CN114065636A - Marine winch brake mechanism fault diagnosis method based on data driving - Google Patents

Marine winch brake mechanism fault diagnosis method based on data driving Download PDF

Info

Publication number
CN114065636A
CN114065636A CN202111393045.6A CN202111393045A CN114065636A CN 114065636 A CN114065636 A CN 114065636A CN 202111393045 A CN202111393045 A CN 202111393045A CN 114065636 A CN114065636 A CN 114065636A
Authority
CN
China
Prior art keywords
data
dae
fault diagnosis
neural network
sampling
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
CN202111393045.6A
Other languages
Chinese (zh)
Inventor
张辉
曹佑忍
吴琦
李俊
胡中泰
张胜文
方喜峰
朱鹏程
程德俊
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.)
Jiangsu University of Science and Technology
Original Assignee
Jiangsu University of Science and 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 Jiangsu University of Science and Technology filed Critical Jiangsu University of Science and Technology
Priority to CN202111393045.6A priority Critical patent/CN114065636A/en
Publication of CN114065636A publication Critical patent/CN114065636A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/048—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
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2111/00—Details relating to CAD techniques
    • G06F2111/10—Numerical modelling

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Data Mining & Analysis (AREA)
  • Artificial Intelligence (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Software Systems (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Mathematical Physics (AREA)
  • Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Computing Systems (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Probability & Statistics with Applications (AREA)
  • Medical Informatics (AREA)
  • Computer Hardware Design (AREA)
  • Geometry (AREA)
  • Testing And Monitoring For Control Systems (AREA)

Abstract

The invention discloses a data-driven fault diagnosis method for a marine winch brake mechanism, which comprises the steps of selecting a pressure sensor and installing the pressure sensor on the winch brake mechanism; collecting pressure signal data borne by a winch brake pin shaft in the operation process and processing the data; and constructing a data driving algorithm model to obtain a fault diagnosis model, and monitoring the state of the winch brake mechanism by using the fault diagnosis model and realizing fault classification. According to the method, the fault diagnosis of the ship winch brake mechanism is realized through a data processing technology, the problems of real-time performance and accuracy of the traditional fault diagnosis are solved, and the running safety of the ship winch brake mechanism is improved.

Description

Marine winch brake mechanism fault diagnosis method based on data driving
Technical Field
The invention relates to a fault diagnosis method for a brake mechanism, in particular to a fault diagnosis method for a marine winch brake mechanism based on data driving.
Background
With the use of strander on ships, more and more stranders are currently used for marine installations, such as flood dragon, sea dragon intermediate strander. The working environment of most ships is comparatively abominable, and vibration and impact etc. extremely easily cause on boats and ships winch brake mechanism because of not hard up influence equipment's normal operating, in case winch brake mechanism breaks down in the operation process, can cause serious influence. Therefore, the braking force of the winch braking mechanism in the using process needs to be monitored in real time, fault troubleshooting of the winch braking mechanism by an operator is facilitated, and normal operation of the winch brake is further ensured.
At present, most ships do not have the function of effectively monitoring and diagnosing the winch brake mechanism in real time. CN212721882U A real-time monitoring system for brake force of an anchor and mooring machine, which is based on the aspect of the structure of the anchor and mooring machine, analyzes the parts of the anchor and mooring machine which may have faults; but the state monitoring of the brake force of the anchor and mooring equipment from the data driving angle is more real-time and efficient. CN109583092A is a fault diagnosis method for an intelligent mechanical system by multi-level multi-mode feature extraction, which does not adopt a fault diagnosis model correction technology on a fault diagnosis model, and an initial fault diagnosis model brings errors to a diagnosis result because of model inaccuracy, thereby causing severe damage to a winch brake mechanism. In the existing fault diagnosis method, a shallow fault diagnosis model cannot extract deeper network characteristics; the traditional fault feature extraction only considers the features on the data space, and does not effectively extract the time series features.
Therefore, it is desired to solve the above problems.
Disclosure of Invention
The purpose of the invention is as follows: the invention aims to provide a data-driven fault diagnosis method for a marine winch brake mechanism.
The technical scheme is as follows: in order to achieve the aim, the invention discloses a data-driven fault diagnosis method for a marine winch brake mechanism, which comprises the following steps of:
(1) selecting a pressure sensor, and installing the pressure sensor on a winch brake mechanism;
(2) collecting pressure signal data borne by the winch brake pin shaft in the operation process and processing the data;
(3) and constructing a data driving algorithm model to obtain a fault diagnosis model, and monitoring the state of the winch brake mechanism by using the fault diagnosis model and realizing fault classification.
Wherein, the pressure sensor in the step (1) is a strain gauge pressure sensor; because the brake force monitoring load shaft is connected with the brake pin shaft through the brake screw rod, when the winch brake mechanism works, the circumferential force acting on the upper arc-shaped steel belt is transmitted to the brake force monitoring load shaft, the strain gauge pressure sensors are installed in the horizontal direction at 3 times and the vertical direction at 6 times on the surface of the brake force monitoring load shaft, the installation positions of the strain gauge pressure sensors are relatively vertical, and the state monitoring of the pressure borne by the winch brake pin shaft in the operation process is realized.
Preferably, the specific method for processing data in step (2) includes the following steps:
(2.1) setting sampling parameters, wherein the sampling parameters comprise sampling frequency f1Sampling frequency f2Sampling frequency f3The sampling time duration t is 10s and the sampling interval Δ t is 1Sampling pressure signals borne by the winch brake pin shaft in the whole life cycle for 0ms, collecting operation data of different stages in the operation process of the winch brake pin shaft, setting a pressure sampling threshold value as Y, and stopping data sampling when a sampling pressure value reaches Y, wherein the operation data in the whole life cycle comprises normal data, slight fault data, moderate fault data and severe fault data; otherwise, continuing sampling;
(2.2) outputting and storing the sampling result in the step (2.1) in a numerical form, connecting a strain gauge pressure sensor with a PLC (programmable logic controller), and transmitting an acting force to a brake force monitoring load shaft during the operation process of a brake pin shaft to cause the brake pin shaft to deform; the brake pin shaft deformation signal acquired by the pressure sensor is converted into a voltage signal through a built-in processing program in the PLC; the output voltage signal is converted through an A/D converter, the converted voltage signal amplifies and outputs a pressure signal value according to a bridge circuit, the numerical value output by each sampling is stored in a CSV format, three different sampling frequencies are respectively sampled, the direction pressure signal is placed in a first column of a CSV file at time 3, the direction pressure signal is placed in a second column of the CSV file at time 6 and is stored as 1.CSV, and the data storage is finished by the analogy;
(2.3) carrying out nonlinear dimensionality reduction on the output data in the step (2.2) by using a weighted Kernel Principal Component Analysis (KPCA) to obtain dimensionality-reduced data, wherein the weighted kernel principal component analysis is improved by using a kernel function idea on the basis of principal component analysis, and a kernel function adopts a Radial Basis Function (RBF); and mapping the data to a high-dimensional space by using a kernel principal component analysis method, and then performing nonlinear dimensionality reduction on the mapped data.
Furthermore, the method for setting the sampling frequency in step (2.1) includes the following steps:
(2.1.1) the standard frequency range of the winch brake mechanism is 0-500 HZ, when the winch brake mechanism initially operates, the pressure sensor acquires signals which are normal signals, the signals are represented by a time domain signal diagram, and the signals show a continuous trend;
(2.1.2) when the sampling frequency is increased from 0 to 500Hz, the acquisition signal corresponding to each sampling frequency is represented by a time domain signal, and when the sampling frequency is 230Hz, the sampling signal undergoes first gradual jump; when the sampling frequency is 350Hz, the sampling signal has second gradual change jump; when the sampling frequency is 450Hz, the sampling signal has a third gradual change jump;
(2.1.3) obtaining three different sampling frequencies through the step (2.1.2), wherein the sampling frequency f _1 is 230Hz, the sampling frequency f _2 is 350Hz, and the sampling frequency f _3 is 450 Hz.
Further, the specific method for performing nonlinear dimensionality reduction on the data in the step (2.3) comprises the following steps:
(2.3.1) principal component analysis, namely converting the raw data acquired in the step (2.2) into a standard matrix, calculating the covariance of the standard matrix to obtain a covariance matrix, further obtaining the eigenvalue, the eigenvector and the contribution rate of the covariance matrix, selecting a larger contribution rate, and determining the number of principal components according to the contribution rate;
(2.3.2) weighting the kernel function, and setting the standard threshold of the kernel function as g0If the function value g is greater than g0Indicating that the original data is more dispersed, let G ═ epsilon0G, so that the data tend to aggregate from dispersion, where G is the weighted function, ε0A weight greater than a standard threshold; if g is less than g0Indicating that the original data is relatively aggregated, let G equal to epsilon1g, so that the data tends to diverge from the aggregation,. epsilon1Is a weight less than a standard threshold; and setting weights to promote the original data to tend to be even and even from divergence and aggregation, so as to obtain high-quality dimension-reduced data.
Moreover, the specific method for constructing the data-driven algorithm model in the step (3) comprises the following steps:
(3.1) according to the operation signal of the typical winch brake pin shaft collected in the step (2), extracting the characteristics of the input signal by using a DAE-GRU neural network combining depth self-coding and a circulating gate, and establishing an initialization fault diagnosis model according to the network structure parameters;
(3.2) testing the fault diagnosis model obtained in the step (3.1) by using a test set, comparing a classification result obtained by a Softmax classifier with the one-hot coded label, judging whether the diagnosis result is consistent, and if the diagnosis result is inconsistent, correcting the fault model and executing the step (3.3); if the diagnosis results are consistent, outputting the diagnosis results to obtain a DAE-GRU neural network-based winch brake mechanism fault diagnosis model;
and (3.3) correcting the initialized fault diagnosis model, improving related parameters of the DAE-GRU neural network structure, wherein the related parameters comprise compression weight and reconstruction error, and the super parameters in the hidden layer comprise an activation function, a classification function, iteration times, a training step length, a learning rate and an error threshold value, so that the initialized fault diagnosis model is corrected, and finally, the fault diagnosis model of the winch brake mechanism is established and fault classification is realized.
Preferably, the specific method for establishing the initialization fault diagnosis model in the step (3.1) includes the following steps:
(3.1.1) further dividing the data processed in the step (2) into a training set and a test set, wherein 70% of the data are used as the training set, 30% of the data are used as the test set, the training set and the test set both comprise normal state monitoring data and fault state monitoring data of a winch brake pin shaft, the training set is used for determining a fault diagnosis model and obtaining model parameters, and the test set is used for verifying whether the fault diagnosis model meets requirements; performing label processing on the training set and the test set by adopting one-hot coding, and representing the fault state of the running data by using 0 and 1, namely encoding by using N states of an N-bit register state device, wherein 1000 is set to be in a normal state, 0100 is in a slight fault state, 0010 is in a moderate fault state, and 0001 is in a severe fault state;
(3.1.2) establishing a DAE-GRU neural network structure, wherein the DAE neural network is encoded and decoded by an input layer, a hidden layer and an output layer, and a full-connection layer encoding network in the hidden layer of the DAE neural network is replaced by a GRU network to form a DAE-GRU encoding network; selecting and determining data dimensionality by adopting a sliding window in the coding network, wherein the size of the data dimensionality of the sliding window is self-adaptively changed according to the size of the data dimensionality required by the decoding network; in the DAE-GRU coding network structure, a decoding network is formed by adding a full connection layer behind a hidden layer; a Dropout layer is added behind the decoding network to prevent the fault diagnosis model from generating an overfitting phenomenon; connecting the hidden layer nodes in the DAE-GRU neural network structure by using a Softmax classifier to classify faults;
(3.1.3) establishing a DAE-GRU neural network model, and setting the number M of hidden layers in the DAE-GRU neural network structure in the step (3.1.2) to be 3 and the number N of hidden layer neurons to be 70; setting the size Z of a sliding window to be 1000; setting an output hidden layer and a current hidden layer of a DAE-GRU neural network structure as 3 layers and 2 layers, wherein P is the other layer, and Q is the other layer;
(3.1.4) taking the training set with the label as the input of the DAE-GRU neural network model, and training in the DAE-GRU neural network structure in the step (3.1.2); training set data provided with labels as input of the DAE-GRU neural network structure through the DAE-GRU neural network model in the step (3.1.3), and outputting labels corresponding to the training set data as output of the DAE-GRU neural network structure after training; setting the iteration times as R to be 100 times, finishing training if the iteration times is more than R, and outputting a fault diagnosis result; otherwise, returning to continue training by using the DAE-GRU neural network model until the iteration requirement is met;
the specific method for training by using the DAE-GRU neural network model comprises the following steps:
inputting the training set data with the label into a DAE-GRU neural network model, selecting the data dimension by adopting a sliding window in a depth self-encoder coding network as the input of the depth self-encoding network, and training the data dimension to be the input of a circulating gate unit; the formula is as follows:
H＝θ(λ1Z+b1) (1)
I＝θ(λ2H+b2) (2)
wherein λ is1、b1、λ2、b2Weight parameters and bias parameter matrixes of an encoder and a decoder respectively; theta is the neuron activation function, Z is the sliding window data size,h is hidden layer information, I is DAE output layer information;
the circulating gate network calculates the input information of the deep self-coding network, firstly, local information is extracted from long-term information, and the formula is as follows:
ht-1′＝ht-1*r (3)
wherein h ist-1' is local information, ht-1Is long time information, r is reset gate;
the current required information is obtained through the local information, and the formula is as follows:
Figure BDA0003369025750000051
where h' is the current information, γ is the weight matrix, xtIs external input information, tanh is used to compress the weight matrix;
the required output signal is obtained through the steps, and the formula is as follows:
ht＝(1-z)*ht-1+z*h′ (5)
wherein h istIs new long-term information, z is the update gate in the cycle gate;
(3.1.5) classifying the fault by using a DAE-GRU neural network; data marked by the one-hot codes are adopted in the step (3.1.1) of classified output by a Softmax classifier in the DAE-GRU neural network structure, then the output data is converted into an output value by utilizing the reverse one-hot codes, an initial fault diagnosis model of the winch brake mechanism based on the DAE-GRU neural network winch brake mechanism is obtained, and fault classification is realized; wherein an output value of 0 represents a normal state, an output value of 1 represents a slight fault state, an output value of 2 represents a moderate fault state, and an output value of 3 represents a severe fault state.
Furthermore, the specific method for correcting the fault diagnosis model in the step (3.3) includes the following steps:
(3.3.1) retraining the DAE-GRU neural network-based fault diagnosis model by adjusting the structural parameters of the DAE-GRU neural network, and adjusting the weight between adjacent neurons through a bias function;
the DAE-GRU neural network structure parameters are adjusted, the weights between the input layer and the hidden layer and between the hidden layer and the adjacent neuron nodes in the output layer are updated, and the formula is as follows:
Figure BDA0003369025750000053
wherein, Wik' is the updated weight, WikIs the updated weight, η, of the next layer of nodesiIs a learning rate, sets an initial learning rate eta0＝0.001，δikIs a learning factor, X, between connected layersiIs the value of node i;
(3.3.2) by updating the weights between adjacent neuron nodes, adopting a cross entropy loss function to judge the error between the output value and the input value, namely judging by using a labeled data set, wherein the formula is as follows:
Figure BDA0003369025750000052
wherein, ynkIs the neural network structure output, tnkThe label representing the correct solution, namely the value of the kth element of the nth data, K is the total data set, E is the error offset value, log is the logarithm, and N is the number of input data;
(3.3.3) setting an error threshold E0(ii) a If the cross entropy loss function calculation error E of the step (3.3.2) is larger than the error threshold value E0Updating the learning rate by using a formula (8), returning to the step (3.3.1) to calculate the error by using back propagation, and otherwise, finishing the correction of the fault diagnosis model of the winch brake mechanism based on the DAE-GRU neural network;
ηi+1＝0.95epoch_num*ηi (8)
wherein eta isi+1The updated learning rate is calculated by substituting it into equation (6), and epoch _ num is the number of iterations.
Has the advantages that: compared with the prior art, the invention has the following remarkable advantages:
(1) according to the method, the fault diagnosis of the ship winch brake mechanism is realized through a data processing technology, the problems of real-time performance and accuracy of the traditional fault diagnosis are solved, and the running safety of the ship winch brake mechanism is improved;
(2) the method adopts a kernel principal component analysis method to carry out high-dimensional mapping on data, realizes that the data can be linearly divided from nonlinearity on a high-dimensional plane through a Gaussian radial kernel function idea, and mines the nonlinear information of deeper layers in the data;
(3) the GRU network replaces a full-connection layer coding network in a DAE hidden layer, the full-connection layer is added behind the coding network to form a decoding network, a Dropout layer is added behind the full-connection layer to prevent a fault model from being over-fitted, and a Softmax classifier is connected with nodes of the hidden layer to realize fault classification and realize the conversion from a shallow network to a deeper network structure; the method not only solves the defect of time sequence problem in the traditional feature extraction, but also overcomes the defects of difficult neural network parameter training, slow training speed, prevention of layout convergence and gradient disappearance and the like in a DAE network hidden layer, enhances the feature extraction capability and strengthens the network training;
(4) the invention optimizes the hyper-parameters in the fault diagnosis model through the DAE network, adjusts the weight between adjacent neurons by using a bias function, corrects the fault model by adopting the DAE-GRU network, classifies different fault types, optimizes the fault diagnosis model of the winch brake mechanism, improves the diagnosis precision of the model and ensures the high efficiency and the robustness of the model.
Drawings
FIG. 1 is a schematic diagram of a fault diagnosis model of the present invention;
fig. 2 is a schematic flow chart of fault diagnosis in the present invention.
Detailed Description
The technical scheme of the invention is further explained by combining the attached drawings.
As shown in fig. 1 and 2, the method for diagnosing the fault of the marine winch brake mechanism based on data driving of the invention comprises the following steps:
(1) selecting a pressure sensor, and installing the pressure sensor on a winch brake mechanism; wherein the pressure sensor is a strain gauge pressure sensor; the brake force monitoring load shaft and the brake pin shaft are connected through the brake screw, when the winch brake mechanism works, the circumferential force acting on the upper arc-shaped steel belt is transmitted to the brake force monitoring load shaft, the strain gauge pressure sensors are installed in the horizontal direction at 3 and the vertical direction at 6 times on the surface of the brake force monitoring load shaft, and the installation positions of the strain gauge pressure sensors are relatively vertical, so that the state monitoring of the pressure borne by the winch brake pin shaft in the operation process is realized;
the invention carries out structural analysis on the brake mechanism of the pair twister: the winch brake mechanism mainly comprises a brake pin shaft, an upper arc-shaped steel belt assembly, a lower arc-shaped steel belt assembly, a brake screw, a brake force monitoring load shaft and a brake base; and judging the position of the winch where the brake is easy to fail according to the running condition among the parts of the winch brake mechanism, so that the sensor is convenient to install. In the operation process of the brake mechanism, pressure load is applied to the brake pin shaft along with the upper and lower arc-shaped steel belt assemblies, and acting force is transmitted to the brake force monitoring load shaft through the brake pin shaft. When the winch brake mechanism works for a long time, the brake pin shaft fails due to factors such as abrasion, impact and vibration;
(2) collecting pressure signal data borne by the winch brake pin shaft in the operation process and processing the data,
the method comprises the following steps:
(2.1) setting sampling parameters, wherein the sampling parameters comprise sampling frequency f1Sampling frequency f2Sampling frequency f3Sampling pressure signals borne by a winch brake pin shaft in a full life cycle, acquiring running data of different stages in the running process of the winch brake pin shaft, setting a pressure sampling threshold value as Y, and stopping data sampling when a sampling pressure value reaches Y, wherein the sampling time length t is 10s and the sampling interval delta t is 10 ms; otherwise, continuing sampling;
the method for setting the sampling frequency comprises the following steps:
(2.1.1) the standard frequency range of the winch brake mechanism is 0-500 HZ, when the winch brake mechanism initially operates, the pressure sensor acquires signals which are normal signals, the signals are represented by a time domain signal diagram, and the signals show a continuous trend;
(2.1.2) when the sampling frequency is increased from 0 to 500Hz, the acquisition signal corresponding to each sampling frequency is represented by a time domain signal, and when the sampling frequency is 230Hz, the sampling signal undergoes first gradual jump; when the sampling frequency is 350Hz, the sampling signal has second gradual change jump; when the sampling frequency is 450Hz, the sampling signal has a third gradual change jump;
(2.1.3) obtaining three different sampling frequencies through the step (2.1.2), wherein the three different sampling frequencies are respectively the sampling frequency f1230Hz, sampling frequency f2350Hz, sampling frequency f3＝450Hz；
(2.2) outputting and storing the sampling result in the step (2.1) in a numerical form, connecting a strain gauge pressure sensor with a PLC (programmable logic controller), and transmitting an acting force to a brake force monitoring load shaft during the operation process of a brake pin shaft to cause the brake pin shaft to deform; the brake pin shaft deformation signal acquired by the pressure sensor is converted into a voltage signal through a built-in processing program in the PLC; the output voltage signal is converted through an A/D converter, the converted voltage signal amplifies and outputs a pressure signal value according to a bridge circuit, the numerical value output by each sampling is stored in a CSV format, three different sampling frequencies are respectively sampled, the direction pressure signal is placed in a first column of a CSV file at time 3, the direction pressure signal is placed in a second column of the CSV file at time 6 and is stored as 1.CSV, and the data storage is finished by the analogy;
(2.3) carrying out nonlinear dimensionality reduction on the output data in the step (2.2) by using a weighted Kernel Principal Component Analysis (KPCA) to obtain dimensionality-reduced data, wherein the weighted kernel principal component analysis is improved by using a kernel function idea on the basis of principal component analysis, and a kernel function adopts a Radial Basis Function (RBF); the collected data is mostly in a dispersed state, and the data cannot be effectively extracted. Therefore, mapping the data to a high-dimensional space by using a kernel principal component analysis method, then performing nonlinear dimensionality reduction on the mapped data, and inputting the dimensionality reduced data into a winch brake mechanism fault diagnosis model;
the specific method for carrying out nonlinear dimensionality reduction on data comprises the following steps:
(2.3.1) principal component analysis, namely converting the raw data acquired in the step (2.2) into a standard matrix, calculating the covariance of the standard matrix to obtain a covariance matrix, further obtaining the eigenvalue, the eigenvector and the contribution rate of the covariance matrix, selecting a larger contribution rate, and determining the number of principal components according to the contribution rate;
(2.3.2) weighting the kernel function, and setting the standard threshold of the kernel function as g0If the function value g is greater than g0Indicating that the original data is more dispersed, let G ═ epsilon0G, so that the data tend to aggregate from dispersion, where G is the weighted function, ε0A weight greater than a standard threshold; if g is less than g0Indicating that the original data is relatively aggregated, let G equal to epsilon1g, so that the data tends to diverge from the aggregation,. epsilon1Is a weight less than a standard threshold; setting weight to make the original data tend to be uniform from divergence and aggregation, so as to obtain high-quality dimension reduction data;
(3) constructing a data driving algorithm model to obtain a fault diagnosis model, and monitoring the state of a winch brake mechanism by using the fault diagnosis model and realizing fault classification;
the method for constructing the data-driven algorithm model comprises the following specific steps:
(3.1) according to the operation signal of the typical winch brake pin shaft collected in the step (2), extracting the characteristics of the input signal by using a DAE-GRU neural network combining depth self-coding and a circulating gate, and establishing an initialization fault diagnosis model according to the network structure parameters; the DAE-GRU neural network is mainly used for integrating a time sequence during feature extraction, so that the problem of data gradient disappearance in the training process is solved; the output data of the hidden layer of the deep self-encoder is interacted through an update gate and a reset gate in the GRU network, so that a unique network structure of the DAE-GRU is formed;
the establishment of the initial fault diagnosis model comprises the following specific steps:
(3.1.1) further dividing the data processed in the step (2) into a training set and a test set, wherein 70% of the data are used as the training set, 30% of the data are used as the test set, the training set and the test set both comprise normal state monitoring data and fault state monitoring data of a winch brake pin shaft, the training set is used for determining a fault diagnosis model and obtaining model parameters, and the test set is used for verifying whether the fault diagnosis model meets requirements; performing label processing on the training set and the test set by adopting one-hot coding, and representing the fault state of the running data by using 0 and 1, namely encoding by using N states of an N-bit register state device, wherein 1000 is set to be in a normal state, 0100 is in a slight fault state, 0010 is in a moderate fault state, and 0001 is in a severe fault state;
(3.1.2) establishing a DAE-GRU neural network structure, wherein the DAE neural network is encoded and decoded by an input layer, a hidden layer and an output layer, and a full-connection layer encoding network in the hidden layer of the DAE neural network is replaced by a GRU network to form a DAE-GRU encoding network; selecting and determining data dimensionality by adopting a sliding window in the coding network, wherein the size of the data dimensionality of the sliding window is self-adaptively changed according to the size of the data dimensionality required by the decoding network; in the DAE-GRU coding network structure, a decoding network is formed by adding a full connection layer behind a hidden layer; a Dropout layer is added behind the decoding network to prevent the fault diagnosis model from generating an overfitting phenomenon; connecting the hidden layer nodes in the DAE-GRU neural network structure by using a Softmax classifier to classify faults;
(3.1.3) establishing a DAE-GRU neural network model, and setting the number M of hidden layers in the DAE-GRU neural network structure in the step (3.1.2) to be 3 and the number N of hidden layer neurons to be 70; setting the size Z of a sliding window to be 1000; setting an output hidden layer and a current hidden layer of a DAE-GRU neural network structure as 3 layers and 2 layers, wherein P is the other layer, and Q is the other layer;
(3.1.4) taking the training set with the label as the input of the DAE-GRU neural network model, and training in the DAE-GRU neural network structure in the step (3.1.2); training set data provided with labels as input of the DAE-GRU neural network structure through the DAE-GRU neural network model in the step (3.1.3), and outputting labels corresponding to the training set data as output of the DAE-GRU neural network structure after training; setting the iteration times as R to be 100 times, finishing training if the iteration times is more than R, and outputting a fault diagnosis result; otherwise, returning to continue training by using the DAE-GRU neural network model until the iteration requirement is met;
the specific method for training by using the DAE-GRU neural network model comprises the following steps:
inputting the training set data with the label into a DAE-GRU neural network model, selecting the data dimension by adopting a sliding window in a depth self-encoder coding network as the input of the depth self-encoding network, and training the data dimension to be the input of a circulating gate unit; the formula is as follows:
H＝θ(λ1Z+b1) (1)
I＝θ(λ2H+b2) (2)
wherein λ is1、b1、λ2、b2Weight parameters and bias parameter matrixes of an encoder and a decoder respectively; theta is a neuron activation function, Z is the size of sliding window data, H is hidden layer information, and I is DAE output layer information;
the circulating gate network calculates the input information of the deep self-coding network, firstly, local information is extracted from long-term information, and the formula is as follows:
ht-1′＝ht-1*r (3)
wherein h ist-1' is local information, ht-1Is long time information, r is reset gate;
the current required information is obtained through the local information, and the formula is as follows:
Figure BDA0003369025750000101
wherein h' is the current informationγ is a weight matrix, xtIs external input information, tanh is used to compress the weight matrix; the required output signal is obtained through the steps, and the formula is as follows:
ht＝(1-z)*ht-1+z*h′ (5)
wherein h istIs new long-term information, z is the update gate in the cycle gate;
(3.1.5) classifying the fault by using a DAE-GRU neural network; data marked by the one-hot codes are adopted in the step (3.1.1) of classified output by a Softmax classifier in the DAE-GRU neural network structure, then the output data is converted into an output value by utilizing the reverse one-hot codes, an initial fault diagnosis model of the winch brake mechanism based on the DAE-GRU neural network winch brake mechanism is obtained, and fault classification is realized; wherein the output value is 0 to represent the normal state, the output value is 1 to represent the slight fault state, the output value is 2 to represent the moderate fault state, and the output value is 3 to represent the severe fault state;
(3.2) testing the fault diagnosis model obtained in the step (3.1) by using a test set, comparing a classification result obtained by a Softmax classifier with the one-hot coded label, judging whether the diagnosis result is consistent, and if the diagnosis result is inconsistent, correcting the fault model and executing the step (3.3); if the diagnosis results are consistent, outputting the diagnosis results to obtain a DAE-GRU neural network-based winch brake mechanism fault diagnosis model;
(3.3) correcting the initialized fault diagnosis model, improving related parameters of the DAE-GRU neural network structure, wherein the related parameters comprise compression weight and reconstruction error, and the super parameters in the hidden layer comprise an activation function, a classification function, iteration times, a training step length, a learning rate and an error threshold value, so that the correction of the initialized fault diagnosis model is realized, and finally, establishing a fault diagnosis model of a winch brake mechanism and realizing fault classification;
correcting the fault diagnosis model, which comprises the following specific steps:
(3.3.1) retraining the DAE-GRU neural network-based fault diagnosis model by adjusting the structural parameters of the DAE-GRU neural network, and adjusting the weight between adjacent neurons through a bias function;
the DAE-GRU neural network structure parameters are adjusted, the weights between the input layer and the hidden layer and between the hidden layer and the adjacent neuron nodes in the output layer are updated, and the formula is as follows:
Figure BDA0003369025750000111
wherein, Wik' is the updated weight, WikIs the updated weight, η, of the next layer of nodesiIs a learning rate, sets an initial learning rate eta0＝0.001，δikIs a learning factor, X, between connected layersiIs the value of node i;
(3.3.2) by updating the weights between adjacent neuron nodes, adopting a cross entropy loss function to judge the error between the output value and the input value, namely judging by using a labeled data set, wherein the formula is as follows:
Figure BDA0003369025750000112
wherein, ynkIs the neural network structure output, tnkThe label representing the correct solution, namely the value of the kth element of the nth data, K is the total data set, E is the error offset value, log is the logarithm, and N is the number of input data;
(3.3.3) setting an error threshold E0(ii) a If the cross entropy loss function calculation error E of the step (3.3.2) is larger than the error threshold value E0Updating the learning rate by using a formula (8), returning to the step (3.3.1) to calculate the error by using back propagation, and otherwise, finishing the correction of the fault diagnosis model of the winch brake mechanism based on the DAE-GRU neural network;
ηi+1＝0.95epoch_num*ηi (8)
wherein eta isi+1The updated learning rate is calculated by substituting it into equation (6), and epoch _ num is the number of iterations.

Claims (8)

1. A data-driven marine winch brake mechanism fault diagnosis method is characterized by comprising the following steps:
(1) selecting a pressure sensor, and installing the pressure sensor on a winch brake mechanism;
(2) collecting pressure signal data borne by the winch brake pin shaft in the operation process and processing the data;
(3) and constructing a data driving algorithm model to obtain a fault diagnosis model, and monitoring the state of the winch brake mechanism by using the fault diagnosis model and realizing fault classification.
2. The marine winch brake mechanism fault diagnosis method based on data driving according to claim 1, characterized in that: the pressure sensor in the step (1) selects a strain gauge pressure sensor; because the brake force monitoring load shaft is connected with the brake pin shaft through the brake screw rod, when the winch brake mechanism works, the circumferential force acting on the upper arc-shaped steel belt is transmitted to the brake force monitoring load shaft, the strain gauge pressure sensors are installed in the horizontal direction at 3 times and the vertical direction at 6 times on the surface of the brake force monitoring load shaft, the installation positions of the strain gauge pressure sensors are relatively vertical, and the state monitoring of the pressure borne by the winch brake pin shaft in the operation process is realized.
3. The marine winch brake mechanism fault diagnosis method based on data driving according to claim 2, characterized in that: the specific method for processing the data in the step (2) comprises the following steps:
(2.1) setting sampling parameters, wherein the sampling parameters comprise sampling frequency f1Sampling frequency f2Sampling frequency f3Sampling pressure signals borne by a winch brake pin shaft in a full life cycle, wherein the sampling time length t is 10s and the sampling interval delta t is 10ms, acquiring operation data of different stages in the operation process of the winch brake pin shaft, setting a pressure sampling threshold value as Y, and when sampling pressure is applied, setting a pressure sampling threshold value as Y, wherein the operation data of the winch brake pin shaft in the full life cycle comprises normal data, slight fault data, moderate fault data and severe fault dataStopping data sampling when the force value reaches Y; otherwise, continuing sampling;
(2.2) outputting and storing the sampling result in the step (2.1) in a numerical form, connecting a strain gauge pressure sensor with a PLC (programmable logic controller), and transmitting an acting force to a brake force monitoring load shaft during the operation process of a brake pin shaft to cause the brake pin shaft to deform; the brake pin shaft deformation signal acquired by the pressure sensor is converted into a voltage signal through a built-in processing program in the PLC; the output voltage signal is converted through an A/D converter, the converted voltage signal amplifies and outputs a pressure signal value according to a bridge circuit, the numerical value output by each sampling is stored in a CSV format, three different sampling frequencies are respectively sampled, the direction pressure signal is placed in a first column of a CSV file at time 3, the direction pressure signal is placed in a second column of the CSV file at time 6 and is stored as 1.CSV, and the data storage is finished by the analogy;
(2.3) carrying out nonlinear dimensionality reduction on the output data in the step (2.2) by using a weighted Kernel Principal Component Analysis (KPCA) to obtain dimensionality-reduced data, wherein the weighted kernel principal component analysis is improved by using a kernel function idea on the basis of principal component analysis, and a kernel function adopts a Radial Basis Function (RBF); and mapping the data to a high-dimensional space by using a kernel principal component analysis method, and then performing nonlinear dimensionality reduction on the mapped data.
4. The marine winch brake mechanism fault diagnosis method based on data driving according to claim 3, characterized in that: the method for setting the sampling frequency in the step (2.1) comprises the following steps:
(2.1.1) the standard frequency range of the winch brake mechanism is 0-500 HZ, when the winch brake mechanism initially operates, the pressure sensor acquires signals which are normal signals, the signals are represented by a time domain signal diagram, and the signals show a continuous trend;
(2.1.2) when the sampling frequency is increased from 0 to 500Hz, the acquisition signal corresponding to each sampling frequency is represented by a time domain signal, and when the sampling frequency is 230Hz, the sampling signal undergoes first gradual jump; when the sampling frequency is 350Hz, the sampling signal has second gradual change jump; when the sampling frequency is 450Hz, the sampling signal has a third gradual change jump;
(2.1.3) obtaining three different sampling frequencies through the step (2.1.2), wherein the sampling frequency f _1 is 230Hz, the sampling frequency f _2 is 350Hz, and the sampling frequency f _3 is 450 Hz.
5. The marine winch brake mechanism fault diagnosis method based on data driving according to claim 4, characterized in that: the specific method for performing nonlinear dimensionality reduction on the data in the step (2.3) comprises the following steps:
(2.3.1) principal component analysis, namely converting the raw data acquired in the step (2.2) into a standard matrix, calculating the covariance of the standard matrix to obtain a covariance matrix, further obtaining the eigenvalue, the eigenvector and the contribution rate of the covariance matrix, selecting a larger contribution rate, and determining the number of principal components according to the contribution rate;
(2.3.2) weighting the kernel function, and setting the standard threshold of the kernel function as g0If the function value g is greater than g0Indicating that the original data is more dispersed, let G ═ epsilon0G, so that the data tend to aggregate from dispersion, where G is the weighted function, ε0A weight greater than a standard threshold; if g is less than g0Indicating that the original data is relatively aggregated, let G equal to epsilon1g, so that the data tends to diverge from the aggregation,. epsilon1Is a weight less than a standard threshold; and setting weights to promote the original data to tend to be even and even from divergence and aggregation, so as to obtain high-quality dimension-reduced data.
6. The marine winch brake mechanism fault diagnosis method based on data driving according to claim 5, characterized in that: the specific method for constructing the data-driven algorithm model in the step (3) comprises the following steps:
(3.1) according to the operation signal of the typical winch brake pin shaft collected in the step (2), extracting the characteristics of the input signal by using a DAE-GRU neural network combining depth self-coding and a circulating gate, and establishing an initialization fault diagnosis model according to the network structure parameters;
(3.2) testing the fault diagnosis model obtained in the step (3.1) by using a test set, comparing a classification result obtained by a Softmax classifier with the one-hot coded label, judging whether the diagnosis result is consistent, and if the diagnosis result is inconsistent, correcting the fault model and executing the step (3.3); if the diagnosis results are consistent, outputting the diagnosis results to obtain a DAE-GRU neural network-based winch brake mechanism fault diagnosis model;
and (3.3) correcting the initialized fault diagnosis model, improving related parameters of the DAE-GRU neural network structure, wherein the related parameters comprise compression weight and reconstruction error, and the super parameters in the hidden layer comprise an activation function, a classification function, iteration times, a training step length, a learning rate and an error threshold value, so that the initialized fault diagnosis model is corrected, and finally, the fault diagnosis model of the winch brake mechanism is established and fault classification is realized.
7. The marine winch brake mechanism fault diagnosis method based on data driving according to claim 6, characterized in that: the specific method for establishing the initialization fault diagnosis model in the step (3.1) comprises the following steps:
(3.1.1) further dividing the data processed in the step (2) into a training set and a test set, wherein 70% of the data are used as the training set, 30% of the data are used as the test set, the training set and the test set both comprise normal state monitoring data and fault state monitoring data of a winch brake pin shaft, the training set is used for determining a fault diagnosis model and obtaining model parameters, and the test set is used for verifying whether the fault diagnosis model meets requirements; performing label processing on the training set and the test set by adopting one-hot coding, and representing the fault state of the running data by using 0 and 1, namely encoding by using N states of an N-bit register state device, wherein 1000 is set to be in a normal state, 0100 is in a slight fault state, 0010 is in a moderate fault state, and 0001 is in a severe fault state;
(3.1.2) establishing a DAE-GRU neural network structure, wherein the DAE neural network is encoded and decoded by an input layer, a hidden layer and an output layer, and a full-connection layer encoding network in the hidden layer of the DAE neural network is replaced by a GRU network to form a DAE-GRU encoding network; selecting and determining data dimensionality by adopting a sliding window in the coding network, wherein the size of the data dimensionality of the sliding window is self-adaptively changed according to the size of the data dimensionality required by the decoding network; in the DAE-GRU coding network structure, a decoding network is formed by adding a full connection layer behind a hidden layer; a Dropout layer is added behind the decoding network to prevent the fault diagnosis model from generating an overfitting phenomenon; connecting the hidden layer nodes in the DAE-GRU neural network structure by using a Softmax classifier to classify faults;
(3.1.3) establishing a DAE-GRU neural network model, and setting the number M of hidden layers in the DAE-GRU neural network structure in the step (3.1.2) to be 3 and the number N of hidden layer neurons to be 70; setting the size Z of a sliding window to be 1000; setting an output hidden layer and a current hidden layer of a DAE-GRU neural network structure as 3 layers and 2 layers, wherein P is the other layer, and Q is the other layer;
(3.1.4) taking the training set with the label as the input of the DAE-GRU neural network model, and training in the DAE-GRU neural network structure in the step (3.1.2); training set data provided with labels as input of the DAE-GRU neural network structure through the DAE-GRU neural network model in the step (3.1.3), and outputting labels corresponding to the training set data as output of the DAE-GRU neural network structure after training; setting the iteration times as R to be 100 times, finishing training if the iteration times is more than R, and outputting a fault diagnosis result; otherwise, returning to continue training by using the DAE-GRU neural network model until the iteration requirement is met;
the specific method for training by using the DAE-GRU neural network model comprises the following steps:
inputting the training set data with the label into a DAE-GRU neural network model, selecting the data dimension by adopting a sliding window in a depth self-encoder coding network as the input of the depth self-encoding network, and training the data dimension to be the input of a circulating gate unit; the formula is as follows:
H＝θ(λ1Z+b1) (1)
I＝θ(λ2H+b2) (2)
wherein λ is1、b1、λ2、b2Weight parameters and bias parameter matrixes of an encoder and a decoder respectively; theta is a neuron activation function, Z is the size of sliding window data, H is hidden layer information, and I is DAE output layer information;
the circulating gate network calculates the input information of the deep self-coding network, firstly, local information is extracted from long-term information, and the formula is as follows:
ht-1′＝ht-1*r (3)
wherein h ist-1’Is local information, ht-1Is long time information, r is reset gate;
the current required information is obtained through the local information, and the formula is as follows:
Figure FDA0003369025740000041
where h' is the current information, γ is the weight matrix, xtIs external input information, tanh is used to compress the weight matrix;
the required output signal is obtained through the steps, and the formula is as follows:
ht＝(1-z)*ht-1+z*h′ (5)
wherein h istIs new long-term information, z is the update gate in the cycle gate;
(3.1.5) classifying the faults by using a DAE-GRU neural network, outputting the labeled data by using a Softmax classifier in a DAE-GRU neural network structure in a classifying way by using a Softmax classifier in the step (3.1.1), and converting the output data into an output value by using inverse unique hot coding to obtain an initial fault diagnosis model of the winch brake mechanism based on the DAE-GRU neural network winch brake mechanism and realize fault classification; wherein an output value of 0 represents a normal state, an output value of 1 represents a slight fault state, an output value of 2 represents a moderate fault state, and an output value of 3 represents a severe fault state.
8. The marine winch brake mechanism fault diagnosis method based on data driving according to claim 7, characterized in that: the specific method for correcting the fault diagnosis model in the step (3.3) comprises the following steps:
(3.3.1) retraining the DAE-GRU neural network-based fault diagnosis model by adjusting the structural parameters of the DAE-GRU neural network, and adjusting the weight between adjacent neurons through a bias function;
the DAE-GRU neural network structure parameters are adjusted, the weights between the input layer and the hidden layer and between the hidden layer and the adjacent neuron nodes in the output layer are updated, and the formula is as follows:
Figure FDA0003369025740000051
wherein, Wik' is the updated weight, WikIs the updated weight, η, of the next layer of nodesiIs a learning rate, sets an initial learning rate eta0＝0.001，δikIs a learning factor, X, between connected layersiIs the value of node i;
(3.3.2) by updating the weights between adjacent neuron nodes, adopting a cross entropy loss function to judge the error between the output value and the input value, namely judging by using a labeled data set, wherein the formula is as follows:
Figure FDA0003369025740000052
wherein, ynkIs the neural network structure output, tnkThe label representing the correct solution, namely the value of the kth element of the nth data, K is the total data set, E is the error offset value, log is the logarithm, and N is the number of input data;
(3.3.3) setting an error threshold E0(ii) a If the cross entropy loss function calculation error E of the step (3.3.2) is larger than the error threshold value E0Updating the learning rate by using the formula (8), and returning to the step (3.3.1) by using the inverseError calculation is carried out on the propagation direction, otherwise, the correction of the fault diagnosis model of the winch brake mechanism based on the DAE-GRU neural network is completed;
ηi+1＝0.95epoch_num*ηi (8)
wherein eta isi+1The updated learning rate is calculated by substituting it into equation (6), and epoch _ num is the number of iterations.
CN202111393045.6A 2021-11-23 2021-11-23 Marine winch brake mechanism fault diagnosis method based on data driving Pending CN114065636A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111393045.6A CN114065636A (en) 2021-11-23 2021-11-23 Marine winch brake mechanism fault diagnosis method based on data driving

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111393045.6A CN114065636A (en) 2021-11-23 2021-11-23 Marine winch brake mechanism fault diagnosis method based on data driving

Publications (1)

Publication Number Publication Date
CN114065636A true CN114065636A (en) 2022-02-18

Family

ID=80279348

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111393045.6A Pending CN114065636A (en) 2021-11-23 2021-11-23 Marine winch brake mechanism fault diagnosis method based on data driving

Country Status (1)

Country Link
CN (1) CN114065636A (en)

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115526215A (en) * 2022-11-24 2022-12-27 光微半导体(吉林)有限公司 Method and system for diagnosing, maintaining, analyzing and optimizing rubber pump fault
CN115891741A (en) * 2022-09-30 2023-04-04 南京邮电大学 Remote fault early warning method and device suitable for electric vehicle charging process
CN116702030A (en) * 2023-05-31 2023-09-05 浙江大学 Blast furnace state monitoring method and device based on sensor reliability analysis
CN117056734A (en) * 2023-10-12 2023-11-14 山东能源数智云科技有限公司 Method and device for constructing equipment fault diagnosis model based on data driving
CN117151196A (en) * 2023-10-26 2023-12-01 湘江实验室 Layer-by-layer increment expected propagation-based interpretable fault diagnosis attribution method

Cited By (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115891741A (en) * 2022-09-30 2023-04-04 南京邮电大学 Remote fault early warning method and device suitable for electric vehicle charging process
CN115891741B (en) * 2022-09-30 2023-09-22 南京邮电大学 Remote fault early warning method and device suitable for electric automobile charging process
CN115526215A (en) * 2022-11-24 2022-12-27 光微半导体(吉林)有限公司 Method and system for diagnosing, maintaining, analyzing and optimizing rubber pump fault
CN116702030A (en) * 2023-05-31 2023-09-05 浙江大学 Blast furnace state monitoring method and device based on sensor reliability analysis
CN116702030B (en) * 2023-05-31 2024-01-23 浙江大学 Blast furnace state monitoring method and device based on sensor reliability analysis
CN117056734A (en) * 2023-10-12 2023-11-14 山东能源数智云科技有限公司 Method and device for constructing equipment fault diagnosis model based on data driving
CN117056734B (en) * 2023-10-12 2024-02-06 山东能源数智云科技有限公司 Method and device for constructing equipment fault diagnosis model based on data driving
CN117151196A (en) * 2023-10-26 2023-12-01 湘江实验室 Layer-by-layer increment expected propagation-based interpretable fault diagnosis attribution method
CN117151196B (en) * 2023-10-26 2024-01-30 湘江实验室 Layer-by-layer increment expected propagation-based interpretable fault diagnosis attribution method

Similar Documents

Publication Publication Date Title
CN114065636A (en) Marine winch brake mechanism fault diagnosis method based on data driving
CN113935460B (en) Intelligent diagnosis method for mechanical faults under unbalanced-like data set
CN110298455B (en) Mechanical equipment fault intelligent early warning method based on multivariate estimation prediction
CN110942101B (en) Rolling bearing residual life prediction method based on depth generation type countermeasure network
CN103824137B (en) A kind of complex mechanical equipment multi-state failure prediction method
CN111026058B (en) Semi-supervised deep learning fault diagnosis method based on Watherstein distance and self-encoder
CN112834211A (en) Fault early warning method for transmission system of wind turbine generator
CN111931625B (en) Product key part residual life prediction method based on asymmetric loss neural network
CN112785092B (en) Switch residual life prediction method based on self-adaptive deep feature extraction
CN113642414A (en) Method for predicting residual service life of rolling bearing based on Transformer model
CN113128338A (en) Intelligent diagnosis method for printing machine roller fault under small sample
CN115563563A (en) Fault diagnosis method and device based on transformer oil chromatographic analysis
CN116578940A (en) Bearing fault new type identification and diagnosis method based on mixed depth self-coding
CN110779988A (en) Bolt life prediction method based on deep learning
CN114722952A (en) Bearing fault detection and classification integrated method based on representation learning
CN116677569A (en) Fan tower stress monitoring and early warning method and system
CN115712817A (en) Fault diagnosis method of industrial motor pump based on convolutional neural network
CN116007940A (en) Rolling bearing residual life prediction method considering matching failure threshold
CN114861349A (en) Rolling bearing RUL prediction method based on model migration and wiener process
CN112986393B (en) Bridge inhaul cable damage detection method and system
CN117664558A (en) Generator gear box abnormality detection method, device, equipment and storage medium
CN116821730B (en) Fan fault detection method, control device and storage medium
CN117763335A (en) Equipment fault prediction method based on large model
CN115901259A (en) Rolling bearing weak fault diagnosis method based on two-dimensional image and CNN
Yang et al. Imbalanced bearing fault diagnosis based on adaptive cost-sensitive neural network

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