CN116776075A - Fan blade health monitoring system and monitoring method thereof - Google Patents
Fan blade health monitoring system and monitoring method thereof Download PDFInfo
- Publication number
- CN116776075A CN116776075A CN202310712040.8A CN202310712040A CN116776075A CN 116776075 A CN116776075 A CN 116776075A CN 202310712040 A CN202310712040 A CN 202310712040A CN 116776075 A CN116776075 A CN 116776075A
- Authority
- CN
- China
- Prior art keywords
- health
- signal
- data
- blade
- vibration
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Withdrawn
Links
- 230000036541 health Effects 0.000 title claims abstract description 57
- 238000012544 monitoring process Methods 0.000 title claims abstract description 53
- 238000000034 method Methods 0.000 title claims abstract description 35
- 238000012549 training Methods 0.000 claims abstract description 20
- 230000009467 reduction Effects 0.000 claims abstract description 15
- 230000004927 fusion Effects 0.000 claims abstract description 12
- 238000001228 spectrum Methods 0.000 claims abstract description 11
- 238000005259 measurement Methods 0.000 claims abstract description 9
- 238000012512 characterization method Methods 0.000 claims abstract description 7
- 230000006870 function Effects 0.000 claims description 18
- 238000012545 processing Methods 0.000 claims description 10
- 238000009826 distribution Methods 0.000 claims description 8
- 230000007547 defect Effects 0.000 claims description 7
- 239000013598 vector Substances 0.000 claims description 7
- 238000000354 decomposition reaction Methods 0.000 claims description 6
- 239000011159 matrix material Substances 0.000 claims description 6
- 238000012360 testing method Methods 0.000 claims description 6
- 238000004458 analytical method Methods 0.000 claims description 5
- 238000007781 pre-processing Methods 0.000 claims description 5
- 230000003750 conditioning effect Effects 0.000 claims description 4
- 230000000694 effects Effects 0.000 claims description 4
- 238000013507 mapping Methods 0.000 claims description 4
- 230000003595 spectral effect Effects 0.000 claims description 4
- 238000001914 filtration Methods 0.000 claims description 3
- 238000005070 sampling Methods 0.000 claims description 3
- 230000009466 transformation Effects 0.000 claims description 3
- 238000013528 artificial neural network Methods 0.000 claims description 2
- 238000013499 data model Methods 0.000 claims description 2
- 230000008034 disappearance Effects 0.000 claims description 2
- 238000010606 normalization Methods 0.000 claims description 2
- 238000002360 preparation method Methods 0.000 claims description 2
- 238000001514 detection method Methods 0.000 abstract description 4
- 238000003745 diagnosis Methods 0.000 description 3
- 230000008569 process Effects 0.000 description 3
- 239000000523 sample Substances 0.000 description 3
- 230000004913 activation Effects 0.000 description 2
- 238000011156 evaluation Methods 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 230000008901 benefit Effects 0.000 description 1
- 238000005336 cracking Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 230000009977 dual effect Effects 0.000 description 1
- 230000003628 erosive effect Effects 0.000 description 1
- 238000007499 fusion processing Methods 0.000 description 1
- 230000003862 health status Effects 0.000 description 1
- 238000007477 logistic regression Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000004092 self-diagnosis Methods 0.000 description 1
- 230000008054 signal transmission Effects 0.000 description 1
- 230000008646 thermal stress Effects 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/10—Pre-processing; Data cleansing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
- G06N3/0442—Recurrent networks, e.g. Hopfield networks characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/0895—Weakly supervised learning, e.g. semi-supervised or self-supervised learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/02—Preprocessing
- G06F2218/04—Denoising
- G06F2218/06—Denoising by applying a scale-space analysis, e.g. using wavelet analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/08—Feature extraction
- G06F2218/10—Feature extraction by analysing the shape of a waveform, e.g. extracting parameters relating to peaks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/12—Classification; Matching
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Computational Linguistics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Evolutionary Biology (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Measurement Of Mechanical Vibrations Or Ultrasonic Waves (AREA)
Abstract
The invention discloses a fan blade health monitoring system and a monitoring method thereof. According to the invention, the fault signals are extracted and collected under the monitoring of multiple sensors, and the collected signals are amplified and noise reduced. And the preprocessed signal data are fused with the frequency spectrum characteristics of the acoustic emission signal and the vibration signal, self-supervision health characterization learning is carried out to generate a serious scarce negative sample, and the measurement data collected by healthy and unhealthy blades are classified through auxiliary tasks, so that supervision learning based on pseudo tags is realized. The wind turbine generator blade online monitoring system based on the model LSTM can realize real-time wind turbine generator blade online monitoring by utilizing the data fusion, EMD noise reduction, self-supervision learning and LSTM model training technology, can diagnose faults with high precision, evaluates the detection of the health state of the current blade and the judgment of the fault type, and meets the requirement of a wind power market on the online monitoring of the wind turbine generator blade.
Description
Technical Field
The invention relates to the field of fan blade fault diagnosis, in particular to a fan blade health monitoring system and a fan blade health monitoring method based on the fusion of acoustic emission signals and vibration signals.
Background
Wind energy is the most widely applied renewable energy at present, but the working environment of a wind turbine generator is bad, and faults of all systems frequently occur. Especially, the fan blade is used as the core equipment of the unit, the cost of the fan blade accounts for 15% -20% of the cost of the whole unit equipment, the fan blade is easy to be damaged by external impacts such as thermal stress, hygroscopicity and lightning strike, and the fan blade has higher operation and maintenance cost. At present, the generator, the gear box, the pitch mechanism and the yaw mechanism of the wind turbine generator are all provided with a remote online monitoring system, so that the running state of components can be monitored in real time, but one of the components, which is easy to be damaged by lightning strike and wind erosion, of a fan blade can generate faults such as cracking and layering of the blade. At present, the remote real-time state monitoring means is not complete, and if the problem is not found in time and the device is started to operate, disastrous results can be caused.
At present, researches on fault diagnosis parts of fan blades are not fully developed, the existing early warning models are mostly concentrated on modeling and analysis of a certain part in a certain monitoring mode, the model precision is not high, the generalization capability is not enough, self-diagnosis and self-warning of the whole fan blades cannot be realized, and comprehensive evaluation of the health degree of the fan cannot be carried out. For example, a fan blade fault diagnosis system based on acoustic emission, which detects the damaged area of the blade through acoustic emission technology, and then adopts a characteristic extraction method of acoustic emission signals of the blade through wavelet analysis to identify different types of damage. However, relying on only a single technique may suffer from technical drawbacks resulting in insufficient detection accuracy and timeliness.
Disclosure of Invention
In view of this situation, the present invention provides a fan blade health monitoring system and a monitoring method thereof, so as to solve the above problems existing in the prior art and application.
The invention is realized by the following technical scheme:
fan blade health monitoring system includes vibration sensor: setting a mounting section at a position which is about 2-3m more away from 1/3 of the root of the fan blade, wherein the vibration sensor is mounted in the section according to the position requirement of the dual-channel vibration sensor, and is about 20-30cm away from the tip of the trailing edge, and the vibration sensor is mounted in a direction perpendicular to the axis so as to ensure that axial and radial vibration is collected; acoustic emission sensor: arranging a mounting section at the center of the blade, and respectively mounting an acoustic emission sensor at the front edge and the rear edge in the mounting section, wherein each blade is provided with two sensors, the pressure side sensor is about 20-30cm away from the tip of the rear edge, the tension side sensor is about 20-30cm away from the tip of the front edge, and the two AE sensors are mounted on the same cross section; and a data acquisition system: the system is responsible for converting vibration signals and acoustic emission signals which are collected by monitoring sensors arranged on all parts of the fan blade into digital signals for subsequent analysis and processing; the output of the plurality of vibration signals and acoustic emission signal sensors is connected with the input of the signal conditioning and amplifying circuit, and then is connected with the data acquisition hardware through the multi-channel ADC sampling module, and digital signals are calculated, controlled and transmitted.
The invention also provides a health monitoring method realized by the fan blade health monitoring system based on the fusion of the acoustic emission signal and the vibration signal, which comprises the steps of extracting and collecting fault signals under the monitoring of a plurality of sensors, and amplifying and denoising the collected signals; and then, the preprocessed signal data are fused with the frequency spectrum characteristics of the acoustic emission signal and the vibration signal, self-supervision health characterization learning is carried out to generate a serious scarce negative sample, and the measurement data collected by healthy and unhealthy blades are classified through auxiliary tasks, so that supervision learning based on pseudo tags is realized.
Advantageous effects
1) The time domain and wavelet feature extraction technology based on empirical mode decomposition filtering noise reduction improves the quality of acoustic emission signals of the blade, and the machine learning acoustic emission blade defect identification technology based on multidimensional feature fusion is a domestic initiative;
2) The blade defect real-time monitoring system based on the fusion of the vibration signal and the sound emission signal is a domestic initiative, can realize the real-time monitoring of the blade defect, and can quantify the fault degree and find early faults;
3) The data end, the feature end and the decision end are subjected to comprehensive application, optimization and comparison of a multi-source data fusion technology;
4) The frequency spectrum characteristic and other characteristic statistics of the acoustic emission signal and the frequency spectrum characteristic and other characteristic statistics of the vibration signal form a characteristic vector, and the signal fusion processing is efficiently completed in this way.
Drawings
FIG. 1 is a flow chart of the present invention;
FIG. 2 is a schematic plan view of the deployment location of acoustic emission signal sensors and vibration signal sensors in accordance with the present invention;
FIG. 3 is a schematic cross-sectional view of the deployment location of the acoustic emission signal sensor and the vibration signal sensor of the present invention;
FIG. 4 is a flow chart of a data acquisition system of the present invention;
FIG. 5 flow chart of hybrid multi-scale convolution combined double-layer LSTM model in the present invention
Detailed Description
Other advantages and effects of the present invention will become readily apparent to those skilled in the art from the following disclosure, when considered in light of the following detailed description of the invention.
The flow of the fan blade defect health monitoring method provided by the invention is shown in the figure 1: fig. 2 is a schematic plan view of deployment positions of the acoustic emission signal sensor and the vibration signal sensor, and fig. 3 is a schematic cross-sectional view of deployment positions of the acoustic emission signal sensor and the vibration signal sensor.
Step one: vibration sensor and acoustic emission sensor mounting
Vibration sensor 2 mounting position: one mounting section is provided at a distance of about 2-3m more than 1/3 of the fan blade root. Within this section the vibration sensor 2 should be mounted about 20-30cm from the trailing edge tip, as required by the location of the dual channel vibration sensor. The vibration sensor 2 should be mounted in a direction perpendicular to the axis to ensure that axial and radial vibrations can be captured. And according to the model and the connection requirement of the vibration sensor, the connection mode of the vibration sensor is determined, and stable signal transmission is ensured.
Acoustic emission sensor 1 mounting position: an acoustic emission sensor 1 is provided in the center of the blade (about 1/3 of the total length from the root of the blade) and has a leading edge and a trailing edge mounted in the mounting section. Two sensors should be installed per blade. Wherein the pressure side sensor is about 20-30cm from the trailing edge tip and the tension side sensor is about 20-30cm from the leading edge tip. The two AE sensors are mounted on the same cross section, which requires special attention here.
Step two: signal acquisition based on acoustic emission sensor and vibration sensor as shown in FIG. 4
In the embodiment, the data acquisition system combines the acoustic emission detection technology and the vibration detection technology, namely, two probes are used, the two probes are integrated in the same channel number conditioning period, and the same data acquisition processing card detects, acquires and processes two different signals, so that the bandwidth of the monitored signals is greatly expanded. The data acquisition system is responsible for converting vibration signals and acoustic emission signals which are monitored and acquired by sensors arranged on all parts of the fan blade into digital signals for subsequent analysis and processing. The output of the plurality of vibration signals and acoustic emission signal sensors is connected with the input of the signal conditioning and amplifying circuit, and then is connected with the data acquisition hardware through the multi-channel ADC sampling module, and digital signals are calculated, controlled and transmitted.
Step three: noise reduction treatment is carried out on the acoustic emission signal and the vibration signal;
noise reduction is carried out on the acoustic emission signals. First, empirical mode decomposition (EmpiricalModeDecomposition, EMD) is carried out, then, correlation coefficients of each intrinsic mode component (IntrinsicModeFunction, IMF) and an original signal are calculated, components with low correlation are removed, residual IMF components are reconstructed, and EMD filtering is completed.
Specifically, the method can be carried out according to the following steps:
1) The original signal is adaptively decomposed into N IMF components by EMD.
2) And calculating the correlation coefficient between each IMF and the original signal.
corr＝IMF i (t)*x(t)
3) And eliminating components with correlation coefficients smaller than a set threshold value, and reconstructing residual components.
Wherein N is the total number of IMFs; m is the number of IMF components that need to be reconstructed.
Then extracting the domain characteristics such as acoustic emission duration, peak amplitude, center frequency, energy and the like, extracting the sub-band peak characteristic frequency after wavelet packet classification through wavelet transformation, and calculating the energy distribution of the frequency band after wavelet decomposition; and finally, combining the characteristics to obtain wavelet packet characteristics to form a characteristic matrix for the next blade fault discrimination model training.
For the noise reduction processing of the vibration signal, discrete Wavelet Transform (DWT) is used. DWT is a wavelet transform method that rapidly processes discrete signals. It breaks down the signal into a number of wavelet sub-bands, each representing a different frequency range. The proper wavelet base can be selected according to the characteristics of the signals, so that the noise reduction of the signals is realized. Specifically, the method can be carried out according to the following steps:
1. the acquired vibration signal is decomposed into a plurality of wavelet sub-bands. DWT is built using a tool function wavedec to decompose the signal into a plurality of wavelet subbands.
2. And selecting a proper threshold value for denoising by utilizing the wavelet coefficient and the corresponding sub-band. Threshold denoising was performed using a wdencmp function. Wherein the threshold size is controlled using the parameter "t" of the wdencmp function.
3. And finally, reconstructing the processed signal by using a waverec function to obtain a noise-reduced vibration signal.
Step four: preprocessing the data after noise reduction
The invention provides a method for preprocessing vibration signals and sound emission signals, which comprises the following steps:
1. and collecting the vibration signal and the sound emission signal after noise reduction.
2. And respectively performing short-time Fourier transform (STFT) on the vibration signal and the sound emission signal to obtain the time-frequency domain characteristics of the vibration signal and the sound emission signal.
3. Characteristic statistics of the vibration signal are collected, including power spectral density, maximum amplitude, etc.
4. Characteristic statistics of the acoustic emission signal are collected, including power spectral density, peak values, and the like.
5. And forming a characteristic vector by the frequency spectrum characteristic and other characteristic statistics of the acoustic emission signal and the frequency spectrum characteristic and other characteristic statistics of the vibration signal.
Step five: generating negative samples for self-supervised health characterization learning
In this step we use two data enhancement strategies, signal component fusion and noise addition in self-supervised health characterization learning (SHRL), to generate the negative samples. Specifically, we treat the preprocessed data (i.e., raw training data) as positive samples, and then generate negative fan blade samples with different health conditions through data enhancement, so as to compensate for the scarcity of the negative fan blade samples. Generating unhealthy samples for auxiliary tasks through data enhancementThe scheme performs a data enhancement strategy of component fusion and uses a pseudo tag for supervision. Specifically, the author performs random replacement of part of the components in the preprocessed training data to achieve a fusion ratio of 0<η<1, and meanwhile, gaussian white noise with standard deviation of E is added to generate false label negative samples with different health states,and assigning pseudo tags to all samples according to the health status. The data enhancement scheme considers differences and similarities among components, ensures the rationality and representativeness of the enhanced unhealthy samples, improves the diversity of data and increases the robustness against attacks.
Step six: supervision of auxiliary tasks and pseudo tags
The auxiliary task in the SHRL method is to achieve pseudo-tag based supervised learning by classifying the measurement data collected by healthy and unhealthy leaves. The training objectives of this auxiliary task can be represented by formula Eq. (5):
expressed by, wherein y 1 ＝[1,0]，y 2 ＝[0,1]Is one-hot-coded (one-hot). In the SHRL method, the high-level of the network is utilized to obtain an efficient representation of the measurements to understand the health of the leaf by supervised learning from the supervisory signals generated from the raw data. In the SHRL method, the network is connected in series via two sub-networks θ map And theta pro The two tasks of mapping and projection are respectively realized. Wherein the mapping network consists of a plurality of convolution layers and an activation function, acquires the representation of the original data in a potential space through nonlinear transformation, and learns the transformed parameter theta based on supervised learning of pseudo tags map . The projection head is a multi-layer perceptron with depth H, which is used for separating the influence of auxiliary and health monitoring tasks. By the auxiliary task of the SHRL method and supervised learning of pseudo tags, we can effectively optimize Eq. (5) and semi-supervised obtain a representation suitable for condition monitoring.
Step seven: on-line monitoring and diagnosing blade health condition
On-line health monitoring of the leaf utilizes a representation pattern in a learned high-level mapping network, and a Kernel Density Estimation (KDE) is utilized to extract the distribution of health data representations. In this case, the core density estimator corresponding to a representation of healthy leaves is:
where K is a kernel function. In the developed method, a gaussian kernel function is used because it has convenient mathematical properties, expressed as:
and determining a bandwidth matrix N according to the covariance matrix of the data representation, thereby obtaining an internal model of the healthy leaf data representation. N represents the difference between the representation distribution (representational distributions) of the health samples and the monitoring samples, revealing the blade health based on the learned representation patterns. Then, the measured values in operation are collected, and negative log likelihood is calculated, so that the difference between the measured values and the health data model is evaluated, and the health condition of the blade can be known. The negative likelihood metric is normalized by means of a Fault Index (FI), so that the interpretability of the monitoring result is improved, and the health condition of the blade is judged:
the greater the value range of FI is at [0,1]. FI, the higher the likelihood of fan blade failure.
Step eight: fault type discrimination for blade defects
At this point we use a model that has been trained to diagnose blade defects. Specifically, according to the evaluation result of the health state of the blade, whether the blade fails or not is judged. If the blade is found to be faulty, the operation data of the blade can be further analyzed, and the fault type can be judged by combining priori knowledge and experience.
In this embodiment, according to the spectrum characteristics of the vibration signal and the acoustic emission signal, priori knowledge and experience are combined to primarily determine the fault type.
First, the data preparation process, training the model divides the original data into a training set and a test set. The training sets are all from the fused signal data after noise reduction and preprocessing in the fourth step, so that the effect of being applied to a real scene is achieved. The test set is signal data and a public data set of different fault types collected in an actual scene. The normalization processing is carried out on each data point, so that the gradient disappearance problem can be avoided. And finally, obtaining the corresponding signal fault type characteristics.
The example proposes a method for classifying faults by adopting a mixed multi-scale convolution combined with a double-layer LSTM model, wherein the mixed multi-scale convolution combined with a double-layer long-short-term memory network model is shown in fig. 5, and mainly comprises a mixed multi-scale and 2 long-short-term memory neural networks formed by 3 multi-scale convolutions and 1 full-connection layer.
And (3) introducing the data fused in the step four as a data set into a mixed multi-scale convolution and double-layer LSTM model for training. After the data set is ready, adjusting the super parameters in the model to be optimal, including continuously testing different learning rates, and finding out the optimal learning rate; and (3) using an early stopping method in the model training process, stopping training when the loss function of the verification set starts to rise, so as to avoid the model from being over-fitted and find the optimal iteration times.
The design adopts cross entropy loss as a loss function, and quantifies the difference between the predicted value and the true value of the target. The loss function of a single word or a single segment in the time series data is as follows:
model outputIs a probability value, and y (t) Is a series of defined values (labels) and therefore y (t) ＝1。
The average value of the cross entropy loss is:
the model is trained using the prepared dataset. Because the fault types are numerous and mutually exclusive and belong to multi-classification tasks, the output of the model is converted into class probability by adopting a multi-classification logistic regression model softmax in the activation function. The softmax function accepts a vector as input and returns a vector having the same dimensions as the input vector, where each element is a probability value between 0 and 1. The functional expression is as follows:
wherein the method comprises the steps ofFor model output of a certain fault type, denominator +.>And outputting a set for models of all fault types, so as to obtain the probability P of the current fault type. Training a neural network model using a dataset for the purpose of approximating the most likely fault type θ T . Finally, after training is completed, the test set is used to evaluate the performance of the model.
Finally, we can combine prior knowledge and experience to make preliminary judgment, use trained mixed multi-scale convolution to combine with double-layer LSTM model to classify new input data, and judge the fault type generated by the fan blade according to the final probability of various faults.
The foregoing description of the preferred embodiments of the invention is not intended to be limiting, but rather is intended to cover all modifications, equivalents, and alternatives falling within the spirit and principles of the invention.
Claims (10)
1. A fan blade health monitoring system is characterized by comprising
Vibration sensor: setting a mounting section at a position which is more than 2-4m away from the root of the fan blade to the tip direction, wherein the vibration sensor is mounted in the section according to the position requirement of the dual-channel vibration sensor, the distance from the rear edge tip is about 20-30cm, and the vibration sensor is mounted in the direction perpendicular to the axis so as to ensure that axial and radial vibration is collected;
acoustic emission sensor: a mounting section is arranged at the center of the blade, an acoustic emission sensor of a front edge and an acoustic emission sensor of a rear edge are respectively mounted in the mounting section, each blade is provided with two sensors, wherein the pressure side sensor is 20-30cm away from the tip of the rear edge, the tension side sensor is 20-30cm away from the tip of the front edge, and the two acoustic emission sensors are mounted on the same cross section;
and a data acquisition system: the vibration signals and acoustic emission signals collected by the monitoring of the sensors arranged on each part of the fan blade are converted into digital signals for subsequent analysis and processing;
the output of the plurality of vibration signals and acoustic emission signal sensors is connected with the input of the signal conditioning and amplifying circuit, and the output of the plurality of vibration signals and acoustic emission signal sensors is connected with the data acquisition hardware through the multichannel ADC sampling module to calculate, control and transmit digital signals.
2. The method for monitoring the health of the fan blade by the health monitoring system according to claim 1, wherein the method comprises the steps of extracting and collecting fault signals under the monitoring of a plurality of sensors, and amplifying and denoising the collected signals;
and then, the preprocessed signal data are fused with the frequency spectrum characteristics of the acoustic emission signal and the vibration signal, self-supervision health characterization learning is carried out to generate a scarce negative sample, and the measurement data collected by healthy and unhealthy blades are classified through auxiliary tasks, so that supervision learning based on pseudo tags is realized.
3. The method for health monitoring implemented by a fan blade health monitoring system according to claim 2, wherein the noise reduction processing comprises noise reduction processing of acoustic emission signals and vibration signals,
the acoustic emission signal noise reduction comprises firstly performing empirical mode decomposition, then calculating correlation coefficients of each intrinsic mode component and the original signal, removing components with lower correlation, reconstructing residual IMF components, completing EMD filtering,
specifically, the method comprises the following steps:
1) The original signal is subjected to EMD self-adaptive decomposition into N IMF components;
2) Calculating the correlation coefficient between each IMF and the original signal
corr＝IMF i (r)*x(t)
3) Removing components with correlation coefficients smaller than a set threshold value, and reconstructing residual components
Wherein N is the total number of IMFs; m is the number of IMF components to be reconstructed;
extracting the time domain characteristics of acoustic emission duration, peak amplitude, center frequency and energy, extracting the sub-frequency band peak characteristic frequency after wavelet packet classification through wavelet transformation, and calculating the energy distribution of the frequency band after wavelet decomposition; and finally, combining the characteristics to obtain wavelet packet characteristics to form a characteristic matrix for the next step of blade fault recognition model training.
4. A fan blade health monitoring system implemented health monitoring method according to claim 3, characterized in that the noise reduction processing of the vibration signal uses discrete wavelet transform, in particular according to the following steps:
decomposing the acquired vibration signal into a plurality of wavelet sub-bands, and establishing a DWT (discrete wavelet transform) by using a tool function 'wavedec' function, so as to decompose the signal into a plurality of wavelet sub-bands; selecting a proper threshold value for denoising by utilizing wavelet coefficients and sub-bands corresponding to the wavelet coefficients, and denoising the threshold value by utilizing a 'wlan cmp' function; and finally, reconstructing the processed signal by using a 'waverec' function to obtain a noise-reduced vibration signal.
5. The method for monitoring the health of a fan blade according to claim 4, wherein the preprocessing of the noise reduced data comprises the steps of:
1) Collecting vibration signals and sound emission signals after noise reduction;
2) Respectively carrying out short-time Fourier transform on the vibration signal and the sound emission signal to obtain time-frequency domain characteristics of the vibration signal and the sound emission signal;
3) Collecting characteristic statistics of vibration signals, including power spectral density and maximum amplitude;
4) Collecting characteristic statistics of acoustic emission signals, including power spectral density and peak value;
5) And forming a characteristic vector by the frequency spectrum characteristic and other characteristic statistics of the acoustic emission signal and the frequency spectrum characteristic and other characteristic statistics of the vibration signal.
6. The fan blade health monitoring system implemented health monitoring method of claim 2, wherein the performing self-supervised health characterization learning generates a negative sample, comprising: generating a negative sample by using two data enhancement strategies of signal component fusion and noise addition in self-supervision health characterization learning, regarding preprocessed original training data as a positive sample, and then generating fan blade negative samples with different health conditions by component fusion and noise addition, so as to make up for the scarcity of the negative samples;
generating unhealthy samples for auxiliary tasks through data enhancementComponent fusion and supervision using pseudo tags, in particular random replacement of partial components in pre-processed training data to achieve fusionRatio of 0<η<1, and meanwhile, adding Gaussian white noise with standard deviation of E, thereby generating false label negative samples with different health states, and distributing false labels to all samples according to the health states.
7. The method for monitoring the health of a fan blade health monitoring system according to claim 2, wherein the auxiliary task and the pseudo-label are supervised, wherein the auxiliary task is supervised learning based on the pseudo-label is realized by classifying measurement data collected by healthy and unhealthy blades, and the training target of the auxiliary task can be represented by a formula (5):
expressed by, wherein y 1 ＝[1,0]，y 2 ＝[0,1]Is one-hot vectors.
8. The fan blade health monitoring system implemented health monitoring method of claim 2, further comprising on-line monitoring and diagnosing blade health: for online health monitoring of blades using learned representation patterns in a high-level mapping network, extracting a distribution of health data representations using a kernel density estimate, the kernel density estimate corresponding to a representation of a healthy blade is:
wherein K is a kernel function; in the developed method, a gaussian kernel function is used because it has convenient mathematical properties, expressed as:
and determining a bandwidth matrix N according to the covariance matrix of the data representation, thereby obtaining an internal model of the healthy blade data representation, wherein N represents the difference between the representation distribution (representation distributions) of the healthy sample and the monitoring sample, revealing the health condition of the blade based on the learned representation mode, then collecting the measurement value in operation, calculating to obtain negative log likelihood, evaluating the difference between the measurement value and the healthy data model, so that the health condition of the blade can be known, the negative likelihood measurement is normalized by means of a Fault Index (FI), the interpretability of the monitoring result is improved, and the health condition of the blade is judged:
the greater the value range of FI is at [0,1]. FI, the higher the likelihood of fan blade failure.
9. The method for monitoring the health of a fan blade according to claim 2, further comprising performing fault type discrimination on a blade defect: according to the frequency spectrum characteristics of the vibration signal and the sound emission signal, combining priori knowledge and experience, primarily judging the fault type, specifically: firstly, in the data preparation process, training is carried out on a model to divide original data into a training set and a testing set, wherein the training set is from fused signal data after noise reduction and preprocessing so as to achieve the effect of being applied to a real scene, the testing set is signal data of different fault types collected in the real scene, normalization processing is carried out on each data point, the gradient disappearance problem is avoided, and finally, corresponding signal fault type characteristics are obtained.
10. The method for monitoring the health of a fan blade according to claim 9, wherein the fault classification adopts a mixed multi-scale convolution combined with a double-layer LSTM model to classify the fault type, and the mixed multi-scale convolution combined with a double-layer long-short-term memory network model mainly comprises a mixed multi-scale and 2 long-short memory neural networks formed by 3 multi-scale convolutions and 1 fully-connected layer.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310712040.8A CN116776075A (en) | 2023-06-15 | 2023-06-15 | Fan blade health monitoring system and monitoring method thereof |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310712040.8A CN116776075A (en) | 2023-06-15 | 2023-06-15 | Fan blade health monitoring system and monitoring method thereof |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116776075A true CN116776075A (en) | 2023-09-19 |
Family
ID=87988999
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310712040.8A Withdrawn CN116776075A (en) | 2023-06-15 | 2023-06-15 | Fan blade health monitoring system and monitoring method thereof |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116776075A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117093854A (en) * | 2023-10-19 | 2023-11-21 | 安徽建筑大学 | Transformer mechanical fault diagnosis method, equipment and storage medium |
| CN117647587A (en) * | 2024-01-30 | 2024-03-05 | 浙江大学海南研究院 | Acoustic emission signal classification method, computer equipment and medium |
-
2023
- 2023-06-15 CN CN202310712040.8A patent/CN116776075A/en not_active Withdrawn
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117093854A (en) * | 2023-10-19 | 2023-11-21 | 安徽建筑大学 | Transformer mechanical fault diagnosis method, equipment and storage medium |
| CN117093854B (en) * | 2023-10-19 | 2024-02-09 | 安徽建筑大学 | Transformer mechanical fault diagnosis method, equipment and storage medium |
| CN117647587A (en) * | 2024-01-30 | 2024-03-05 | 浙江大学海南研究院 | Acoustic emission signal classification method, computer equipment and medium |
| CN117647587B (en) * | 2024-01-30 | 2024-04-09 | 浙江大学海南研究院 | Acoustic emission signal classification method, computer equipment and medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Tran et al. | Effective multi-sensor data fusion for chatter detection in milling process | |
| CN116776075A (en) | Fan blade health monitoring system and monitoring method thereof | |
| CN109949823B (en) | DWPT-MFCC and GMM-based in-vehicle abnormal sound identification method | |
| CN109213121B (en) | Method for diagnosing clamping cylinder fault of fan braking system | |
| CN113505655B (en) | Intelligent bearing fault diagnosis method for digital twin system | |
| CN108256556A (en) | Wind-driven generator group wheel box method for diagnosing faults based on depth belief network | |
| CN112257530B (en) | Rolling bearing fault diagnosis method based on blind signal separation and support vector machine | |
| CN106650071A (en) | Intelligent fault diagnosis method for rolling bearing | |
| CN109858345B (en) | Intelligent fault diagnosis method suitable for pipe expansion equipment | |
| CN115035913B (en) | Sound abnormity detection method | |
| CN112101142A (en) | Slurry pump running state evaluation method, monitoring terminal and computer readable storage medium | |
| CN113707175B (en) | Acoustic event detection system based on feature decomposition classifier and adaptive post-processing | |
| Zhao et al. | A novel deep fuzzy clustering neural network model and its application in rolling bearing fault recognition | |
| CN114487129A (en) | Flexible material damage identification method based on acoustic emission technology | |
| CN117033928A (en) | Satellite internet flow identification system based on machine learning | |
| CN113919525A (en) | Power station fan state early warning method, system and application thereof | |
| CN113361372A (en) | Main reducer multi-fault intelligent diagnosis method based on multi-channel data deep mining | |
| Lu et al. | Early fault warning and identification in condition monitoring of bearing via wavelet packet decomposition coupled with graph | |
| CN114118219A (en) | Data-driven real-time abnormal detection method for health state of long-term power-on equipment | |
| CN116453544A (en) | Industrial equipment operation state monitoring method based on voiceprint recognition | |
| CN112560674B (en) | Method and system for detecting sound signal quality | |
| CN110222386A (en) | A kind of planetary gear degenerate state recognition methods | |
| Kim et al. | Physics-informed time-frequency fusion network with attention for noise-robust bearing fault diagnosis | |
| CN114263621A (en) | Test method and system for diagnosing and simulating cavitation fault of centrifugal pump | |
| CN111539381B (en) | Construction method of wind turbine bearing fault classification diagnosis model |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| WW01 | Invention patent application withdrawn after publication | ||
| WW01 | Invention patent application withdrawn after publication |
Application publication date: 20230919 |