CN112101532B - Self-adaptive multi-model driving equipment fault diagnosis method based on edge cloud cooperation - Google Patents
Self-adaptive multi-model driving equipment fault diagnosis method based on edge cloud cooperation Download PDFInfo
- Publication number
- CN112101532B CN112101532B CN202011289732.9A CN202011289732A CN112101532B CN 112101532 B CN112101532 B CN 112101532B CN 202011289732 A CN202011289732 A CN 202011289732A CN 112101532 B CN112101532 B CN 112101532B
- Authority
- CN
- China
- Prior art keywords
- model
- diagnosis
- cloud
- edge
- branch
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- 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
- G01—MEASURING; TESTING
- G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
- G01M13/00—Testing of machine parts
- G01M13/04—Bearings
- G01M13/045—Acoustic or vibration analysis
-
- 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
- G06N3/044—Recurrent networks, e.g. Hopfield 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
- G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Acoustics & Sound (AREA)
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
Abstract
The invention discloses a self-adaptive multi-model driving equipment fault diagnosis method based on edge cloud cooperation. Firstly, the invention provides a time tolerance factor on-demand multi-model branch selection method, a plurality of diagnosis model branches are arranged, and a branch model with the highest accuracy is selected within the range of the time tolerance factor; further adopting a diagnosis model division method based on edge cloud cooperation to divide a fault diagnosis model based on deep learning between edge clouds by taking a layer as granularity; and finally, providing a cross-working-condition diagnosis method based on edge cloud cooperation, training a general working condition model by the cloud end, issuing the general working condition model to the edge end, and diagnosing the data of the individualized working condition by the edge end. Aiming at the problems that the cloud computing fault diagnosis mode is insufficient in instantaneity, the computing power resources and the storage capacity of edge equipment are limited, and a deep learning fault diagnosis model cannot be directly deployed, the traditional cloud computing diagnosis mode is improved in a side-cloud cooperation mode, time delay is effectively reduced, and cross-working-condition diagnosis is achieved.
Description
Technical Field
The invention relates to the technical field of fault diagnosis of industrial equipment and edge cloud cooperation, in particular to a fault diagnosis method of self-adaptive multi-model driving equipment based on edge cloud cooperation.
Background
With the development of artificial intelligence, internet of things and industrial internet technology, the industrial manufacturing industry is moving to digitization and intellectualization, and mechanical equipment is also moving towards increasingly complex and integrated directions. Once the precise key mechanical parts with complex structures break down, the normal operation of mechanical equipment is seriously influenced, and great loss is caused. Therefore, the predictive diagnosis and prevention of the failure of large-scale machinery is an important subject for the development of the industrial manufacturing industry.
The traditional fault diagnosis method mainly comprises a manual mode feature extraction method, a signal processing method and a deep learning method. The manual extraction of the signal features has low efficiency and cannot deal with rapidly-increased massive industrial data; the signal processing method mainly comprises the methods of wavelet transformation, Fourier transformation, empirical mode decomposition and the like, data features are extracted through data processing, and then the feature vector of the vibration signal is obtained to carry out fault classification diagnosis. However, the method cannot extract deep abstract features among multiple fault modes, and is not suitable for processing massive complex heterogeneous industrial equipment data. In order to solve the problems, some domestic and foreign scholars propose a fault diagnosis method based on deep learning, a deep neural network can model complex and heterogeneous industrial data with high time-varying property, multi-dimensional nonlinearity and the like, complex data transformation and feature extraction are not needed, and an original vibration signal can be directly used to realize end-to-end fault diagnosis. As a typical deep learning model, the Convolutional Neural Network (CNN) is very suitable for processing nonlinear and non-stationary signals, and has a great application prospect in the field of fault diagnosis. An article [ Liqiang, data-driven fault diagnosis method based on deep convolutional neural network research [ D ]. Shandong university, 2018 ]) provides a data-driven fault diagnosis method based on EHHT-CNNs, and automatic extraction of deep features of data is achieved without expert experience. An intelligent rolling bearing composite fault diagnosis method based on MWT and CNN [ J ] mechanical transmission, 2016(12) ] proposes a fault diagnosis method based on wavelet transformation and CNN, wherein the original vibration signals are processed by the wavelet transformation, and feature maps converted after processing are input into the CNN for fault identification and classification. The deep learning algorithm realizes efficient extraction of data features and high-precision diagnosis, however, most data in industrial production are time-sequence data, correlation between previous data and subsequent data is needed, and the algorithm does not consider the problem. A long-short term memory network (LSTM) has a memory function and is suitable for processing time sequence data, and an article [ Tangsai, bearing fault diagnosis algorithm research [ D ] based on the long-short term memory network, Chongqing university, 2018 ] provides a bearing fault diagnosis method based on the LSTM, and high-precision diagnosis is realized from historical fault data of a system. However, the diagnosis method based on the LSTM has the problem of frequency domain information loss, and for the above problem, an article [ yellow business chang, bearing fault state identification method based on deep learning research [ D ]. science and technology university in china, 2019 ] proposes a bearing fault diagnosis method based on a time-frequency image and 2DCNN, which converts an original signal into a time-frequency signal image and inputs the time-frequency image into a model for identification and classification. The problem of frequency domain information loss in bearing fault diagnosis of the LSTM model is solved, and the diagnosis model has stronger robustness. In summary, the current device fault diagnosis method based on the deep neural network has become a research hotspot.
With the exponential increase of equipment state data, a centralized cloud computing mode can cause larger end-to-end time delay and energy consumption, is not beneficial to data privacy protection, and cannot meet the requirements of real-time performance, reliability and safety in the field of industrial equipment fault diagnosis; the edge computing mode can well solve the problems of the cloud computing mode, but because the edge computing power and the storage capacity are limited, a deep neural network-based diagnostic model cannot be directly deployed. These problems present a significant challenge to industrial equipment fault diagnosis.
Disclosure of Invention
In order to solve the problems in the prior art, the invention provides a fault diagnosis method for self-adaptive multi-model driving equipment based on edge cloud cooperation. Firstly, a diagnostic model partitioning method based on edge cloud cooperation is adopted to partition the model among edge clouds by taking layers as granularity, so that the model diagnosis time delay is reduced; secondly, a time tolerance factor multi-model branch selection method is provided, a plurality of diagnosis branches are arranged, and the diagnosis accuracy and the execution time of the diagnosis model gradually increase along with the gradual increase of the layer number and the complexity of the diagnosis model of the plurality of branches. And the model branches are selected in a self-adaptive manner according to the time tolerance factor, so that the balance between the execution time and the accuracy of the diagnosis model is realized, and the tolerance requirement of a user on the diagnosis time is met. And finally, realizing edge cloud data cooperation and cross-working condition diagnosis.
In order to solve the technical problems, the technical scheme adopted by the invention is as follows:
a self-adaptive multi-model driving equipment fault diagnosis method based on edge cloud cooperation comprises the following steps:
s1: the cloud firstly trains a diagnosis model of a general working condition and sends the diagnosis model to the edge terminal to serve as the edge terminal diagnosis model, so that cross-working condition diagnosis is realized. The system deploys deep learning algorithm models at the cloud end and the edge end, and establishes a regression prediction model based on logarithm for different types of layers in the models so as to predict service response time delay of diagnosis model branches;
s2: after the edge terminal obtains the user time tolerance factor value, self-adaptive multi-model selection is carried out, and a diagnosis model branch decision maker selects a model branch which meets the user time tolerance factor and has the highest accuracy and obtains a corresponding edge cloud division point;
s3: according to the division point of the selected model branch, the edge end firstly executes the front end part of the branch model, after the intermediate result is uploaded to the cloud end, the cloud end executes the rear end part of the model, and the final diagnosis result is returned to the edge end;
s4: the edge terminal uploads and gathers the data set of the personalized working conditions to a cloud terminal to form a cloud terminal data set; the cloud retrains the model periodically, optimizes and adjusts the weight of the model, and improves the diagnosis precision and generalization of the model.
Further, in step S1, three diagnostic branch models, which are all fault diagnostic models based on deep learning, are designed first, and the number of layers and complexity of the diagnostic models of the three branches are gradually increased, so that the diagnostic accuracy and execution time of the branch models are gradually increased. The Branch1 (Branch 1) model employs four convolutional layers, the Branch2 (Branch 2) model employs five convolutional layers, and the Branch3 (Branch 3) model employs three convolutional layers and three LSTM networks.
Further, in step S2, the method for selecting multiple models using time tolerance factors includes the following steps:
1-1) establishing a logarithm-based regression model at a cloud end and an edge end according to the type of a network layer, predicting the execution time of each layer of a diagnosis model according to the regression model, and summing and accumulating to obtain the total execution time of the model;
1-2) before selecting the multiple models, predicting the execution time of the 3 branch models by the regression model established in the 1-1);
1-3) receiving a diagnosis time tolerance factor value of a user as a basis for selecting a diagnosis model;
1-4) selecting a branch with a prediction delay slightly lower than but closest to the time tolerance factor according to the user time tolerance factor so as to maximize the accuracy of the diagnosis model within the range of the time tolerance factor.
Further, in step S3, a diagnostic model partitioning method based on edge cloud cooperation is adopted to partition the fault diagnostic model based on deep learning between edge clouds with a layer as a granularity, and the specific steps are as follows:
2-1) analyzing data and computational characteristics of each layer of the diagnostic model by taking the layer as granularity, and unloading the layer with longer execution time or higher computational resource consumption to a cloud so as to reduce end-to-end waiting time;
2-2) the output data amount of the simultaneous division point layer should be small to reduce the data transmission amount;
2-3) executing the layer before the division point by the edge end, and uploading the intermediate result to the cloud end after the execution is finished;
2-4) the rear end part of the cloud execution model, after execution is finished, a diagnosis result is returned to the edge end, and diagnosis is finished;
adopt the produced beneficial effect of above-mentioned technical scheme to lie in:
the invention provides a self-adaptive multi-model driving equipment fault diagnosis method based on edge cloud cooperation, which is used for diagnosing equipment faults by adopting the idea based on the edge cloud cooperation aiming at the problems that the real-time performance of a cloud computing fault diagnosis mode is insufficient, the computing power resources and the storage capacity of edge equipment are limited, and a deep learning fault diagnosis model cannot be directly deployed.
The equipment fault diagnosis method provided by the invention is called as an adaptive multi-model drive surface diagnosis method (MFDE for short) based on edge cloud cooperation. In the method, a cloud end with stronger computing resources and storage resources undertakes a public computing task with larger computing capacity and is responsible for mining analysis of mass data and training and updating of a model; the edge terminal is responsible for collecting and processing the data of the equipment terminal and making real-time service response, so that the requirements that the equipment fault diagnosis model based on deep learning needs larger computing power and storage resources and has good real-time diagnosis performance are met.
Compared with the traditional equipment fault diagnosis method, the invention has the advantages that: (1) by adopting a diagnostic model dividing method based on edge cloud cooperation, a fault diagnostic model based on deep learning is divided among edge clouds by taking layers as granularity, so that the data transmission quantity is reduced, and the model diagnosis delay is reduced; (2) providing a time tolerance factor on-demand multi-model branch driving method, setting a plurality of diagnosis model branches, and selecting a branch model with the highest accuracy within the time tolerance factor range to realize the balance of the execution time and the accuracy of the diagnosis model and meet the tolerance requirement of a user on the diagnosis time; (3) and providing a cross-working-condition diagnosis method based on edge cloud cooperation, and realizing the edge cloud data cooperation and the cross-working-condition diagnosis.
The method provided by the invention is applied to bearing fault diagnosis, and the effectiveness of the provided fault diagnosis method in the aspects of reducing time delay and cross-working condition diagnosis is verified through test analysis; in the aspect of execution time, compared with a fault diagnosis method based on cloud computing, the fault diagnosis delay is reduced to a certain extent; in the aspect of diagnosis accuracy, the 3 diagnosis branch models are superior to the traditional BP and RNN diagnosis models; meanwhile, after the branch model is issued to the edge end from the cloud end, a better diagnosis effect is shown under each edge node test set, the diagnosis accuracy difference of each edge node is smaller, and a better cross-working condition diagnosis effect is achieved.
Drawings
FIG. 1 is an overall framework diagram of a fault diagnosis method for an adaptive multi-model driving device based on edge cloud coordination;
FIG. 2 is a flow chart of a fault diagnosis method for an adaptive multi-model driving device based on edge cloud coordination;
FIG. 3 is a diagram of a multi-branch diagnostic model architecture;
FIG. 4 is a side cloud collaborative cross-condition diagnostic overview framework;
FIG. 5 is a comparison of accuracy versus loss values for diagnostic branch model training of FIG. 1;
FIG. 6 is a comparison of accuracy versus loss values for diagnostic branch model training of FIG. 2;
FIG. 7 is a comparison of accuracy of diagnostic branch model validation with loss values of FIG. 1;
FIG. 8 is a comparison of accuracy of diagnostic branch model validation with loss values FIG. 2;
FIG. 9 is a graph of model test accuracy versus execution delay of FIG. 1;
FIG. 10 is a graph of model test accuracy versus execution delay of FIG. 2;
FIG. 11 is a graph of the effect of classification of a confusion matrix diagnostic branch model FIG. 1;
FIG. 12 is a graph of the effect of classification of a confusion matrix diagnostic branching model FIG. 2;
FIG. 13 is a graph of the effect of classification of a confusion matrix diagnostic branching model FIG. 3;
FIG. 14 is a diagnostic branch model middle layer visualization FIG. 1;
FIG. 15 is a diagnostic branch model middle layer visualization FIG. 2;
FIG. 16 is a diagnostic branching model middle layer visualization FIG. 3;
FIG. 17 is a diagnostic total latency graph 1 of a branch diagnostic model;
FIG. 18 is a diagnostic total latency graph 2 of a branch diagnostic model;
FIG. 19 is a diagnostic total latency graph 3 of a branch diagnostic model;
FIG. 20 is a graph of diagnostic accuracy for a single-case training branch diagnostic model;
FIG. 21 is a graph of diagnostic accuracy for a partial mixed-regime training branch diagnostic model;
FIG. 22 is a graph of diagnostic accuracy for a fully mixed-regime trained branch diagnostic model;
FIG. 23 is a Branch1 model network architecture parameter diagram;
FIG. 24 is a Branch2 model network architecture parameter diagram;
FIG. 25 is a Branch3 model network architecture parameter diagram;
FIG. 26 is a graph of independent variables in a regression model;
FIG. 27 is a parameter diagram of an edge termination platform configuration;
FIG. 28 is a cloud platform configuration parameter graph;
FIG. 29 is a bearing failure experimental data sample information plot;
FIG. 30 is a graph of Branches test accuracy versus execution time;
FIG. 31 is a graph of diagnostic accuracy for a single-case trained Branches model;
FIG. 32 is a graph of diagnostic accuracy for a part of a mixed regime trained Branches model;
FIG. 33 is a graph of diagnostic accuracy for a fully mixed-regime trained Branches model.
Detailed Description
The present invention will be described in further detail with reference to the accompanying drawings and specific embodiments.
The MFDEC method provided by the invention is designed by adopting the idea of edge cloud cooperation as a whole. The cloud has the following characteristics: the computing power and the storage resources are rich, and the complex computing task can be quickly executed and mass data samples can be stored. However, the centralized cloud computing fault diagnosis mode needs to upload mass equipment state data and cannot meet the real-time diagnosis requirement; the edge end has the following characteristics: the method is closer to a data source, has good real-time response and is more suitable for personalized service customization. However, the computation resources and storage resources of the edge end are limited, and the large-scale deep learning diagnosis model cannot be directly executed at the edge end. By combining the advantages and the disadvantages of the cloud end and the edge end, the invention adopts a side-cloud cooperation-based mode to diagnose equipment faults. The overall framework diagram is shown in fig. 1, and the main work is completed by the cooperation of the cloud end and the edge end. The whole body is divided into three parts, namely an equipment end, an edge end and a cloud end.
The equipment terminal is mainly responsible for uploading vibration signal data to the edge terminal and receiving a diagnosis result returned by the edge terminal.
And after the edge terminal obtains the user time tolerance factor value, selecting the branch model which meets the user time tolerance factor value and has the maximum accuracy and obtaining the corresponding edge cloud dividing point. And then, loading the model weight of the corresponding branch sent by the cloud, inputting the edge node test data, and executing the layer before the division point of the selected model branch. And after the execution is finished, uploading the intermediate result of the division point layer to the cloud.
The cloud task is mainly divided into two parts, wherein the first part trains a general working condition diagnosis model with strong generalization performance by utilizing self-abundant computing resources and massive training data, and the model is finally issued to an edge end to serve as an edge end diagnosis model; and after receiving the intermediate result of the division point layer, the cloud end of the second part executes the layer behind the branch division point of the selected diagnostic model, wherein the part belongs to the layer with higher computational resource consumption in the model. And after the execution is finished, the diagnosis result is returned to the edge equipment from the cloud.
The MFDEC method of the present invention, a flowchart of which is shown in fig. 2, includes the following steps:
and S1, the cloud end firstly trains the diagnosis model of the general working condition and sends the diagnosis model to the edge end to be used as the edge end diagnosis model, so that cross-working condition diagnosis is realized. The system deploys deep learning algorithm models at the cloud end and the edge end, and establishes a regression prediction model based on logarithm for different types of layers in the models so as to predict service response time delay of diagnosis model branches;
the diagnosis models adopted by the invention are three fault diagnosis branch models based on deep learning, and the number of layers and the complexity of the diagnosis models of the three branches are gradually increased, so that the diagnosis accuracy and the execution time of the branch models are gradually increased. The Branch model structure is shown in fig. 3, the Branch1 (Branch 1) model adopts four convolutional layers, the Branch2 (Branch 2) model adopts five convolutional layers, and the Branch3 (Branch 3) model adopts three convolutional layers and three LSTM networks. The first-layer CNN network of the Branch1 (Branch 1) model adopts a convolution kernel with a larger size, and the rest CNN networks adopt convolution kernels with smaller sizes to acquire local features so as to improve the diagnosis precision; the Branch2 (Branch 2) model is subjected to parameter adjustment, the number of CNN network layers is increased, and the accuracy of model diagnosis is further improved; branch3 (Branch 3) adopts two CNN networks to perform characteristic preprocessing, and the output of the CNN network at the third layer is used as the input of the LSTM, so that the accuracy of fault diagnosis is further improved. Fig. 23 shows the network configuration parameters of Branch1 (Branch 1), fig. 24 shows the network configuration parameters of Branch2 (Branch 2), and fig. 25 shows the network configuration parameters of Branch3 (Branch 3).
The specific process of the edge cloud data collaborative cross-working-condition diagnosis method provided by the invention is described as follows, and an overall frame of edge cloud collaborative cross-working-condition diagnosis is shown in fig. 4.
The edge terminal collects a data set of individual working conditions, the working conditions of the data of different edge terminals are different, and each edge terminal only has equipment state data of one working condition. And the data set of the cloud end is uploaded and gathered by each edge end data, so that the cloud end has data of all working conditions.
By utilizing the abundant computing resources and mass training data, the cloud end can train a general working condition diagnosis model with strong generalization, and the model can be finally issued to the edge end to serve as the diagnosis model. In addition, the cloud can retrain the diagnosis model regularly, revise the model parameter to improve diagnosis precision and the generalization of model.
The edge terminal is responsible for collecting real-time vibration signal data of the individualized working condition to form an edge terminal data set. During fault diagnosis, the edge end loads the universal working condition diagnosis model issued by the cloud as a diagnosis model, and carries out real-time diagnosis on data of individualized working conditions, so that cross-working condition diagnosis is realized. In addition, the personalized working condition sample data sets can be uploaded to the cloud end from the edge end periodically so as to enrich the cloud end data sets.
S2: in order to meet the tolerance requirement of a user on diagnosis time, a time tolerance factor on-demand multi-model branch selection method is provided. After the edge terminal obtains the user time tolerance factor value, self-adaptive multi-model selection is carried out, and a diagnosis model branch decision maker selects a model branch which meets the user time tolerance factor and has the highest accuracy and obtains a corresponding edge cloud division point;
the steps of the time tolerance factor on-demand multi-model branch selection method provided by the invention are described as follows.
1-1) establishing a logarithm-based regression model at the cloud end and the edge end according to the type of the network layer, wherein independent variables determining the time delay of each type of network layer in the regression model are shown in figure 26, the execution time of each layer of the diagnosis model is predicted according to the regression model, and the total execution time of the model is obtained after summation and accumulation. In addition, the diagnosis delay of the diagnosis model also comprises model loading delays at the cloud end and the edge end;
1-2) before selecting multiple models, predicting the execution time of 3 branch models by the regression model established in 1-1), wherein the formula is as follows:
predictTime=
wherein there are 3 model branches in total, andia branch model has
A layer of a material selected from the group consisting of,pexpressing the index value of the division point layer, predicting the operation time delay of each network layer by using a regression model established by 1-1),
is shown asiThe first of the model branchesjA layer of a material selected from the group consisting of,
is shown asiThe first of the model branches
The layer (the last layer) is,
is shown asjThe network layer with the layer deployed in the cloud is predicted to delay,
is shown asjThe network layer with the layer deployed at the edge end is predicted time delay.
Layer is edge cloud division point layeriIs branched topThe data output of the layer is carried out,Bwhich represents the bandwidth of a particular network,Inputwhich represents the input of the diagnostic data and,predictTimerepresenting the execution time prediction value of the current model branch.
1-3) receiving a diagnosis time tolerance factor value of a user as a basis for selecting a plurality of diagnosis model branches;
1-4) selecting a branch with a prediction delay slightly lower than but closest to the time tolerance factor according to the user time tolerance factor so as to maximize the accuracy of the diagnosis model within the range of the time tolerance factor.
S3, according to the division points of the selected model branches, the edge end executes the front end part of the branch model at first, after the intermediate result is uploaded to the cloud end, the cloud end executes the rear end part of the model, and the final diagnosis result is returned to the edge end;
the diagnostic model partitioning method based on edge cloud cooperation provided by the invention has the following steps.
2-1) analyzing data and computational characteristics of each layer of the diagnostic model by taking the layer as granularity, and unloading the layer with longer execution time or higher computational resource consumption to the cloud so as to reduce end-to-end waiting time. In terms of hierarchical delay, the convolutional layer and the fully-connected layer are layers with higher delay; in the aspect of output data size, the convolutional layer has the largest output data size due to the fact that a filter is adopted to extract a large number of features, the output data size after passing through the active layer is kept unchanged, the output data size after passing through the full connection layer is gradually reduced, and the output data size can be remarkably reduced through the pooling layer;
2-2) the amount of output data of the simultaneous division point layer should be small to reduce the amount of data transmission. Because the execution time delay of the pooling layer is small and the output data amount is small, the last pooling layer is finally selected as a dividing point layer;
2-3) executing the layer before the division point by the edge end, and uploading the intermediate result to the cloud end after the execution is finished;
2-4) the cloud executes the layers behind the division point layer, the layers have longer execution time or higher computational resource consumption, and after the execution is finished, the diagnosis result is returned to the edge end, and the diagnosis is finished;
s4, uploading and summarizing the data set of the personalized working conditions to a cloud end by the edge end to form a cloud end data set; the cloud retrains the model periodically, optimizes and adjusts the weight of the model, and improves the diagnosis precision and generalization of the model.
Based on the steps, the MFDEC method provided by the invention firstly adopts a diagnostic model division method based on edge cloud cooperation to divide the model between edge clouds by taking layers as granularity, so that the model diagnosis time delay is reduced; secondly, a time tolerance factor multi-model branch selection method is provided, a plurality of diagnosis branches are arranged, and the diagnosis accuracy and the execution time of the diagnosis model gradually increase along with the gradual increase of the layer number and the complexity of the diagnosis model of the plurality of branches. And the model branches are selected in a self-adaptive manner according to the time tolerance factor, so that the balance between the execution time and the accuracy of the diagnosis model is realized, and the tolerance requirement of a user on the diagnosis time is met. Finally, edge cloud data cooperation is realized, and cross-working condition diagnosis is realized. The invention effectively reduces the diagnosis delay and meets the tolerance requirement of the user on the diagnosis time on the premise of ensuring the accuracy of the fault diagnosis model. In addition, cross-working condition fault diagnosis is realized.
The invention discloses a test verification of a self-adaptive multi-model driving equipment fault diagnosis method based on edge cloud cooperation, which comprises the following steps:
the rolling bearing is a key and easily-damaged precise mechanical part, and the rolling bearing is taken as a research object to perform experimental demonstration on the self-adaptive multi-model driving equipment fault diagnosis method based on edge cloud cooperation.
1. Test environment
The experimental platform set up by the invention consists of a cloud end and an edge end. The configuration parameters of the edge platform are shown in fig. 27, and the configuration parameters of the cloud platform are shown in fig. 28.
The deep learning framework adopts Tensorflow and keras, a real-time fault diagnosis program with cooperation of a cloud end and an edge end is realized by using python, communication between the cloud end and the edge end is realized by using an open-source RPC interface triple, a plurality of edge nodes are simulated by operating systems mounted by a plurality of Vmware virtual machines, and the edge nodes are connected in the same local area network. After the edge terminal selects the model branch according to the time tolerance factor and determines the division point, the edge terminal executes the front end part of the model, the output of the division point layer is uploaded to the cloud server, the cloud server executes the rear end part of the model, and the diagnosis result is sent back to the edge terminal.
2. Description of data
The data set adopted in the test is a public data set of the university of Keyssjohn, and comprises normal data, Inner circle (Inner radius Fault) Fault data, Outer circle (Outer radius Fault) Fault data and rolling element (Ball Fault) Fault data. The failure diameters are respectively 0.007 inches, 0.014 inches and 0.021 inches, the total number of the four rotating speeds is 1730r/min (3 HP), 1750r/min (2 HP), 1772r/min (1 HP) and 1797r/min (0 HP), and data collected at the four rotating speeds are used as data of four working conditions. Each operating condition (rotational speed) data contains ten failure type status data, which are normal status, inner ring failure-diameter 0.007 inches, outer ring failure-diameter 0.007 inches, rolling element failure-diameter 0.007 inches, inner ring failure-diameter 0.014 inches, outer ring failure-diameter 0.014 inches, rolling element failure-diameter 0.014 inches, inner ring failure-diameter 0.021 inches, outer ring failure-diameter 0.021 inches, and rolling element failure-diameter 0.021 inches status data, respectively. Experimental data sample information is shown in fig. 29.
Test one: classification diagnosis effect of multi-branch diagnosis model
Original vibration signal data under ten different states are respectively marked by 0-9, a data set is divided into a training set and a test set according to the dividing ratio of 3:1, the training set is further divided into the training set and a verification set according to the ratio of 3:1, the training set, the verification set and the test set are randomly disordered, and the randomness of data selection is guaranteed. When the model is trained, a training set is divided into a plurality of batches, data of one batch is input in each iteration training, the cross entropy of softmax is used as a loss function, and Adam is used as an optimizer to adjust training parameters. And then calculating the loss value and the accuracy in the training process.
And respectively training the three branch models at the cloud, wherein the training data set is a mixed data set of 0,1,2 and 3HP working conditions, and the ratio of the training set to the verification set is 3: 1. In addition, in order to verify the diagnostic effect of the three branch models provided by the invention, the diagnosis is compared with the traditional neural network BP and RNN diagnostic models. Fig. 5, 6, 7, and 8 show the training and validation accuracy and loss values of the BP, RNN, Branch1, Branch2, and Branch3 models as a function of iteration rounds during the training process.
The results show that the model has already tended to converge over 50 iterations. The accuracy of the BP diagnosis model is about 74%, and the loss value is about 0.72; the accuracy of the RNN diagnostic model is about 90%, and the loss value is about 0.3; the accuracy rates of the Branch1, Branch2 and Branch3 diagnosis models are all above 96%, and the loss values are all below 0.10, so that the 3 diagnosis Branch models provided by the invention are superior to the traditional BP and RNN diagnosis models in fault diagnosis.
The classical BP network model, the RNN network model, and the 3 diagnostic branch models are tested on the test set 10 times, and the average value is taken to obtain the accuracy and execution time of each branch model, as shown in fig. 30.
Fig. 9 and 10 show that: the test accuracy of the 3 branch models designed by the invention is higher than that of the traditional BP and RNN diagnostic models, the diagnostic accuracy is higher, the test accuracy and the execution time of the 3 branch models are gradually increased, and the balance between the execution time and the accuracy of the diagnostic models is realized.
The confusion matrices shown in fig. 11, 12 and 13 respectively show the classification diagnosis effect of the diagnosis Branch model on the test set samples, wherein the test set classification accuracy of Branch1 is 98.52%, the test set classification accuracy of Branch2 is 98.91%, and the test set classification accuracy of Branch3 is 99.23%.
Next, fig. 14, 15 and 16 will visually show the variation tendency of the vibration signal data through the middle layers of the Branch1, Branch2 and Branch3, respectively. In the figure, the input original vibration signal data is firstly displayed visually, and then the output results of the original vibration signal data passing through the CNN layer, the LSTM layer and the FC layer are displayed.
And (2) test II: total time delay contrast test based on cloud computing fault diagnosis and MFDEC method provided by the invention
The 3 Branch diagnostic models of Branch1, Branch2 and Branch3 are taken as experimental models, and the delays of the three diagnostic Branch models in a cloud computing mode and the MFDEC method provided by the invention are respectively researched under the bandwidth of 8000 kbs.
In the aspect of total time delay, when reasoning diagnosis is carried out in a cloud computing mode, firstly, the equipment side uploads data to the cloud side, then the cloud side loads the model weight for diagnosis, and finally, a diagnosis result is returned to the equipment side. The total delay of the cloud reasoning diagnosis is composed of the three parts of time; when reasoning diagnosis is carried out in the edge cloud collaborative model division mode, the time delay comprises the time of the front end part of the edge end execution model, the time of the edge end uploading a model division point layer output result to the cloud end, the time of the rear end part of the cloud end execution model and the time of the diagnosis result returning to the edge end from the cloud end. The time delay of each part of the two diagnosis modes is classified and summarized, and the following results can be obtained: the total time delay of the diagnosis mode based on cloud computing and the MFDEC provided by the invention is composed of two time delays of the diagnosis model calculation time delay and the data transmission time delay.
Fig. 17, fig. 18 and fig. 19 show the total delay of the 3 model branches in two modes and the model calculation delay and data transmission delay in each mode, respectively. For the Branch1 model, the total latency in the cloud-based computing mode was 4.99s, with the model computation time accounting for 24.15% of the total latency. The total time delay of the MFDEC method provided by the invention is 4.34s, wherein the model calculation time accounts for 36.02% of the total time delay; for the Branch2 model, the total latency in the cloud-based computing mode was 5.44s, with the model computation time accounting for 25.43% of the total latency. The total time delay of the MFDEC method provided by the invention is 4.53s, wherein the calculation time of the model accounts for 39.51% of the total time delay; for the Branch3 model, the total latency in the cloud-based computing mode was 10.77s, where the model computation time was 27.76% of the total latency. The total time delay of the MFDEC method provided by the invention is 8.49s, wherein the model calculation time accounts for 43.59% of the total time delay.
As shown in fig. 17, 18 and 19, the experiment can be concluded that: the total time delay of the MFDEC method provided by the invention in 3 diagnosis model branches is lower than that of a diagnosis method based on cloud computing. This is due to: for the diagnosis method based on cloud computing, the cloud end is responsible for executing all layers of the model, the total execution time of the model is less than that of the MFDEC method provided by the invention due to rich computing resources, and the data transmission quantity is much higher than that of the MFDEC method provided by the invention; for the MFDEC method provided by the invention, the cloud end and the edge end both participate in the execution of the model, the cloud end is responsible for the layer with higher execution time delay and larger calculated amount, although the execution time of the model is slightly higher than that of the diagnosis method based on cloud computing, the transmitted data amount is far smaller than that of the diagnosis method based on cloud computing, and the data transmission time accounts for a larger time delay in the total time delay, so the total time delay is smaller than that of the diagnosis method based on cloud computing. In addition, the end-to-end time delay of the two diagnosis methods comprises model calculation time and data transmission time, and the proportion of the data transmission time to the total time delay is larger.
The following conclusions can be drawn: compared with the traditional cloud computing diagnosis mode, the MFDEC method provided by the invention can reduce the data transmission quantity and reduce the model diagnosis time delay.
And (3) test III: edge cloud data collaborative cross-working condition diagnosis contrast test
The experimental data are equipment state data under four working conditions, and the loads corresponding to different working conditions are 1730r/min (3 HP), 1750r/min (2 HP), 1772r/min (1 HP) and 1797r/min (0 HP) respectively. In the MFDEC method provided by the invention, the edge end has a data set of individual working conditions, the working conditions of the data acquired by different edge ends are different, and each edge end only has equipment state data of one working condition. The data set of the cloud is uploaded and summarized from the data of each edge end, so that the data of all four working conditions are possessed, namely the training data of the diagnosis model are the data of all four working conditions of the cloud.
In order to verify that the MFDEC method provided by the invention has strong cross-working-condition diagnosis capability, three tests are combined with data of various working conditions for experimental comparison and verification. Firstly, the training data of the cloud 3 branch diagnosis models only comprises a working condition, namely 1HP, namely, the data under the working condition of 1HP is adopted for model training, and the training accuracy is obtained. Then, after the cloud branch model is issued to each edge end, the test is performed on each edge data set for 10 times, and the average value is taken to obtain the accuracy of the test as shown in fig. 31 and fig. 20.
From fig. 20, it can be derived that: the accuracy of the 3 branch diagnosis models under the edge node 2 test set is similar to that of the model training, and the accuracy under the other edge node test sets is lower than that of the cloud model training, because the training data of the cloud 3 branch diagnosis models only contain data of 1HP working conditions, namely the working conditions of the edge node 2 data. This gives: if the cloud model training set data only contains one working condition, the model has a good diagnosis effect only on the edge equipment containing the working condition, and the diagnosis effect is general on the edge equipment containing other working conditions.
On the basis of the experiment, the training data of the cloud 3 branch diagnosis model only comprises two working conditions, namely 1HP and 2HP, namely the 1HP working condition data and the 2HP working condition data are used for carrying out cloud model training, and the training accuracy is obtained. After the cloud model is issued to each edge end, the test is performed on each edge data set for 10 times, and the average value is taken to obtain the accuracy of the test as shown in fig. 32 and 21.
It is shown in fig. 21 that the accuracy of the 3 branch diagnostic models in the edge node 2 and edge node 3 test sets is similar to the accuracy of the model training, and the accuracy in the other edge node test sets is lower than the accuracy of the cloud model training, because the training data of the cloud 3 branch diagnostic models only include data of 1HP and 2HP operating conditions, that is, the operating conditions of the data of the edge node 2 and the edge node 3. This gives: if the model training set data only contains two working conditions, the cloud model has a good diagnosis effect only on the edge devices containing the two working conditions, and the diagnosis effect is general on the edge devices containing other working conditions.
In the MFDEC method provided by the invention, the training data of the diagnosis model are the data of all four working conditions of the cloud. Finally, model training is performed on the cloud by using data under all four working conditions, then the cloud model is issued to each edge end, testing is performed on each edge data set for 10 times, and an average value is taken to obtain the accuracy of the testing as shown in fig. 33 and 22.
As can be seen from fig. 22, in the MFDEC method provided by the present invention, 3 branch diagnostic models all exhibit a better diagnostic effect under each edge node test set, and the difference between the diagnostic accuracy of each edge node is small, and is all over 97%. This gives: the cloud diagnosis model branches have a good diagnosis effect on each edge device with different working conditions.
Through experimental comparison, the following conclusions can be drawn: the more data working condition types contained in the cloud training set, the more sufficient the diagnostic model learns the distribution rule of the fault data, the better the universality of the model is, the better the diagnostic effect of the model after the model is issued to each edge end is, and the stronger the cross-working condition diagnostic capability is. In the MFDEC method provided by the invention, all the edge node data are uploaded to the cloud and used as cloud diagnosis model training data, so that the MFDEC method provided by the invention has strong cross-working condition diagnosis capability.
5. Conclusion
The invention provides a self-adaptive multi-model driving fault diagnosis method based on edge cloud cooperation. 3 diagnosis model branches are set, and the model branches are selected in a self-adaptive mode according to a time tolerance factor value provided by a user, so that the balance between the execution time and the accuracy of diagnosis is realized; during fault diagnosis, diagnosis model edge cloud division is performed, a layer with large calculation amount is executed by a cloud end with rich calculation resources, and the rest model layers are executed by the edge end, so that the diagnosis instantaneity is improved. The self-adaptive multi-model driving fault diagnosis method based on edge cloud cooperation is lower than the equipment fault diagnosis method based on cloud computing in the aspect of total time delay and has higher diagnosis accuracy. In addition, in order to realize cross-working-condition diagnosis, the invention provides a side cloud data collaborative cross-working-condition diagnosis method. The cloud end is responsible for training a diagnosis model of the general working condition and collecting data of each edge side as a training set; and the edge terminal acquires data of the individualized working condition and diagnoses by using the universal model of the cloud as a diagnosis model. The experimental results show that: the diagnosis accuracy of the test data of each edge node of the 3 diagnosis branch models is more than 96%, and the obtained diagnosis model has good cross-working condition diagnosis capability. Future work will further study how to make the diagnostic model trained by the laboratory environment data migrate to the actual industrial scene, and have better diagnostic effect.
Claims (3)
1. A self-adaptive multi-model driving equipment fault diagnosis method based on edge cloud cooperation is characterized by comprising the following steps: the method comprises the following steps:
s1: the system deploys deep learning algorithm models at a cloud end and an edge end, establishes a logarithm-based regression model at the cloud end and the edge end according to the type of network layers, predicts the execution time of each layer of a diagnosis model according to the regression model, and obtains the total execution time of the model after summation and accumulation, before multi-model selection, predicts the execution time of 3 branch models by the established regression model, wherein the 3 branch models are a branch model 1 adopting four convolution layers, a branch model 2 adopting five convolution layers and a branch model 3 adopting three convolution layers and three LSTM network layers;
s2: after the edge terminal obtains the user time tolerance factor value, self-adaptive multi-model selection is carried out, and a diagnosis model branch decision maker selects a branch with a prediction time delay slightly lower than but closest to the time tolerance factor so as to maximize the accuracy of the diagnosis model within the time tolerance factor range and obtain a corresponding edge cloud division point;
s3: according to the division point of the selected model branch, the edge end firstly executes the front end part of the branch model, after the intermediate result is uploaded to the cloud end, the cloud end executes the rear end part of the model, and the final diagnosis result is returned to the edge end;
s4: the edge terminal uploads and gathers the data set of the personalized working conditions to a cloud terminal to form a cloud terminal data set; the cloud retrains the model periodically, optimizes and adjusts the weight of the model, and improves the diagnosis precision and generalization of the model.
2. The method for diagnosing the fault of the self-adaptive multi-model driving equipment based on the edge cloud coordination as claimed in claim 1, wherein the method comprises the following steps: in step S2, a time tolerance factor multi-model branch selection method is adopted, and the steps are as follows:
1-1) establishing a logarithm-based regression model at a cloud end and an edge end according to the type of a network layer, predicting the execution time of each layer of a diagnosis model according to the regression model, and summing and accumulating to obtain the total execution time of the model;
1-2) before selecting the multiple models, respectively predicting the execution time of the 3 branch models by the regression model established in the 1-1);
1-3) receiving a diagnosis time tolerance factor value of a user as a basis for selecting a diagnosis model;
1-4) selecting a branch with a prediction delay slightly lower than but closest to the time tolerance factor according to the user time tolerance factor so as to maximize the accuracy of the diagnosis model within the range of the time tolerance factor.
3. The method for diagnosing the fault of the self-adaptive multi-model driving equipment based on the edge cloud coordination as claimed in claim 1, wherein the method comprises the following steps: in step S3, a diagnostic model partitioning method based on edge cloud coordination is adopted, and specifically:
the layers with long execution time or large computational resource consumption are unloaded to the cloud so as to reduce end-to-end waiting time, meanwhile, the output data volume of the division point layer is small so as to reduce data transmission volume, the layers before the division points are executed by the edge end firstly, after the execution is finished, intermediate results are uploaded to the cloud, the layers after the division point layer are executed by the cloud, the execution time of the layers is long or the computational resource consumption is large, and after the execution is finished, diagnosis results are returned to the edge end.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011289732.9A CN112101532B (en) | 2020-11-18 | 2020-11-18 | Self-adaptive multi-model driving equipment fault diagnosis method based on edge cloud cooperation |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011289732.9A CN112101532B (en) | 2020-11-18 | 2020-11-18 | Self-adaptive multi-model driving equipment fault diagnosis method based on edge cloud cooperation |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112101532A CN112101532A (en) | 2020-12-18 |
| CN112101532B true CN112101532B (en) | 2021-02-12 |
Family
ID=73784707
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011289732.9A Active CN112101532B (en) | 2020-11-18 | 2020-11-18 | Self-adaptive multi-model driving equipment fault diagnosis method based on edge cloud cooperation |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112101532B (en) |
Families Citing this family (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| 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 |
| CN112837282A (en) * | 2021-01-27 | 2021-05-25 | 上海交通大学 | Small sample image defect detection method based on cloud edge cooperation and deep learning |
| CN112835684B (en) * | 2021-03-02 | 2024-03-22 | 浙江工业大学 | Virtual machine deployment method for mobile edge computing |
| CN112989712B (en) * | 2021-04-27 | 2021-08-03 | 浙大城市学院 | Aeroengine fault diagnosis method based on 5G edge calculation and deep learning |
| CN112947290B (en) * | 2021-05-16 | 2021-08-20 | 北京赛博联物科技有限公司 | Edge cloud cooperation-based equipment state monitoring method and system and storage medium |
| CN113222148B (en) * | 2021-05-20 | 2022-01-11 | 浙江大学 | Neural network reasoning acceleration method for material identification |
| CN113783716B (en) * | 2021-07-27 | 2024-06-28 | 国网冀北电力有限公司信息通信分公司 | Flow prediction method and device based on cloud edge collaborative framework |
| CN114112399B (en) * | 2021-11-26 | 2022-12-02 | 江南大学 | Combined fault diagnosis system and fault diagnosis method for wind turbine generator bearing |
| CN114626807A (en) * | 2022-03-01 | 2022-06-14 | 中广核工程有限公司 | Nuclear power scene management method, system, device, computer equipment and storage medium |
| CN114564312A (en) * | 2022-03-03 | 2022-05-31 | 北京航空航天大学云南创新研究院 | Cloud edge-side cooperative computing method based on adaptive deep neural network |
| CN115297035B (en) * | 2022-08-04 | 2024-06-18 | 杭州杰牌传动科技有限公司 | Intelligent operation and maintenance system with edge cloud cooperation |
| CN116106005B (en) * | 2023-04-12 | 2023-06-06 | 北京理工大学 | Rotary equipment fault diagnosis method based on end Bian Yun collaborative machine learning |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109635958A (en) * | 2018-12-12 | 2019-04-16 | 成都航天科工大数据研究院有限公司 | A kind of predictive industrial equipment maintaining method and maintenance system based on edge calculations |
| CN109933004A (en) * | 2019-03-27 | 2019-06-25 | 苏芯物联技术(南京)有限公司 | The machine failure diagnosis and prediction method and system cooperateed with based on edge calculations and cloud |
| CN110070060A (en) * | 2019-04-26 | 2019-07-30 | 天津开发区精诺瀚海数据科技有限公司 | A kind of method for diagnosing faults of bearing apparatus |
| CN110162018A (en) * | 2019-05-31 | 2019-08-23 | 天津开发区精诺瀚海数据科技有限公司 | The increment type equipment fault diagnosis method that knowledge based distillation is shared with hidden layer |
| CN110216680A (en) * | 2019-07-05 | 2019-09-10 | 山东大学 | A kind of service robot cloud ground collaborative fault diagnosis system and method |
| CN110336845A (en) * | 2019-04-02 | 2019-10-15 | 中国联合网络通信集团有限公司 | Industrial product quality method of real-time, equipment and system |
| CN110334580A (en) * | 2019-05-04 | 2019-10-15 | 天津开发区精诺瀚海数据科技有限公司 | The equipment fault classification method of changeable weight combination based on integrated increment |
| CN111142049A (en) * | 2020-01-16 | 2020-05-12 | 合肥工业大学 | Intelligent transformer fault diagnosis method based on edge cloud cooperation mechanism |
| CN111353482A (en) * | 2020-05-25 | 2020-06-30 | 天津开发区精诺瀚海数据科技有限公司 | LSTM-based fatigue factor recessive anomaly detection and fault diagnosis method |
| CN111507489A (en) * | 2020-04-20 | 2020-08-07 | 电子科技大学中山学院 | Cloud-edge-coordinated amusement equipment fault prediction and health management system and method |
-
2020
- 2020-11-18 CN CN202011289732.9A patent/CN112101532B/en active Active
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109635958A (en) * | 2018-12-12 | 2019-04-16 | 成都航天科工大数据研究院有限公司 | A kind of predictive industrial equipment maintaining method and maintenance system based on edge calculations |
| CN109933004A (en) * | 2019-03-27 | 2019-06-25 | 苏芯物联技术(南京)有限公司 | The machine failure diagnosis and prediction method and system cooperateed with based on edge calculations and cloud |
| CN110336845A (en) * | 2019-04-02 | 2019-10-15 | 中国联合网络通信集团有限公司 | Industrial product quality method of real-time, equipment and system |
| CN110070060A (en) * | 2019-04-26 | 2019-07-30 | 天津开发区精诺瀚海数据科技有限公司 | A kind of method for diagnosing faults of bearing apparatus |
| CN110334580A (en) * | 2019-05-04 | 2019-10-15 | 天津开发区精诺瀚海数据科技有限公司 | The equipment fault classification method of changeable weight combination based on integrated increment |
| CN110162018A (en) * | 2019-05-31 | 2019-08-23 | 天津开发区精诺瀚海数据科技有限公司 | The increment type equipment fault diagnosis method that knowledge based distillation is shared with hidden layer |
| CN110216680A (en) * | 2019-07-05 | 2019-09-10 | 山东大学 | A kind of service robot cloud ground collaborative fault diagnosis system and method |
| CN111142049A (en) * | 2020-01-16 | 2020-05-12 | 合肥工业大学 | Intelligent transformer fault diagnosis method based on edge cloud cooperation mechanism |
| CN111507489A (en) * | 2020-04-20 | 2020-08-07 | 电子科技大学中山学院 | Cloud-edge-coordinated amusement equipment fault prediction and health management system and method |
| CN111353482A (en) * | 2020-05-25 | 2020-06-30 | 天津开发区精诺瀚海数据科技有限公司 | LSTM-based fatigue factor recessive anomaly detection and fault diagnosis method |
Non-Patent Citations (6)
| Title |
|---|
| A Check Value Fault Diagnosis Method Based on Variational Mode Decomposition and Permutation Entropy;Zhen Pan等;《2019 IEEE 8th Data Driven Control and Learning Systems Conference》;20190527;第650-655页 * |
| Fault diagnosis and fault-tolerant control for a nonlinear electro-hydraulic system;Liang Chen等;《2010 Conference on Control and Fault Tolerant Systems》;20101008;第269-274页 * |
| Neurosurgeon: Collaborative Intelligence Between the Cloud and Mobile Edge;Yiping Kang等;《ACM SIGPLAN Notices》;20170430;第1-15页 * |
| 云/边缘协同的轴承故障诊断方法;张文龙等;《计算机集成制造系统》;20200331;第26卷(第3期);第589-599页 * |
| 关于边缘计算和边云协同，看这一篇就够了~;边缘计算社区;《CSDN》;20190914;https://blog.csdn.net/weixin_41033724/article/details/100841011 * |
| 对数回归-ARMA周期预测模型及其应用;赵凌等;《水资源与水工程学报》;20101231;第21卷(第6期);第19-21页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112101532A (en) | 2020-12-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112101532B (en) | Self-adaptive multi-model driving equipment fault diagnosis method based on edge cloud cooperation | |
| WO2020077672A1 (en) | Method and device for training service quality evaluation model | |
| CN106934497A (en) | Wisdom cell power consumption real-time predicting method and device based on deep learning | |
| CN109597965A (en) | Data processing method, system, terminal and medium based on deep neural network | |
| CN114154646A (en) | Efficiency optimization method for federal learning in mobile edge network | |
| CN113125960A (en) | Vehicle-mounted lithium ion battery charge state prediction method based on random forest model | |
| CN112200391B (en) | Power distribution network edge side load prediction method based on k-nearest neighbor mutual information feature simplification | |
| CN108304925A (en) | A kind of pond computing device and method | |
| CN115392481A (en) | Federal learning efficient communication method based on real-time response time balancing | |
| CN108304926A (en) | A kind of pond computing device and method suitable for neural network | |
| CN115562244A (en) | Equipment fault diagnosis method based on dynamic federal learning | |
| CN115713026A (en) | Time series data prediction model construction method, device, equipment and readable storage medium | |
| CN112527547B (en) | Mechanical intelligent fault prediction method based on automatic convolution neural network | |
| CN115766475B (en) | Semi-asynchronous power federal learning network based on communication efficiency and communication method thereof | |
| CN117436485A (en) | Multi-exit point end-edge-cloud cooperative system and method based on trade-off time delay and precision | |
| Gao et al. | Short-term traffic flow prediction based on time-Space characteristics | |
| CN116774089A (en) | Convolutional neural network battery state of health estimation method and system based on feature fusion | |
| Cui et al. | Network traffic prediction based on Hadoop | |
| CN107273970B (en) | Reconfigurable platform of convolutional neural network supporting online learning and construction method thereof | |
| CN114781598A (en) | Fault prediction method based on hierarchical neural network distributed training | |
| CN115169660A (en) | Cutter wear prediction method based on multi-scale space-time feature fusion neural network | |
| CN112685176A (en) | Resource-constrained edge computing method for improving DDNN (distributed neural network) | |
| CN113098707B (en) | Virtual network function demand prediction method in edge network | |
| Cai et al. | Calculation Method of Available Transfer Capacity Based on Graph Convolutional Network | |
| Yang et al. | Fedprem: A novel federated reinforcement learning framework for predictive maintenance |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |