CN113591638A - Planetary gearbox fault diagnosis method based on convolution capsule network - Google Patents

Planetary gearbox fault diagnosis method based on convolution capsule network Download PDF

Info

Publication number
CN113591638A
CN113591638A CN202110818877.1A CN202110818877A CN113591638A CN 113591638 A CN113591638 A CN 113591638A CN 202110818877 A CN202110818877 A CN 202110818877A CN 113591638 A CN113591638 A CN 113591638A
Authority
CN
China
Prior art keywords
layer
capsule
convolution
fault
pooling
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
Application number
CN202110818877.1A
Other languages
Chinese (zh)
Inventor
张冕
黎德才
古震岳
马跃
康天博
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Tianjin University of Technology
Original Assignee
Tianjin University of Technology
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Tianjin University of Technology filed Critical Tianjin University of Technology
Priority to CN202110818877.1A priority Critical patent/CN113591638A/en
Publication of CN113591638A publication Critical patent/CN113591638A/en
Pending legal-status Critical Current

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
    • G01—MEASURING; TESTING
    • G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
    • G01M13/00—Testing of machine parts
    • G01M13/02—Gearings; Transmission mechanisms
    • G01M13/021—Gearings
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
    • G01M13/00—Testing of machine parts
    • G01M13/02—Gearings; Transmission mechanisms
    • G01M13/028—Acoustic or vibration analysis
    • 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
    • G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
    • 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
    • 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
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/08—Feature extraction

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Evolutionary Computation (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Signal Processing (AREA)
  • Acoustics & Sound (AREA)
  • Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)

Abstract

The invention discloses a planetary gearbox fault diagnosis method based on a convolution capsule network, which belongs to the field of fault diagnosis and comprises the following steps of S1, acquiring planetary gearbox data under various health states, and constructing a sample set; s2, utilizing a convolution pooling structure to realize automatic extraction and dimension reduction of fault characteristics; s3, representing the characteristics and transmitting information through the vectorization of the capsule structure, and meanwhile, calculating the correlation degree between different capsule layer vectors by adopting a dynamic routing mechanism; and S4, continuously optimizing model parameters through an interval loss function and input data, and realizing accurate intelligent diagnosis of faults of different gear parts. The invention provides a network model capable of improving the fault diagnosis capability of the network model by combining the convolutional neural network and the capsule network, and the efficiency is high.

Description

Planetary gearbox fault diagnosis method based on convolution capsule network
Technical Field
The invention belongs to the field of fault diagnosis, relates to the technical field of fault diagnosis of planetary gearboxes, and particularly relates to a planetary gearbox fault diagnosis method based on a convolution capsule network.
Background
The planetary gear box is a core transmission link of large-scale mechanical equipment, and due to the severe service environment of the planetary gear box, if a plurality of factors such as frequent change of working conditions, heavy load, extreme weather and the like exist, key components in the planetary gear system are easy to break down, and then the whole body is driven by pulling one to cause catastrophic accidents. Therefore, the development of fault diagnosis research of the planetary gear box has important significance for ensuring safe and stable operation of large-scale mechanical equipment and avoiding life and property loss of people.
Since the advent of deep learning, the method shows powerful automatic feature learning and extracting capabilities, and is widely applied to the fields of image recognition, natural language processing, voice recognition, fault diagnosis and the like. Typical Deep learning methods, such as Convolutional Neural Network (CNN), Generative Adaptive Network (GAN), Deep Belief Network (DBN), etc., are gradually extended to fault diagnosis methods of planetary gearboxes: researchers diagnose the faults of the planet wheels in different health states through a CNN model; researchers also provide a GAN diagnosis model which adopts a stacked noise reduction automatic encoder as a discriminator, diagnoses sun gears in various different health states, and has stronger diagnosis capability under the condition of a small number of samples and noise; researchers also propose a DBN method with optimized parameters based on the locust algorithm to identify multiple fault types of a single gear. Currently, a deep learning algorithm-based diagnostic model is gaining wide attention from researchers, however, most of the current research is being conducted around different faults of a single gear part. Due to the fact that gear fault occurrence positions and forms in the planetary gear box are various, fault diagnosis of a single gear is only conducted, and safe operation of mechanical equipment is difficult to guarantee comprehensively. In particular, a deep learning model with CNN as a core has a good effect on identifying a failure of a single component, but when applied to failure data including different components, it is difficult to ensure high diagnostic performance.
A Capsule Network (CN) model is proposed for the first time in 2017, the one-sidedness of information transmission of a traditional scalar neuron is broken through, and spatial information of features is further excavated by taking vectors as input and output of the Network. In recent years, capsule networks show strong detail feature extraction capability, and sharp awns are originally revealed in the field of fault diagnosis of mechanical equipment. However, few researchers at home and abroad currently apply the capsule network to fault diagnosis of the planetary gearbox.
Disclosure of Invention
This section is for the purpose of summarizing some aspects of embodiments of the invention and to briefly introduce some preferred embodiments. In this section, as well as in the abstract and the title of the invention of this application, simplifications or omissions may be made to avoid obscuring the purpose of the section, the abstract and the title, and such simplifications or omissions are not intended to limit the scope of the invention.
The present invention has been made in view of the above problems occurring in the description of the related art, and therefore it is an object of the present invention to provide a method for diagnosing a fault of a planetary gearbox based on a convolutional capsule network, which combines a convolutional neural network and a capsule network to provide a network model capable of improving the fault diagnosis capability thereof with high efficiency.
In order to solve the technical problems, the invention adopts the technical scheme that: a planetary gearbox fault diagnosis method based on a convolution capsule network comprises the following steps,
s1, acquiring the data of the planetary gear box under various health states, and constructing a sample set;
s2, utilizing a convolution pooling structure to realize automatic extraction and dimension reduction of fault characteristics;
s3, representing the characteristics and transmitting information through the vectorization of the capsule structure, and meanwhile, calculating the correlation degree between different capsule layer vectors by adopting a dynamic routing mechanism;
and S4, continuously optimizing model parameters through an interval loss function and input data, and realizing accurate intelligent diagnosis of faults of different gear parts.
Further, the convolutional capsule network model comprises 9 layers: 1 input layer, 2 convolutional layers, 2 pooling layers, 1 Dropout layer, 1 pre-capsule layer, 1 digital capsule layer and 1 output layer, wherein:
an input layer: the raw data is subjected to z-score standardization to be used as convolution capsule network input data, and the standardization formula is as follows:
Figure BDA0003171151940000021
in the formula, YiThe normalized data are obtained; xiIs original data;
Figure BDA0003171151940000022
is the mean value of the original data; sigma is the variance of the original data;
the convolutional layer 1: extracting input features by adopting a wide convolution kernel so as to reduce the influence of noise;
and (3) convolutional layer 2: a large number of narrow convolution kernels are adopted to fully extract the bottom layer characteristics of the features;
a pooling layer: a pooling layer is attached behind each convolution layer and used for reducing model training parameters and improving the training speed;
dropout layer: to discard part of the neurons to prevent overfitting during the training phase;
pre-capsule layer: convolution operation is carried out, and meanwhile, a capsule is constructed by a convolution result in a vector form and is used as the input of a digital capsule layer;
digital capsule layer: calculating the correlation degree between the capsule layers through a dynamic routing mechanism to realize accurate classification of fault characteristics;
an output layer: and performing two-norm solution on the output vector to obtain probability distribution of different fault types.
Further, the convolution capsule network model adopts an interval loss function, and the expression of the interval loss function is as follows:
Lk＝Tkmax(0，m+-||vk||)2+λ(1-Tk)max(0，||vk||-m-)2
in the formula: subscript k is a fault category k; l iskInterval loss for class k; t iskA classification indication function (1 is taken for correct classification, and 0 is taken for incorrect classification); v. ofkRepresenting the probability of being identified as a fault category k; m is the upper bound, 0.9 is selected; m is the lower bound, 0.1 is taken; for the scaling factor, take 0.5 and the total loss is the sum of the losses for the various cases.
Furthermore, the raw data is standardized by z-score to be used as the input data of the convolution capsule network, and the standardized formula is as follows:
Figure BDA0003171151940000031
wherein, YiThe normalized data are obtained; xiIs original data;
Figure BDA0003171151940000032
is the mean value of the original data; σ is the raw data variance.
Further, in S2, the raw data z-score is normalized and then input into the convolutional layer, and feature extraction is performed on the input data through convolution operation and an activation function: firstly, carrying out convolution operation on input data through n convolution kernels with the same dimensionality, and simultaneously adding an offset term to obtain a convolution result; then, nonlinear mapping is carried out on the convolution result by using an activation function, n feature matrices are output, and a calculation formula is as follows:
Figure BDA0003171151940000033
in the formula:
Figure BDA0003171151940000034
outputting a jth element of the nth feature matrix for the convolutional layer; mjIs the jth convolution region of the input data; xiIs MjThe ith element of (1);
Figure BDA0003171151940000035
a weight matrix of the nth convolution kernel; bnA bias term corresponding to the nth convolution kernel; f (-) represents the activation function.
Further, in S2, the extracted initial features are input to a pooling layer, and the dimension reduction process for the feature matrix is implemented through pooling operation: dividing the output data of the convolutional layer into q pooling units with equal width without overlapping, and performing pooling operation on elements in each unit to obtain an output characteristic matrix of the pooling layer.
Further, the Pooling operation used is Max Pooling (Max Pooling), and the expression of the Pooling process is:
Figure BDA0003171151940000041
in the formula:
Figure BDA0003171151940000042
outputting the q element in the nth feature matrix for the pooling layer;
Figure BDA0003171151940000043
the qth pooling unit of the nth feature matrix is input for the pooling layer,
Figure BDA0003171151940000044
wherein W is the width of the pooling unit;
after the elements are subjected to pooling operation, inputting the elements into a Dropout layer for processing, and abandoning part of neurons to prevent overfitting in a training stage;
wherein Dropout ratio is 0.25.
Further, data enter a pre-capsule layer after passing through a Dropout layer, a capsule is constructed by a convolution result in a vector form and serves as an input of the digital capsule layer, and a calculation formula is as follows:
Figure BDA0003171151940000045
wherein, g1(t) isOutputting the capsule network convolution layer at the time t; e.g. of the typei(t) is the output of the primary capsule layer at time t, which is a series of 8-dimensional vectors;
Figure BDA0003171151940000046
weight matrix and offset vector of the capsule network convolution layer at the time t;
Figure BDA0003171151940000047
a weight matrix and a bias vector of a primary capsule layer at the time t; relu () is a Rectified Linear Unit.
Further, the output of the primary capsule layer is processed by a compression and dynamic routing algorithm to obtain the output of the digital capsule layer, and the specific formula of the processing is as follows:
Figure BDA0003171151940000048
wherein v isj(t) is the output of the jth capsule of the digital capsule layer at time t, j ∈ [0, 7 ]]It is the index value corresponding to each planetary gearbox health status; lj(t) is the total input vector from the primary capsule layer to the digital capsule layer at time t;
Figure BDA0003171151940000049
for prediction vectors at time t, by multiplying the output of the primary capsule layer by a weight matrix Wij(t) obtaining that i is the number of 8-dimensional vectors output by the primary capsule layer; a isij(t) is the coupling coefficient at time t; bij(t) iterative update parameters of the dynamic routing algorithm at the time t, wherein the initial value is 0; r is the iteration number of the dynamic routing algorithm;
calculating the vector vjAnd (t) comparing the lengths of all vectors at the current time t, taking the state of the motor corresponding to the index value of the longest vector as the state of the planet wheel or the sun wheel corresponding to the current data, and outputting the gear state.
Further, according to the output v of the digital capsule layerj(t) building a loss function for optimizing a fault diagnosis model, whichThe calculation expression is as follows:
Lk(t)＝Tk(t)max(0，m+-||vk(t)||)2+λ(1-Tk(t))max(0，||vk(t)||-m-)2
in the formula: subscript k is a fault category k; l isk(t) interval loss for class k; t isk(t) is a classification indication function (1 is taken for correct classification, and 0 is taken for incorrect classification otherwise); v. ofk(t) represents the probability of identifying as a failure category k; m is the upper bound, 0.9 is selected; m is the lower bound, 0.1 is taken; for the proportionality coefficient, take 0.5. The total loss is the sum of all losses;
calculating the loss function value L at time tk(t) loss function value L at time t-1k(t-1) comparing if Lk(t) is less than Lk(t-1), directly optimizing the weights and the offsets of the convolutional neural network and the capsule network by using a gradient descent method; otherwise, optimizing the weights and the offsets of the convolutional neural network and the capsule network by using an Adam Optimizer;
judging whether the current iteration time T reaches the maximum iteration time T, if so, stopping iteration, taking the updated weight and offset as parameters of the convolution capsule network, and then outputting the final state; otherwise, adding 1 to the current iteration times t, returning to the convolution layer 1, and continuously selecting the next Batch for training;
and dividing the collected vibration data of the planetary gearbox into a training set and a testing set according to a proportion, and inputting the training set and the testing set into a model so as to output the probability distribution of the health state of the planetary gearbox.
Compared with the prior art, the invention has the following advantages and positive effects.
1. The invention provides a fault diagnosis method, which can diagnose various faults of different gear parts of a planetary gear box and realize end-to-end automatic diagnosis from original data to fault categories;
2. the method combines the deep extraction capability of the convolution pooling structure on the fault characteristics and the characteristic of vectorization excavation of characteristic space information of the capsule structure, and realizes accurate and intelligent diagnosis on the fault of the planetary gearbox;
3. compared with the traditional convolutional neural network and the capsule network, the convolutional capsule network has higher diagnosis accuracy rate under the mixed condition of single-class component multi-fault data and multi-class component multi-fault data, and has good application prospect.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings needed to be used in the description of the embodiments will be briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without inventive exercise.
FIG. 1 is a flow chart of an embodiment of a planetary gearbox fault diagnosis method based on an enhanced capsule network disclosed by the invention;
FIG. 2 is a diagram of a convolutional capsule network model architecture of the present invention;
FIG. 3 is a waveform of the original vibration signal of the planetary gearbox in a healthy state and 7 fault states;
FIG. 4 is a graph of the diagnostic accuracy and loss for the training set and validation set of the convolutional capsule network of the present invention.
Detailed Description
In order to make the aforementioned objects, features and advantages of the present invention comprehensible, embodiments accompanied with figures are described in detail below.
In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present invention, but the present invention may be practiced in other ways than those specifically described and will be readily apparent to those of ordinary skill in the art without departing from the spirit of the present invention, and therefore the present invention is not limited to the specific embodiments disclosed below.
Next, the present invention will be described in detail with reference to the drawings, wherein the cross-sectional views illustrating the structure of the device are not enlarged partially according to the general scale for convenience of illustration when describing the embodiments of the present invention, and the drawings are only examples, which should not limit the scope of the present invention. In addition, the three-dimensional space dimensions including length, width and depth should be included in the actual manufacturing
Again, it should be noted that the embodiments and features of the embodiments of the present invention may be combined with each other without conflict.
In the description of the present invention, it is to be understood that the terms "center", "longitudinal", "lateral", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", and the like, indicate orientations or positional relationships based on those shown in the drawings, and are used only for convenience in describing the present invention and for simplicity in description, and do not indicate or imply that the referenced devices or elements must have a particular orientation, be constructed and operated in a particular orientation, and thus, are not to be construed as limiting the present invention. Furthermore, the terms "first", "second", etc. are used for descriptive purposes only and are not to be construed as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first," "second," etc. may explicitly or implicitly include one or more of that feature. In the description of the present invention, "a plurality" means two or more unless otherwise specified.
In the description of the present invention, it should be noted that, unless otherwise explicitly specified or limited, the terms "mounted," "connected," and "connected" are to be construed broadly, e.g., as meaning either a fixed connection, a removable connection, or an integral connection; can be mechanically or electrically connected; they may be connected directly or indirectly through intervening media, or they may be interconnected between two elements. The specific meaning of the above terms in the present invention can be understood by those of ordinary skill in the art through specific situations.
In order to make the objects, technical solutions and advantages of the present invention more apparent, specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.
As shown in FIG. 1, a method for diagnosing the fault of the planetary gearbox based on the convolution capsule network comprises the following steps:
s1, acquiring the data of the planetary gear box under various health states, and constructing a sample set;
s2, utilizing a convolution pooling structure to realize automatic extraction and dimension reduction of fault characteristics;
s3, representing the characteristics and transmitting information through the vectorization of the capsule structure, and meanwhile, calculating the correlation degree between different capsule layer vectors by adopting a dynamic routing mechanism;
and S4, continuously optimizing model parameters through an interval loss function and input data, and realizing accurate intelligent diagnosis of faults of different gear parts.
The main structure of the convolutional capsule network provided by the present invention is shown in fig. 2, and comprises: an input layer, a convolution layer, a pooling layer, a Dropout layer, a pre-capsule layer, a digital capsule layer and an output layer. The input layer is used for standardizing original data through z-score to serve as convolution capsule network input data, the convolution layer is used for extracting input features and a large number of narrow convolution kernels through wide convolution kernels in sequence, the Dropout layer is used for abandoning part of neurons to prevent overfitting in a training stage, the pre-capsule layer is used for constructing capsules according to convolution results in a vector mode to serve as input of a digital capsule layer, the digital capsule layer calculates the correlation degree between the capsule layers through a dynamic routing mechanism to achieve accurate classification of fault features, and the output layer is used for carrying out two-norm solution on output vectors to obtain probability distribution of different fault types.
Specifically, the method comprises the following steps:
the formula for the z-score normalization process for the input layer is:
Figure BDA0003171151940000071
in the formula, YiThe normalized data are obtained; xiIs original data;
Figure BDA0003171151940000072
is the mean value of the original data; σ is the raw data variance.
Inputting a convolution layer after the z-score standardization processing of original data, and realizing the feature extraction of the input data through convolution operation and an activation function: firstly, carrying out convolution operation on input data through n convolution kernels with the same dimensionality, and simultaneously adding an offset term to obtain a convolution result; then, nonlinear mapping is carried out on the convolution result by using an activation function, n feature matrices are output, and a calculation formula is as follows:
Figure BDA0003171151940000081
in the formula:
Figure BDA0003171151940000082
outputting a jth element of the nth feature matrix for the convolutional layer; mjIs the jth convolution region of the input data; xiIs MjThe ith element of (1);
Figure BDA0003171151940000083
a weight matrix of the nth convolution kernel; bnA bias term corresponding to the nth convolution kernel; f (-) represents the activation function.
Inputting the extracted initial features into a pooling layer, and realizing the dimension reduction process of the feature matrix through pooling operation: dividing the output data of the convolutional layer into q pooling units with equal width without overlapping, and performing pooling operation on elements in each unit to obtain an output characteristic matrix of the pooling layer. The Pooling operation used is Max Pooling (Max Pooling), and the Pooling process expression is:
Figure BDA0003171151940000084
in the formula:
Figure BDA0003171151940000085
outputting the q element in the nth feature matrix for the pooling layer;
Figure BDA0003171151940000086
the qth pooling unit of the nth feature matrix is input for the pooling layer,
Figure BDA0003171151940000087
wherein W is the width of the pooling unit.
After the elements are subjected to pooling operation, inputting the elements into a Dropout layer for processing, and abandoning part of neurons to prevent overfitting in a training stage;
wherein Dropout ratio is 0.25.
Data enter a pre-capsule layer after passing through a Dropout layer, a capsule is constructed by a convolution result in a vector form and is used as input of a digital capsule layer, and a calculation formula is as follows:
Figure BDA0003171151940000088
wherein, g1(t) is the output of the capsule network convolution layer at time t; e.g. of the typei(t) is the output of the primary capsule layer at time t, which is a series of 8-dimensional vectors;
Figure BDA0003171151940000089
weight matrix and offset vector of the capsule network convolution layer at the time t;
Figure BDA00031711519400000810
a weight matrix and a bias vector of a primary capsule layer at the time t; relu () is a Rectified Linear Unit;
the output of the primary capsule layer is processed through a compression and dynamic routing algorithm to obtain the output of the digital capsule layer, and the specific formula of the processing is as follows:
Figure BDA0003171151940000091
wherein v isj(t) is the output of the jth capsule of the digital capsule layer at time t, j ∈ [0, 7 ]]It is the index value corresponding to each planetary gearbox health status; lj(t) is the total input vector from the primary capsule layer to the digital capsule layer at time t;
Figure BDA0003171151940000092
for prediction vectors at time t, by multiplying the output of the primary capsule layer by a weight matrix Wij(t) obtaining that i is the number of 8-dimensional vectors output by the primary capsule layer; a isij(t) is the coupling coefficient at time t; bij(t) iterative update parameters of the dynamic routing algorithm at the time t, wherein the initial value is 0; and r is the iteration number of the dynamic routing algorithm.
Calculating the vector vjAnd (t) comparing the lengths of all vectors at the current time t, taking the state of the motor corresponding to the index value of the longest vector as the state of the planet wheel or the sun wheel corresponding to the current data, and outputting the gear state.
According to the output v of the digital capsule layerj(t) building a loss function for optimizing the fault diagnosis model, wherein the calculation expression of the loss function is as follows:
Lk(t)＝Tk(t)max(0，m+-||vk(t)||)2+λ(1-Tk(t))max(0，||vk(t)||-m-)2
in the formula: subscript k is a fault category k; l isk(t) interval loss for class k; t isk(t) is a classification indication function (1 is taken for correct classification, and 0 is taken for incorrect classification otherwise); v. ofk(t) represents the probability of identifying as a failure category k; m is the upper bound, 0.9 is selected; m is the lower bound, 0.1 is taken; for the proportionality coefficient, take 0.5. The total loss is the sum of the losses of the various cases.
Calculating the loss function value L at time tk(t) loss function value L at time t-1k(t-1) comparing if Lk(t) is less than Lk(t-1), directly optimizing the weights and the offsets of the convolutional neural network and the capsule network by using a gradient descent method; otherwise, optimizing the weights and the bias of the convolutional neural network and the capsule network by using an Adam Optimizer, wherein the concrete formula of the Adam Optimizer optimization is as follows:
Figure BDA0003171151940000093
wherein g (t) is the gradient at time t; m istIs an exponential moving average；htIs a squared gradient; beta is a1、β2Is a constant number, beta1、β2∈[0，1](ii) a E is a constant, and defaults to 1 e-8; thetatRepresenting updated model parameters including updated weights and biases of the convolutional neural network and the capsule network.
Judging whether the current iteration time T reaches the maximum iteration time T, if so, stopping iteration, taking the updated weight and offset as parameters of the convolution capsule network, and then outputting the final state; otherwise, adding 1 to the current iteration times t, returning to the convolution layer 1, and continuously selecting the next Batch for training;
and dividing the collected vibration data of the planetary gearbox into a training set and a testing set according to a proportion, and inputting the training set and the testing set into a model so as to output the probability distribution of the health state of the planetary gearbox.
In order to verify the effectiveness of the method disclosed by the invention, a dynamic fault and fault transmission experiment table (Dynamci driving trian Simulator, DDS) is used for acquiring vibration signals of the planetary gearbox under different faults. The experiment table mainly comprises a servo driving motor; one input torque encoder and one output torque encoder; a planetary gear box; acquiring data by an acceleration sensor (100mv/g) in the horizontal direction and the vertical direction; a dead axle gear box and a programmable magnetic brake.
TABLE 1 planetary gearbox Assembly Gear parameters
Gear parameters Numerical value
Number of sun gear teeth 28
Number of teeth of planetary gear 36
Number of teeth of gear ring 100
Number of planet wheels 4
Under the two working conditions that the rotating speed of the motor is 1800r/min and 3000r/min respectively, vibration data of the health state of 8 types of gears are collected respectively, the collection time is 19.2s, and the sampling frequency is set to be 10240 Hz. The vibration signal is intercepted as one sample with a length of 1024 data points, and the ratio of 4: the ratio of 1 divides the training set and the test set, and the data distribution is shown in table 2.
TABLE 2 training set, test set data distribution
Figure BDA0003171151940000111
The time domain signal of the above 8 types of fault type samples is shown in fig. 3.
The design of the invention aims at the fault diagnosis experimental analysis under the condition of mixing fault data of the sun wheel and the planet wheel, and the sample distribution is shown in the table 3.
TABLE 3 Gear Fault sample distribution Table
Figure BDA0003171151940000112
The label of the data set adopts a coding mode of 'one-hot': for the fault diagnosis of X categories, the labels are converted into X-dimensional vectors, one value is 1, the rest are 0, and the labels of each category of faults are different.
The input sample dimension of the convolution capsule network model is 1024 multiplied by 1, and the output dimension is the number of fault types. The activation function adopts a Relu function, and the Loss function adopts a space Loss (Margin Loss) function. The model specific structure parameter settings are shown in table 4.
TABLE 4 convolution capsule network model structural parameter set-up
Figure BDA0003171151940000121
The invention is utilized to carry out fault diagnosis on the planetary gear box with 8 types of health states, and the training set is divided into 9 parts: the proportion of 1 is divided into a training set and a verification set, the single processing data volume batch-size of the two networks is set to be 10, 50 epochs are trained totally, an Adam (adaptive motion) optimization function is adopted, and the learning rate r is set to be 0.001.
As shown in FIG. 4, the training and verification process of the convolutional capsule network model for fault diagnosis of the planetary gear box in the 8 types of health states under the working conditions that the rotating speed of the motor is 1800r/min (left) and 3000r/min (right) is given.
In order to verify the advantages of the method provided by the invention, the invention is used for carrying out multi-class fault diagnosis on the planetary gear box in 8-class health states by comparing and analyzing a Convolutional Neural Network (CNN) and a traditional Capsule Network (CN). In order to make the experimental result more convincing, the comparison experiment adopts a ten-fold cross validation training mode for 10 times, the training set is equally divided into 10 parts, 9 parts are selected for training without repeating every time, and 1 part is selected for validation. The single training data volume batch-size was set to 10, 5 epochs per training, for a total of 50 epochs per 10 trains. The two models adopt Adam algorithm to optimize model parameters, and the learning rate r is set to be 0.001. The accuracy of the test set was used as a reference for comparison, and the average of the 5 results was taken, and the diagnostic results for the 3-class models are shown in table 5.
TABLE 5 accuracy (%) of various types of failure tests of different gear parts
Figure BDA0003171151940000131
As shown in Table 5, under the working condition that the rotating speed of the motor is 1800r/min, the test accuracy of the TC-CN model reaches 98.19 percent, which is superior to 93.01 percent of the CNN model and 92.96 percent of the CN model; under the working condition that the rotating speed of the motor is 3000r/min, the TC-CN model achieves the test accuracy of 97.36 percent, which is respectively higher than that of the CN model by 4.47 percent and that of the CNN model by 10.98 percent.
Experimental results show that the convolution capsule network model has higher identification precision for various faults of different gear parts.
It should be noted that the above-mentioned embodiments are only for illustrating the technical solutions of the present invention and not for limiting, and although the present invention has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that modifications or equivalent substitutions may be made on the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, which should be covered by the claims of the present invention.

Claims (10)

1. A planetary gearbox fault diagnosis method based on a convolution capsule network is characterized by comprising the following steps: comprises the following steps of (a) carrying out,
s1, acquiring the data of the planetary gear box under various health states, and constructing a sample set;
s2, utilizing a convolution pooling structure to realize automatic extraction and dimension reduction of fault characteristics;
s3, representing the characteristics and transmitting information through the vectorization of the capsule structure, and meanwhile, calculating the correlation degree between different capsule layer vectors by adopting a dynamic routing mechanism;
and S4, continuously optimizing model parameters through an interval loss function and input data, and realizing accurate intelligent diagnosis of faults of different gear parts.
2. The method for diagnosing the fault of the planetary gearbox based on the convolution capsule network as claimed in claim 1, wherein the method comprises the following steps: the convolutional capsule network model comprises 9 layers: 1 input layer, 2 convolutional layers, 2 pooling layers, 1 Dropout layer, 1 pre-capsule layer, 1 digital capsule layer and 1 output layer, wherein:
an input layer: the raw data is subjected to z-score standardization to be used as convolution capsule network input data, and the standardization formula is as follows:
Figure FDA0003171151930000011
in the formula, YiThe normalized data are obtained; xiIs original data;
Figure FDA0003171151930000012
is the mean value of the original data; sigma is the variance of the original data;
the convolutional layer 1: extracting input features by adopting a wide convolution kernel so as to reduce the influence of noise;
and (3) convolutional layer 2: a large number of narrow convolution kernels are adopted to fully extract the bottom layer characteristics of the features;
a pooling layer: a pooling layer is attached behind each convolution layer and used for reducing model training parameters and improving the training speed;
dropout layer: to discard part of the neurons to prevent overfitting during the training phase;
pre-capsule layer: convolution operation is carried out, and meanwhile, a capsule is constructed by a convolution result in a vector form and is used as the input of a digital capsule layer;
digital capsule layer: calculating the correlation degree between the capsule layers through a dynamic routing mechanism to realize accurate classification of fault characteristics;
an output layer: and performing two-norm solution on the output vector to obtain probability distribution of different fault types.
3. The method for diagnosing the fault of the planetary gearbox based on the convolution capsule network as claimed in claim 1, wherein the method comprises the following steps: the convolution capsule network model adopts an interval loss function, and the expression is as follows:
Lk＝Tkmax(0，m+-||vk||)2+λ(1-Tk)max(0，||vk||-m-)2
in the formula: subscript k is a fault category k; l iskInterval loss for class k; t iskA classification indication function (1 is taken for correct classification, and 0 is taken for incorrect classification); v. ofkRepresenting the probability of being identified as a fault category k; m is the upper bound, 0.9 is selected; m is the lower bound, 0.1 is taken; for the proportionality coefficient, take 0.5 and the total loss as the loss of each sampleThe sum of the losses.
4. The method for diagnosing the fault of the planetary gearbox based on the convolution capsule network as claimed in claim 1, wherein the method comprises the following steps: the raw data is subjected to z-score standardization to be used as convolution capsule network input data, and the standardization formula is as follows:
Figure FDA0003171151930000021
wherein, YiThe normalized data are obtained; xiIs original data;
Figure FDA0003171151930000022
is the mean value of the original data; σ is the raw data variance.
5. The method for diagnosing the fault of the planetary gearbox based on the convolution capsule network as claimed in claim 4, wherein the method comprises the following steps: in S2, the raw data z-score is normalized and input to the convolution layer, and feature extraction is performed on the input data by convolution operation and an activation function: firstly, carrying out convolution operation on input data through n convolution kernels with the same dimensionality, and simultaneously adding an offset term to obtain a convolution result; then, nonlinear mapping is carried out on the convolution result by using an activation function, n feature matrices are output, and a calculation formula is as follows:
Figure FDA0003171151930000023
in the formula:
Figure FDA0003171151930000024
outputting a jth element of the nth feature matrix for the convolutional layer; mjIs the jth convolution region of the input data; xiIs MiThe ith element of (1);
Figure FDA0003171151930000025
a weight matrix of the nth convolution kernel; bnA bias term corresponding to the nth convolution kernel; f (-) represents the activation function.
6. The method for diagnosing the fault of the planetary gearbox based on the convolution capsule network as claimed in claim 1, wherein the method comprises the following steps: in S2, the extracted initial features are input to a pooling layer, and the dimension reduction process for the feature matrix is implemented by pooling operation: dividing the output data of the convolutional layer into q pooling units with equal width without overlapping, and performing pooling operation on elements in each unit to obtain an output characteristic matrix of the pooling layer.
7. The method for diagnosing the fault of the planetary gearbox based on the convolution capsule network as claimed in claim 6, wherein the method comprises the following steps: the Pooling operation used is Max Pooling (Max Pooling), and the Pooling process expression is:
Figure FDA0003171151930000026
in the formula:
Figure FDA0003171151930000031
outputting the q element in the nth feature matrix for the pooling layer;
Figure FDA0003171151930000032
the qth pooling unit of the nth feature matrix is input for the pooling layer,
Figure FDA0003171151930000033
wherein W is the width of the pooling unit;
after the elements are subjected to pooling operation, inputting the elements into a Dropout layer for processing, and abandoning part of neurons to prevent overfitting in a training stage;
wherein Dropout ratio is 0.25.
8. The method for diagnosing the fault of the planetary gearbox based on the convolution capsule network as claimed in claim 6, wherein the method comprises the following steps: data enter a pre-capsule layer after passing through a Dropout layer, a capsule is constructed by a convolution result in a vector form and is used as input of a digital capsule layer, and a calculation formula is as follows:
Figure FDA0003171151930000034
wherein, g1(t) is the output of the capsule network convolution layer at time t; e.g. of the typei(t) is the output of the primary capsule layer at time t, which is a series of 8-dimensional vectors;
Figure FDA0003171151930000035
weight matrix and offset vector of the capsule network convolution layer at the time t;
Figure FDA0003171151930000036
a weight matrix and a bias vector of a primary capsule layer at the time t; relu () is a Rectified Linear Unit.
9. The method for diagnosing the fault of the planetary gearbox based on the convolution capsule network as claimed in claim 6, wherein the method comprises the following steps: the output of the primary capsule layer is processed through a compression and dynamic routing algorithm to obtain the output of the digital capsule layer, and the specific formula of the processing is as follows:
Figure FDA0003171151930000037
wherein v isj(t) is the output of the jth capsule of the digital capsule layer at time t, j ∈ [0, 7 ]]It is the index value corresponding to each planetary gearbox health status; lj(t) is the total input vector from the primary capsule layer to the digital capsule layer at time t;
Figure FDA0003171151930000038
for prediction vectors at time t, by multiplying the output of the primary capsule layer by a weight matrix Wij(t) obtaining that i is the number of 8-dimensional vectors output by the primary capsule layer; a isij(t) is the coupling coefficient at time t; bij(t) iterative update parameters of the dynamic routing algorithm at the time t, wherein the initial value is 0; r is the iteration number of the dynamic routing algorithm;
calculating the vector vjAnd (t) comparing the lengths of all vectors at the current time t, taking the state of the motor corresponding to the index value of the longest vector as the state of the planet wheel or the sun wheel corresponding to the current data, and outputting the gear state.
10. The method for diagnosing the fault of the planetary gearbox based on the convolution capsule network as claimed in claim 1, wherein the method comprises the following steps: according to the output v of the digital capsule layerj(t) building a loss function for optimizing the fault diagnosis model, wherein the calculation expression of the loss function is as follows:
Lk(t)＝Tk(t)max(0，m+-||vk(t)||)2+λ(1-Tk(t))max(0，||vk(t)||-m-)2
in the formula: subscript k is a fault category k; l isk(t) interval loss for class k; t isk(t) is a classification indication function (1 is taken for correct classification, and 0 is taken for incorrect classification otherwise); v. ofk(t) represents the probability of identifying as a failure category k; m is the upper bound, 0.9 is selected; m is the lower bound, 0.1 is taken; for the proportionality coefficient, take 0.5. The total loss is the sum of all losses;
calculating the loss function value L at time tk(t) loss function value L at time t-1k(t-1) comparing if Lk(t) is less than Lk(t-1), directly optimizing the weights and the offsets of the convolutional neural network and the capsule network by using a gradient descent method; otherwise, optimizing the weights and the offsets of the convolutional neural network and the capsule network by using an Adam Optimizer;
judging whether the current iteration time T reaches the maximum iteration time T, if so, stopping iteration, taking the updated weight and offset as parameters of the convolution capsule network, and then outputting the final state; otherwise, adding 1 to the current iteration times t, returning to the convolution layer 1, and continuously selecting the next Batch for training;
and dividing the collected vibration data of the planetary gearbox into a training set and a testing set according to a proportion, and inputting the training set and the testing set into a model so as to output the probability distribution of the health state of the planetary gearbox.
CN202110818877.1A 2021-07-20 2021-07-20 Planetary gearbox fault diagnosis method based on convolution capsule network Pending CN113591638A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110818877.1A CN113591638A (en) 2021-07-20 2021-07-20 Planetary gearbox fault diagnosis method based on convolution capsule network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110818877.1A CN113591638A (en) 2021-07-20 2021-07-20 Planetary gearbox fault diagnosis method based on convolution capsule network

Publications (1)

Publication Number Publication Date
CN113591638A true CN113591638A (en) 2021-11-02

Family

ID=78248351

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110818877.1A Pending CN113591638A (en) 2021-07-20 2021-07-20 Planetary gearbox fault diagnosis method based on convolution capsule network

Country Status (1)

Country Link
CN (1) CN113591638A (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114227382A (en) * 2022-01-18 2022-03-25 湖北汽车工业学院 Cutter damage monitoring system and method based on novel capsule network
CN114548153A (en) * 2022-01-21 2022-05-27 电子科技大学 Planetary gearbox fault diagnosis method based on residual error-capsule network
CN114757239A (en) * 2022-06-15 2022-07-15 浙江大学 Fan fault migratable diagnosis method based on data enhancement and capsule neural network

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114227382A (en) * 2022-01-18 2022-03-25 湖北汽车工业学院 Cutter damage monitoring system and method based on novel capsule network
CN114548153A (en) * 2022-01-21 2022-05-27 电子科技大学 Planetary gearbox fault diagnosis method based on residual error-capsule network
CN114757239A (en) * 2022-06-15 2022-07-15 浙江大学 Fan fault migratable diagnosis method based on data enhancement and capsule neural network
CN114757239B (en) * 2022-06-15 2022-08-30 浙江大学 Fan fault migratable diagnosis method based on data enhancement and capsule neural network

Similar Documents

Publication Publication Date Title
CN113591638A (en) Planetary gearbox fault diagnosis method based on convolution capsule network
US20210270244A1 (en) Method and system for fault diagnosis of gearbox of wind turbine generator
CN113255078A (en) Bearing fault detection method and device under unbalanced sample condition
CN111810124A (en) Oil pumping well fault diagnosis method based on characteristic re-calibration residual convolution neural network model
CN112147432A (en) BiLSTM module based on attention mechanism, transformer state diagnosis method and system
CN113505655A (en) Bearing fault intelligent diagnosis method for digital twin system
CN111413091A (en) Gear box fault diagnosis method under strong noise interference based on data driving
CN115508073B (en) Prototype adaptive mechanical equipment fault diagnosis method based on multi-scale attention
CN116358871B (en) Rolling bearing weak signal composite fault diagnosis method based on graph rolling network
CN112257528A (en) Wind power gear box fault diagnosis method based on wavelet transformation and dense connection expansion convolutional neural network
CN112686372A (en) Product performance prediction method based on depth residual GRU neural network
CN112305388B (en) On-line monitoring and diagnosing method for insulation partial discharge faults of generator stator winding
CN114091504A (en) Rotary machine small sample fault diagnosis method based on generation countermeasure network
CN115859142A (en) Small sample rolling bearing fault diagnosis method based on convolution transformer generation countermeasure network
CN115099278A (en) Equipment state diagnosis method based on wavelet packet time-frequency diagram and ResNet migration network
CN115290326A (en) Rolling bearing fault intelligent diagnosis method
CN116026569A (en) Mechanical equipment unsupervised migration intelligent fault diagnosis method for releasing source data
CN115221958A (en) Mechanical equipment selective migration fault diagnosis method based on composite weight
CN115204272A (en) Industrial system fault diagnosis method and equipment based on multi-sampling rate data
CN113758709A (en) Rolling bearing fault diagnosis method and system combining edge calculation and deep learning
CN116644273A (en) Fault diagnosis method and system based on interpretability multiplication convolution network
CN113591625B (en) Gearbox fault diagnosis method based on unbalanced sample importance weighting
CN113409213B (en) Method and system for enhancing noise reduction of time-frequency diagram of fault signal of plunger pump
CN115375968A (en) Fault diagnosis method for planetary gearbox
CN112414446B (en) Data-driven transmission sensor fault diagnosis method

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