative optimization method in a convolutional neural network model to obtain parameter weight values ​​of the AlexNet-EAFU model.

[0128] Step S503: using the trained AlexNet-EAFU model to extract the image feature vector of the newly measured leaf envelope contour image.

[0129] Step S504: using a sigmoid function to determine the classification corresponding to the image features of the newly measured blade envelope contour image, and to determine whether the blade belongs to a cracked fault class or a crack-free intact class.

[0130] Specifically, after the envelope contour of the blade is input into the AlexNet-EAFU model, the process of determining whether it is a cracked blade is as follows:

[0131] The image of the leaf envelope contour map is input into the EAFU module and the AlexNet main part respectively, and image processing is performed separately to form two branches.

[0132] First branch: After the image is input into the EAFU module, the edges of the input envelope contour map are first extracted and converted into a 227×227 edge map. The edge map is compressed into features after passing through a 7×7 convolution layer, a 3×3 max pooling layer, a 3×3 dilated convolution layer with a dilation rate of 2, a 3×3 convolution layer, a 3×3 dilated convolution layer with a dilation rate of 4, and a 3×3 convolution layer. feature Then the edge branch features are output through the spatial attention module

[0133] Second branch: The image of the leaf envelope is input into the main part of AlexNet, and the main part of AlexNet compresses the input image into a 27×27×256 feature map. For example, the main part of the AlexNet network can be compressed through two convolutional layers and one pooling layer.

[0134] The output features of the two branches are fused through a 1×1 convolution kernel:

[0135]

[0136] Output new feature F out .

[0137] New feature F out Continue through Figure 7 The continuous convolutional layer and pooling layer shown in the figure are processed and finally compressed into a feature vector z through a fully connected layer. The sigmoid function is used to classify the feature vector z:

[0138]

[0139] When P>0.5, the envelope contour of the blade is determined to belong to the crack class, and when P<0.5, the envelope contour of the blade is determined to belong to the intact class.

[0140] When a blade is detected to be cracked, the equipment is shut down and the blade can be processed to improve the safety of equipment operation.

[0141] The EAFU module designed in this paper enhances the network model's boundary perception by combining edge information from the envelope contour image with an attention mechanism. During image recognition, the module focuses on envelope edge features and leverages the inherent dynamic characteristics of the vibration signal envelope to improve the model's accuracy in envelope contour image classification.

[0142] Table 1

[0143]

[0144] Although the principles of the present invention are described in detail above in conjunction with the preferred embodiments of the present invention, those skilled in the art should understand that the above embodiments are merely explanations of exemplary implementations of the present invention and are not intended to limit the scope of the present invention.

[0145] The present invention is not intended to be limiting in any way and should not be construed as departing from the spirit and scope of the present invention.

[0146] In this case, any obvious equivalent transformation, simple replacement, etc. based on the technical solution of the present invention shall not be

[0147] Any changes shall fall within the scope of protection of the present invention.

Claims

1. A single-probe blade crack online detection method based on blade tip vibration envelope contour diagram, characterized in that: The blade crack online detection method comprises the following steps: S100, collecting a pulse voltage waveform when the blade rotates through the capacitive sensor, and recording the time it takes for the device to complete each rotation; S200, calculating the vibration characteristics of the blade tip when the blade sweeps across the capacitance sensor triggering pulse voltage waveform during the blade rotation process based on the pulse voltage waveform; The vibration characteristics include blade tip vibration displacement and instantaneous vibration velocity; S300, performing noise reduction processing on the solved blade tip vibration displacement and instantaneous vibration velocity; S400, constructing an envelope contour diagram of the displacement response of the resonance region based on the blade tip vibration displacement and the instantaneous vibration velocity; S500: Inputting the envelope contour image of the fused vibration characteristics into a neural network model for crack diagnosis.

2. The single-probe blade crack online detection method based on blade tip vibration envelope contour diagram according to claim 1 is characterized in that: In step S400, the following steps are used to construct an envelope contour diagram of the displacement response of the resonance region: S401, selecting a resonance region time window; S402, intercepting the instantaneous vibration velocity within the time window, and converting the instantaneous vibration velocity into a set of blade tip vibration displacements within the resonance region; S403, the first blade tip vibration displacement d solved from the measurement results in the time window B The second blade tip vibration displacement obtained after conversion is Perform a symmetric transformation along the x-axis and plot the signal d in the same coordinate system B (t),–d B (t), and S404, finding the outermost point set of the area covered by all scattered points consisting of the four groups of first blade tip vibration displacement and second blade tip vibration displacement; S405, using a smooth curve to fit the upper and lower boundary points of the outermost point set, the area enclosed by the smooth curve is the envelope contour of the vibration signal of the resonance area; S406 , filling the area enclosed by the envelope contour image and converting it into a standard binary envelope contour image with pixels of H×W, where H is the height of the image and W is the width of the image.

3. The single-probe blade crack online detection method based on blade tip vibration envelope contour diagram according to claim 2 is characterized in that: In step S402, the instantaneous vibration velocity v captured in the time window is converted into B Converted into a set of blade tip vibration displacements in the resonance region Among them, D AM is the peak-to-peak value of the blade tip vibration displacement measurement signal within the time window, V AM is the peak-to-peak value of the velocity signal in the time window, v B is the instantaneous vibration speed.

4. The single-probe blade crack online detection method based on blade tip vibration envelope contour diagram according to claim 2 is characterized in that: In step S404, the Alpha Shapes non-convex hull algorithm is used to identify the signal d B (t),–d B (t), and The outermost point of the scattered point set.

5. The single-probe blade crack online detection method based on blade tip vibration envelope contour diagram according to claim 2 is characterized in that: In step S401, the process of selecting the resonance region time window is: After the blade resonates, record the time T corresponding to the peak value of the blade tip vibration displacement peak ; T peak The starting time T is when the vibration displacement of the blade tip near the blade resonance zone begins to rise. start , the displacement signal decays after passing the resonance peak T peak -T start The final time T end , the middle section is the time window containing the blade resonance area; the time window satisfies T peak -T start =T end -T peak .

6. The single-probe blade crack online detection method based on blade tip vibration envelope contour diagram according to claim 1 is characterized in that: In step S500, the neural network model is an AlexNet-EAFU model, which includes an EAFU module, an AlexNet main part, a 1×1 convolution kernel, a convolution layer, a pooling layer, and a fully connected layer; the ends of the EAFU module and the AlexNet main part are respectively connected to the 1×1 convolution kernel data, and the 1×1 convolution kernel data is input into the convolution layer, the pooling layer, and the fully connected layer in sequence; wherein, the EAFU module includes a convolution layer, a pooling layer, a hole convolution layer, and a spatial attention module, and the AlexNet main part includes a convolution layer and a pooling layer.

7. The single-probe blade crack online detection method based on blade tip vibration envelope contour diagram according to claim 6 is characterized in that: In step S500, the specific steps of determining whether the blade has cracks using the neural network model are as follows: S501, constructing a training data set; For the blade to be tested, a finite element dynamics model is established based on the set parameter information, and numerical simulation is performed to calculate multiple groups of vibration signals of crack-free blades and multiple groups of vibration signals of cracked blades. In step S400, multiple binary envelope contour images are constructed as training data sets; S502, training the training data set samples using an iterative optimization method in a neural network model to obtain parameter weight values ​​of an AlexNet-EAFU model; S503, using the trained AlexNet-EAFU model to extract the image feature vector of the newly measured leaf envelope contour image; S504 , using a sigmoid function to determine the classification corresponding to the image features of the newly measured blade envelope contour image, and to determine whether the blade belongs to a cracked fault class or a crack-free intact class.

8. The single-probe blade crack online detection method based on blade tip vibration envelope contour diagram according to claim 7 is characterized in that: The process of determining whether the blade is a cracked blade in step S504 is as follows: The image of the leaf envelope contour is input into the AlexNet-EAFU model and divided into two branches; First branch: After the leaf envelope image is input into the EAFU module, the edges of the input envelope image are first extracted and converted into a 227×227 edge map; the edge map is compressed into features after passing through a 7×7 convolution layer, a 3×3 maximum pooling layer, a 3×3 dilated convolution layer with a dilation rate of 2, a 3×3 convolution layer, a 3×3 dilated convolution layer with a dilation rate of 4, and a 3×3 convolution layer. feature Then the edge branch features are output through the spatial attention module Second branch: The image of the leaf envelope is input into the main part of AlexNet, and the main part of AlexNet compresses the input image into a 27×27×256 feature map. The output features of the two branches are fused through a 1×1 convolution kernel: Output new feature F out ; New feature F out Continue processing through successive convolutional layers and pooling layers, and finally compress it into a feature vector z through a fully connected layer; use the sigmoid function to classify the feature vector z: When P>0.5, the envelope contour of the blade is determined to belong to the crack class, and when P<0.5, the envelope contour of the blade is determined to belong to the intact class.

9. The single-probe blade crack online detection method based on blade tip vibration envelope contour diagram according to claim 1 is characterized in that: In step S300, the calculated signal sequence d B (t) and v B (t) Use the smooth function command to perform noise reduction.

10. The single-probe blade crack online detection method based on blade tip vibration envelope contour diagram according to claim 1, characterized in that: In step S200, the steps for solving the blade tip vibration displacement and instantaneous vibration velocity are as follows: Step S201: Calculating the bending angle of the blade tip caused by vibration by fitting polynomial coefficients; The pulse voltage u(t) collected by the capacitance sensor and the blade tip angle The relationship is as follows: Among them, U m is the peak value of the pulse voltage, λ is a constant related to the capacitance sensor, and t is the time; The corner of the leaf tip in, is the initial angle of the blade, is the angle at which the blade sweeps across the sensing area of ​​the capacitive sensor, which is proportional to the rotational speed ω. is the bending angle of the blade tip caused by vibration; The blade tip bending angle within the excitation pulse voltage period is approximated to a quadratic curve: Where a, b, and c are the coefficients of the blade tip bending angle expression obtained by curve fitting; Step S202: Calculate the tip vibration displacement based on the bending angle of the blade tip vibration; Where R is the distance from the center of the blade to the tip of the blade; Step S203: According to the blade tip vibration displacement d B Solve for the instantaneous vibration velocity v of the blade tip B ; v B =d′ B =R(2at+b)。

Images