Voiceprint and stress wave multi-mode identification method for icing state of fan blade

Abstract

The invention discloses a voiceprint and stress wave multi-mode identification method for a fan blade icing state, and the method comprises the steps: constructing a quantum Mel filter group, and converting a voiceprint signal of the fan blade icing state into a quantum state code; quantum convolution feature extraction operation is carried out on the quantum state code, and the quantum state code is converted into a classic feature vector according to a quantum convolution feature extraction result; constructing an icing state classification model, and inputting the classic feature vector into the icing state classification model for processing; by constructing the icing state classification model, the voiceprint and the stress wave in the icing state of the fan can be considered, and an efficient and accurate method is provided for icing detection of the wind driven generator blade.

Description

Technical Field

[0001] The present invention relates to the technical field of wind blade monitoring, and in particular to a method for multi-modal recognition of sound prints and stress waves of an icing state of a wind turbine blade. Background Art

[0002] As wind turbines expand to high-latitude and complex climate areas, blade ice monitoring has become a core technical challenge to ensure the safe operation of wind turbines. The current mainstream monitoring methods are mainly based on acoustic vibration analysis and guided wave detection technology. The dynamic response signal of the blade is collected through an acoustic emission sensor array, and the modal parameter offset characteristics when the ice layer is formed are extracted by combining time-frequency analysis. In recent years, scholars have proposed a voiceprint recognition method based on Mel-frequency cepstral coefficients (MFCC). By using the filter bank design of the bionic auditory system, the enhanced extraction of micro-vibration characteristics in the early stage of icing is achieved in the specific frequency band of the wind turbine (80Hz-8kHz). At the same time, the Lamb wave detection technology excited by the piezoelectric ceramic array can invert the distribution of ice thickness by analyzing the dispersion characteristic distortion on the stress wave propagation path. At the same time, at the data processing level, the fusion architecture of deep convolutional networks and support vector machines has been proven to effectively improve the classification accuracy of multi-source heterogeneous data.

[0003] However, the existing technical system still has significant limitations. First, the traditional Mel filter group faces the contradiction between frequency domain resolution and computational complexity in the process of quantization reconstruction. Its linear triangular window function is difficult to adapt to the nonlinear frequency deviation effect caused by the dynamic growth of the ice layer, resulting in an increase in the feature loss rate of high-frequency harmonic components. In addition, the classical convolutional network has a Hilbert space mismatch problem when extracting quantum state features. Conventional tensor dimensionality reduction operations will destroy the phase-sensitive information contained in the quantum entangled state and cause an increase in recognition errors. Summary of the invention

[0004] The purpose of this section is to summarize some aspects of embodiments of the present invention and briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this section and the specification abstract and the invention title of this application to avoid blurring the purpose of this section, the specification abstract and the invention title, and such simplifications or omissions cannot be used to limit the scope of the present invention.

[0005] In view of the above existing problems, the present invention is proposed. Therefore, the present invention provides a method for multi-modal recognition of the soundprint and stress wave of the icing state of a wind turbine blade, which is used to solve the problems raised in the background technology.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions: a method for multimodal recognition of soundprint and stress wave of icing state of wind turbine blades, comprising:

[0007] Construct a quantum Mel filter bank to convert the voiceprint signal of the ice state of the fan blade into a quantum state encoding;

[0008] Perform an extraction operation of quantum convolution features on the quantum state encoding, and convert it into a classical feature vector according to the result of the quantum convolution feature extraction;

[0009] Construct an ice state classification model, input the classical feature vector into the ice state classification model for processing, so as to realize the recognition result of the original fan blade voiceprint signal and stress wave in the ice state.

[0010] As a preferred solution of the voiceprint and stress wave multimodal recognition method for the ice state of the fan blade according to the present invention, wherein: the construction of the quantum Mel filter bank includes:

[0011] Obtain the frequency range of the original fan blade voiceprint signal, and regard the frequency range as the frequency space;

[0012] Map the frequency space to the quantum state space through the Mel scale to obtain the Mel frequency interval corresponding to each quantum state in the quantum state space.

[0013] As a preferred solution of the voiceprint and stress wave multimodal recognition method for the ice state of the fan blade according to the present invention, wherein: the frequency range of the original fan blade voiceprint signal includes:

[0014] Use a metasurface acoustic array to collect the vibration signal of the fan blade, and extract the voiceprint signal of 0.1 - 20 kHz through a lock-in amplifier circuit.

[0015] As a preferred solution of the voiceprint and stress wave multimodal recognition method for the ice state of the fan blade according to the present invention, wherein: converting the voiceprint signal of the ice state of the fan blade into a quantum state encoding includes:

[0016] According to the ice thickness of the fan blade, perform non-uniform sampling on the high-frequency change and low-frequency change caused by icing, and allocate them to the Mel frequency interval corresponding to each quantum state for encoding.

[0017] As a preferred solution of the voiceprint and stress wave multimodal recognition method for the ice state of the fan blade according to the present invention, wherein: the extraction operation of quantum convolution features on the quantum state encoding includes:

[0018] Construct a quantum convolution kernel for sensitive ice features, and apply controlled rotation gates to the quantum convolution kernel respectively according to the high-frequency change and low-frequency change caused by icing to obtain a quantum convolution output state.

[0019] As a preferred embodiment of the method for multi-modal recognition of acoustic and stress wave of the icing state of a fan blade according to the present invention, it includes: converting the result of quantum convolution feature extraction into a classical feature vector, including:

[0020] Through quantum measurement, converting the quantum convolution output state into a set of classical probability values, and mapping the probability values to obtain a classical feature vector.

[0021] As a preferred embodiment of the method for multi-modal recognition of acoustic and stress wave of the icing state of a fan blade according to the present invention, it includes: constructing an icing state classification model, and inputting the classical feature vector into the icing state classification model for processing, including:

[0022] Taking the classical feature vector and the classical stress wave feature as the inputs of the icing state classification model, and fusing them in a reproducing kernel Hilbert space.

[0023] As a preferred embodiment of the method for multi-modal recognition of acoustic and stress wave of the icing state of a fan blade according to the present invention, it further includes:

[0024] Adopting a support vector machine in the icing state classification model, and embedding aerodynamic constraint conditions in the support vector machine.

[0025] As a preferred embodiment of the method for multi-modal recognition of acoustic and stress wave of the icing state of a fan blade according to the present invention, the aerodynamic constraint conditions include:

[0026] The pressure gradient value on the surface of the fan blade.

[0027] Compared with the prior art, the beneficial effects of the invention are:

[0028] 1. Through the non-linear mapping mechanism of the quantum Mel filter bank, the traditional Mel scale is extended to the quantum state space, breaking through the limitation range of the Shannon sampling theorem, and can effectively identify the acoustic signals of ice layers at high and low frequencies;

[0029] 2. By applying a controlled rotation gate to the quantum convolution kernel, the equivalent extraction of multi-dimensional classical feature vectors can be achieved with a small number of quantum bits, retaining most of the quantum phase-sensitive information and reducing the computational complexity. Description of the Drawings

[0030] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts. Among them:

[0031] Figure 1 This is the overall flowchart of the acoustic fingerprint and stress wave multimodal recognition method for the icing state of the fan blade according to an embodiment of the present invention. Detailed implementation manners

[0032] To make the above objects, features, and advantages of the present invention more obvious and understandable, the following will describe the detailed implementation manners of the present invention with reference to the accompanying drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of them. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0033] In the following description, many specific details are set forth to fully understand the present invention. However, the present invention can also be implemented in other ways different from those described herein. Those skilled in the art can make similar extensions without departing from the connotation of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0034] Secondly, the so-called "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that can be included in at least one implementation manner of the present invention. The "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment that excludes other embodiments.

[0035] The present invention will be described in detail with reference to the schematic diagrams. When detailing the embodiments of the present invention, for the convenience of description, the cross-sectional views showing the device structure will be enlarged locally out of the general scale, and the schematic diagrams are only examples and should not limit the protection scope of the present invention here. In addition, in actual production, three-dimensional spatial dimensions including length, width, and depth should be included.

[0036] At the same time, in the description of the present invention, it should be noted that the orientation or positional relationship indicated by terms such as "upper, lower, inner, and outer" is based on the orientation or positional relationship shown in the drawings, and is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation to the present invention. In addition, the terms "first, second, or third" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0037] Unless otherwise clearly defined and limited in the present invention, the terms "installation, connection, and coupling" shall be understood in a broad sense. For example, it may be a fixed connection, a detachable connection, or an integral connection; it may also be a mechanical connection, an electrical connection, or a direct connection, or it may be indirectly connected through an intermediate medium, or it may be the communication inside two components. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific situations.

[0038] Embodiment 1

[0039] Referring to Figure 1 , which is the first embodiment of the present invention. This embodiment provides a method for multi-modal recognition of acoustic fingerprints and stress waves in the icing state of fan blades, including:

[0040] S1. Construct a quantum Mel filter bank to convert the acoustic fingerprint signal of the fan blade icing state into a quantum state encoding;

[0041] Specifically, use a metasurface acoustic array to collect the vibration signal of the fan blade, and extract the acoustic fingerprint signal of 0.1 - 20 kHz through a lock-in amplifier circuit;

[0042] Specifically, the metasurface acoustic array is composed of a topological insulator material, and its acoustic wave focusing efficiency η satisfies the formula:

[0043]

[0044] where P f represents the focused sound pressure, and P0 represents the ambient noise sound pressure;

[0045] It should be noted that the harmonic distortion caused by the icing of the fan blade is non-uniform. For example, the frequency shift has a non-linear relationship with time, and this change is closely related to the ice layer growth rate, while traditional methods are difficult to capture this dynamic characteristic;

[0046] Furthermore, obtain the frequency range of the original fan blade acoustic fingerprint signal, and regard the frequency range as the frequency space;

[0047] It should be explained that the frequency space refers to the frequency domain representation of the signal, usually converted from the time-domain signal through Fourier transform, and it contains the frequency range of the signal; for example, from 0 Hz to a certain maximum frequency f max Hz;

[0048] Furthermore, map the frequency space to the quantum state space through the Mel scale to obtain the Mel frequency interval corresponding to each quantum state in the quantum state space;

[0049] It should be noted that the Mel scale is a scale based on the non - linear perception of sound frequency by the human ear. It converts the actual frequency into Mel frequency (unit: Mel), which is more in line with the perception characteristics of the human ear for different frequencies;

[0050] It should be noted that the quantum state space is the Hilbert space spanned by qubits. For n qubits, the space dimension can be expressed as 2 n , that is, there are 2 n ground states, denoted as |k>, where k = 0, 1, 2, …, 2 n - 1;

[0051] Specifically, using the Mel scale formula to convert the actual frequency range in the frequency space into the Mel frequency range, we can obtain: Mel(0) = 0, Mel(f max ) and the corresponding value is also the maximum Mel frequency value;

[0052] Furthermore, divide the Mel frequency range into 2 n intervals, and each interval corresponds to a quantum ground state. Then the width of each interval is expressed as:

[0053]

[0054] Then, the range of the k - th Mel frequency interval can be expressed as:

[0055] (k·ΔMel,(k + 1)·ΔMel]

[0056] Specifically, each quantum ground state is assigned to the k - th Mel frequency interval. Then the ground state |k> is equivalent to the feature of the original fan blade voiceprint signal within this Mel frequency interval;

[0057] Furthermore, according to the ice thickness of the fan blade, non - uniformly sample the high - frequency changes and low - frequency changes caused by icing, and assign them to the Mel frequency intervals corresponding to each quantum state for encoding;

[0058] It should be noted that since the icing of the fan blade will change its surface acoustic characteristics, resulting in the change of the voiceprint signal spectrum, it needs to be considered; for example, under high - frequency changes (under a relatively thin ice layer), it shows that icing can cause distortion or attenuation of high - frequency harmonics; under low - frequency changes (under a relatively thick ice layer), it shows that icing can enhance low - frequency resonance or change the low - frequency response;

[0059] Specifically, as can be seen from the above, in the case of a relatively thin ice layer, it mainly affects the high - frequency part of the voiceprint signal, which corresponds to a larger k value in the Mel frequency interval. For example, |m + 1> ~ |2 n-1>. In the case of a relatively thick ice layer, it mainly affects the low-frequency part of the voiceprint signal, which corresponds to a relatively small k value in the Mel frequency range, such as |0> to |m>;

[0060] It should be noted that setting m and representing it as a demarcation point mainly serves to determine the relationship between the ice thickness and the frequency change;

[0061] Specifically, considering the ice thickness on the fan blade, the encoded quantum state |ψ> can be expressed as:

[0062]

[0063] Among them, the amplitude a k is related to the signal characteristics of the k-th Mel frequency range;

[0064] It should be noted that according to the different ice thicknesses, the amplitude a also needs to be dynamically adjusted during encoding k , then it can be adjusted by non-uniform sampling as follows:

[0065] a = 0.03h 2 + 1.7h + 2.4

[0066] Among them, h is the ice thickness, that is, the ice layer thickness;

[0067] S2. Perform an extraction operation on the quantum convolution features of the quantum state encoding, and convert it into a classical feature vector according to the result of the quantum convolution feature extraction;

[0068] Furthermore, construct a quantum convolution kernel for sensitive ice formation features, and apply controlled rotation gates to the quantum convolution kernel respectively according to the high-frequency change and low-frequency change caused by the ice formation to obtain a quantum convolution output state;

[0069] It should be explained that the quantum convolution kernel is a specially designed quantum operator, which is mainly used in the solution of the present invention to capture the voiceprint signal of the fan blade caused by ice formation and provide a basis for applying the controlled rotation gate subsequently;

[0070] It should be explained that the controlled rotation gate is a quantum gate that can apply a rotation operation to the target qubit according to the control condition (quantum convolution kernel);

[0071] Specifically, the quantum convolution kernel contains the encoded quantum state |ψ> and the weight value corresponding to the quantum state |ψ>;

[0072] Furthermore, using the quantum convolution kernel as the control condition, adjust the rotation angle of the rotation gate according to the high-frequency change and low-frequency change caused by the ice formation;

[0073] Specifically, in high-frequency variations, increase the rotation angle corresponding to the Mel frequency range where the k value is located, and in low-frequency variations, decrease the rotation angle corresponding to the Mel frequency range where the k value is located;

[0074] It should be noted that by increasing and decreasing the rotation angle corresponding to the Mel frequency range where the k value is located, the controlled rotation gate can increase the sensitivity of the quantum convolution kernel to the icing characteristics;

[0075] Specifically, a quantum convolution operation is constructed in the application of the controlled rotation gate to the quantum convolution kernel. The quantum convolution operation is implemented by a unitary operator U conv and applied to the input encoded quantum state |ψ> to obtain |ψ out > = U conv |ψ>, where |ψ out > represents the quantum convolution output state, which contains the extracted icing feature information;

[0076] Furthermore, through quantum measurement, the quantum convolution output state is converted into a set of classical probability values, and the probability values are mapped to obtain a classical feature vector;

[0077] Specifically, the quantum convolution output state is measured by selecting a computational basis, where the computational basis refers to the basis composed of all possible qubit states, to obtain the probability of each basis state |k>:

[0078] P k = |<k|ψ out >| 2

[0079] where P k represents the probability that the measurement result is the basis state |k>;

[0080] Specifically, by performing quantum measurement on the quantum convolution output state, a set of classical probability values can be expressed as:

[0081]

[0082] It should be noted that due to the randomness of quantum measurement, multiple measurements are required in actual operations; for example, if several quantum measurements are performed, the frequency of each |k> appearance is statistically counted to obtain an estimated value to judge whether the is close to the true P k ;

[0083] Furthermore, the obtained classical probability values are mapped to obtain a 2 n -dimensional classical feature vector f as follows:

[0084]

[0085] It should be noted that the probability value can be transformed according to different specific tasks; for example, calculating the expected value or weighted sum of certain ground states;

[0086] S3. Construct an icing state classification model, input the classical feature vector into the icing state classification model for processing, so as to realize the recognition result of the original fan blade acoustic fingerprint signal and stress wave in the icing state;

[0087] Furthermore, the classical feature vector and classical stress wave feature are used as the input of the icing state classification model, and they are fused in the reproducing kernel Hilbert space;

[0088] Specifically, the classical stress wave feature is the feature extracted from the stress wave signal by traditional signal processing methods (such as time-frequency analysis), which reflects the physical characteristics of the fan blade surface (changes in stress distribution or vibration mode);

[0089] It should be explained that the reproducing kernel Hilbert space (RKHS) is a mathematical framework suitable for dealing with nonlinear relationships and mapping features from different sources to a high-dimensional space for fusion;

[0090] Specifically, the fusion method adopts weighted summation:

[0091]

[0092] where, x i and x j represent the feature vectors of the i-th and j-th samples respectively; x i and s j represent the classical stress wave features of the i-th and j-th samples respectively; SWD represents the sliced Wasserstein distance of the stress wave signal; σ represents the activation function;

[0093] It should be noted that after the fusion is completed, a support vector machine SVM is used for the classification of the icing state. The SVM finds the optimal hyperplane in the high-dimensional space to distinguish different categories, and is very suitable for processing the data in the RKHS;

[0094] Specifically, a support vector machine is adopted in the icing state classification model, and aerodynamic constraint conditions are embedded in the support vector machine;

[0095] It should be explained that the goal of the SVM support vector machine is to find a hyperplane that minimizes the classification error and maximizes the margin between categories, and then this optimization problem is expressed as:

[0096]

[0097] Among them, w is the weight vector of the hyperplane, b is the bias term, and ξ i is expressed as a slack variable, used to handle inseparable samples. C is the penalty parameter, which controls the tolerance of the icing state classification model to errors. φ(x i ) is the feature mapping function, which maps the input data into a high-dimensional space. y i is the true label of the sample, and η is the physical constraint strength coefficient. is the pressure gradient value on the surface of the wind turbine blade;

[0098] It should be noted that by constructing model-based data-driven and considering the physical drive of the wind turbine blade, the icing state classification model can not only fit the data but also conform to the aerodynamic laws of the wind turbine blade, which can improve the accurate classification of the icing state of the wind turbine blade.

[0099] Those skilled in the art should understand that the embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present application can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The solutions in the embodiments of the present application can be implemented in various computer languages. For example, object-oriented programming languages such as Java and interpreted scripting languages such as JavaScript.

[0100] The present application is described with reference to the flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each flow and / or block in the flowchart and / or block diagram can be implemented by computer program instructions, and the combination of the flows and / or blocks in the flowchart and / or block diagram can also be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate means for implementing the functions specified in Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks.

[0101] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory generate a manufactured article including instruction means, and the instruction means implements the functions specified in Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks.

[0102] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus, so that a series of operation steps are executed on the computer or other programmable apparatus to produce a computer-implemented process, thereby providing instructions for implementing the functions specified in one process or a plurality of processes and / or boxes Figure 1 one process or a plurality of processes and / or boxes Figure 1 steps in one box or a plurality of boxes.

[0103] Although the preferred embodiments of the present application have been described, additional changes and modifications can be made to these embodiments by those skilled in the art once they learn the basic creative concept. Therefore, the appended claims are intended to be construed to include the preferred embodiments as well as all changes and modifications falling within the scope of the present application.

[0104] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalent technologies, the present application is also intended to include these modifications and variations.

Claims

1. A method for multimodal recognition of the soundprint and stress wave of the icing state of a fan blade, characterized in that: include: Construct a quantum Mel filter bank to convert the soundprint signal of the icing state of the wind turbine blades into quantum state encoding; Performing a quantum convolution feature extraction operation on the quantum state encoding, and converting the result of the quantum convolution feature extraction into a classical feature vector; An icing state classification model is constructed, and the classical feature vector is input into the icing state classification model for processing, thereby realizing the recognition result of the original fan blade sound print signal and stress wave in the icing state.

2. The method for multi-modal recognition of the soundprint and stress wave of the icing state of a wind turbine blade according to claim 1, characterized in that: The construction of the quantum Mel filter bank comprises: Obtaining the frequency range of the original fan blade soundprint signal, and regarding the frequency range as the frequency space; The frequency space is mapped to the quantum state space through the Mel scale to obtain the Mel frequency interval corresponding to each quantum state in the quantum state space.

3. The method for multi-modal recognition of the soundprint and stress wave of the icing state of a wind turbine blade according to claim 2, characterized in that: The frequency range of the original fan blade soundprint signal includes: The vibration signals of fan blades are collected using a metasurface acoustic array, and the soundprint signals of 0.1 to 20 kHz are extracted through a phase-locked amplifier circuit.

4. The method for multi-modal recognition of the soundprint and stress wave of the icing state of a wind turbine blade according to claim 2, characterized in that: The acoustic signal of the wind turbine blade icing state is converted into quantum state encoding, including: According to the ice thickness of the wind turbine blades, the high-frequency and low-frequency changes caused by ice are sampled non-uniformly and allocated to the Mel frequency interval corresponding to each quantum state for encoding.

5. The method for multi-modal recognition of the soundprint and stress wave of the icing state of a wind turbine blade according to claim 4, characterized in that: The quantum state encoding is subjected to a quantum convolution feature extraction operation, comprising: A quantum convolution kernel with sensitive ice characteristics is constructed, and a controlled rotating gate is applied to the quantum convolution kernel according to the high-frequency changes and low-frequency changes caused by the ice to obtain a quantum convolution output state.

6. The method for multi-modal recognition of the soundprint and stress wave of the icing state of a wind turbine blade according to claim 5, characterized in that: The results of quantum convolution feature extraction are converted into classical feature vectors, including: Through quantum measurement, the quantum convolution output state is converted into a set of classical probability values, and the probability values ​​are mapped to obtain classical eigenvectors.

7. The method for multi-modal recognition of the soundprint and stress wave of the icing state of a wind turbine blade according to claim 6, characterized in that: Constructing an icing state classification model, and inputting the classical feature vector into the icing state classification model for processing, including: The classical eigenvectors and classical stress wave features are taken as inputs of the icing state classification model and fused in the reproducing kernel Hilbert space.

8. The method for multi-modal recognition of the soundprint and stress wave of the icing state of a wind turbine blade according to claim 7, characterized in that: Also includes: A support vector machine is used in the icing state classification model, and aerodynamic constraints are embedded in the support vector machine.

9. The method for multi-modal recognition of the soundprint and stress wave of the icing state of a wind turbine blade according to claim 8, characterized in that: The aerodynamic constraints include: The pressure gradient value on the fan blade surface.

Images