CN113409213B - Method and system for enhancing noise reduction of time-frequency diagram of fault signal of plunger pump - Google Patents
Method and system for enhancing noise reduction of time-frequency diagram of fault signal of plunger pump Download PDFInfo
- Publication number
- CN113409213B CN113409213B CN202110692924.2A CN202110692924A CN113409213B CN 113409213 B CN113409213 B CN 113409213B CN 202110692924 A CN202110692924 A CN 202110692924A CN 113409213 B CN113409213 B CN 113409213B
- Authority
- CN
- China
- Prior art keywords
- time
- frequency
- model
- layer
- fault
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000010586 diagram Methods 0.000 title claims abstract description 71
- 238000000034 method Methods 0.000 title claims abstract description 45
- 230000009467 reduction Effects 0.000 title claims abstract description 17
- 230000002708 enhancing effect Effects 0.000 title claims abstract description 12
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 31
- 238000003745 diagnosis Methods 0.000 claims abstract description 25
- 238000012549 training Methods 0.000 claims description 45
- 238000012360 testing method Methods 0.000 claims description 31
- 230000004913 activation Effects 0.000 claims description 27
- 238000013145 classification model Methods 0.000 claims description 25
- 230000006870 function Effects 0.000 claims description 24
- 239000011159 matrix material Substances 0.000 claims description 19
- 238000012545 processing Methods 0.000 claims description 15
- 238000006243 chemical reaction Methods 0.000 claims description 12
- 238000005457 optimization Methods 0.000 claims description 8
- 230000009466 transformation Effects 0.000 claims description 8
- 238000007781 pre-processing Methods 0.000 claims description 7
- 230000008569 process Effects 0.000 claims description 7
- 238000004422 calculation algorithm Methods 0.000 claims description 6
- 230000001133 acceleration Effects 0.000 claims description 5
- 238000004458 analytical method Methods 0.000 description 7
- 238000010606 normalization Methods 0.000 description 7
- 238000000354 decomposition reaction Methods 0.000 description 5
- 238000000605 extraction Methods 0.000 description 4
- 230000001965 increasing effect Effects 0.000 description 3
- 238000012958 reprocessing Methods 0.000 description 3
- 238000013135 deep learning Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 230000036039 immunity Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000005070 sampling Methods 0.000 description 2
- 238000001228 spectrum Methods 0.000 description 2
- 230000002159 abnormal effect Effects 0.000 description 1
- 238000005299 abrasion Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000007635 classification algorithm Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 230000008034 disappearance Effects 0.000 description 1
- 230000036541 health Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 210000002569 neuron Anatomy 0.000 description 1
- 230000000737 periodic effect Effects 0.000 description 1
- 238000007637 random forest analysis Methods 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 238000010183 spectrum analysis Methods 0.000 description 1
- 210000002784 stomach Anatomy 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/70—Denoising; Smoothing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
-
- 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
- G06N3/045—Combinations of networks
-
- 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
- G06N3/047—Probabilistic or stochastic networks
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Probability & Statistics with Applications (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Measurement Of Mechanical Vibrations Or Ultrasonic Waves (AREA)
- Complex Calculations (AREA)
Abstract
The application provides a method and a system for enhancing noise reduction of a time-frequency diagram of a fault signal of a plunger pump, comprising the following steps: firstly, collecting fault vibration signals of a plunger pump, then converting original data into a time-frequency diagram, then carrying out noise reduction and enhancement treatment on the time-frequency diagram by using the method, finally extracting characteristics in the time-frequency diagram by using a trained CNN model, carrying out fault diagnosis, searching key areas in the time-frequency diagram by using the gradient of a characteristic diagram of a specific network layer of a convolutional neural network, and clearing non-key areas, thereby enhancing the time-frequency diagram of the original signal. According to the method, the vibration data characteristics are converted into the characteristics of the time-frequency diagram, and the time sequence signals are converted into the picture classification signals, so that fault diagnosis with high accuracy is realized; meanwhile, the noise reduction enhancement method for the time-frequency diagram improves the noise resistance and the robustness of model diagnosis.
Description
Technical Field
The application relates to the technical field of fault diagnosis of plunger pumps, in particular to a method and a system for enhancing noise reduction of a time-frequency chart of a fault signal of a plunger pump.
Background
The aviation hydraulic pump is a key element of an airplane hydraulic system, and meanwhile, the airplane hydraulic system widely adopts the plunger pump because of the characteristics of compact structure, small moment of inertia, large flow, easiness in control and the like of the plunger pump. In order to further increase the power density of the plunger pump, increasing the rotational speed is an effective method. Common types of faults for high-speed plunger pumps include cavitation, abrasion and the like, and the faults easily cause bad consequences such as shell damage, abnormal vibration and the like and even cause safety accidents. Therefore, the method has important significance for fault diagnosis of the aviation plunger pump.
The traditional fault diagnosis mainly compares the health state with the operation state of the pump when faults occur, and comprises data acquisition, feature extraction and fault classification identification. The method mainly utilizes spectrum analysis to extract relevant characteristics, and combines relevant classification algorithms such as SVM, random forest and other models to perform fault recognition. The following disadvantages currently exist: 1) The accuracy of diagnosis is severely dependent on feature extraction, which requires manual design, is time-consuming and depends on experience; 2) The manually extracted features cannot guarantee adequate representation of the features when the fault occurs; 3) At present, a plurality of diagnosis methods do not perform well under the condition that acquired signals are noisy.
The deep learning technology has strong characteristic representation capability, can automatically extract characteristics, and has wide application in voice recognition and image processing. Currently, some students apply a deep learning method to fault diagnosis of various mechanical devices. However, the traditional fault analysis method can also provide a certain reference, and particularly has a plurality of mature methods and technologies in the aspect of time-frequency analysis. The application provides a method for enhancing signal noise reduction in fault diagnosis by combining a time-frequency diagram and a convolutional neural network, thereby improving the reliability and the accuracy of diagnosis.
Patent document CN106404386a (application number: CN 201610757230.1) discloses a method for collecting, extracting and diagnosing early fault characteristic signals of a gear box, an acoustic emission sensor is installed at a position to be monitored of gear box equipment, and a gear box bearing seat is generally selected for collecting acoustic emission signals in a working state of the gear box. The noise-containing signals with different signal to noise ratios are selected to calculate the singular spectrum slope under different decomposition layer numbers, the singular spectrum slope is gradually increased along with the increase of the decomposition layer numbers, and the acquired acoustic emission signals are utilized to select the optimal decomposition layer numbers according to the optimal realization process of the decomposition layer numbers. And according to the selected optimal decomposition layer number, performing redundant lifting wavelet analysis processing on the acquired acoustic emission signals to obtain a time domain diagram and a frequency domain diagram of the signals. And judging the equipment fault condition through analysis of the time domain diagram and the frequency domain diagram.
Disclosure of Invention
Aiming at the defects in the prior art, the application aims to provide a method and a system for enhancing noise reduction of a time-frequency diagram of a fault signal of a plunger pump.
The method for enhancing noise reduction of the time-frequency diagram of the fault signal of the plunger pump comprises the following steps:
step 1: collecting vibration signals of the plunger pump under the condition of different fault grades through a triaxial vibration acceleration sensor arranged on the plunger pump housing;
step 2: slicing and segmenting the acquired vibration signals according to a preset size;
step 3: performing time-frequency transformation on the sliced sample to obtain a time-frequency diagram of the sample, and dividing all the time-frequency diagrams into a training set and a testing set according to a preset proportion;
step 4: training a convolutional neural network classification model by using training set data, and performing fault classification on a test set;
step 5: acquiring class activation views of a training set time-frequency chart through a trained convolutional neural network classification model, and integrating the class activation views to obtain a key area identification matrix;
step 6: performing enhancement processing on all the time-frequency graphs according to the key region identification matrix;
step 7: and retraining the convolutional neural network classification model by using the training set after the enhancement processing to obtain a fault diagnosis model, and predicting the fault degree of the sample.
Preferably, the network structure of the convolutional neural network classification model is composed of three convolutional layers:
the first layer convolution kernel is set to 3*3 in size and 32 in number, and the output size and the input size after convolution are the same; the number of convolution kernels of the second layer and the third layer is set to 16;
and outputting classification results by using two full-connection layers, and converting the forward propagation result of the convolutional neural network classification model into a probability distribution value of the classification label by using a softMax activation function in the last full-connection layer.
Preferably, the error between the prediction result and the real label is estimated by using a cross entropy loss function, and an RMSProp optimization algorithm is adopted to perform iterative training to optimize a convolutional neural network classification model, wherein the loss function has the formula:
Loss(p,q)＝-∑ x p(x)q(x)
where p is the sample realism value, q is the predicted tag value, x is the sequence number, p (x) is the probability distribution of the sample realism value, and q (x) is the probability distribution of the model predicted tag value.
Preferably, sequentially inputting the time-frequency diagrams of the training set samples into a preprocessing model to obtain an output feature diagram of a second layer of convolution layer in the model, and weighting each channel in the feature diagram by using the gradient of the category relative to the channel to obtain a space diagram of the activation intensity of the input image to the category;
and sequentially accumulating the space diagrams of the activation intensity, only reserving data greater than eighty bits in the space diagrams, adopting the same method for each category to obtain key identification areas of different categories, and overlapping the key identification areas of different categories to obtain a key area identification matrix.
Preferably, the enhancing process includes: performing time-frequency conversion on the sample signal to obtain a time-frequency converted two-dimensional array, performing dot multiplication on the two-dimensional array and the key region identification matrix to obtain a processed two-dimensional array, and converting the processed two-dimensional array into a picture.
The application provides a plunger pump fault signal time-frequency diagram noise reduction enhancement system, which comprises:
module M1: collecting vibration signals of the plunger pump under the condition of different fault grades through a triaxial vibration acceleration sensor arranged on the plunger pump housing;
module M2: slicing and segmenting the acquired vibration signals according to a preset size;
module M3: performing time-frequency transformation on the sliced sample to obtain a time-frequency diagram of the sample, and dividing all the time-frequency diagrams into a training set and a testing set according to a preset proportion;
module M4: training a convolutional neural network classification model by using training set data, and performing fault classification on a test set;
module M5: acquiring class activation views of a training set time-frequency chart through a trained convolutional neural network classification model, and integrating the class activation views to obtain a key area identification matrix;
module M6: performing enhancement processing on all the time-frequency graphs according to the key region identification matrix;
module M7: and retraining the convolutional neural network classification model by using the training set after the enhancement processing to obtain a fault diagnosis model, and predicting the fault degree of the sample.
Preferably, the network structure of the convolutional neural network classification model is composed of three convolutional layers:
the first layer convolution kernel is set to 3*3 in size and 32 in number, and the output size and the input size after convolution are the same; the number of convolution kernels of the second layer and the third layer is set to 16;
and outputting classification results by using two full-connection layers, and converting the forward propagation result of the convolutional neural network classification model into a probability distribution value of the classification label by using a softMax activation function in the last full-connection layer.
Preferably, the error between the prediction result and the real label is estimated by using a cross entropy loss function, and an RMSProp optimization algorithm is adopted to perform iterative training to optimize a convolutional neural network classification model, wherein the loss function has the formula:
Loss(p,q)＝-∑ x p(x)q(x)
where p is the sample realism value, q is the predicted tag value, x is the sequence number, p (x) is the probability distribution of the sample realism value, and q (x) is the probability distribution of the model predicted tag value.
Preferably, sequentially inputting the time-frequency diagrams of the training set samples into a preprocessing model to obtain an output feature diagram of a second layer of convolution layer in the model, and weighting each channel in the feature diagram by using the gradient of the category relative to the channel to obtain a space diagram of the activation intensity of the input image to the category;
and sequentially accumulating the space diagrams of the activation intensity, only reserving data greater than eighty bits in the space diagrams, adopting the same method for each category to obtain key identification areas of different categories, and overlapping the key identification areas of different categories to obtain a key area identification matrix.
Preferably, the enhancing process includes: performing time-frequency conversion on the sample signal to obtain a time-frequency converted two-dimensional array, performing dot multiplication on the two-dimensional array and the key region identification matrix to obtain a processed two-dimensional array, and converting the processed two-dimensional array into a picture.
Compared with the prior art, the application has the following beneficial effects:
the method combines time-frequency analysis and convolutional neural network, performs feature extraction by using time-frequency analysis, classifies images by using the convolutional neural network, and enhances the time-frequency diagram by using the class activation intensity diagram, so that the model diagnosis performance is good and the accuracy is high under the noise condition.
Drawings
Other features, objects and advantages of the present application will become more apparent upon reading of the detailed description of non-limiting embodiments, given with reference to the accompanying drawings in which:
FIG. 1 is a flow chart of the present application;
FIG. 2 is a diagram of the effect of sample time-frequency transformation;
FIG. 3 is a flowchart of a time-frequency diagram enhancement process;
FIG. 4 is a diagram of a model network architecture;
FIG. 5 is an original time-frequency transform diagram;
FIG. 6 is a time-frequency transform and enhancement processing diagram;
FIG. 7 is a graph of accuracy and loss during training;
fig. 8 is a graph of accuracy and loss for test sets at different noise.
Detailed Description
The present application will be described in detail with reference to specific examples. The following examples will assist those skilled in the art in further understanding the present application, but are not intended to limit the application in any way. It should be noted that variations and modifications could be made by those skilled in the art without departing from the inventive concept. These are all within the scope of the present application.
Example 1:
the application aims to solve the problem of providing a fault diagnosis method of a high-speed aviation plunger pump, which can realize signal noise reduction enhancement, thereby realizing signal characteristic extraction and fault diagnosis and improving the diagnosis accuracy of noise-containing signals.
In order to achieve the above object, the application provides a plunger pump fault signal time-frequency diagram noise reduction enhancement method based on Grad-CAM, which is characterized by comprising the following steps:
s1: collecting vibration signals of a shell when a plunger pump fails
Collecting vibration signals of the plunger pump under different inlet pressures at a sampling frequency fs
S2: data sample segmentation
The method adopted by the application needs to perform time-frequency conversion on the original signal to obtain a time-frequency diagram of the sample, wherein the time-frequency diagram is the input of the subsequent model training and testing. The original signal is divided into N sections, and the number of samples in each type is as follows: l (L) i ＝n i Total number of samples/s:according to the above relation, the vibration signal becomes an N-segment signal, each segment having a tag. Is marked asWhere s is the length of each segment of the sample and the outermost subscript indicates the category.
m represents the number of sample categories; i represents the subscript of a certain sample class;representing the segmented first segment of the original signal.
S3: time-frequency conversion of divided samples
Obtaining a time-frequency diagram by adopting short-time Fourier transform; the data is divided into training and testing sets, and in the pictures after time-frequency conversion of each group of sample data, the samples are divided into training sets X according to a certain proportion train And test set X test ；
S4: pretreatment model training
And constructing a CNN classification model for classifying and diagnosing the time-frequency diagram, and learning fault characteristics in the time-frequency diagram mainly through three convolution layers and two full connection layers. The construction details of the network structure are as follows:
the convolution layer conv2d_1 is the first layer to extract features of the picture and needs to be able to extract the features sufficiently to provide an output for the subsequent layers containing sufficient useful information. The number of convolution kernels (filters) is set to 32, the size is set to (3, 3), and the plane size after convolution is set to be the same as the original (padding=same), so that the dimension and the input of the subsequent feature map are consistent. The Conv2d_1 layer weight initialization adopts random normal; the activation function adopts a relu activation function, so that the convergence speed can be increased, and the condition of gradient disappearance is avoided. The convolution layer is followed by a batch normalization layer (batch_normalization_1) and a random discard layer (drop_1).
The parameter setting of the convolution layer conv2d_2 is consistent with conv_1 except that the number of filters is different, and the number of convolution kernels is set to 16. After the convolution layer, a batch normalization layer (batch_normalization_2) and a random discard layer (drop_2) are provided
The parameter setting of the convolution layer conv2d_3 is consistent with the conv2d_1, and the batch normalization layer batch_normalization_3 and the random discard layer (dropout_3) are set in the same way later
And finally, outputting a predicted classification result by the network, outputting the classification result by utilizing two full-connection layers, and converting a model forward propagation result into a probability distribution value of a classification label by adopting softMax as an activation function of the last full-connection layer in order to obtain a multi-classification prediction result.
Error between the predicted result and the real label is evaluated using a cross entropy loss function. The optimization objective is to minimize the loss function, the model training optimization method adopts an RMSROP optimization algorithm, and enough rounds are iterated to obtain a trained model. The formula of the loss function is:
where p (x) is the probability distribution of the true values of the samples and q (x) is the probability distribution of the model predictive labels.
S5: important area identification
After the pre-processing model is obtained, an effective diagnosis of the acquired signals can be made. However, the time-frequency diagram of the noise signal is obviously different from that of the noise-free signal, so that the fault diagnosis accuracy of the noise-containing signal is obviously reduced. Without solving the problems, the application provides a method for carrying out noise reduction enhancement treatment on a time-frequency diagram by using a class activation intensity diagram of a model, which comprises the following steps:
inputting a time-frequency diagram into a model to obtain a characteristic diagram output of a conv2d_2 layer in the model;
calculating the gradient of model class output corresponding to the feature map by using the function provided by Keras;
the gradient of the characteristic map channel is averaged, and the gradient average value is used as the weight of the characteristic map;
multiplying the feature map and the gradient mean value to obtain a gradient weighted feature map, and then obtaining a class activation intensity map of a sample by channel mean value of the weighted feature map;
superposing class activation intensity graphs of all samples, and only reserving data greater than eighty bits to obtain a key identification area of a certain class;
repeating the steps in different categories to finally obtain key identification areas with different fault grades, and superposing a plurality of areas to obtain a signal enhancement processing matrix.
S6: data reprocessing
And (3) carrying out time-frequency conversion on the sample signal, carrying out element corresponding multiplication on the sample signal and the signal enhancement processing matrix obtained in the step (S5), and then converting the processed time-frequency conversion array into a time-frequency diagram. Procedure see the accompanying drawings
S7: model retraining
The model is retrained by using the training set after the retrained, and the obtained model can more accurately identify the heavy point area.
S8: fault diagnosis
Test set X test And inputting a trained CNN model, and predicting the fault degree of the sample.
Example 2:
example 2 is a preferred example of example 1.
Referring to fig. 1, which is a flowchart of the method for enhancing noise reduction of a plunger pump fault signal time-frequency diagram based on Grad-CAM, the method for diagnosing cavitation fault of a high-speed aviation plunger pump comprises the following steps:
s1: a vibration sensor is arranged on a shell of the plunger pump, a collection device is connected, vibration signals generated when the pump generates cavitation of different degrees under different inlet pressures are collected, and the sampling frequency is 10240Hz.
S2: the original vibration signal is divided into N segments, each segment serving as a sample. Each segment of sample was scored for cavitation level according to flow loss. The present embodiment is divided into four classes, namely severe cavitation, medium cavitation, slight cavitation and no cavitation, and 256 data points are adopted for each section of sample length in the embodiment.
The severity of cavitation is measured by flow loss and expressed as:
wherein: q t Is the theoretical flow rate, q in Is the actual inlet flow.
Dividing the training set and the test set, dividing the converted picture into the training set and the test set according to the test set proportion of 0.2, namely 80% of samples are used as the training set X train The rest are taken as test set X test 。
S3: sample time-frequency conversion
The vibration signal collected under the single rotating speed working condition of the operation working condition of the embodiment has more periodic components, so that the time-frequency analysis is more suitable by adopting the short-time Fourier transformation. And selecting proper short-time Fourier transform parameters, wherein the effect is shown in figure 2.
S4: establishing a pretreatment model
Model structure as shown in fig. 4, the input is a time-frequency plot after sample time-frequency transformation, the first layer is a convolution layer conv2d_1, the convolution kernel size set (3, 3), and the number is 32. After which the batch normalization layer and the random discard layer are passed. And then to the next convolutional layer conv2d_2. The convolution kernel size of the second convolution layer is set to (3, 3), the number of convolution kernels is set to 16, and then the convolution kernels are connected to a third convolution layer conv2d_3 through a batch normalization layer and a random discard layer, and the third convolution layer is set to be consistent with the second convolution layer. The expansion is performed through a batch normalization layer and a random discard layer. After expansion, a full connection layer with the number of neurons of 32 is arranged, and finally, the full connection layer is connected to an output layer through a batch normalization and random discarding layer. The output layer outputs the category of cavitation fault.
And establishing a loss function, and selecting an optimization algorithm to train the model. The loss function employs cross entropy (Cross Entropy Loss) as the loss function. For the four classes of failure levels of this example, assuming that the true value of the sample is q (x) = (0, 1, 0), the predicted value of the model is p (x) = (a, b, c, d), then:
Loss(p,q)＝-0·loga-1·logb-0·logc-0·logd
where a, b, c, d represent the inverse of the probability that the model estimates for the four classes of fault levels, respectively.
Training and testing the pretreatment model to obtain the pretreatment model with good performance on the test set.
S5: important area identification
And obtaining a characteristic diagram of the conv2d_2 layer in the model by utilizing the preprocessing model, and outputting gradients between the class pair characteristic diagram by the model. And (5) obtaining the key identification area of the preprocessing model according to the description S5 by utilizing the characteristic diagram and the gradient.
S6: data reprocessing
And carrying out point multiplication on the data after sample time-frequency transformation and the key identification area matrix, converting the data into a time-frequency diagram, and carrying out reprocessing on the training set and the testing set, referring to fig. 3.
S7: diagnostic model training
The model structure and the training method are the same as those of the pretreatment model.
S8: and inputting the test set data into a diagnosis model, and predicting the empty call degree of the test set.
More specifically, the application utilizes a fault simulation experiment table of the plunger pump, and vibration signals under different inlet pressures are collected through a triaxial vibration acceleration sensor arranged on a pump housing of the plunger pump. The experiment table can measure inlet and outlet flow, and the cavitation degree is divided by the flow loss degree, as shown in the following table:
| inlet pressure (Mpa) | 0.25 | 0.15 | 0.10 | 0 |
| Flow loss | 1.0％ | 2.0％ | 8.0％ | 76.0％ |
| Severity of cavitation | No empty stomach | Slight cavitation | Medium idle call | Severe cavitation |
The acquisition frequency of the vibration signal is 10240Hz, and the acquired original signals are divided into four types. Segmenting each type of data and converting the segmented data into a time-frequency diagram, dividing a training set and a testing set according to the proportion, wherein the data set is as follows:
| category(s) | Training set | Test set |
| Severe cavitation | 192@(128,128,3) | 48@(128,128,3) |
| Medium cavitation | 192@(128,128,3) | 48@(128,128,3)) |
| Slight cavitation | 192@(128,128,3) | 48@(128,128,3) |
| Cavitation-free | 192@(128,128,3) | 48@(128,128,3) |
| Totals to | 768 | 192 |
Experimental results
(1) Sample time-frequency diagram conversion result and enhancement processing
According to the method, the samples are subjected to time-frequency conversion and stored into pictures, and meanwhile, the noise reduction enhancement method provided by the application is utilized for processing, and the processing results are shown in fig. 5 and 6.
(2) Model prediction accuracy
Model performance is verified on a test set by using python and TensorFlow to build and train, the accuracy of the trained model in the verification set can reach 99.5%, and the accuracy and loss curves in the training process are shown in figure 7.
(3) Model noise immunity
In order to better simulate the complex working condition of real fault monitoring, white noise with different signal-to-noise ratios (SNR) is added into the original signals of the test set. The model is trained on data without white noise, and the data processing enhancement mode of the application ensures that the method has good anti-noise performance, as shown in figure 8, and is an accuracy and loss curve chart of a test set under different noises.
The formula for calculating the signal-to-noise ratio of the vibration signal is as follows:
wherein: p represents the power level of the signal, P signal Representing the magnitude of the signal power without noise, P nosie Representing the power level of the noise.
Experiments prove that the noise immunity under the condition of SNR= -4-10 dB, and the result is as follows:
| SNR | 0 | 2 | 4 | 6 | 8 | 10 |
| accuracy rate of | 63.54％ | 78.12％ | 90.10％ | 95.31％ | 97.91％ | 98.43％ |
| Loss of | 0.90 | 0.60 | 0.37 | 0.22 | 0.13 | 0.08 |
The result shows that the cavitation fault diagnosis method provided by the application has good anti-noise capability, the accuracy of the test set is higher than 80% under the condition that the signal-to-noise ratio is higher than 2dB, and the corresponding loss function value is smaller, so that the diagnosis reliability is high and the robustness is strong.
Those skilled in the art will appreciate that the systems, apparatus, and their respective modules provided herein may be implemented entirely by logic programming of method steps such that the systems, apparatus, and their respective modules are implemented as logic gates, switches, application specific integrated circuits, programmable logic controllers, embedded microcontrollers, etc., in addition to the systems, apparatus, and their respective modules being implemented as pure computer readable program code. Therefore, the system, the apparatus, and the respective modules thereof provided by the present application may be regarded as one hardware component, and the modules included therein for implementing various programs may also be regarded as structures within the hardware component; modules for implementing various functions may also be regarded as being either software programs for implementing the methods or structures within hardware components.
The foregoing describes specific embodiments of the present application. It is to be understood that the application is not limited to the particular embodiments described above, and that various changes or modifications may be made by those skilled in the art within the scope of the appended claims without affecting the spirit of the application. The embodiments of the application and the features of the embodiments may be combined with each other arbitrarily without conflict.
Claims (2)
1. The method for enhancing noise reduction of the time-frequency diagram of the fault signal of the plunger pump is characterized by comprising the following steps of:
step 1: collecting vibration signals of the plunger pump under the condition of different fault grades through a triaxial vibration acceleration sensor arranged on the plunger pump housing;
step 2: slicing and segmenting the acquired vibration signals according to a preset size;
step 3: performing time-frequency transformation on the sliced sample to obtain a time-frequency diagram of the sample, and dividing all the time-frequency diagrams into a training set and a testing set according to a preset proportion;
step 4: training a convolutional neural network classification model by using training set data, and performing fault classification on a test set;
step 5: acquiring class activation views of a training set time-frequency chart through a trained convolutional neural network classification model, and integrating the class activation views to obtain a key area identification matrix;
step 6: performing enhancement processing on all the time-frequency graphs according to the key region identification matrix;
step 7: retraining the convolutional neural network classification model by using the training set after the enhancement treatment to obtain a fault diagnosis model, and predicting the fault degree of the sample;
the network structure of the convolutional neural network classification model is composed of three convolutional layers:
the first layer convolution kernel is set to 3*3 in size and 32 in number, and the output size and the input size after convolution are the same; the number of convolution kernels of the second layer and the third layer is set to 16;
outputting classification results by using two full-connection layers, and converting the forward propagation result of the convolutional neural network classification model into a probability distribution value of a classification label by using a softMax activation function by using the last full-connection layer;
estimating the error between the prediction result and the real label by using a cross entropy loss function, and performing iterative training to optimize a convolutional neural network classification model by adopting an RMSProp optimization algorithm, wherein the loss function has the formula:
Loss(p,q)＝-Σ x p(x)q(x)
wherein p is a sample true value, q is a predicted tag value, x is a sequence number, p (x) is a probability distribution of the sample true value, and q (x) is a probability distribution of a model predicted tag value;
sequentially inputting the time-frequency images of the training set samples into a preprocessing model to obtain an output feature image of a second layer of convolution layer in the model, and weighting each channel in the feature image by using the gradient of the category relative to the channel to obtain a space image of the activation intensity of the input image to the category;
sequentially accumulating the space diagrams of the activation intensity, only reserving data greater than eighty bits in the space diagrams, adopting the same method for each category to obtain key identification areas of different categories, and overlapping the key identification areas of different categories to obtain a key area identification matrix;
the enhancement process includes: performing time-frequency conversion on the sample signal to obtain a time-frequency converted two-dimensional array, performing dot multiplication on the two-dimensional array and the key region identification matrix to obtain a processed two-dimensional array, and converting the processed two-dimensional array into a picture.
2. The utility model provides a plunger pump fault signal time-frequency diagram noise reduction enhancement system which characterized in that includes:
module M1: collecting vibration signals of the plunger pump under the condition of different fault grades through a triaxial vibration acceleration sensor arranged on the plunger pump housing;
module M2: slicing and segmenting the acquired vibration signals according to a preset size;
module M3: performing time-frequency transformation on the sliced sample to obtain a time-frequency diagram of the sample, and dividing all the time-frequency diagrams into a training set and a testing set according to a preset proportion;
module M4: training a convolutional neural network classification model by using training set data, and performing fault classification on a test set;
module M5: acquiring class activation views of a training set time-frequency chart through a trained convolutional neural network classification model, and integrating the class activation views to obtain a key area identification matrix;
module M6: performing enhancement processing on all the time-frequency graphs according to the key region identification matrix;
module M7: retraining the convolutional neural network classification model by using the training set after the enhancement treatment to obtain a fault diagnosis model, and predicting the fault degree of the sample;
the network structure of the convolutional neural network classification model is composed of three convolutional layers:
the first layer convolution kernel is set to 3*3 in size and 32 in number, and the output size and the input size after convolution are the same; the number of convolution kernels of the second layer and the third layer is set to 16;
outputting classification results by using two full-connection layers, and converting the forward propagation result of the convolutional neural network classification model into a probability distribution value of a classification label by using a softMax activation function by using the last full-connection layer;
estimating the error between the prediction result and the real label by using a cross entropy loss function, and performing iterative training to optimize a convolutional neural network classification model by adopting an RMSProp optimization algorithm, wherein the loss function has the formula:
Loss(p,q)＝-∑ x p(x)q(x)
wherein p is a sample true value, q is a predicted tag value, x is a sequence number, p (x) is a probability distribution of the sample true value, and q (x) is a probability distribution of a model predicted tag value;
sequentially inputting the time-frequency images of the training set samples into a preprocessing model to obtain an output feature image of a second layer of convolution layer in the model, and weighting each channel in the feature image by using the gradient of the category relative to the channel to obtain a space image of the activation intensity of the input image to the category;
sequentially accumulating the space diagrams of the activation intensity, only reserving data greater than eighty bits in the space diagrams, adopting the same method for each category to obtain key identification areas of different categories, and overlapping the key identification areas of different categories to obtain a key area identification matrix;
the enhancement process includes: performing time-frequency conversion on the sample signal to obtain a time-frequency converted two-dimensional array, performing dot multiplication on the two-dimensional array and the key region identification matrix to obtain a processed two-dimensional array, and converting the processed two-dimensional array into a picture.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110692924.2A CN113409213B (en) | 2021-06-22 | 2021-06-22 | Method and system for enhancing noise reduction of time-frequency diagram of fault signal of plunger pump |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110692924.2A CN113409213B (en) | 2021-06-22 | 2021-06-22 | Method and system for enhancing noise reduction of time-frequency diagram of fault signal of plunger pump |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113409213A CN113409213A (en) | 2021-09-17 |
| CN113409213B true CN113409213B (en) | 2023-11-14 |
Family
ID=77682245
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110692924.2A Active CN113409213B (en) | 2021-06-22 | 2021-06-22 | Method and system for enhancing noise reduction of time-frequency diagram of fault signal of plunger pump |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113409213B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117270482A (en) * | 2023-11-22 | 2023-12-22 | 博世汽车部件(苏州)有限公司 | Automobile factory control system based on digital twin |
Citations (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101188453A (en) * | 2006-11-20 | 2008-05-28 | 阿尔卡特朗讯 | Method and system for wireless cellular indoor communications |
| KR20120114776A (en) * | 2011-04-08 | 2012-10-17 | (주)한세로에프에이 | Inspecting method of color similarity using complementary color relationship of chromatic color and achromatic color, subject detecting method and heat treatment inspecting method using the same |
| CN107915303A (en) * | 2016-10-10 | 2018-04-17 | 天津有序环境科技发展有限公司 | Can connecting shaft pump Ammonia adding System and control method more |
| CN108694298A (en) * | 2018-07-04 | 2018-10-23 | 温州大学 | A kind of axial plunger pump bearing inner race and outer ring method for diagnosing faults |
| CN108757426A (en) * | 2018-07-04 | 2018-11-06 | 温州大学 | Oilfield water filling plunger pump trouble diagnostic method |
| CN108830127A (en) * | 2018-03-22 | 2018-11-16 | 南京航空航天大学 | A kind of rotating machinery fault feature intelligent diagnostic method based on depth convolutional neural networks structure |
| CN109214402A (en) * | 2017-07-04 | 2019-01-15 | 华北电力大学（保定） | A kind of people having the same aspiration and interest unit grouping method of combination WAVELET FUZZY entropy and GG fuzzy clustering |
| CN110766027A (en) * | 2019-10-22 | 2020-02-07 | 腾讯科技（深圳）有限公司 | Image area positioning method and training method of target area positioning model |
| CN111160167A (en) * | 2019-12-18 | 2020-05-15 | 北京信息科技大学 | Spindle fault classification and identification method based on S-transform deep convolutional neural network |
| CN111401136A (en) * | 2020-02-22 | 2020-07-10 | 上海交通大学 | Plunger pump cavitation degree detection method and device and terminal |
| CN111428548A (en) * | 2019-09-16 | 2020-07-17 | 中国石油大学(华东) | Vibration signal processing method based on Matlab time-frequency analysis |
| CN111796342A (en) * | 2020-06-05 | 2020-10-20 | 山东科技大学 | Tiny fault diagnosis method and device for cam type absolute gravimeter |
| WO2020248471A1 (en) * | 2019-06-14 | 2020-12-17 | 华南理工大学 | Aggregation cross-entropy loss function-based sequence recognition method |
| CN112566174A (en) * | 2020-12-02 | 2021-03-26 | 中国电子科技集团公司第五十二研究所 | Abnormal I/Q signal identification method and system based on deep learning |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11544422B2 (en) * | 2019-09-16 | 2023-01-03 | Palo Alto Research Center Incorporated | Machine learning based systems and methods for real time, model based diagnosis |
-
2021
- 2021-06-22 CN CN202110692924.2A patent/CN113409213B/en active Active
Patent Citations (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101188453A (en) * | 2006-11-20 | 2008-05-28 | 阿尔卡特朗讯 | Method and system for wireless cellular indoor communications |
| KR20120114776A (en) * | 2011-04-08 | 2012-10-17 | (주)한세로에프에이 | Inspecting method of color similarity using complementary color relationship of chromatic color and achromatic color, subject detecting method and heat treatment inspecting method using the same |
| CN107915303A (en) * | 2016-10-10 | 2018-04-17 | 天津有序环境科技发展有限公司 | Can connecting shaft pump Ammonia adding System and control method more |
| CN109214402A (en) * | 2017-07-04 | 2019-01-15 | 华北电力大学（保定） | A kind of people having the same aspiration and interest unit grouping method of combination WAVELET FUZZY entropy and GG fuzzy clustering |
| CN108830127A (en) * | 2018-03-22 | 2018-11-16 | 南京航空航天大学 | A kind of rotating machinery fault feature intelligent diagnostic method based on depth convolutional neural networks structure |
| CN108694298A (en) * | 2018-07-04 | 2018-10-23 | 温州大学 | A kind of axial plunger pump bearing inner race and outer ring method for diagnosing faults |
| CN108757426A (en) * | 2018-07-04 | 2018-11-06 | 温州大学 | Oilfield water filling plunger pump trouble diagnostic method |
| WO2020248471A1 (en) * | 2019-06-14 | 2020-12-17 | 华南理工大学 | Aggregation cross-entropy loss function-based sequence recognition method |
| CN111428548A (en) * | 2019-09-16 | 2020-07-17 | 中国石油大学(华东) | Vibration signal processing method based on Matlab time-frequency analysis |
| CN110766027A (en) * | 2019-10-22 | 2020-02-07 | 腾讯科技（深圳）有限公司 | Image area positioning method and training method of target area positioning model |
| CN111160167A (en) * | 2019-12-18 | 2020-05-15 | 北京信息科技大学 | Spindle fault classification and identification method based on S-transform deep convolutional neural network |
| CN111401136A (en) * | 2020-02-22 | 2020-07-10 | 上海交通大学 | Plunger pump cavitation degree detection method and device and terminal |
| CN111796342A (en) * | 2020-06-05 | 2020-10-20 | 山东科技大学 | Tiny fault diagnosis method and device for cam type absolute gravimeter |
| CN112566174A (en) * | 2020-12-02 | 2021-03-26 | 中国电子科技集团公司第五十二研究所 | Abnormal I/Q signal identification method and system based on deep learning |
Non-Patent Citations (6)
| Title |
|---|
| A new bearing fault diagnosis method based on signal-to-image mapping and convolutional neural network;Jing Zhao 等;《Measurement》;20210531;第176卷;1-15 * |
| Vision-Based Fault Diagnostics Using Explainable Deep Learning With Class Activation Maps;Kyung Ho Sun 等;《IEEE Access 》;第8卷;98962-98972 * |
| 一种具有强抗噪性的深度学习故障诊断模型;吴春志 等;《西安交通大学学报》;20210430;第55卷(第4期);摘要、第2节 * |
| 基于卷积神经网络和灰色模型的轴承健康评估及预测研究;闫康;《中国优秀硕士学位论文全文数据库 工程科技Ⅱ辑》;第2021年卷(第1期);C029-288 * |
| 基于时频图和 CNN 的滚动轴承故障诊断;刘炳集 等;《南昌航空大学学报(自然科学版)》;20180630;第32卷(第2期);摘要、第2.1-2.2节 * |
| 基于机器学习的轴承机械振动信号分析方法研究;马鎏豪;《中国优秀硕士学位论文全文数据库 工程科技Ⅱ辑》;20201015;第2020年卷(第10期);C029-12 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113409213A (en) | 2021-09-17 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Pan et al. | A novel deep learning network via multiscale inner product with locally connected feature extraction for intelligent fault detection | |
| CN109035149B (en) | License plate image motion blur removing method based on deep learning | |
| CN111523509B (en) | Equipment fault diagnosis and health monitoring method integrating physical and depth expression characteristics | |
| CN111238814B (en) | Rolling bearing fault diagnosis method based on short-time Hilbert transform | |
| CN110617966A (en) | Bearing fault diagnosis method based on semi-supervised generation countermeasure network | |
| CN113673346B (en) | Motor vibration data processing and state identification method based on multiscale SE-Resnet | |
| CN111401136B (en) | Plunger pump cavitation degree detection method and device and terminal | |
| CN111751133A (en) | Intelligent fault diagnosis method of deep convolutional neural network model based on non-local mean embedding | |
| CN115290326A (en) | Rolling bearing fault intelligent diagnosis method | |
| CN113409213B (en) | Method and system for enhancing noise reduction of time-frequency diagram of fault signal of plunger pump | |
| CN114964782A (en) | Rolling bearing fault diagnosis acoustic emission signal processing method and system based on compressed sensing | |
| CN114861349A (en) | Rolling bearing RUL prediction method based on model migration and wiener process | |
| CN113758709A (en) | Rolling bearing fault diagnosis method and system combining edge calculation and deep learning | |
| CN112149804B (en) | Novel convolutional neural network model and application | |
| CN114065809A (en) | Method and device for identifying abnormal sound of passenger car, electronic equipment and storage medium | |
| CN117592543A (en) | Aeroengine rolling bearing fault diagnosis method based on self-supervision learning | |
| CN117473414A (en) | Bearing fault position identification method based on low-noise time-frequency image | |
| CN117037841A (en) | Acoustic signal hierarchical cavitation intensity identification method based on hierarchical transition network | |
| CN116772122A (en) | Natural gas pipeline leakage fault diagnosis method, system, equipment and medium | |
| CN116541771A (en) | Unbalanced sample bearing fault diagnosis method based on multi-scale feature fusion | |
| Li et al. | A robust fault diagnosis method for rolling bearings based on deep convolutional neural network | |
| CN116484176A (en) | Bearing fault diagnosis method, system and storage medium based on ultra-wavelet | |
| CN111047537A (en) | System for recovering details in image denoising | |
| CN116361723A (en) | Bearing fault diagnosis and classification method based on multi-scale characteristics and attention | |
| CN115017939A (en) | Intelligent diagnosis method and device for faults of aircraft fuel pump and storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |