[go: up one dir, main page]
More Web Proxy on the site http://driver.im/

CN109782603A - Detection method and monitoring system of rotating machinery coupling fault - Google Patents

Detection method and monitoring system of rotating machinery coupling fault Download PDF

Info

Publication number
CN109782603A
CN109782603A CN201910109012.0A CN201910109012A CN109782603A CN 109782603 A CN109782603 A CN 109782603A CN 201910109012 A CN201910109012 A CN 201910109012A CN 109782603 A CN109782603 A CN 109782603A
Authority
CN
China
Prior art keywords
fault
layer
training
convolutional neural
neural network
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
CN201910109012.0A
Other languages
Chinese (zh)
Inventor
盛立
牟大伟
高明
周东华
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.)
China University of Petroleum East China
Shandong University of Science and Technology
Original Assignee
China University of Petroleum East China
Shandong 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 China University of Petroleum East China, Shandong University of Science and Technology filed Critical China University of Petroleum East China
Priority to CN201910109012.0A priority Critical patent/CN109782603A/en
Publication of CN109782603A publication Critical patent/CN109782603A/en
Pending legal-status Critical Current

Links

Landscapes

  • Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)

Abstract

本发明涉及一种旋转机械耦合故障的检测方法及监测系统，所述检测方法的步骤为：利用旋转机械在正常工况和故障工况下采集的振动信号数据作为训练数据集，建立深度卷积神经网络模型，将振动数据直接作为输入，引入自归一化策略将神经元激活值标准化，对深度卷积神经网络模型参数进行训练，并保存训练后的参数数据，采集实时工况下的数据作为测试数据，通过深度卷积神经网络模型实现故障检测。本发明无需工业过程精确的数学模型，便于实际应用；同时实现了故障检测和故障工况的类别区分，能够有效监测出产生机械损坏的具体部件，检测准确率高。

The invention relates to a detection method and a monitoring system for a coupling fault of a rotating machinery. The steps of the detection method are: using vibration signal data collected by the rotating machinery under normal working conditions and faulty working conditions as a training data set to establish a deep convolution The neural network model takes the vibration data directly as input, introduces a self-normalization strategy to standardize the activation values of neurons, trains the parameters of the deep convolutional neural network model, saves the trained parameter data, and collects data under real-time conditions. As test data, fault detection is achieved through a deep convolutional neural network model. The invention does not need an accurate mathematical model of the industrial process, which is convenient for practical application; meanwhile, it realizes the classification of fault detection and fault working conditions, can effectively monitor the specific components that cause mechanical damage, and has high detection accuracy.

Description

The detection method and monitoring system of rotating machinery coupling fault
Technical field
The invention belongs to industrial machinery monitoring and fault diagnosis technology fields, are related to a kind of inspection of rotating machinery coupling fault Survey method and monitoring system.
Background technique
In modern industrial production, it is examined by the vibration signal of collection machinery component to carry out failure to complex electromechanical systems Disconnected is one of the diagnostic method being most widely used in current rotary machinery fault diagnosis.Conventional vibration diagnostic method and theory are Tend to be mature, core is generally divided into two large divisions: first is that the feature extraction of vibration signal, second is that pattern classification.
Rotating machinery, which refers to, relies primarily on the machinery that specific function is completed in spinning movement, is answered extensively in mechanical equipment With, wherein gear is that the higher critical component of utility ratio, state and geometrical characteristic play the normal operation of mechanical equipment with axis To vital effect.In mechanical equipment operational process, inevitably there is equipment component aging and wear problem, The failure mode of resulting gear and bearing is varied, such as: bearing is made of inner ring, outer ring and ball, any bit Setting generation problem all can lead to bearing fault；The gear distresses such as fracture, hypodontia, abrasion and the scratch of gear.Rotating machinery event The complexity of barrier is embodied in the diversification of the characteristic polymorphic and failure generational verctor of different fault types, especially work as because When ageing equipment wear reason makes gear and bearing that coupling fault occur, the feature of vibration signal has more complexity.Pass through event Characteristic signal is extracted in barrier diagnosis, trouble unit is determined, to avoid bigger loss.
Traditional data-driven fault detection method needs expertise and extensive people using hand-made feature Power, each module are both needed to gradually train, and can not model to large-scale data, and the accuracy rate of coupling fault detection is poor.
Summary of the invention
There is a problem of fault detection accuracy rate difference when the present invention is for existing fault detection method detection coupling fault, mentions Detection method and monitoring system for a kind of rotating machinery coupling fault that fault detection accuracy rate is high.
In order to achieve the above object, the present invention provides a kind of detection methods of rotating machinery coupling fault, containing following Step:
(1) it acquires to be used as under industrial process nominal situation with the multistage sensor measurement data occurred under coupling fault and instruct Practice data, and establishes training dataset；
(2) depth convolutional neural networks model is established, which is equipped with Floor 12 hidden layer, using five layers of convolutional layer, five The down-sampled pond layer of layer and two layers of full articulamentum composition model frame, convolutional layer are arranged alternately with down-sampled pond layer, complete to connect Rear set of the layer in depth convolutional neural networks model；
(3) training dataset is input in depth convolutional neural networks model, neuronal activation value is carried out from normalizing Change operation, carry out gradient backpropagation training, update each network layer of depth convolutional neural networks model weight matrix parameter and Lay particular stress on matrix parameter；
(4) the weight parameter matrix of each network layer and weighting parameter square after storage depth convolutional neural networks model training Battle array；
(5) on-line sensor data are acquired, using depth convolutional neural networks model to the on-line sensor data of acquisition The affiliated type detection of failure is carried out, faulty generation is judged whether according to the affiliated type of failure.
Further, in step (1), the specific steps of training dataset are established are as follows: under acquisition industrial process nominal situation With malfunction lower bearing and gear coupling failure time domain vibration signal, including normal condition and 11 kinds of malfunctions totally 12 kinds of shapes State adds label according to different fault types, and every 2048 sampled points make training dataset as a data sample.
Further, in step (2), in depth convolutional neural networks model, the output of convolutional layer are as follows:
In formula,Indicate the output of convolutional layer,Indicating linear operation, ξ () indicates activation primitive, × indicate convolution behaviour Make,Indicate that convolution kernel, J indicate the number of convolution kernel, M indicates that the width of convolution kernel, N indicate the length of convolution kernel Degree,Indicate one layer of convolutional layer output,It indicates to lay particular stress on parameter；
The maximum pond of pond layer is defined as:
In formula,Indicate the pondization output of l i-th of neuron of layer, c indicates pond size；
Full articulamentum is identical structure with the network layer in bp neural network, and the full connection of full articulamentum calculates is defined as:
In formula,Indicate the linear activation of full articulamentum,Indicate the weight parameter matrix of full articulamentum, xl-1In expression One layer of network layer output.
Further, in step (3), the nerve to each convolutional layer is realized using scaling index linear unit activation function First activation value is carried out from normalizing operation, the scaling index linear unit activation function representation are as follows:
In formula, λ=1.050700987355480493419, α=1.673263242354377284817.
Further, in step (3), the step of backpropagation training are as follows:
(1) training dataset is input in depth convolutional neural networks model, calculates depth convolutional neural networks model Target loss function；
(2) gradient is calculated using gradient descent method, adaptability moments estimation algorithm updates weight parameter matrix；
(3) whether training Epoch reaches required value, otherwise the return step (1) if not up to terminates to train, and saves instruction Weight parameter matrix after white silk.
Further, in step (3), if given fault data collection isWherein xeIt is e-th of data sample, e =1,2 ..., E indicates number of samples,For the vector of one-hot type, different health are indicated The label of situation；Classified using softmax separator to coupling fault time domain vibration signal, sample xeNeural network forecast knot FruitIt indicates are as follows:
In formula,It is the weight matrix of the full articulamentum of most last layer, xe,l-1It is the output matrix of l-1 layers of convolutional layer, blIt indicates L layers of weighting matrix；
The then target loss function of depth convolutional neural networks model is defined as:
Further, in step (3), adaptability moments estimation algorithm updates the specific steps of weight parameter matrix are as follows:
By learning rate α, single order moments estimation attenuation rate β1, second order moments estimation attenuation rate β2, numerical stability constant ε, depth convolution Weight parameter matrix θ, target loss function L (θ), frequency of training n, batch size s, the first moment of neural network model are estimated Meter m and second order moments estimation v is input in depth convolutional neural networks model；
Initialize learning rate α=0.001, single order moments estimation attenuation rate β1It is 0.9, second order moments estimation attenuation rate β2It is 0.99, Numerical stability constant ε=10-8；Initialize neural network weight parameter matrix θ, use standard deviation for 0.1 random initializtion； Initialize that single order moments estimation m is 0, second order moments estimation v is 0 simultaneously, frequency of training n is 0；
When not reaching trained termination condition, i.e. target loss function L (θ) is not converged or frequency of training not up to provides When number:
n←n+1
Gradient descent method calculates gradient,
Update inclined single order moments estimation, m ← β1m+(1-β1)g
Update inclined second order moments estimation, v ← β2v+(1-β2)g⊙g
First moment deviation is calculated,
Second moment deviation is calculated,
Weight is updated,
Reach trained termination condition, returns to weighting parameter θ.
In order to achieve the above object, the present invention also provides a kind of rotating machinery coupling fault detection systems, comprising:
Data acquisition module, for acquire under industrial process nominal situation with occur coupling fault under vibration signal；
The training dataset generation module connecting with data acquisition module, the vibration signal for that will acquire generate training number According to collection；
The depth convolutional neural networks model being connect with the training dataset generation module, for diagnosing fault；
Model training module, for training depth convolutional neural networks model；
Parameter memory module, for storing the parameter after the depth convolutional neural networks model training；
Breakdown judge module is connect with the data acquisition module and depth convolutional neural networks model, for using deeply It spends convolutional neural networks model and breakdown judge is carried out to the data of data collecting module collected.
Compared with prior art, the beneficial effects of the present invention are:
The vibration signal data that the present invention is acquired under nominal situation and fault condition using rotating machinery is as training number According to collection, depth convolutional neural networks model is established, by training dataset directly as input, to depth convolutional neural networks model Parameter is trained, and saves the supplemental characteristic after training, is acquired the data under real-time working condition as test data, is passed through depth Convolutional neural networks model realization fault detection.The present invention is not necessarily to the accurate mathematical model of industrial process, is convenient for practical application；Together When realize the class discrimination of fault detection and fault condition, can effectively detect the specific component for generating mechanical failure, inspection It is high to survey accuracy rate.
Detailed description of the invention
Fig. 1 is the flow chart of the detection method of rotating machinery coupling fault of the present invention；
Fig. 2 is the framework of depth convolutional neural networks model in the detection method of rotating machinery coupling fault of the present invention Figure；
Fig. 3 is the training flow chart of depth convolutional neural networks model in rotating machinery coupling fault of the present invention；
Fig. 4 is that 12 kinds of different fault types vibrate letter in the detection method of rotating machinery coupling fault of the present invention Number time-domain diagram；
Fig. 5 is that the test error of the embodiment of the present invention one loses curve graph；
Fig. 6 is the failure detection result confusion matrix figure of the embodiment of the present invention one；
Fig. 7 is the structure diagram that rotating machinery coupling fault of the present invention monitors system.
Specific embodiment
In the following, the present invention is specifically described by illustrative embodiment.It should be appreciated, however, that not into one In the case where step narration, element, structure and features in an embodiment can also be advantageously incorporated into other embodiments In.
Referring to Fig. 1, present invention discloses a kind of detection method of rotating machinery coupling fault, this method directly reads sensing The original time domain vibration signal acquired on device, the depth convolutional neural networks model after training being capable of real time on-line monitoring machine The health status of rotating machinery in tool system excavates high-level abstract characterization, obtains accurate fault diagnosis result, contain Following steps:
It is used as and instructs with the multistage sensor measurement data occurred under coupling fault under S101, acquisition industrial process nominal situation Practice data, and establishes training dataset；
S102, establish depth convolutional neural networks model, which is equipped with Floor 12 hidden layer, using five layers of convolutional layer, Five layers of down-sampled pond layer and two layers of full articulamentum composition model frame, convolutional layer are arranged alternately with down-sampled pond layer, Quan Lian Layer is connect in the rear set of depth convolutional neural networks model；
S103, training dataset is input in depth convolutional neural networks model, neuronal activation value return certainly One changes operation, carries out gradient backpropagation training, updates the weight matrix parameter of each network layer of depth convolutional neural networks model With weighting matrix parameter；
The weight parameter matrix of each network layer and weighting parameter square after S104, storage depth convolutional neural networks model training Battle array；
S105, acquisition on-line sensor data, using depth convolutional neural networks model to the on-line sensor number of acquisition According to the affiliated type detection of failure is carried out, faulty generation is judged whether according to the affiliated type of failure.
In the above-mentioned detection method of the present invention, the specific steps of training dataset are established are as follows: acquisition industrial process nominal situation Lower and malfunction lower bearing and gear coupling failure time domain vibration signal, including normal condition and totally 12 kinds of 11 kinds of malfunctions State adds label according to different fault types, and every 2048 sampled points make training data as a data sample Collection.
Convolutional neural networks imply in convolutional neural networks generally by this up of three-layer of input layer, hidden layer and output layer Layer generally comprises the composition such as convolutional layer, pond layer and full articulamentum, and multilayer convolution replaces with pond layer, constitutes one completely Depth convolutional neural networks model.Convolutional layer can carry out weight to weight parameter w and share when carrying out convolution operation, can Reduce the memory headroom of the calculation amount and occupancy during model training.
Usual convolutional neural networks are widely used in computer vision field, are suitable for recognition of face, digital handwriting body The Classification and Identification of the two dimensional images such as identification, the input of model are generally two dimension input.The present invention shakes for the time domain of rotating machinery Dynamic signal carries out fault diagnosis, since time domain vibration signal is one-dimensional data, needs two-dimensional convolution operations improvement to be one-dimensional volume Product operation, to adapt to the input of original vibration signal.
In the depth convolutional neural networks model constructed in the above-mentioned detection method of the present invention, referring to fig. 2, which includes defeated Enter layer, hidden layer and output layer, wherein hidden layer is Floor 12, using five layers of convolutional layer, five layers of down-sampled pond layer and two layers Full articulamentum composition model frame, convolutional layer are arranged alternately with down-sampled pond layer, and full articulamentum is in depth convolutional neural networks The rear set of model.Specifically, referring to table 1, the framework of depth convolutional neural networks model hidden layer are as follows: the convolution being sequentially connected Layer C1, pond layer P2, convolutional layer C3, pond layer P4, convolutional layer C5, pond layer P6, convolutional layer C7, pond layer P8, convolutional layer C9, Pond layer P10, full articulamentum F11 and full articulamentum F12.Data processing and calculating are executed by node between layers.
Table 1
Network layer Core size Nuclear volume Output size
C1 64×1 16 512×16
P2 2×1 16 256×16
C3 3×1 32 256×32
P4 2×1 32 128×32
C5 3×1 64 128×64
P6 2×1 64 64×64
C7 3×1 64 64×64
P8 2×1 64 32×64
C9 3×1 64 32×64
P10 2×1 64 16×64
F11 100 1 100×1
F12 12 1 12×1
To adapt to rotating machinery periodic vibration signal, other convolutional layers are different from, the convolution kernel of convolutional layer C1 uses 64 × 1 big convolution kernel, convenient for extracting feature；The convolution kernel of convolutional layer C3, C5, C7 and C9 are 3 × 1 greatly, and depth is respectively 16,32,64,64；Pond layer P2, P4, P6, P8 and P10 are using maximum pondization operation, and P2 layers are 4 × 4, remaining is 2 × 2 Structure；It is 12 × 1 that the output of full articulamentum F11, which is 100 × 1, F12 layers, eventually passes through softmax classifier and is divided into 12 classes.Most Eventually by the vector for the one-hot types that the final output of convolutional neural networks is 12 dimensions, if it is that vibration signal is corresponding A kind of failure, then output is [0,1,0,0,0,0,0,0,0,0,0,0], if corresponding second of failure, output for [0,0,1,0, 0,0,0,0,0,0,0,0], if there is no failure, equipment normal operation exports [1,0,0,0,0,0,0,0,0,0,0,0].
In convolutional layer, convolution kernel is a series of parameter matrixs for capableing of shared parameter, and one group of connection can be shared same Weight, rather than each it is connected with a different weight.The output of convolutional layer are as follows:
In formula,Indicate the output of convolutional layer,Indicating linear operation, ξ () indicates activation primitive, × indicate convolution behaviour Make,Indicate that convolution kernel, J indicate the number of convolution kernel, M indicates that the width of convolution kernel, N indicate the length of convolution kernel Degree,Indicate one layer of convolutional layer output,It indicates to lay particular stress on parameter.
In convolutional layer output, the effect of activation primitive is that linear operation is carried out to non-linear words, enhances fitting effect.Needle It is difficult to the feature extracted to rotating machinery coupling fault, using normalization strategy, the neuronal activation value of each convolutional layer is carried out From normalizing operation, is realized using scaling index linear unit (referred to as: SELUs) activation primitive and the neuron of each convolutional layer is swashed Value living is carried out from normalizing operation, the scaling index linear unit activation function representation are as follows:
In formula, λ=1.050700987355480493419, α=1.673263242354377284817.
It carries out avoiding extracting height from normalizing operation using scaling index linear unit (referred to as: SELUs) activation primitive Gradient suddenly disappears or the problem of explosive increase when dimensional feature.
The above-mentioned detection method of the present invention, selection increase pond layer after convolutional layer and carry out pondization operation, mainly carry out Down-sampling is further reduced number of parameters by removing unessential sample in the characteristic information obtained after convolution.In the present invention It states in detection method, carries out the method for down-sampling using maximum pond method, the maximum pond of pond layer is defined as:
In formula,Indicate the pondization output of l i-th of neuron of layer, c indicates pond size.
The above-mentioned detection method of the present invention, full articulamentum add the rear set in entire depth convolutional neural networks model, entirely Articulamentum is identical structure, the mark of the Feature Mapping for mainly arriving e-learning to sample with the network layer in bp neural network Remember in space, the two dimensional character figure that convolutional layer exports is converted into an one-dimensional vector, is convenient for subsequent failure modes.Quan Lian The full connection for connecing layer calculates is defined as:
In formula,Indicate the linear activation of full articulamentum,Indicate the weight parameter matrix of full articulamentum, xl-1In expression One layer of network layer output.
Referring to Fig. 3, in above-mentioned detection method, the step of depth convolutional neural networks model backpropagation training are as follows:
(1) training dataset is input in depth convolutional neural networks model, calculates depth convolutional neural networks model Target loss function；
(2) gradient is calculated using gradient descent method, adaptability moments estimation algorithm updates weight parameter matrix；
(3) whether training Epoch reaches required value, otherwise the return step (1) if not up to terminates to train, and saves instruction Weight parameter matrix after white silk.
Specifically, if given fault data collection isWherein xeIt is e-th of data sample, e=1,2 ..., E table Show number of samples,For the vector of one-hot type, the label of different health status is indicated； Classified using softmax separator to coupling fault time domain vibration signal, sample xeNeural network forecast resultIt indicates are as follows:
In formula,It is the weight matrix of the full articulamentum of most last layer, xe,l-1It is the output matrix of l-1 layers of convolutional layer, blIt indicates L layers of weighting matrix；
The then target loss function of depth convolutional neural networks model is defined as:
Depth convolutional neural networks model optimizes above-mentioned target loss function using gradient descent method, is updated with this Weight parameter matrix and weighting parameter matrix.
In the above-mentioned detection method of the present invention, the depth convolutional neural networks model of use, since the number of plies is 12 layers, the number of plies compared with It is deep, training data is based on using adaptability moments estimation algorithm and iteratively updates neural network weight parameter matrix.Specifically, it adapts to Property moments estimation algorithm update weight parameter matrix specific steps are as follows:
By learning rate α, single order moments estimation attenuation rate β1, second order moments estimation attenuation rate β2, numerical stability constant ε, depth convolution Weight parameter matrix θ, target loss function L (θ), frequency of training n, batch size s, the first moment of neural network model are estimated Meter m and second order moments estimation v is input in depth convolutional neural networks model；
Initialize learning rate α=0.001, single order moments estimation attenuation rate β1It is 0.9, second order moments estimation attenuation rate β2It is 0.99, Numerical stability constant ε=10-8；Initialize neural network weight parameter matrix θ, use standard deviation for 0.1 random initializtion； Initialize that single order moments estimation m is 0, second order moments estimation v is 0 simultaneously, frequency of training n is 0；
When not reaching trained termination condition, i.e. target loss function L (θ) is not converged or frequency of training not up to provides When number:
n←n+1
Gradient descent method calculates gradient,
Update inclined single order moments estimation, m ← β1m+(1-β1)g
Update inclined second order moments estimation, v ← β2v+(1-β2)g⊙g
First moment deviation is calculated,
Second moment deviation is calculated,
Weight is updated,
Reach trained termination condition, returns to weighting parameter θ.
The above-mentioned detection method of the present invention, the vibration signal number acquired under nominal situation and fault condition using rotating machinery According to as training dataset, depth convolutional neural networks model is established, by training dataset directly as input, to depth convolution Neural network model parameter is trained, and saves the supplemental characteristic after training, acquires the data under real-time working condition as test Data pass through depth convolutional neural networks model realization fault detection.The above-mentioned detection method of the present invention is accurate without industrial process Mathematical model, be convenient for practical application；The class discrimination for realizing fault detection and fault condition simultaneously, can effectively detect The specific component of mechanical failure is generated, Detection accuracy is high.
Referring to Fig. 7, the present invention also provides a kind of rotating machinery coupling fault detection systems, comprising:
Data acquisition module 1, for acquire under industrial process nominal situation with occur coupling fault under vibration signal；
The training dataset generation module 2 connecting with data acquisition module 1, the vibration signal for that will acquire generate training Data set；
The depth convolutional neural networks model 3 being connect with the training dataset generation module 2, for diagnosing fault；
Model training module 4, for training depth convolutional neural networks model；
Parameter memory module 5, for storing the parameter after the depth convolutional neural networks model training；
Breakdown judge module 6 is connect, for utilizing with the data acquisition module 1 and depth convolutional neural networks model 3 Depth convolutional neural networks model carries out breakdown judge to the data that data acquisition module 1 acquires.
The above-mentioned rotating machinery coupling fault monitoring system of the present invention passes through in data collecting module collected machine driven system The coupled vibrations sensing data of parallel-shaft gearbox middle (center) bearing and gear, as on-line testing data, the test data and mould Training dataset data in type training module are consistent.Can detecte using the above-mentioned monitoring system of the present invention recognize it is trained Gear-bearing coupling fault, and detect the affiliated type of the failure, judge whether faulty generation, fault detection accuracy rate Height, it is practical.
In order to verify the detection method of the above-mentioned rotating machinery coupling fault of the present invention to the effective of coupling rotating machinery fault Property, and the above-mentioned rotating machinery coupling fault monitoring system of the verifying present invention, its progress is distinguished below with two specific embodiments Explanation.
Embodiment one: wind turbine power transmission fault diagnosis comprehensive experiment table is used, the acceleration on experimental bench is passed through The bearing and gear coupling fault vibration signal of sensor acquisition parallel-shaft gearbox are spent, acquires 11 kinds of fault conditions and normal altogether Mechanical oscillation signal under operating condition makes training dataset and test data set.
The vibration signal of acquisition totally 12 kinds of different fault types, referring to table 2.As shown in Table 2,12 kinds of fault type difference Are as follows: (1) normal condition N, (2) inner-ring bearing failure IF, (3) roller bearing fault condition RF, (4) race bearing failure OF, (5) Coupling fault IRO occurs for inner ring, ball and outer ring condition, and coupling event occurs for (6) ball bearing and cut-off wheel state (RCH) Coupling fault, (8) race bearing and teeth-missing gear state (OCH) occur for barrier, (7) ball bearing and Gear with Crack state (RCR) Shaft coupling failure occurs, coupling fault, (10) inner-ring bearing and hypodontia occur for (9) race bearing and Gear with Crack state (OCR) Shaft coupling failure occurs for gear condition (ICH), and coupling fault, (12) occur for (11) inner-ring bearing and Gear with Crack state (ICR) In ball bearing, crackle and cut-off wheel state coupling fault RCC.The time domain vibration signal of every kind of failure is referring to fig. 4.
Table 2
In order to prove the present invention using depth convolutional neural networks model on rotating machinery coupling fault test problems Validity, the present invention by with BP neural network, SCNN (activation primitive be sigmoid function) and sparse self-encoding encoder come pair Than verifying.
Setting learning rate is 0.001, dropout 0.5, and every 2048 sampled points are as a data sample, using 696 A data sample is trained, and experiment carries out 1500 training in total.Depth convolutional neural networks model uses 12 layer networks altogether The structure of layer, using small batch training.
It is tested by training, by the loss function curve of test process as shown in figure 5, curve has had reached convergence shape State.Comparative experiments classifies situation to the identification of 12 kinds of different faults, the standard of the rotating machinery coupling fault diagnosis under algorithms of different True rate is referring to table 4.As shown in Table 4, the present invention has reached 95.8% using the correct recognition rata of depth convolutional neural networks model, It is compared to the 60.3% of traditional BP neural network, the 81.5% of 76.2% and SCNN of sparse self-encoding encoder, is promoted respectively 30.3%, 19.4% and 8.1%.
Table 4
Algorithm Sample type Sample size Accuracy rate
ANN Time-domain signal 2048×1 60.3%
Sparse self-encoding encoder Time-domain signal 2048×1 76.2%
SCNN(Sigmoid) Time-domain signal 2048×1 81.5%
DCNN Time-domain signal 2048×1 95.8%
In order to which the classification results of every kind of health status are described in detail, the confusion matrix of the measuring accuracy of the experiment is drawn, is mixed Matrix confuse referring to Fig. 6.It will be appreciated from fig. 6 that single failure is easier identified and can obtain higher standard compared to coupling fault True rate, close to 100%, and the coupling fault classification accuracy of bearing and gear is slightly worse, has certain probability to assign to and is coupled Single failure in, but still ensure that 90% or more accuracy rate.
Embodiment two: in order to verify rotating machinery coupling fault monitoring system of the present invention to the reality that can satisfy on-line monitoring Time when monitoring system diagnostic data and the relationship of batch size are tested in the requirement of when property.By being surveyed in difference batch size It tries, the time required for available proposed one signal of system diagnostics is about 44ms, used wind turbine mechanomotive force The sample frequency that drive failures diagnose the acceleration transducer of collection machinery vibration signal in comprehensive experiment table is 5.12KMz, often A sample signal takes 2048 sampled points, and the relationship of time and batch size is referring to table 5, as shown in Table 5, rotating machinery coupling of the present invention Closing fault monitoring system can be very good the requirement for meeting real-time.
Table 5
Criticize size 11 23 35 47 59 71 83
Time (s) 0.0439 0.0437 0.0423 0.0445 0.0451 0.0438 0.0427
Embodiment provided above only with illustrating the present invention for convenience, and it is not intended to limit the protection scope of the present invention, Technical solution scope of the present invention, person of ordinary skill in the field make various simple deformations and modification, should all include In the above claim.

Claims (8)

1.一种旋转机械耦合故障的检测方法，其特征在于，含有以下步骤：1. a detection method for a rotating machinery coupling fault, is characterized in that, contains the following steps: (一)采集工业过程正常工况下与发生耦合故障下的多段传感器测量数据作为训练数据，并建立训练数据集；(1) Collecting multi-segment sensor measurement data under normal working conditions and coupling faults in the industrial process as training data, and establishing a training data set; (二)建立深度卷积神经网络模型，该模型设有十二层隐含层，采用五层卷积层、五层降采样池化层与两层全连接层组成模型框架，卷积层与降采样池化层交替设置，全连接层在深度卷积神经网络模型的后置位；(2) Establish a deep convolutional neural network model. The model has twelve hidden layers. Five convolutional layers, five downsampling pooling layers and two fully connected layers are used to form the model framework. The convolutional layer and the The downsampling pooling layers are alternately set, and the fully connected layer is placed after the deep convolutional neural network model; (三)将训练数据集输入到深度卷积神经网络模型中，对神经元激活值进行自归一化操作，进行梯度反向传播训练，更新深度卷积神经网络模型各网络层的权重矩阵参数和偏重矩阵参数；(3) Input the training data set into the deep convolutional neural network model, perform self-normalization operation on neuron activation values, perform gradient backpropagation training, and update the weight matrix parameters of each network layer of the deep convolutional neural network model. and the weighted matrix parameters; (四)存储深度卷积神经网络模型训练后各网络层的权重参数矩阵和偏重参数矩阵；(4) storing the weight parameter matrix and the bias parameter matrix of each network layer after the training of the deep convolutional neural network model; (五)采集在线传感器数据，利用深度卷积神经网络模型对采集的在线传感器数据进行故障所属类型检测，根据故障所属类型判断是否有故障发生。(5) Collect online sensor data, use the deep convolutional neural network model to detect the type of fault on the collected online sensor data, and judge whether there is a fault according to the type of fault. 2.如权利要求1所述的旋转机械耦合故障的检测方法，其特征在于，步骤(一)中，建立训练数据集的具体步骤为：采集工业过程正常工况下与故障状态下轴承与齿轮耦合故障时域振动信号，包括正常状态与11种故障状态共12种状态，根据不同的故障类型添加标签，每2048个采样点作为一个数据样本，制作训练数据集。2. The detection method of rotating machinery coupling fault as claimed in claim 1, characterized in that, in step (1), the specific step of establishing a training data set is: collecting bearings and gears under normal working conditions and fault conditions of industrial processes Coupling fault time domain vibration signals, including normal state and 11 fault states, a total of 12 states, add labels according to different fault types, every 2048 sampling points are used as a data sample to create a training data set. 3.如权利要求2所述的旋转机械耦合故障的检测方法，其特征在于，步骤(二)中，在深度卷积神经网络模型中，卷积层的输出为：3. the detection method of rotating machinery coupling fault as claimed in claim 2, is characterized in that, in step (2), in deep convolutional neural network model, the output of convolution layer is: 式中，表示卷积层的输出，表示线性操作，ξ(·)表示激活函数，×表示卷积操作，表示卷积核，J表示卷积核的数目，M表示卷积核的宽度，N表示卷积核的长度，表示上一层的卷积层输出，表示偏重参数；In the formula, represents the output of the convolutional layer, represents the linear operation, ξ( ) represents the activation function, × represents the convolution operation, Represents the convolution kernel, J represents the number of convolution kernels, M represents the width of the convolution kernel, N represents the length of the convolution kernel, represents the output of the convolutional layer of the previous layer, Represents a biased parameter; 池化层的最大池化定义为：The max pooling of the pooling layer is defined as: 式中，表示第l层第i个神经元的池化输出，c表示池化大小；In the formula, represents the pooling output of the ith neuron in the lth layer, and c represents the pooling size; 全连接层与bp神经网络中的网络层为相同结构，全连接层的全连接计算定义为：The fully connected layer has the same structure as the network layer in the bp neural network. The fully connected calculation of the fully connected layer is defined as: 式中，表示全连接层的线性激活，表示全连接层的权重参数矩阵，xl-1表示上一层的网络层输出。In the formula, represents the linear activation of the fully connected layer, Represents the weight parameter matrix of the fully connected layer, and x l-1 represents the network layer output of the previous layer. 4.如权利要求3所述的旋转机械耦合故障的检测方法，其特征在于，步骤(三)中，采用缩放指数线性单位激活函数实现对各卷积层的神经元激活值进行自标准化操作，所述缩放指数线性单位激活函数表示为：4. the detection method of rotating machinery coupling fault as claimed in claim 3, is characterized in that, in step (3), adopts scaling exponential linear unit activation function to realize that the neuron activation value of each convolution layer is carried out self-standardization operation, The scaling exponential linear unit activation function is expressed as: 式中，λ＝1.050700987355480493419，α＝1.673263242354377284817。In the formula, λ=1.050700987355480493419, α=1.673263242354377284817. 5.如权利要求3或4所述的旋转机械耦合故障的检测方法，其特征在于，步骤(三)中，反向传播训练的步骤为：5. the detection method of rotating machinery coupling fault as claimed in claim 3 or 4, is characterized in that, in step (3), the step of back propagation training is: (1)训练数据集输入至深度卷积神经网络模型中，计算深度卷积神经网络模型的目标损失函数；(1) The training data set is input into the deep convolutional neural network model, and the target loss function of the deep convolutional neural network model is calculated; (2)采用梯度下降法计算梯度，适应性矩估计算法更新权重参数矩阵；(2) The gradient is calculated by the gradient descent method, and the weight parameter matrix is updated by the adaptive moment estimation algorithm; (3)训练Epoch是否达到要求值，若未达到则返回步骤(1)，否则结束训练，保存训练后的权重参数矩阵。(3) Whether the training Epoch reaches the required value, if not, return to step (1), otherwise end the training, and save the weight parameter matrix after training. 6.如权利要求5所述的旋转机械耦合故障的检测方法，其特征在于，步骤(三)中，设给定故障数据集为其中xe是第e个数据样本，e＝1,2,...,E表示样本数目， 为one-hot类型的向量，表示不同健康状况的标签；采用softmax分离器对耦合故障时域振动信号进行分类，样本xe的网络预测结果表示为：6. The detection method of rotating machinery coupling fault as claimed in claim 5, characterized in that, in step (3), the given fault data set is set as where x e is the e-th data sample, e=1,2,...,E represents the number of samples, is a one-hot type vector, representing the labels of different health conditions; the softmax separator is used to classify the coupled fault time-domain vibration signal, and the network prediction result of the sample x e Expressed as: 式中，是最末层全连接层的权重矩阵，xe,l-1是l-1层卷积层的输出矩阵，bl表示第l层的偏重矩阵；In the formula, is the weight matrix of the last fully connected layer, x e,l-1 is the output matrix of the l-1 layer convolutional layer, b l represents the weight matrix of the lth layer; 则深度卷积神经网络模型的目标损失函数定义为：Then the objective loss function of the deep convolutional neural network model is defined as: 7.如权利要求5所述的旋转机械耦合故障的检测方法，其特征在于，步骤(三)中，适应性矩估计算法更新权重参数矩阵的具体步骤为：7. the detection method of rotating machinery coupling fault as claimed in claim 5, is characterized in that, in step (3), the concrete step of adaptive moment estimation algorithm updating weight parameter matrix is: 将学习率α、一阶矩估计衰减率β1、二阶矩估计衰减率β2、数值稳定常数ε、深度卷积神经网络模型的权重参数矩阵θ、目标损失函数L(θ)、训练次数n、batch sizes、一阶矩估计m以及二阶矩估计v输入至深度卷积神经网络模型中；The learning rate α, the estimated decay rate of the first-order moment β 1 , the estimated decay rate of the second-order moment β 2 , the numerical stability constant ε, the weight parameter matrix θ of the deep convolutional neural network model, the objective loss function L(θ), the number of training n, batch sizes, first-order moment estimates m and second-order moment estimates v are input into the deep convolutional neural network model; 初始化学习率α＝0.001，一阶矩估计衰减率β1为0.9，二阶矩估计衰减率β2为0.99，数值稳定常数ε＝10-8；初始化神经网络的权重参数矩阵θ，采用标准差为0.1的随机初始化；同时初始化一阶矩估计m为0、二阶矩估计v为0，训练次数n为0；The initialized learning rate α=0.001, the first -order moment estimation decay rate β1 is 0.9, the second-order moment estimation decay rate β2 is 0.99, and the numerical stability constant ε=10 −8 ; the weight parameter matrix θ of the initialized neural network adopts the standard deviation It is a random initialization of 0.1; at the same time, initialize the first-order moment estimation m as 0, the second-order moment estimation v as 0, and the training times n as 0; 当没有达到训练终止条件，即目标损失函数L(θ)未收敛或者训练次数未达到规定次数时：n←n+1When the training termination condition is not reached, that is, the target loss function L(θ) does not converge or the number of training times does not reach the specified number of times: n←n+1 梯度下降法计算梯度， Gradient descent method calculates the gradient, 更新偏一阶矩估计，m←β1m+(1-β1)gUpdate the partial first moment estimate, m←β 1 m+(1-β 1 )g 更新偏二阶矩估计，v←β2v+(1-β2)g⊙gUpdate the partial second moment estimate, v←β 2 v+(1-β 2 )g⊙g 计算一阶矩偏差， Calculate the first moment deviation, 计算二阶矩偏差， Calculate the second moment deviation, 更新权值， update the weights, 达到训练终止条件，返回权值参数θ。When the training termination condition is reached, the weight parameter θ is returned. 8.一种旋转机械耦合故障监测系统，其特征在于，包括：8. A rotating machinery coupling fault monitoring system, characterized in that, comprising: 数据采集模块，用于采集工业过程正常工况下与发生耦合故障下的振动信号；The data acquisition module is used to collect vibration signals under normal working conditions and coupling faults in industrial processes; 与数据采集模块连接的训练数据集生成模块，用于将采集的振动信号生成训练数据集；The training data set generation module connected with the data acquisition module is used to generate the training data set from the collected vibration signals; 与所述训练数据集生成模块连接的深度卷积神经网络模型，用于诊断故障；A deep convolutional neural network model connected with the training data set generation module for diagnosing faults; 模型训练模块，用于训练深度卷积神经网络模型；A model training module for training a deep convolutional neural network model; 参数存储模块，用于存储所述深度卷积神经网络模型训练后的参数；a parameter storage module for storing the parameters after the training of the deep convolutional neural network model; 故障判断模块，与所述数据采集模块和深度卷积神经网络模型连接，用于利用深度卷积神经网络模型对数据采集模块采集的数据进行故障判断。The fault judgment module is connected with the data acquisition module and the deep convolutional neural network model, and is used for performing fault judgment on the data collected by the data acquisition module by using the deep convolutional neural network model.
CN201910109012.0A 2019-02-03 2019-02-03 Detection method and monitoring system of rotating machinery coupling fault Pending CN109782603A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910109012.0A CN109782603A (en) 2019-02-03 2019-02-03 Detection method and monitoring system of rotating machinery coupling fault

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910109012.0A CN109782603A (en) 2019-02-03 2019-02-03 Detection method and monitoring system of rotating machinery coupling fault

Publications (1)

Publication Number Publication Date
CN109782603A true CN109782603A (en) 2019-05-21

Family

ID=66504247

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910109012.0A Pending CN109782603A (en) 2019-02-03 2019-02-03 Detection method and monitoring system of rotating machinery coupling fault

Country Status (1)

Country Link
CN (1) CN109782603A (en)

Cited By (22)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110163793A (en) * 2019-05-27 2019-08-23 湃方科技(天津)有限责任公司 Convolutional calculation acceleration method and device
CN110243590A (en) * 2019-06-25 2019-09-17 中国民航大学 A Rotor System Fault Diagnosis Method Based on Principal Component Analysis and Width Learning
CN110428004A (en) * 2019-07-31 2019-11-08 中南大学 Component of machine method for diagnosing faults under data are unbalance based on deep learning
CN110443488A (en) * 2019-07-31 2019-11-12 中国石油大学(华东) The recognition methods of drilling well spill hazard, system and equipment based on convolutional neural networks
CN110779746A (en) * 2019-10-24 2020-02-11 西安理工大学 Improved diagnostic method for composite faults of rotating machinery in deep sparse autoencoder networks
CN110969244A (en) * 2019-12-18 2020-04-07 长安大学 A construction safety monitoring method based on convolutional neural network
CN111062271A (en) * 2019-11-29 2020-04-24 中国船舶工业系统工程研究院 Method and device for establishing fault identification model of rotating equipment
CN111126603A (en) * 2019-12-25 2020-05-08 江苏远望仪器集团有限公司 Equipment fault prediction method, device and equipment based on neural network model
CN111242099A (en) * 2020-02-28 2020-06-05 红云红河烟草(集团)有限责任公司 System and method for identifying disordered cigarettes
CN111505424A (en) * 2020-05-06 2020-08-07 哈尔滨工业大学 Large experimental device power equipment fault diagnosis method based on deep convolutional neural network
CN111768022A (en) * 2020-05-09 2020-10-13 中国煤矿机械装备有限责任公司 Equipment detection method and device for coal machine production equipment
CN112462734A (en) * 2020-11-03 2021-03-09 贵州江南航天信息网络通信有限公司 Industrial production equipment fault prediction analysis method and model
CN112561097A (en) * 2020-12-23 2021-03-26 济南浪潮高新科技投资发展有限公司 Bearing monitoring method and system based on cloud and fog edge cooperation
CN113180670A (en) * 2021-05-24 2021-07-30 北京测态培元科技有限公司 Method for identifying mental state of depression patient based on finger pulse signals
CN113392862A (en) * 2020-03-12 2021-09-14 中国移动通信集团山东有限公司 Self-healing management and control method and device for perception data, computer equipment and storage medium
CN113887461A (en) * 2021-10-13 2022-01-04 深圳潜翔定制设计有限公司 A pattern recognition method, device, device and storage medium for mechanical parts
CN114057053A (en) * 2022-01-18 2022-02-18 杭州浅水数字技术有限公司 Method for monitoring fatigue degree of component of special machine
CN114154400A (en) * 2021-11-15 2022-03-08 中国人民解放军63963部队 Unmanned vehicle health state detection system and detection method
CN114386526A (en) * 2022-01-18 2022-04-22 郑州轻工业大学 A joint convolutional neural network diagnosis method for rotating machinery faults
CN114705467A (en) * 2022-04-01 2022-07-05 深圳市玄羽科技有限公司 A kind of artificial intelligence mechanical equipment monitoring system and monitoring method
CN116257747A (en) * 2022-12-22 2023-06-13 中国人民解放军海军工程大学 An Early Anomaly Detection Method for Rotor Systems Fused with Kurtosis and Convolutional Autoencoders
CN116821650A (en) * 2023-04-28 2023-09-29 重庆大学 Range extender fault detection model construction method, fault detection and diagnosis method and maintenance method

Citations (14)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPH08159928A (en) * 1994-12-01 1996-06-21 Posukon:Kk Device and method for diagnosing presence or absence of abnormality of rotary machine utilizing neural network
TW201100825A (en) * 2009-06-22 2011-01-01 Nat Univ Chin Yi Technology Inverter fault diagnosis method
CN103454113A (en) * 2013-09-16 2013-12-18 中国人民解放军国防科学技术大学 Method for monitoring health of rotary machine suitable for working condition changing condition
US20150238148A1 (en) * 2013-10-17 2015-08-27 Siemens Aktiengesellschaft Method and system for anatomical object detection using marginal space deep neural networks
CN105841961A (en) * 2016-03-29 2016-08-10 中国石油大学(华东) Bearing fault diagnosis method based on Morlet wavelet transformation and convolutional neural network
CN106485324A (en) * 2016-10-09 2017-03-08 成都快眼科技有限公司 A kind of convolutional neural networks optimization method
KR20170093613A (en) * 2016-02-05 2017-08-16 울산대학교 산학협력단 Method for bearing fault diagnosis
US20170364800A1 (en) * 2016-06-16 2017-12-21 Qatar University Method and apparatus for performing motor-fault detection via convolutional neural networks
US20180107925A1 (en) * 2016-10-19 2018-04-19 Samsung Electronics Co., Ltd. Method and apparatus for neural network quantization
CN108090871A (en) * 2017-12-15 2018-05-29 厦门大学 A kind of more contrast MR image reconstruction methods based on convolutional neural networks
CN108108768A (en) * 2017-12-29 2018-06-01 清华大学 Photovoltaic glass defect classification method and device based on convolutional neural networks
CN108444708A (en) * 2018-04-16 2018-08-24 长安大学 The method for building up of rolling bearing intelligent diagnostics model based on convolutional neural networks
CN108830127A (en) * 2018-03-22 2018-11-16 南京航空航天大学 A kind of rotating machinery fault feature intelligent diagnostic method based on depth convolutional neural networks structure
CN109033450A (en) * 2018-08-22 2018-12-18 太原理工大学 Lift facility failure prediction method based on deep learning

Patent Citations (14)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPH08159928A (en) * 1994-12-01 1996-06-21 Posukon:Kk Device and method for diagnosing presence or absence of abnormality of rotary machine utilizing neural network
TW201100825A (en) * 2009-06-22 2011-01-01 Nat Univ Chin Yi Technology Inverter fault diagnosis method
CN103454113A (en) * 2013-09-16 2013-12-18 中国人民解放军国防科学技术大学 Method for monitoring health of rotary machine suitable for working condition changing condition
US20150238148A1 (en) * 2013-10-17 2015-08-27 Siemens Aktiengesellschaft Method and system for anatomical object detection using marginal space deep neural networks
KR20170093613A (en) * 2016-02-05 2017-08-16 울산대학교 산학협력단 Method for bearing fault diagnosis
CN105841961A (en) * 2016-03-29 2016-08-10 中国石油大学(华东) Bearing fault diagnosis method based on Morlet wavelet transformation and convolutional neural network
US20170364800A1 (en) * 2016-06-16 2017-12-21 Qatar University Method and apparatus for performing motor-fault detection via convolutional neural networks
CN106485324A (en) * 2016-10-09 2017-03-08 成都快眼科技有限公司 A kind of convolutional neural networks optimization method
US20180107925A1 (en) * 2016-10-19 2018-04-19 Samsung Electronics Co., Ltd. Method and apparatus for neural network quantization
CN108090871A (en) * 2017-12-15 2018-05-29 厦门大学 A kind of more contrast MR image reconstruction methods based on convolutional neural networks
CN108108768A (en) * 2017-12-29 2018-06-01 清华大学 Photovoltaic glass defect classification method and device based on convolutional neural networks
CN108830127A (en) * 2018-03-22 2018-11-16 南京航空航天大学 A kind of rotating machinery fault feature intelligent diagnostic method based on depth convolutional neural networks structure
CN108444708A (en) * 2018-04-16 2018-08-24 长安大学 The method for building up of rolling bearing intelligent diagnostics model based on convolutional neural networks
CN109033450A (en) * 2018-08-22 2018-12-18 太原理工大学 Lift facility failure prediction method based on deep learning

Non-Patent Citations (5)

* Cited by examiner, † Cited by third party
Title
GÜNTER KLAMBAUER 等: "Self-Normalizing Neural Networks", 《31 ST CONFERENCE ON NEURAL INFORMATION PROCESSING SYSTEMS》 *
张传雷 等: "《基于图像分析的植物及其病虫害识别方法研究》", 31 October 2018, 中国经济出版社 *
张神林: "基于卷积神经网络的滚动轴承及行星齿轮箱故障诊断方法", 《中国优秀硕士学位论文全文数据库工程科技Ⅱ辑》 *
李恒 等: "基于短时傅里叶变换和卷积神经网络的轴承故障诊断方法", 《振动与冲击》 *
邢晓羊 等: "基于特征损失的医学图像超分辨率重建", 《计算机工程与应用》 *

Cited By (35)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110163793A (en) * 2019-05-27 2019-08-23 湃方科技(天津)有限责任公司 Convolutional calculation acceleration method and device
CN110163793B (en) * 2019-05-27 2023-05-23 湃方科技(天津)有限责任公司 Convolution calculation acceleration method and device
CN110243590A (en) * 2019-06-25 2019-09-17 中国民航大学 A Rotor System Fault Diagnosis Method Based on Principal Component Analysis and Width Learning
CN110443488A (en) * 2019-07-31 2019-11-12 中国石油大学(华东) The recognition methods of drilling well spill hazard, system and equipment based on convolutional neural networks
CN110428004A (en) * 2019-07-31 2019-11-08 中南大学 Component of machine method for diagnosing faults under data are unbalance based on deep learning
CN110779746A (en) * 2019-10-24 2020-02-11 西安理工大学 Improved diagnostic method for composite faults of rotating machinery in deep sparse autoencoder networks
CN110779746B (en) * 2019-10-24 2021-08-10 西安理工大学 Diagnosis method for improving composite fault of deep sparse self-encoder network rotating machinery
CN111062271A (en) * 2019-11-29 2020-04-24 中国船舶工业系统工程研究院 Method and device for establishing fault identification model of rotating equipment
CN110969244A (en) * 2019-12-18 2020-04-07 长安大学 A construction safety monitoring method based on convolutional neural network
CN110969244B (en) * 2019-12-18 2023-04-25 长安大学 A construction safety monitoring method based on convolutional neural network
CN111126603A (en) * 2019-12-25 2020-05-08 江苏远望仪器集团有限公司 Equipment fault prediction method, device and equipment based on neural network model
CN111242099A (en) * 2020-02-28 2020-06-05 红云红河烟草(集团)有限责任公司 System and method for identifying disordered cigarettes
CN111242099B (en) * 2020-02-28 2024-10-29 红云红河烟草(集团)有限责任公司 System and method for identifying disordered cigarettes
CN113392862A (en) * 2020-03-12 2021-09-14 中国移动通信集团山东有限公司 Self-healing management and control method and device for perception data, computer equipment and storage medium
CN113392862B (en) * 2020-03-12 2022-12-09 中国移动通信集团山东有限公司 Self-healing management and control method and device for sensing data, computer equipment and storage medium
CN111505424A (en) * 2020-05-06 2020-08-07 哈尔滨工业大学 Large experimental device power equipment fault diagnosis method based on deep convolutional neural network
CN111768022A (en) * 2020-05-09 2020-10-13 中国煤矿机械装备有限责任公司 Equipment detection method and device for coal machine production equipment
CN112462734B (en) * 2020-11-03 2022-06-17 贵州江南航天信息网络通信有限公司 Industrial production equipment fault prediction analysis method and model
CN112462734A (en) * 2020-11-03 2021-03-09 贵州江南航天信息网络通信有限公司 Industrial production equipment fault prediction analysis method and model
CN112561097A (en) * 2020-12-23 2021-03-26 济南浪潮高新科技投资发展有限公司 Bearing monitoring method and system based on cloud and fog edge cooperation
CN112561097B (en) * 2020-12-23 2023-04-21 山东浪潮科学研究院有限公司 Bearing monitoring method and system based on cloud and mist edge cooperation
CN113180670B (en) * 2021-05-24 2023-03-21 北京测态培元科技有限公司 Method for identifying mental state of depression patient based on finger pulse signals
CN113180670A (en) * 2021-05-24 2021-07-30 北京测态培元科技有限公司 Method for identifying mental state of depression patient based on finger pulse signals
CN113887461A (en) * 2021-10-13 2022-01-04 深圳潜翔定制设计有限公司 A pattern recognition method, device, device and storage medium for mechanical parts
CN113887461B (en) * 2021-10-13 2024-08-23 深圳潜翔定制设计有限公司 Method, device, equipment and storage medium for identifying mode of mechanical part
CN114154400A (en) * 2021-11-15 2022-03-08 中国人民解放军63963部队 Unmanned vehicle health state detection system and detection method
CN114154400B (en) * 2021-11-15 2023-12-05 中国人民解放军63963部队 Unmanned vehicle health state detection system and detection method
CN114057053A (en) * 2022-01-18 2022-02-18 杭州浅水数字技术有限公司 Method for monitoring fatigue degree of component of special machine
CN114386526A (en) * 2022-01-18 2022-04-22 郑州轻工业大学 A joint convolutional neural network diagnosis method for rotating machinery faults
CN114057053B (en) * 2022-01-18 2022-04-26 杭州浅水数字技术有限公司 Method for monitoring fatigue degree of component of special machine
CN114705467A (en) * 2022-04-01 2022-07-05 深圳市玄羽科技有限公司 A kind of artificial intelligence mechanical equipment monitoring system and monitoring method
CN114705467B (en) * 2022-04-01 2025-04-25 深圳市玄羽科技有限公司 An artificial intelligence mechanical equipment monitoring system and monitoring method
CN116257747A (en) * 2022-12-22 2023-06-13 中国人民解放军海军工程大学 An Early Anomaly Detection Method for Rotor Systems Fused with Kurtosis and Convolutional Autoencoders
CN116821650A (en) * 2023-04-28 2023-09-29 重庆大学 Range extender fault detection model construction method, fault detection and diagnosis method and maintenance method
CN116821650B (en) * 2023-04-28 2025-04-11 重庆大学 Range extender fault detection model construction method, fault detection and diagnosis method and maintenance method

Similar Documents

Publication Publication Date Title
CN109782603A (en) Detection method and monitoring system of rotating machinery coupling fault
CN112417954B (en) A bearing fault mode diagnosis method and system for small sample data sets
CN110110768B (en) Rolling bearing fault diagnosis method based on parallel feature learning and multiple classifiers
WO2022037068A1 (en) Method for diagnosis of fault in machine tool bearing
CN111046945B (en) Fault type and damage degree diagnosis method based on combined convolutional neural network
CN112257530B (en) Rolling bearing fault diagnosis method based on blind signal separation and support vector machine
CN111967486A (en) Complex equipment fault diagnosis method based on multi-sensor fusion
CN109655259A (en) Combined failure diagnostic method and device based on depth decoupling convolutional neural networks
CN113469060A (en) Multi-sensor fusion convolution neural network aeroengine bearing fault diagnosis method
CN112665852B (en) Variable working condition planetary gearbox fault diagnosis method and device based on deep learning
CN112461543B (en) Rotary machine fault diagnosis method based on multi-classification support vector data description
CN114778112A (en) Audio Identification and Fault Diagnosis Method for Mechanical Faults of Wind Turbine Generators
CN110334764A (en) Intelligent fault diagnosis method for rotating machinery based on integrated deep autoencoder
CN114818774A (en) Intelligent gearbox fault diagnosis method based on multi-channel self-calibration convolutional neural network
CN114034481A (en) Fault diagnosis system and method for rolling mill gearbox
CN110297479B (en) A fault diagnosis method for hydroelectric units based on convolutional neural network information fusion
CN113916535B (en) Bearing diagnosis method, system, equipment and medium based on time frequency and CNN
CN116593157A (en) Gear fault diagnosis method for complex working conditions based on matching meta-learning under few samples
CN116358871A (en) Composite fault diagnosis method for weak signals of rolling bearings based on graph convolutional network
CN115758233A (en) Fault diagnosis method for main transmission gear box of high-speed precision strip steel cold rolling mill
CN111783941B (en) Mechanical equipment diagnosis classification method based on probability confidence convolutional neural network
CN117708574B (en) CNN variable-speed rolling bearing fault diagnosis method embedded with physical information
CN117786549A (en) Speed reducer health state evaluation method, system, equipment and medium based on combined weighting method
CN115808305A (en) Intelligent fault diagnosis method for multi-scale gearbox
Guo et al. A Novel Bearing Fault Diagnosis Method Based on the DLM-CNN Framework

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
RJ01 Rejection of invention patent application after publication

Application publication date: 20190521

RJ01 Rejection of invention patent application after publication