CN114818774A - Intelligent gearbox fault diagnosis method based on multi-channel self-calibration convolutional neural network - Google Patents
Intelligent gearbox fault diagnosis method based on multi-channel self-calibration convolutional neural network Download PDFInfo
- Publication number
- CN114818774A CN114818774A CN202210254062.XA CN202210254062A CN114818774A CN 114818774 A CN114818774 A CN 114818774A CN 202210254062 A CN202210254062 A CN 202210254062A CN 114818774 A CN114818774 A CN 114818774A
- Authority
- CN
- China
- Prior art keywords
- calibration
- self
- neural network
- data
- convolutional neural
- 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.)
- Pending
Links
Images
Classifications
-
- 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
-
- 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
- G06N3/048—Activation functions
-
- 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)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Computational Linguistics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Signal Processing (AREA)
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
Abstract
The invention discloses a gearbox intelligent fault diagnosis method based on a multi-channel self-calibration convolutional neural network, which comprises the following steps of: and (3) increasing the dimension of the data of the angle field of the Gerami, converting one-dimensional vibration signals of the sensors into two-dimensional data, converting the two-dimensional data into gray images as input, establishing a data set, and dividing the data set into a training set and a test set. And constructing a self-calibration convolutional neural network, and extracting data characteristics. And setting a fusion layer, converting the output of the self-calibration convolution neural network into one-dimensional data, and fusing characteristic information. And setting a full connection layer, and mapping the distributed features to a sample mark space. And constructing a Softmax feature classifier to classify the extracted features. And learning the network by using the training set, and testing the trained network by using the test set to realize fault diagnosis of the gearbox. The self-calibration convolutional neural network model provided by the invention is combined with an information fusion method, and can be used for effectively diagnosing the single fault of the gearbox under the working condition of the same rotating speed.
Description
Technical Field
The invention belongs to the technical field of intelligent fault diagnosis of a vibration signal of a gearbox, and relates to an intelligent fault diagnosis method of the gearbox based on a multi-channel self-calibration convolutional neural network.
Background
With the rapid advance of Chinese information technology and industrial generation technology, modern mechanical industrial equipment has the characteristics of large scale, high speed, complexity, intellectualization and integration, the functions of the equipment are continuously improved, and the internal structure of the equipment is increasingly precise. Gearboxes are the most widely used important components in mechanical equipment, and smooth operation of the gearboxes is crucial to the normal operation of the whole equipment. However, the gear box is often in harsh working conditions such as variable rotation speed and high load, so that damage and even failure are easily caused, and huge economic loss and casualties are caused.
Classical fault diagnosis techniques include expert systems, model parsing, and signal processing. However, the methods have the defects of poor universality, complex modeling, low robustness, poor characterization capability and the like, so that the application range of the methods is greatly limited. Under the background of the era of big data development, the intelligent fault diagnosis method becomes a hotspot and a difficult problem of research in the field of fault diagnosis by virtue of the universality, end-to-end property and strong characterization capability.
In the failure diagnosis, data acquired by the sensors are single, and converted image data is also acquired by each sensor, so that all characteristics of the failure cannot be reflected from a plurality of angles. The data-level information fusion ensures the integrity of the original information. Therefore, data fusion of a plurality of sensors has important significance for fault diagnosis.
The traditional convolutional neural network cannot learn the characteristics of each channel in a differentiated mode, and the network can learn similar characteristic forms. The self-calibration convolutional neural network is provided, so that the network can learn better difference characteristics conveniently, and the attention and generalization capability of the neural network are improved.
Disclosure of Invention
The invention aims to overcome the defects of the prior art and provides an intelligent fault diagnosis method for a gearbox of a multi-channel self-calibration convolutional neural network, which combines sensor information fusion and the self-calibration convolutional neural network and can effectively diagnose the single fault of the gearbox under the same rotating speed working condition. The invention can realize the data fusion of a plurality of sensors and a single fault, reflect all the characteristics of the fault from a plurality of angles, ensure the integrity of the original information and effectively improve the attention and generalization capability of the neural network.
In order to solve the technical problems, the invention adopts the following technical scheme.
A gearbox intelligent fault diagnosis method based on a multi-channel self-calibration convolutional neural network is characterized in that a plurality of three-way contact type acceleration sensors are fixed on a gearbox and used for respectively measuring vibration signals of an input shaft, an intermediate shaft and an output shaft of a motor in the y direction and the z direction at a certain constant rotating speed; connecting an LMS vibration and noise test system in Belgian with each sensor to obtain the value of the vibration signal measured by the sensor;
the method comprises the following steps:
step 1, performing Granami angle field data dimension increase on an original vibration signal of each sensor: scaling the one-dimensional sequence data, converting the one-dimensional sequence data from a rectangular coordinate system to a polar coordinate system, and then recognizing the time correlation of different time points by considering the angle sum of different points; converting the two-dimensional matrix data into a gray-scale image, establishing a data set, and dividing a training set and a test set;
step 4, setting a full connection layer: mapping the distributed features to a sample label space;
step 6, training and testing the network model: and learning the network by using the training set, and testing the trained network by using the test set to realize fault diagnosis of the gearbox.
Further, the step 1 performs the gray scale angle field data dimension increasing on the raw vibration signal of each sensor, and the process includes:
step 1.1 scaling: performing normalization processing on the one-dimensional sequence data, and mapping the one-dimensional data to [0,1] by utilizing a minmax normalization method;
step 1.2, coordinate conversion: converting the scaled time-series data into polar coordinates, namely: the numerical value is regarded as a cosine value of an included angle, and the converted angle range is [0, pi/2 ];
step 1.3, angle conversion: converting the converted data angle in the polar coordinate, and calculating an angle sum;
if the length of the original time sequence data is 2n, obtaining a matrix with the size of [2n, 2n ] after angle and conversion; adopting a segmentation aggregation approximation to segment the sequence, and then compressing the subsequences in each segment into a value by averaging, thereby reducing the parameters; 2 x 2 is selected as a matrix block and the 2n x 2n matrix is converted to an n x n matrix.
Further, in the step 2, a self-calibration convolution module is established, a self-calibration convolution neural network is established, and data features are extracted, wherein the process comprises the following steps:
(1) for the input original characteristic diagram F 1 ∈R H×W×C/2 Performing Down-sampling once, wherein the size of the characteristic diagram is changed from H multiplied by W multiplied by C/2 to H multiplied by 2W multiplied by 2C/2, and the operation is defined as Down (); h, W, C, wherein the height, width and number of channels of the picture input into the neural network are represented respectively;
(2) convolving the down-sampled feature map, denoted as f 1 (*)；
(3) Performing one-time upsampling operation on the feature map after convolution in the step (2), and recovering the feature map size from H/2 xW/2 xC/2 to H xW xC/2, wherein the operation is marked as Up ();
(4) to be inputted original feature map F 1 ∈R H×W×C/2 And (3) adding the feature map in (3), and obtaining a feature map with the scale of H × W × C/2 after passing through an activation function Sigmoid, wherein the operation is defined as σ (), and a calculation formula of the activation function Sigmoid can be expressed as:
wherein e is a natural constant having a value of about 2.718;
(5) for the original characteristic diagram F 1 ∈R H×W×C/2 Performing a convolution operation, which is denoted as f 2 (*)；
(6) Multiplying the characteristic diagram obtained in (4) by the characteristic diagram obtained in (5), and this operation is described as
(7) Performing convolution operation on the characteristic diagram obtained in the step (6) to obtain an output F 1 '∈R H×W×C/2 The operation is denoted as f 3 () c; the output of the self-calibrating convolution operation can be represented as
Further, in step 3, a fusion layer is set to fuse the feature information extracted by the self-calibration convolution module according to a certain weight, and the process includes:
the output of the self-calibration convolutional neural network is three-dimensional data, and the fusion layer is used for weighting and fusing the features extracted from different channels so as to obtain feature information extracted from a plurality of sensor signals;
converting 6 groups of vibration signals measured by the sensor into gray-scale images and processing the gray-scale images by a self-calibration convolution neural network to obtain 6 new gray-scale images; the value of each pixel point at the same position on the gray scale map is x 1 、x 2 、x 3 、x 4 、x 5 、x 6 The weights of the fusion are respectively set to w 1 、w 2 、w 3 、w 4 、w 5 、w 6 If the value of the pixel point at the corresponding position on the gray-scale image after information fusion is x 1 w 1 +x 2 w 2 +x 3 w 3 +x 4 w 4 +x 5 w 5 +x 6 w 6 。
Further, in the step 5, a Softmax feature classifier is constructed to classify the extracted features, and the process includes:
when the self-calibration convolutional neural network classifies images, an input image is transmitted into a Softmax feature classifier after neural network feature extraction, a parameter matrix theta is obtained after the Softmax feature classifier is trained, the theta is multiplied by an image feature column vector, probability values of the image to which the image belongs are output, wherein the category corresponding to the maximum value is the judgment category of the image;
the Softmax feature classifier maps the input vectors from the N-dimensional space to classes, and the result is given in the form of probability, as shown below:
in the formula (I), the compound is shown in the specification,
as the weight, the classifier parameters corresponding to the classes, the model parameter θ is as follows:
theta is obtained by the training of a Softmax classifier, and the class of the item to be classified is determined by calculating the probability of all possible classes; given a data set comprising n training samples: { (x) (1) ,y (1) ),(x (2) ,y (2) ),...,(x (n) ,y (n) )}，x (i) Representing an input vector, y (i) For each x (i) A category label of (1); for a given test sample x (i) The Softmax classifier estimates the probability that it belongs to each class, which is calculated as follows:
in the formula, h θ (x (i) ) Is a vector of elements p (y) (i) ＝k|x (i) (ii) a Theta) represents x (i) Probability of belonging to class k, h θ (x (i) ) The sum of the individual elements in (1); for x (i) Selecting k corresponding to the maximum probability value as a classification result of the current image; the value of the parameter θ can be found by minimizing a cost function, which is defined as:
wherein {. is an indicative function, equal to 1 for true and equal to 0 for false; j (theta) is minimized and a classifier parameter theta is derived.
Further, in step 6, the process of training and testing the network model includes:
using a vibration signal measured by an acceleration sensor of a motor at a certain constant rotating speed to perform Gramami angular field data dimension increase on an original signal to obtain a two-dimensional gray image; dividing the image into a training set and a testing set according to a certain proportion, inputting the training set into a network model, training the network after setting training parameters, iterating for 30 times, and drawing an accuracy value and a loss rate of each iteration into an accuracy rate and loss rate curve graph;
and inputting the test set into a trained network for testing, and drawing the accuracy result of each test into a test accuracy chart after 30 times of tests.
Compared with the prior art, the invention has the following advantages and beneficial effects:
1. the method adopts a Grarami angular field algorithm to convert one-dimensional vibration signal data of different sensors into two-dimensional matrix data, then converts the two-dimensional matrix data into gray level images, establishes a data set, and divides the data set into a training set and a test set; after the one-dimensional sequence data are converted into two-dimensional data through the Grarami angle field algorithm, the original signal time sequence characteristics are reserved, and the fault characteristics are enhanced.
2. The invention utilizes the self-calibration convolution module to extract the picture characteristic information, the structure can enlarge the receptive field of the self-calibration convolution, each channel of the convolution neural network learns different characteristics differently, multi-scale information is extracted better, and the attention and generalization capability of the neural network are improved. The self-calibration convolutional neural network is used for replacing the traditional fault diagnosis method, labor force is reduced, meanwhile, the fault diagnosis accuracy is improved, and a brand-new theoretical method and a technology implementation means are provided for state monitoring and fault diagnosis of a modern complex large system.
3. The invention adopts a plurality of sensors to collect vibration signals of the gearbox, compared with the signals collected by a single sensor, the fault information reflected by the signals collected by the plurality of sensors is more comprehensive and specific, the fault information can reflect the characteristics of the fault from a plurality of angles, the probability of fault diagnosis errors caused by incomplete signals collected by the single sensor is reduced, and the fault information is input into the self-calibration convolution neural network for training and testing, so that the fault diagnosis can be effectively carried out.
Drawings
FIG. 1 is a schematic diagram of a sensor arrangement and information acquisition according to an embodiment of the present invention.
FIG. 2 is a flow chart of a gearbox fault diagnosis method based on a multi-channel self-calibration convolutional neural network.
FIG. 3 is a Grammin angular field dimension increase schematic of one embodiment of the present invention.
FIG. 4 is a schematic diagram of a self-calibrating convolutional neural network structure according to an embodiment of the present invention.
FIG. 5 is a schematic diagram of information fusion in accordance with an embodiment of the present invention.
FIG. 6 is a graph of training accuracy for one embodiment of the present invention.
FIG. 7 is a training loss rate graph of an embodiment of the present invention.
FIG. 8 is a test accuracy chart for one embodiment of the present invention.
Detailed Description
The invention provides a gearbox intelligent fault diagnosis method based on a multi-channel self-calibration convolutional neural network, which is used for effectively diagnosing single faults of a gearbox under the same rotating speed working condition by combining information fusion and the self-calibration convolutional neural network. The method comprises the following steps: and (3) converting one-dimensional vibration signals measured by a plurality of sensors into two-dimensional data through the data dimension increase of the Granami angular field, converting the two-dimensional data into gray images as input, establishing a data set, and dividing the data set into a training set and a test set. And establishing a self-calibration convolution module, constructing a self-calibration convolution neural network, and extracting data characteristics. And setting a fusion layer, converting the output of the self-calibration convolution neural network into one-dimensional data, and fusing characteristic information. And setting a full connection layer, and mapping the distributed features to a sample mark space. And constructing a Softmax feature classifier to classify the extracted features. And learning the network by using the training set, and testing the trained network by using the test set to realize fault diagnosis of the gearbox. The self-calibration convolutional neural network model provided by the invention combines an information fusion method, and can effectively diagnose the single fault of the gearbox under the working condition of the same rotating speed.
The present invention will be described in further detail with reference to the accompanying drawings.
As shown in figure 1, the gear box of the invention is a bevel gear box, which comprises an input shaft, a middle shaft, an output shaft, 4 planetary gears and a plurality of bearings. The invention uses a plurality of three-way contact acceleration sensors which are fixed on a gear box through magnetic seats to respectively measure vibration signals of an input shaft, an intermediate shaft and an output shaft of a motor at a certain constant rotating speed, and each sensor can measure signals of the corresponding shaft in the y direction and the z direction. The invention uses the Belgian LMS vibration and noise test system to collect signals, and connects the signals with each sensor, thereby obtaining the value of the signals measured by the sensors.
As shown in FIG. 2, the intelligent fault diagnosis method for the gearbox based on the multi-channel self-calibration convolutional neural network comprises the following steps:
step 1, performing Granami angle field data dimension increase on an original vibration signal of each sensor: and scaling the one-dimensional sequence data, and converting the rectangular coordinate system into a polar coordinate system. The temporal correlation of the different points in time is then identified by taking into account the sum of the angles between the different points. And converting the two-dimensional matrix data into a gray-scale image, establishing a data set, and dividing a training set and a test set.
And 3, setting a fusion layer, and fusing the characteristic information extracted by the self-calibration convolution module according to certain weight. And converting the output of the self-calibration convolution module into one-dimensional data, wherein a certain weight refers to that each weight value is a number between 0 and 1, and the sum of the weights is 1.
And 4, setting a full connection layer, and mapping the distributed features to a sample mark space.
And 5, constructing a Softmax feature classifier to classify the extracted features, mapping the input vectors to classes from the N-dimensional space, and giving results in a probability form.
Step 6, training and testing the network model: and learning the network by using the training set, and testing the trained network by using the test set to realize fault diagnosis of the gearbox.
In the step 1, the original vibration signal of each sensor is subjected to the data dimension increment of the Grarami angle field, and the process comprises the following steps:
the method of using the grammine angular field is used to convert one-dimensional data into two-dimensional data, and the principle is shown in fig. 3. The method scales one-dimensional sequence data, and converts the one-dimensional sequence data from a rectangular coordinate system to a polar coordinate system, and identifies the time correlation of different time points by considering the angle sum between different points, and the method is implemented by the following steps:
step 1.1 scaling
And performing normalization processing on the one-dimensional sequence data, and mapping the one-dimensional data to [0,1] by utilizing a minmax normalization method.
Step 1.2 coordinate conversion
And converting the zoomed time sequence data into polar coordinates, namely regarding the numerical value as an included angle cosine value, wherein the converted angle range is [0, pi/2 ].
Step 1.3 Angle Change
The converted data angle is transformed in polar coordinates, and the angle sum is calculated.
If the length of the original time series data is 2n, after angle and conversion, a matrix with the size of [2n, 2n ] is obtained. The sequence is segmented using a segmentation-aggregation approximation, and then the sub-sequences within each segment are compressed to a value by averaging, thereby reducing the parameters. 2 x 2 is selected as a matrix block, and the 2n x 2n matrix is converted into an n x n matrix.
In the step 2, a self-calibration convolution module is established, a self-calibration convolution neural network is established, and data characteristics are extracted, wherein the process comprises the following steps:
the convolutional neural network has the advantage that the spatial correlation of the input data is extracted, and the simple one-dimensional vector spatial relationship is not obvious. The one-dimensional vibration signal sequence data is converted into two-dimensional data, so that a good diagnosis effect can be achieved in the aspect of intelligent fault diagnosis.
Convolution operation is common operation for extracting feature information in a neural network, but the uniform convolution kernel size makes it difficult to extract feature information of different scales in one convolution operation, and the problem is effectively solved by the self-calibration convolution. The network structure is shown in fig. 4.
The self-calibration convolution operation steps are as follows:
(1) for the input original characteristic diagram F 1 ∈R H×W×C/2 And performing Down-sampling once, wherein the size of the characteristic diagram is changed from H multiplied by W multiplied by C/2 to H multiplied by 2W multiplied by C/2, and the operation is defined as Down (). Wherein H, W, C represents the height, width, and number of channels, respectively, of the picture input into the neural network.
(2) Convolving the down-sampled feature map, denoted as f 1 (*)。
(3) And (3) performing one-time upsampling operation on the feature map after the convolution in the step (2), and recovering the feature map size from H/2 xW/2 xC/2 to H xW xC/2, wherein the operation is marked as Up ().
(4) To be inputted original feature map F 1 ∈R H×W×C/2 And (4) adding the feature maps in the step (3), and calculating a formula of the feature maps to obtain the feature map with the scale of H multiplied by W multiplied by C/2 after passing through an activation function Sigmoid, wherein the operation is defined as sigma (#). The activation function Sigmoid calculation formula can be expressed as:
where e is a natural constant, and has a value of about 2.718.
(5) For the original characteristic diagram F 1 ∈R H×W×C/2 Performing a convolution operation, which is denoted as f 2 (*)。
(6) Multiplying the characteristic diagram obtained in (4) by the characteristic diagram obtained in (5), and this operation is described as
(7) Performing convolution operation on the characteristic diagram obtained in the step (6) to obtain an output F 1 '∈R H×W×C/2 The operation is denoted as f 3 (*). The output of the self-calibrating convolution operation can be represented as
In step 3, a fusion layer is set, and the characteristic information extracted by the self-calibration convolution module is fused according to a certain weight, wherein the process comprises the following steps:
the output of the self-calibration convolutional neural network is three-dimensional data, the fusion layer is used for weighting and fusing the features extracted from different channels, so as to obtain feature information extracted from a plurality of sensor signals, and an information fusion schematic diagram is shown in fig. 5.
6 groups of vibration signals measured by the sensor are converted into gray level images, and the gray level images are processed by a self-calibration convolution neural network to obtain 6 new gray level images, wherein the numerical value of each pixel point at the same position on the gray level images is x 1 、x 2 、x 3 、x 4 、x 5 、x 6 The weight of the fusion is set to w 1 、w 2 、w 3 、w 4 、w 5 、w 6 If the value of the pixel point at the corresponding position on the gray-scale image after information fusion is x 1 w 1 +x 2 w 2 +x 3 w 3 +x 4 w 4 +x 5 w 5 +x 6 w 6 。
Step 4 set up the full tie layer, include:
the fully-connected layer maps the distributed features to a sample-label space, which essentially linearly transforms one feature space to another.
And 5, constructing a Softmax feature classifier to classify the extracted features, wherein the process comprises the following steps:
the principle of the Softmax classifier is probability computation. When the self-calibration convolutional neural network classifies images, an input image is transmitted into a classifier after neural network characteristic extraction, a parameter matrix theta can be obtained after the Softmax classifier trains the input image, the theta is multiplied by an image characteristic column vector, and probability values of the image classification belong to various types are output. Wherein, the category corresponding to the maximum value is the judgment category of the image.
The Softmax regression is an extension of logistic regression, provides more possibilities for class labels, and is suitable for multi-classification problems. The Softmax classifier maps the input vectors from the N-dimensional space to classes, and the result is given in the form of probability, as shown below:
in the formula (I), the compound is shown in the specification,
as the weight, the classifier parameters corresponding to the classes, the model parameter θ is as follows:
theta is obtained by training a Softmax classifier, and the class of the item to be classified is determined by calculating all possible class probabilities of the item. Given a data set comprising n training samples: { (x) (1) ,y (1) ),(x (2) ,y (2) ),...,(x (n) ,y (n) )}，x (i) Representing an input vector, y (i) For each x (i) The category label of (1). For a given test sample x (i) The Softmax classifier estimates the probability that it belongs to each class, and the calculation principle is as follows:
in the formula, h θ (x (i) ) Is a vector of elements p (y) (i) ＝k|x (i) (ii) a Theta) represents x (i) Probability of belonging to class k, h θ (x (i) ) The sum of the individual elements in (1). For x (i) And selecting k corresponding to the maximum probability value as a classification result of the current image. The value of the parameter θ can be found by minimizing a cost function defined as:
where {. is an indicative function, is true or equal to 1, and false or equal to 0. J (theta) is minimized and a classifier parameter theta is derived.
Step 6, the process of training and testing the network model includes:
and learning the network by using the training set, and testing the trained network by using the test set to realize fault diagnosis.
The vibration sensor arrangement is shown in figure 1. And respectively acquiring vibration signals of the input shaft, the intermediate shaft and the output shaft in the directions of the y axis and the z axis. The used data comprises 9 fault data and normal data, wherein the 9 fault data respectively comprise the crack width of the inner ring and the outer ring of the bearing is 1.2mm, the crack depth is 2mm, the crack width of the inner ring and the outer ring of the bearing is 0.6mm, the crack depth is 2mm, 2 tooth breakage faults of the bevel gear, the pitting faults of 4 pits and 6 pits, the crack fault of 0.5 times of tooth thickness and the abrasion fault of 0.8 times of tooth thickness.
Vibration signals measured by an acceleration sensor at a certain constant rotating speed of a motor are used, and the original signals are subjected to Gramami angular field data dimension increase to be changed into two-dimensional gray images. Dividing the image into a training set and a testing set according to a certain proportion, inputting the training set into a network model, training the network after setting training parameters, iterating for 30 times, and drawing the accuracy value and the loss rate of each iteration into a graph to obtain the accuracy rate and loss rate curve graphs shown in fig. 6 and 7.
And inputting the test set into the trained network for testing, and drawing the accuracy result of each test in a graph after 30 times of tests to obtain a test accuracy graph shown in FIG. 8.
Claims (6)
1. A gearbox intelligent fault diagnosis method based on a multi-channel self-calibration convolutional neural network is characterized in that a plurality of three-way contact acceleration sensors are fixed on a gearbox and used for respectively measuring vibration signals of an input shaft, an intermediate shaft and an output shaft of a motor in the y direction and the z direction at a certain constant rotating speed; connecting an LMS vibration and noise test system in Belgian with each sensor to obtain the value of the vibration signal measured by the sensor;
the method comprises the following steps:
step 1, performing Granami angle field data dimension increase on an original vibration signal of each sensor: scaling the one-dimensional sequence data, converting the one-dimensional sequence data from a rectangular coordinate system to a polar coordinate system, and then recognizing the time correlation of different time points by considering the angle sum of different points; converting the two-dimensional matrix data into a gray-scale image, establishing a data set, and dividing a training set and a test set;
step 2, establishing a self-calibration convolution module, establishing a self-calibration convolution neural network, and extracting data characteristics: splitting the input feature F into two parts { F } uniformly 1 ,F 2 Sending each part into a special path respectively for paying attention to different types of characteristic information; in the first path, different convolution operations { Conv }are utilized 1 ,Conv 2 ,Conv 3 F pair 1 Performing a self-calibration operation to obtain F 1 '; in the second path, Conv is operated using convolution 4 Performing a simple convolution operation yields F 2 ', original spatial feature information is maintained; will { F over Concate connect operation 1 ′，F′ 2 Connected together as an output;
step 3, setting a fusion layer, and fusing the characteristic information extracted by the self-calibration convolution module according to certain weight: converting the output of the self-calibration convolution module into one-dimensional data; the certain weight refers to that each weight value is a number between 0 and 1, and the sum of the weights is 1;
step 4, setting a full connection layer: mapping the distributed features to a sample label space;
step 5, constructing a Softmax feature classifier to classify the extracted features: mapping the input vector from the N-dimensional space to the category, and giving a result in a probability form;
step 6, training and testing the network model: and learning the network by using the training set, and testing the trained network by using the test set to realize fault diagnosis of the gearbox.
2. The intelligent gearbox fault diagnosis method based on the multi-channel self-calibration convolutional neural network as claimed in claim 1, wherein the step 1 of performing the data dimension increment of the Grarami angular field on the raw vibration signal of each sensor comprises the following steps:
step 1.1 scaling: performing normalization processing on the one-dimensional sequence data, and mapping the one-dimensional data to [0,1] by utilizing a minmax normalization method;
step 1.2, coordinate conversion: converting the scaled time-series data into polar coordinates, namely: the numerical value is regarded as a cosine value of an included angle, and the converted angle range is [0, pi/2 ];
step 1.3, angle conversion: converting the converted data angle in the polar coordinate, and calculating an angle sum;
if the length of the original time sequence data is 2n, obtaining a matrix with the size of [2n, 2n ] after angle and conversion; adopting a segmentation aggregation approximation to segment the sequence, and then compressing the subsequences in each segment into a value by averaging, thereby reducing the parameters; 2 x 2 is selected as a matrix block and the 2n x 2n matrix is converted to an n x n matrix.
3. The intelligent gearbox fault diagnosis method based on the multi-channel self-calibration convolutional neural network as claimed in claim 1, wherein in the step 2, a self-calibration convolutional module is established, a self-calibration convolutional neural network is established, and data characteristics are extracted, and the process comprises the following steps:
(1) for the input original characteristic diagram F 1 ∈R H×W×C/2 Performing Down-sampling once, wherein the size of the characteristic diagram is changed from H multiplied by W multiplied by C/2 to H multiplied by 2W multiplied by 2C/2, and the operation is defined as Down (); h, W, C, wherein the height, width and number of channels of the picture input into the neural network are represented respectively;
(2) convolving the down-sampled feature map, denoted as f 1 (*)；
(3) Performing one-time upsampling operation on the feature map after convolution in the step (2), and recovering the feature map size from H/2 xW/2 xC/2 to H xW xC/2, wherein the operation is marked as Up ();
(4) to be inputted original feature map F 1 ∈R H×W×C/2 And (3) adding the feature map in (3), and obtaining a feature map with the scale of H × W × C/2 after passing through an activation function Sigmoid, wherein the operation is defined as σ (), and a calculation formula of the activation function Sigmoid can be expressed as:
wherein e is a natural constant having a value of about 2.718;
(5) for the original characteristic diagram F 1 ∈R H×W×C/2 Performing a convolution operation, which is denoted as f 2 (*)；
(6) Multiplying the characteristic diagram obtained in (4) by the characteristic diagram obtained in (5), and this operation is described as
(7) Performing convolution operation on the characteristic diagram obtained in the step (6) to obtain an output F 1 ′∈R H×W×C/2 The operation is denoted as f 3 () c; the output of the self-calibrating convolution operation can be represented as
4. The intelligent gearbox fault diagnosis method based on the multi-channel self-calibration convolutional neural network as claimed in claim 1, wherein in the step 3, a fusion layer is arranged, and the characteristic information extracted by the self-calibration convolutional module is fused according to a certain weight, and the process comprises the following steps:
the output of the self-calibration convolutional neural network is three-dimensional data, and the fusion layer is used for weighting and fusing the features extracted from different channels so as to obtain feature information extracted from a plurality of sensor signals;
converting 6 groups of vibration signals measured by the sensors into gray-scale images, and processing the gray-scale images by a self-calibration convolution neural network to obtain 6 new gray-scale images; the value of each pixel point at the same position on the gray scale map is x 1 、x 2 、x 3 、x 4 、x 5 、x 6 The weights of the fusion are respectively set to w 1 、w 2 、w 3 、w 4 、w 5 、w 6 If the value of the pixel point at the corresponding position on the gray-scale image after information fusion is x 1 w 1 +x 2 w 2 +x 3 w 3 +x 4 w 4 +x 5 w 5 +x 6 w 6 。
5. The intelligent gearbox fault diagnosis method based on the multi-channel self-calibration convolutional neural network as claimed in claim 1, wherein in the step 5, a Softmax feature classifier is constructed to classify the extracted features, and the process comprises:
when the self-calibration convolutional neural network classifies images, an input image is transmitted into a Softmax feature classifier after neural network feature extraction, a parameter matrix theta is obtained after the Softmax feature classifier is trained, the theta is multiplied by an image feature column vector, probability values of the image to which the image belongs are output, wherein the category corresponding to the maximum value is the judgment category of the image;
the Softmax feature classifier maps the input vectors from the N-dimensional space to classes, and the result is given in the form of probability, as shown below:
in the formula (I), the compound is shown in the specification,
as the weight, the classifier parameters corresponding to the classes, the model parameter θ is as follows:
theta is obtained by the training of a Softmax classifier, and the class of the item to be classified is determined by calculating the probability of all possible classes; given a data set comprising n training samples: { (x) (1) ,y (1) ),(x (2) ,y (2) ),...,(x (n) ,y (n) )}，x (i) Representing an input vector, y (i) For each x (i) A category label of (1); for a given test sample x (i) The Softmax classifier estimates the probability that it belongs to each class, which is calculated as follows:
in the formula, h θ (x (i) ) Is a vector of elements p (y) (i) ＝k|x (i) (ii) a Theta) represents x (i) Probability of belonging to class k, h θ (x (i) ) The sum of the individual elements in (1); for x (i) Selecting k corresponding to the maximum probability value as a classification result of the current image; the value of the parameter θ can be found by minimizing a cost function defined as:
wherein {. is an indicative function, equal to 1 for true and equal to 0 for false; j (theta) is minimized and a classifier parameter theta is derived.
6. The intelligent gearbox fault diagnosis method based on the multi-channel self-calibration convolutional neural network as claimed in claim 1, wherein the step 6 of training and testing the network model comprises:
using a vibration signal measured by an acceleration sensor of a motor at a certain constant rotating speed to perform Gramami angular field data dimension increase on an original signal to obtain a two-dimensional gray image; dividing the image into a training set and a testing set according to a certain proportion, inputting the training set into a network model, training the network after setting training parameters, iterating for 30 times, and drawing an accuracy value and a loss rate of each iteration into an accuracy rate and loss rate curve graph;
and inputting the test set into a trained network for testing, and drawing the accuracy result of each test into a test accuracy chart after 30 times of tests.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210254062.XA CN114818774A (en) | 2022-03-15 | 2022-03-15 | Intelligent gearbox fault diagnosis method based on multi-channel self-calibration convolutional neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210254062.XA CN114818774A (en) | 2022-03-15 | 2022-03-15 | Intelligent gearbox fault diagnosis method based on multi-channel self-calibration convolutional neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114818774A true CN114818774A (en) | 2022-07-29 |
Family
ID=82528912
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210254062.XA Pending CN114818774A (en) | 2022-03-15 | 2022-03-15 | Intelligent gearbox fault diagnosis method based on multi-channel self-calibration convolutional neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114818774A (en) |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115114965A (en) * | 2022-08-29 | 2022-09-27 | 中国长江三峡集团有限公司 | Wind turbine generator gearbox fault diagnosis model, method, equipment and storage medium |
| CN115238833A (en) * | 2022-09-22 | 2022-10-25 | 中国民航大学 | Method for extracting performance degradation characteristics of air turbine starter bearing |
| CN115937217A (en) * | 2023-03-09 | 2023-04-07 | 安格利(成都)仪器设备有限公司 | Method for automatically detecting corrosion pits by using 3D convolution and correlation operator |
| CN115964360A (en) * | 2023-03-14 | 2023-04-14 | 山东省地震工程研究院 | Earthquake safety evaluation database construction method and system |
| CN116109897A (en) * | 2023-04-14 | 2023-05-12 | 中国科学院自动化研究所 | Robot fish sensor fault detection method and device based on airspace image fusion |
| CN116628591A (en) * | 2023-05-26 | 2023-08-22 | 安徽大学 | Gear box compound fault diagnosis method and system |
| CN116840759A (en) * | 2023-07-04 | 2023-10-03 | 成都泰格微电子研究所有限责任公司 | Rapid calibration system and method suitable for hybrid integrated circuit test system |
| CN117372261A (en) * | 2023-12-04 | 2024-01-09 | 无锡日联科技股份有限公司 | Resolution reconstruction method, device, equipment and medium based on convolutional neural network |
| CN117516925A (en) * | 2023-11-16 | 2024-02-06 | 兰州理工大学 | MSDC-Swin-T-based gearbox fault diagnosis method |
| CN117934963A (en) * | 2024-02-07 | 2024-04-26 | 山东工商学院 | Gas sensor drift compensation method |
-
2022
- 2022-03-15 CN CN202210254062.XA patent/CN114818774A/en active Pending
Cited By (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115114965A (en) * | 2022-08-29 | 2022-09-27 | 中国长江三峡集团有限公司 | Wind turbine generator gearbox fault diagnosis model, method, equipment and storage medium |
| CN115238833A (en) * | 2022-09-22 | 2022-10-25 | 中国民航大学 | Method for extracting performance degradation characteristics of air turbine starter bearing |
| CN115238833B (en) * | 2022-09-22 | 2022-11-25 | 中国民航大学 | Method for extracting performance degradation characteristics of air turbine starter bearing |
| CN115937217A (en) * | 2023-03-09 | 2023-04-07 | 安格利(成都)仪器设备有限公司 | Method for automatically detecting corrosion pits by using 3D convolution and correlation operator |
| CN115964360B (en) * | 2023-03-14 | 2023-05-16 | 山东省地震工程研究院 | Method and system for building earthquake safety evaluation database |
| CN115964360A (en) * | 2023-03-14 | 2023-04-14 | 山东省地震工程研究院 | Earthquake safety evaluation database construction method and system |
| CN116109897A (en) * | 2023-04-14 | 2023-05-12 | 中国科学院自动化研究所 | Robot fish sensor fault detection method and device based on airspace image fusion |
| CN116109897B (en) * | 2023-04-14 | 2023-08-15 | 中国科学院自动化研究所 | Robot fish sensor fault detection method and device based on airspace image fusion |
| CN116628591A (en) * | 2023-05-26 | 2023-08-22 | 安徽大学 | Gear box compound fault diagnosis method and system |
| CN116628591B (en) * | 2023-05-26 | 2024-08-23 | 安徽大学 | Gear box compound fault diagnosis method and system |
| CN116840759A (en) * | 2023-07-04 | 2023-10-03 | 成都泰格微电子研究所有限责任公司 | Rapid calibration system and method suitable for hybrid integrated circuit test system |
| CN117516925A (en) * | 2023-11-16 | 2024-02-06 | 兰州理工大学 | MSDC-Swin-T-based gearbox fault diagnosis method |
| CN117372261A (en) * | 2023-12-04 | 2024-01-09 | 无锡日联科技股份有限公司 | Resolution reconstruction method, device, equipment and medium based on convolutional neural network |
| CN117372261B (en) * | 2023-12-04 | 2024-02-27 | 无锡日联科技股份有限公司 | Resolution reconstruction method, device, equipment and medium based on convolutional neural network |
| CN117934963A (en) * | 2024-02-07 | 2024-04-26 | 山东工商学院 | Gas sensor drift compensation method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN114818774A (en) | Intelligent gearbox fault diagnosis method based on multi-channel self-calibration convolutional neural network | |
| CN113436169B (en) | Industrial equipment surface crack detection method and system based on semi-supervised semantic segmentation | |
| CN111695598B (en) | Monitoring data abnormity diagnosis method based on generation countermeasure network | |
| CN112964469B (en) | Online fault diagnosis method for rolling bearing under variable load of transfer learning | |
| CN110648310B (en) | Weak supervision casting defect identification method based on attention mechanism | |
| CN113205466A (en) | Incomplete point cloud completion method based on hidden space topological structure constraint | |
| CN112818969A (en) | Knowledge distillation-based face pose estimation method and system | |
| CN112257862B (en) | Semi-supervised identification method based on relation network marking sample expansion | |
| CN116593157A (en) | Complex working condition gear fault diagnosis method based on matching element learning under small sample | |
| CN114564987B (en) | Rotary machine fault diagnosis method and system based on graph data | |
| CN115859077A (en) | Multi-feature fusion motor small sample fault diagnosis method under variable working conditions | |
| CN117009916A (en) | Actuator fault diagnosis method based on multi-sensor information fusion and transfer learning | |
| Xue et al. | A novel framework for motor bearing fault diagnosis based on multi-transformation domain and multi-source data | |
| CN113537152A (en) | Flow field state fault detection method based on deep neural network | |
| CN114714145A (en) | Method for enhancing, comparing, learning and monitoring tool wear state by using Gelam angular field | |
| CN113530850B (en) | Centrifugal pump fault diagnosis method based on ESA and stacked capsule self-encoder | |
| Chou et al. | SHM data anomaly classification using machine learning strategies: A comparative study | |
| CN117708574B (en) | CNN variable-speed rolling bearing fault diagnosis method embedded with physical information | |
| CN113984389A (en) | Rolling bearing fault diagnosis method based on multi-receptive-field and improved capsule map neural network | |
| CN113723592A (en) | Fault diagnosis method based on wind power gear box monitoring system | |
| CN117312948A (en) | Rotary machine fault diagnosis method based on multi-sensor fusion and ECA-CNN | |
| CN116934820A (en) | Cross-attention-based multi-size window Transformer network cloth image registration method and system | |
| CN116907838A (en) | Gear fault diagnosis method based on enhanced relation network | |
| CN114383846B (en) | Bearing composite fault diagnosis method based on fault label information vector | |
| CN115099286A (en) | Attention-enhanced harmonic reducer tiny fault diagnosis method and system |
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 |