WO2023093177A1 - 设备故障诊断方法、装置、电子设备及存储介质 - Google Patents

设备故障诊断方法、装置、电子设备及存储介质 Download PDF

Info

Publication number
WO2023093177A1
WO2023093177A1 PCT/CN2022/115738 CN2022115738W WO2023093177A1 WO 2023093177 A1 WO2023093177 A1 WO 2023093177A1 CN 2022115738 W CN2022115738 W CN 2022115738W WO 2023093177 A1 WO2023093177 A1 WO 2023093177A1
Authority
WO
WIPO (PCT)
Prior art keywords
model
equipment
fault diagnosis
local
global
Prior art date
Application number
PCT/CN2022/115738
Other languages
English (en)
French (fr)
Inventor
郭盛
刘辉
赵书宝
Original Assignee
新智我来网络科技有限公司
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 新智我来网络科技有限公司 filed Critical 新智我来网络科技有限公司
Publication of WO2023093177A1 publication Critical patent/WO2023093177A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/02—Preprocessing
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/22—Matching criteria, e.g. proximity measures
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/12—Classification; Matching

Definitions

  • the present disclosure relates to the technical field of equipment fault diagnosis, and in particular to an equipment fault diagnosis method, device, electronic equipment and storage medium.
  • the embodiments of the present disclosure provide a device fault diagnosis method, device, electronic device and storage medium to solve the existing needs of the prior art to train a separate model for each device, the model generalization ability is poor, and the device fault diagnosis The results are not accurate enough to be applicable to complex application scenarios, and the efficiency of equipment fault diagnosis is low.
  • the first aspect of the embodiments of the present disclosure provides a method for diagnosing equipment faults, including: determining multiple participants based on a pre-created joint learning
  • the square contains at least one device; use the preset signal analysis algorithm to perform preprocessing operations on the vibration signal to obtain the signal characteristic data corresponding to the vibration signal, and perform clustering operations on the devices based on the signal characteristic data to obtain multiple similar devices group; integrate the signal feature data corresponding to the equipment in the same similar equipment group in each participant to obtain training data, and use the training data to train the multi-fault diagnosis model to obtain a local model; based on the corresponding The local model and the preset model aggregation method perform aggregation operations on the local model to obtain the global model, and use the global model to update the local model, and perform fault diagnosis on the device based on the updated local model.
  • the second aspect of the embodiments of the present disclosure provides an apparatus for fault diagnosis of equipment, including: an acquisition module configured to determine multiple participants based on a pre-created joint learning architecture, and acquire the Vibration signals, wherein the participants include at least one device; the clustering module is configured to use a preset signal analysis algorithm to perform preprocessing operations on the vibration signals to obtain signal characteristic data corresponding to the vibration signals, and based on the signal characteristic data , perform a clustering operation on the devices to obtain multiple similar device groups; the training module is configured to integrate the signal feature data corresponding to the devices in the same similar device group in each participant to obtain training data, and use the training data
  • the multi-fault diagnosis model is trained to obtain a local model; the diagnosis module is configured to perform an aggregation operation on the local model based on the local model corresponding to each similar equipment group and the preset model aggregation method to obtain a global model, and use the global The model updates the local model, and performs fault diagnosis on the equipment based on the updated local model.
  • the preprocessing operation obtains the signal characteristic data corresponding to the vibration signal, and based on the signal characteristic data, performs a clustering operation on the equipment to obtain multiple similar equipment groups; the signal corresponding to the equipment in the same similar equipment group in each participant
  • the feature data is integrated to obtain training data, and the multi-fault diagnosis model is trained using the training data to obtain a local model; based on the local model corresponding to each similar equipment group and the preset model aggregation method, the aggregation operation is performed on the local model to obtain
  • the global model is used to update the local model, and the device is diagnosed based on the updated local model.
  • the disclosure does not need to train a separate model for each device, is applicable to device fault diagnosis in complex scenarios, and improves the accuracy of device fault diagnosis results and the efficiency of device fault diagnosis.
  • FIG. 1 is a schematic diagram of a joint learning architecture provided by an embodiment of the present disclosure
  • FIG. 2 is a schematic flowchart of a device fault diagnosis method provided by an embodiment of the present disclosure
  • FIG. 3 is a schematic diagram of the overall flow of a method for diagnosing equipment faults in an actual scenario provided by an embodiment of the present disclosure
  • Fig. 4 is a schematic waveform diagram of a device vibration signal in an actual scene provided by an embodiment of the present disclosure
  • FIG. 5 is a spectrum diagram obtained by preprocessing a device vibration signal in an actual scene provided by an embodiment of the present disclosure
  • FIG. 6 is a schematic diagram of a continuous wavelet coefficient matrix obtained by preprocessing equipment vibration signals in an actual scene provided by an embodiment of the present disclosure
  • FIG. 7 is a schematic diagram of the principle of a privacy clustering algorithm provided by an embodiment of the present disclosure.
  • FIG. 8 is a schematic structural diagram of a multi-fault diagnosis model based on a convolutional neural network provided by an embodiment of the present disclosure
  • FIG. 9 is a schematic diagram of hierarchical aggregation and update operations of a personalized joint learning model provided by an embodiment of the present disclosure.
  • FIG. 10 is a schematic structural diagram of an equipment fault diagnosis device provided by an embodiment of the present disclosure.
  • Fig. 11 is a schematic structural diagram of an electronic device provided by an embodiment of the present disclosure.
  • this disclosure proposes a mechanical equipment fault diagnosis method based on personalized joint learning.
  • joint learning has the ability to acquire knowledge from multiple participants , while protecting the advantages of data privacy.
  • the federated learning framework that maintains the same model for all participants may not have a good effect on all machines, so it needs to be based on the fault diagnosis of rotating machinery equipment The actual situation to improve joint learning.
  • Federated learning refers to the comprehensive utilization of various AI (Artificial Intelligence, artificial intelligence) technologies on the premise of ensuring data security and user privacy, and joint multi-party cooperation to jointly mine data value and generate new intelligent business models and models based on joint modeling.
  • Federated learning has at least the following characteristics:
  • Participating nodes control the weakly centralized joint training mode of their own data to ensure data privacy and security in the process of co-creating intelligence.
  • FIG. 1 is a schematic diagram of a joint learning architecture provided by an embodiment of the present disclosure.
  • the architecture of joint learning may include a server (central node) 101 , and participants 102 , 103 , and 104 .
  • the basic model can be established by the server 101, and the server 101 sends the model to the participant 102, the participant 103 and the participant 104 with which a communication connection is established.
  • the basic model can also be uploaded to the server 101 after being created by any participant, and the server 101 sends the model to other participants that have established communication connections with it.
  • Participant 102, participant 103 and participant 104 build a model according to the downloaded basic structure and model parameters, use local data for model training, obtain updated model parameters, and encrypt and upload the updated model parameters to the server 101.
  • the server 101 aggregates the model parameters sent by the participant 102 , the participant 103 and the participant 104 to obtain the global model parameters, and returns the global model parameters to the participant 102 , the participant 103 and the participant 104 .
  • the participant 102, the participant 103 and the participant 104 iterate their models according to the received global model parameters until the models finally converge, thereby realizing the training of the models.
  • the data uploaded by participant 102, participant 103, and participant 104 are model parameters, local data will not be uploaded to server 101, and all participants can share the final model parameters, so data can be guaranteed Co-modeling is achieved on the basis of privacy.
  • Fig. 2 is a schematic flowchart of a device fault diagnosis method provided by an embodiment of the present disclosure.
  • the device fault diagnosis method in FIG. 2 can be executed by a federated learning server.
  • the device fault diagnosis method may specifically include:
  • S201 Determine multiple participants based on the pre-created joint learning architecture, and acquire vibration signals generated by devices of the participants during operation, where the participants include at least one device;
  • the equipment in the embodiments of the present disclosure can be regarded as mechanical equipment in current industrial production, such as rotating mechanical equipment (steam turbine, fan, pump, etc.).
  • Mechanical equipment is usually installed in factories. Therefore, the participants in the joint learning architecture of the present disclosure can be considered as factories, that is, the present disclosure is to diagnose the faults of mechanical equipment in the factory.
  • the following embodiments of the present disclosure are described by taking rotating mechanical equipment as an example. However, it should be understood that the application scenarios of the embodiments of the present disclosure are not limited to fault diagnosis of rotating mechanical equipment. All devices are applicable to the technical solutions of the present disclosure, and the application scenarios of the following embodiments do not limit the technical solutions of the present disclosure.
  • each participant corresponds to a node in the joint learning framework, and the participants perform data interaction with the joint learning server through clients installed on smart terminals such as PCs, tablets, and smart phones.
  • the federated learning framework also has a node that provides services for the client (that is, the server).
  • the server can be a server for performing aggregation operations.
  • the server can coordinate multiple clients to perform joint learning to obtain a joint learning model.
  • the server may be an independent physical server, or a server cluster or cloud computing server composed of multiple physical servers.
  • multiple participants are determined based on the pre-created joint learning architecture, and the vibration signals generated by the equipment of the participants are obtained during operation, wherein the participants include at least one device; using preset The signal analysis algorithm performs preprocessing on the vibration signal to obtain the signal characteristic data corresponding to the vibration signal, and based on the signal characteristic data, performs a clustering operation on the equipment to obtain multiple similar equipment groups; the same similar equipment in each participant
  • the signal feature data corresponding to the equipment in the equipment group is integrated to obtain training data, and the multi-fault diagnosis model is trained using the training data to obtain a local model; based on the local model corresponding to each similar equipment group, and the preset model aggregation method , perform an aggregation operation on the local model to obtain the global model, and use the global model to update the local model, and perform fault diagnosis on the device based on the updated local model.
  • the disclosure does not need to train a separate model for each device, is applicable to device fault diagnosis in complex scenarios, and
  • FIG. 3 is a schematic diagram of the overall flow of the device fault diagnosis method in an actual scene provided by an embodiment of the present disclosure.
  • the implementation process of the equipment fault diagnosis method based on joint learning may specifically include the following contents:
  • the factory is used as the participant, and the mechanical equipment in the factory is used as the object of data collection.
  • the vibration signal generated during the operation of the equipment is collected, and according to the
  • the time-domain indicators and frequency-domain indicators obtained from vibration signal preprocessing are used to cluster equipment from different parties (ie factories) and divide them into multiple similar equipment groups.
  • the corresponding eigenvectors are used as the input of the multi-task fault diagnosis model (that is, the multi-fault diagnosis model), and the multi-task fault diagnosis model is used
  • the multi-task fault classification module judges whether the fault corresponding to each type of the device will occur.
  • determining multiple participants based on a pre-created joint learning framework, and acquiring vibration signals generated by devices of the participants during operation includes: using a preset target object as a participant, using the The above-mentioned participants build a joint learning framework for equipment fault diagnosis, and use vibration sensors installed on the equipment to use different sampling frequencies to collect vibration signals generated by the equipment at different sampling frequencies; wherein, the Default target objects include plant objects.
  • a vibration sensor is installed on each equipment to collect vibration signals generated during the operation of the equipment.
  • the three factories respectively have N 1 , N 2 and N 3 pieces of rotating mechanical equipment.
  • Vibration signals of the equipment are obtained by installing vibration velocity and acceleration sensors on each equipment.
  • the acceleration vibration sensor can be installed on the driving end or the non-driving end of the equipment.
  • the signal feature data includes time-domain features, frequency-domain features, spectrograms, and continuous wavelet coefficient matrices; using a preset signal analysis algorithm, a preprocessing operation is performed on the vibration signal to obtain the signal feature corresponding to the vibration signal Data, including: using the preset feature extraction method to extract the feature of the vibration signal, obtain the time domain feature and frequency domain feature corresponding to the vibration signal, perform Fourier transform on the vibration signal, obtain the frequency spectrum corresponding to the vibration signal, and use The wavelet transformation method obtains the continuous wavelet coefficient matrix of the vibration signal.
  • Fig. 4 is a schematic waveform diagram of the device vibration signal in the actual scene provided by the embodiment of the present disclosure
  • Fig. 5 is a spectrum diagram obtained by preprocessing the device vibration signal in the actual scene provided by the embodiment of the present disclosure
  • Fig. 6 is an embodiment of the present disclosure
  • the vibration time-domain features extracted from the vibration signal include peak value, peak-to-peak value, effective value, kurtosis, skewness, margin, etc.
  • the frequency domain features include the 0.5-fold frequency amplitude and 1-fold frequency amplitude of the equipment speed , 2-octave frequency amplitude, 3-octave frequency amplitude, 4-octave frequency amplitude, 5-octave frequency amplitude, etc., and then the time-domain features and frequency-domain features corresponding to the vibration signals of the above three different sampling frequencies can be combined into A 30-dimensional feature vector a.
  • Fourier transform is performed on the vibration signals of the above three different sampling frequencies to obtain the spectrum diagrams of the three vibration signals, and the spectrum diagrams are intercepted into three one-dimensional feature vectors f 1 , f 2 and f 3 .
  • the lengths of f 1 , f 2 and f 3 may be 128, 128 and 256, respectively.
  • the continuous wavelet transform is performed on the vibration signals of the above three different sampling frequencies to obtain the continuous wavelet coefficient matrices of the three different frequency vibration signals, and the continuous wavelet coefficient matrices are intercepted as three two-dimensional characteristic square matrices C 1 , C2 and C3 .
  • Morlet wavelet function can be selected as the wavelet function of continuous wavelet transform, and the decomposition scales of the vibration signals of three different frequencies can be 128, 256 and 384 respectively.
  • the sizes of C 1 , C 2 and C 3 correspond to 128 ⁇ 128, 256 ⁇ 256 and 384 ⁇ 384.
  • the equipment is clustered to obtain multiple similar equipment groups, including: combining the time domain features and frequency domain features corresponding to the vibration signals at different sampling frequencies to obtain the corresponding equipment
  • the feature vector is used as the input of the privacy clustering algorithm, and the privacy clustering algorithm is used to cluster the devices, so as to divide the devices with similar feature vectors into the same similar device group.
  • Fig. 7 is a schematic diagram of the principle of the privacy clustering algorithm provided by the embodiment of the present disclosure. As shown in Fig. 7, the process of dividing devices based on the privacy clustering algorithm and obtaining similar device groups may include the following contents:
  • the feature vector a of each device is used as the input of the privacy clustering algorithm.
  • only the normal data of each device is used to cluster similar device groups, and the category to which a device belongs is determined by the category to which most of its data belongs.
  • federated k-means as a privacy clustering algorithm, and its basic principle and training process are shown in FIG. 7 .
  • Each client represents a factory participating in federated learning.
  • the client performs k-means training locally and uploads the obtained cluster centers to the server.
  • the server receives the cluster centers from the client, and further aggregates the cluster centers to obtain the global cluster centers represented by the squares in the figure.
  • the client downloads the global cluster center from the server and continues local training.
  • multiple rounds of iterations are used to achieve the convergence of the federated k-means clustering algorithm until the distances between the local cluster centers of all clients and the corresponding global cluster centers are sufficiently small.
  • using the training data to train the multi-fault diagnosis model to obtain the local model includes: for similar equipment groups in each participant, obtaining the spectrogram and continuous wavelet of the vibration signal corresponding to each equipment in the similar equipment group
  • the coefficient matrix based on the sampling frequency of the vibration signal, intercepts the spectrogram and the continuous wavelet coefficient matrix into a one-dimensional feature vector and a two-dimensional feature matrix respectively; the vibration signals, one-dimensional feature vector and two-dimensional feature matrix of different sampling frequencies
  • a multi-fault diagnosis model based on convolutional neural network is constructed; the training data is used to train the multi-fault diagnosis model, and the corresponding to each similar equipment group is obtained.
  • the local model wherein the multi-fault diagnosis model includes a feature extraction module and a multi-fault classification module.
  • FIG. 8 is a schematic structural diagram of a multi-fault diagnosis model based on a convolutional neural network provided by an embodiment of the present disclosure.
  • the multi-fault diagnosis model based on a convolutional neural network may specifically include the following:
  • the embodiments of the present disclosure construct a multi-task fault diagnosis model for rotating machinery based on convolutional neural networks, and divide the model into It is a general vibration feature extraction module (ie feature extraction module) and a multi-task fault diagnosis module (ie multi-fault classification module).
  • the number and size of input features are determined according to the size of vibration signal, spectrogram and continuous wavelet coefficient matrix, and the number of input features is determined according to the type of fault to be diagnosed.
  • the number of outputs of the task fault diagnosis model is determined according to the number of outputs of the task fault diagnosis model.
  • input 1, input 2 and input 3 respectively correspond to x 1 , x 2 and x 3 of the above-mentioned embodiment
  • Input 4 input 5 and input 6 respectively correspond to f 1 , f 2 and f 3 of the above embodiment
  • input 7, input 8 and input 9 respectively correspond to C 1 , C 2 and C 3 of the above embodiment.
  • the multi-task fault diagnosis model in the embodiment of the present disclosure is a local model constructed for different similar equipment groups in each factory (participant), that is, each similar equipment group in each factory corresponds to a local Model.
  • the local model is aggregated to obtain the global model
  • the global model is used to update the local model, including: obtaining all The local model corresponding to the similar device group, the feature extraction modules in the local models of all similar device groups are aggregated to obtain the global feature extraction module, and the local model corresponding to the same similar device group is obtained, and the local model of the same similar device group
  • the multi-fault classification modules in the model are aggregated to obtain the global multi-fault classification module corresponding to the same similar device group; for the similar device groups in each participant, the global features corresponding to the similar device group identifiers are downloaded from the server of the joint learning architecture
  • the extraction module and the global multi-fault classification module and use the global feature extraction module and the global multi-fault classification module to update the feature extraction module and the multi-fault classification module in the local model of the similar equipment group respectively; in the local model of the similar equipment group After the update, the aggregation
  • Fig. 9 is a schematic diagram of the hierarchical aggregation and update operation of the personalized joint learning model provided by the embodiment of the present disclosure. As shown in Fig. 9, the hierarchical aggregation and update operation of the personalized joint learning model may specifically include the following contents:
  • the multi-task fault classification module in the local model corresponding to the similar equipment group obtained by all factory training is aggregated to obtain the global multi-task fault classification module of the similar equipment group;
  • the personalized joint learning model will reach convergence, and finally a global general feature extraction module and a multi-task fault classification module corresponding to each similar device group will be obtained .
  • the local model aggregation may adopt an average aggregation manner.
  • performing fault diagnosis on the device based on the updated local model includes: classifying the device according to the time-domain features and frequency-domain features corresponding to the vibration signals generated by the device when performing fault diagnosis on the device , to determine the similar device group corresponding to the device; based on the similar device group identifier, download the global feature extraction module and the global multi-fault classification module corresponding to the similar device group identifier from the server of the joint
  • the multi-fault classification module is combined into a complete multi-fault diagnosis model, and the multi-fault diagnosis model is used as the local model of the equipment; according to the spectrogram and continuous wavelet coefficient matrix obtained by preprocessing the vibration signal of the equipment, the corresponding one-dimensional The eigenvector, the two-dimensional characteristic square matrix corresponding to the continuous wavelet coefficient matrix, and the vibration signals of different sampling frequencies are used as the input of the local model, so that the local model can make a judgment on the occurrence of each fault of the equipment.
  • the equipment when performing fault diagnosis on the equipment of the original participant or the equipment of the new participant, the equipment can be divided according to the time domain and frequency domain characteristics of the vibration signal of the equipment Go to a similar device group, and then download the model corresponding to the similar device group from the server to diagnose the fault of this device.
  • the information of its similar equipment group can be obtained directly; if the equipment has not participated in the training of the joint learning model, then By extracting the time-domain and frequency-domain features of its vibration signal, it can be divided into a similar device group based on the cluster center obtained by the privacy clustering algorithm. Then, according to the similar equipment group information of the equipment, the global general feature extraction module and the multi-task fault classification module corresponding to the similar equipment group are downloaded from the server, and the two are combined into a complete multi-task fault diagnosis model for rotating mechanical equipment .
  • the vibration waveform of the equipment is preprocessed to obtain the spectrogram and continuous wavelet coefficient matrix that match the input size of the fault diagnosis model.
  • the vibration waveform, spectrogram and continuous wavelet coefficient matrix are input into the multi-task fault diagnosis model of rotating machinery to obtain the occurrence of each fault.
  • the equipment in multiple factories is divided into similar equipment groups by means of private equipment clustering, and a multi-task fault diagnosis model for rotating machinery equipment based on convolutional neural networks is constructed, and finally Based on hierarchical model aggregation, personalized model training and aggregation for each similar equipment group is realized, which solves the data heterogeneity problem of different equipment in the fault diagnosis of rotating machinery equipment.
  • This disclosure uses a joint clustering algorithm to cluster similar equipment for a large number of rotating mechanical equipment in multiple factories, and uses a personalized joint learning algorithm based on hierarchical model aggregation to achieve accurate multi-type faults for multi-type rotating equipment in multiple factories Diagnosis, realizes the full coverage of the application scenarios of the equipment fault diagnosis method, improves the accuracy of the equipment fault diagnosis results and the efficiency of the equipment fault diagnosis.
  • Fig. 10 is a schematic structural diagram of an apparatus for diagnosing equipment faults provided by an embodiment of the present disclosure.
  • the equipment fault diagnosis device includes:
  • the obtaining module 1001 is configured to determine multiple participants based on the pre-created joint learning architecture, and obtain vibration signals generated by devices of the participants during operation, wherein the participants include at least one device;
  • the clustering module 1002 is configured to use a preset signal analysis algorithm to perform a preprocessing operation on the vibration signal to obtain signal characteristic data corresponding to the vibration signal, and perform a clustering operation on the device based on the signal characteristic data to obtain multiple similar device group;
  • the training module 1003 is configured to integrate signal feature data corresponding to devices in the same similar device group in each participant to obtain training data, and use the training data to train the multi-fault diagnosis model to obtain a local model;
  • the diagnosis module 1004 is configured to perform an aggregation operation on the local model based on the local model corresponding to each similar device group and a preset model aggregation method to obtain the global model, and update the local model by using the global model, based on the updated The resulting local model performs fault diagnosis on the device.
  • the acquisition module 1001 in FIG. 10 takes the preset target object as a participant, uses the participant to construct a joint learning framework for equipment fault diagnosis, and utilizes the vibration sensor installed on the equipment to use different sampling Frequency, collecting vibration signals generated by equipment at different sampling frequencies; wherein, the preset target objects include factory objects.
  • the signal feature data includes time-domain features, frequency-domain features, spectrograms, and continuous wavelet coefficient matrices; the clustering module 1002 in Figure 10 uses a preset feature extraction method to perform feature extraction on vibration signals to obtain vibration The corresponding time-domain features and frequency-domain features of the signal, the Fourier transform of the vibration signal is obtained to obtain the corresponding frequency spectrum of the vibration signal, and the continuous wavelet coefficient matrix of the vibration signal is obtained by using the wavelet transformation method.
  • the clustering module 1002 in FIG. 10 combines the time-domain features and frequency-domain features corresponding to the vibration signals at different sampling frequencies to obtain a feature vector corresponding to the device, and use the feature vector as a privacy clustering algorithm
  • the devices are clustered using the privacy clustering algorithm, so that devices with similar feature vectors can be divided into the same similar device group.
  • the training module 1003 in FIG. 10 obtains the spectrogram and the continuous wavelet coefficient matrix of the vibration signals corresponding to each device in the similar device group for the similar device groups in each participant, based on the sampling frequency of the vibration signals, The spectrogram and the continuous wavelet coefficient matrix are intercepted into one-dimensional feature vector and two-dimensional feature matrix respectively; vibration signals of different sampling frequencies, one-dimensional feature vector and two-dimensional feature matrix are used as the input of the convolutional neural network model, according to Preset equipment fault types, build a multi-fault diagnosis model based on convolutional neural network; use the training data to train the multi-fault diagnosis model, and obtain a local model corresponding to each similar equipment group, among them, the multi-fault diagnosis model Contains feature extraction module and multiple fault classification module.
  • the diagnostic module 1004 in FIG. 10 obtains the local models corresponding to all similar device groups, aggregates the feature extraction modules in the local models of all similar device groups to obtain a global feature extraction module, and obtains the same similar device
  • the local model corresponding to the group aggregate the multi-fault classification modules in the local model of the same similar device group to obtain the global multi-fault classification module corresponding to the same similar device group; for the similar device groups in each participant, from the joint
  • the server of the learning architecture downloads the global feature extraction module and the global multi-fault classification module corresponding to the similar device group identifiers, and uses the global feature extraction module and the global multi-fault classification module to respectively analyze the feature extraction module and the global multi-fault classification module in the local model of the similar device group.
  • the multi-fault classification module is updated; after the local models of similar equipment groups are updated, the aggregation and update operations are repeated multiple times until the joint learning model reaches convergence.
  • the diagnosis module 1004 in FIG. 10 classifies the equipment according to the time-domain features and frequency-domain features corresponding to the vibration signals generated by the equipment when performing fault diagnosis on the equipment, so as to determine the similar equipment group corresponding to the equipment ; Based on the similar equipment group identification, download the global feature extraction module and the global multi-fault classification module corresponding to the similar equipment group identification from the server of the federated learning architecture, and combine the global feature extraction module and the global multi-fault classification module into a complete multi-fault classification module.
  • the fault diagnosis model uses the multi-fault diagnosis model as the local model of the equipment; according to the spectrogram and continuous wavelet coefficient matrix obtained by preprocessing the vibration signal of the equipment, the one-dimensional eigenvector corresponding to the spectrogram and the continuous wavelet coefficient matrix corresponding to The two-dimensional feature matrix and the vibration signals of different sampling frequencies are used as the input of the local model, so that the local model can make a judgment on the occurrence of each fault of the equipment.
  • FIG. 11 is a schematic structural diagram of an electronic device 11 provided by an embodiment of the present disclosure.
  • the electronic device 11 of this embodiment includes: a processor 1101 , a memory 1102 , and a computer program 1103 stored in the memory 1102 and capable of running on the processor 1101 .
  • the processor 1101 executes the computer program 1103, the steps in the foregoing method embodiments are implemented.
  • the processor 1101 executes the computer program 1103, the functions of the modules/units in the foregoing device embodiments are realized.
  • the computer program 1103 can be divided into one or more modules/units, and one or more modules/units are stored in the memory 1102 and executed by the processor 1101 to complete the present disclosure.
  • One or more modules/units may be a series of computer program instruction segments capable of accomplishing specific functions, and the instruction segments are used to describe the execution process of the computer program 1103 in the electronic device 11 .
  • the electronic equipment 11 may be electronic equipment such as desktop computers, notebooks, palmtop computers, and cloud servers.
  • the electronic device 11 may include but not limited to a processor 1101 and a memory 1102 .
  • FIG. 11 is only an example of the electronic device 11, and does not constitute a limitation to the electronic device 11. It may include more or less components than shown in the figure, or combine certain components, or different components. , for example, an electronic device may also include an input and output device, a network access device, a bus, and the like.
  • the processor 1101 can be a central processing unit (Central Processing Unit, CPU), and can also be other general-purpose processors, digital signal processors (Digital Signal Processor, DSP), application specific integrated circuits (Application Specific Integrated Circuit, ASIC), on-site Field-Programmable Gate Array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.
  • a general-purpose processor may be a microprocessor, or the processor may be any conventional processor, or the like.
  • the storage 1102 may be an internal storage unit of the electronic device 11 , for example, a hard disk or a memory of the electronic device 11 .
  • the memory 1102 can also be an external storage device of the electronic device 11, for example, a plug-in hard disk equipped on the electronic device 11, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash memory card ( Flash Card), etc.
  • the memory 1102 may also include both an internal storage unit of the electronic device 11 and an external storage device.
  • the memory 1102 is used to store computer programs and other programs and data required by the electronic device.
  • the memory 1102 can also be used to temporarily store data that has been output or will be output.
  • the disclosed apparatus/computer equipment and methods may be implemented in other ways.
  • the device/computer device embodiments described above are only illustrative, for example, the division of modules or units is only a logical function division, and there may be other division methods in actual implementation, and multiple units or components can be Incorporation may either be integrated into another system, or some features may be omitted, or not implemented.
  • the mutual coupling or direct coupling or communication connection shown or discussed may be through some interfaces, and the indirect coupling or communication connection of devices or units may be in electrical, mechanical or other forms.
  • a unit described as a separate component may or may not be physically separated, and a component displayed as a unit may or may not be a physical unit, that is, it may be located in one place, or may be distributed to multiple network units. Part or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.
  • each functional unit in each embodiment of the present disclosure may be integrated into one processing unit, each unit may exist separately physically, or two or more units may be integrated into one unit.
  • the above-mentioned integrated units can be implemented in the form of hardware or in the form of software functional units.
  • an integrated module/unit is realized in the form of a software function unit and sold or used as an independent product, it can be stored in a computer-readable storage medium.
  • the present disclosure realizes all or part of the processes in the methods of the above embodiments, and can also be completed by instructing related hardware through computer programs.
  • the computer programs can be stored in computer-readable storage media, and the computer programs can be processed. When executed by the controller, the steps in the above-mentioned method embodiments can be realized.
  • a computer program may include computer program code, which may be in source code form, object code form, executable file, or some intermediate form or the like.
  • the computer-readable medium may include: any entity or device capable of carrying computer program code, recording medium, U disk, removable hard disk, magnetic disk, optical disk, computer memory, read-only memory (Read-Only Memory, ROM), random access Memory (Random Access Memory, RAM), electrical carrier signal, telecommunication signal and software distribution medium, etc. It should be noted that the content contained in computer readable media may be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, computer readable media may not Including electrical carrier signals and telecommunication signals.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Artificial Intelligence (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • Signal Processing (AREA)
  • Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)

Abstract

本公开提供了一种设备故障诊断方法、装置、电子设备及存储介质。该方法包括：基于联合学习架构确定多个参与方，并获取参与方的设备在运行过程中产生的振动信号；对振动信号执行预处理操作得到信号特征数据，并对设备执行聚类操作，得到多个相似设备组；将每个参与方中同一相似设备组内的设备所对应的信号特征数据进行整合，得到训练数据，利用训练数据对多故障诊断模型进行训练得到本地模型；基于每个相似设备组对应的本地模型以及预设的模型聚合方法，对本地模型执行聚合操作，得到全局模型，利用全局模型对本地模型进行更新，基于更新后的本地模型对设备进行故障诊断。本公开能够提升设备故障诊断结果的精确度以及设备故障诊断的效率。

Description

设备故障诊断方法、装置、电子设备及存储介质 技术领域
本公开涉及设备故障诊断技术领域，尤其涉及一种设备故障诊断方法、装置、电子设备及存储介质。
背景技术
随着工业4.0和智慧能源的提出，工业领域对生产设备的智能控制和管理的要求越来越高，生产管理者和操作人员需要实时掌握设备的运行状态和健康程度，因此对机械设备进行实时在线监测和智能诊断的要求也越来越高。近些年来，随着在线监测设备的普及，工业领域已经积累了大量的在线监测数据，但对这些数据的使用只停留于阈值报警层面；如何更加充分利用这些监测数据进行实时智能的故障诊断已成为一个紧迫而极具挑战性的问题。
目前现有技术中已经提出了许多基于人工智能算法的机械设备故障诊断方法，但是大多数基于人工智能的机械设备故障诊断方法，需要为每台机器设备训练单独的模型，并且模型的泛化能力比较差，无法解决设备间因结构和工作条件的不同引起的数据分布不同的问题，导致机械设备故障诊断结果不够精准。而基于转移学习的故障诊断方法，则需要大量的故障数据和完整的故障类型来支持训练过程，这在实际的工业应用中无法满足，因此无法应用于复杂场景下的设备故障诊断。
鉴于上述现有技术中的问题，需要提供一种无需为每台设备训练单独模型，提升设备故障诊断结果的精确度，应用场景范围广，设备故障诊断效率高的设备故障诊断方法。
发明内容
有鉴于此，本公开实施例提供了一种设备故障诊断方法、装置、电子设备及存储介质，以解决现有技术存在的需要为每台设备训练单独模型，模型泛化能力差，设备故障诊断结果不够精准，无法适用于复杂应用场景，设备故障诊断效率低的问题。
本公开实施例的第一方面，提供了一种设备故障诊断方法，包括：基于预先创建的联合学习架构确定多个参与方，并获取参与方的设备在运行过程中产生的振动信号，其中参与方中包含至少一个设备；利用预设的信号分析算法，对振动信号执行预处理操作，得到振动信号对应的信号特征数据，并基于信号特征数据，对设备执行聚类操作，得到多个相似设备组；将每个参与方中同一相似设备组内的设备所对应的信号特征数据进行整合，得到训练数据，利用训练数据对多故障诊断模型进行训练得到本地模型；基于每个相似设备组对应的本 地模型，以及预设的模型聚合方法，对本地模型执行聚合操作，得到全局模型，并利用全局模型对本地模型进行更新，基于更新后得到的本地模型对设备进行故障诊断。
本公开实施例的第二方面，提供了一种设备故障诊断装置，包括：获取模块，被配置为基于预先创建的联合学习架构确定多个参与方，并获取参与方的设备在运行过程中产生的振动信号，其中参与方中包含至少一个设备；聚类模块，被配置为利用预设的信号分析算法，对振动信号执行预处理操作，得到振动信号对应的信号特征数据，并基于信号特征数据，对设备执行聚类操作，得到多个相似设备组；训练模块，被配置为将每个参与方中同一相似设备组内的设备所对应的信号特征数据进行整合，得到训练数据，利用训练数据对多故障诊断模型进行训练得到本地模型；诊断模块，被配置为基于每个相似设备组对应的本地模型，以及预设的模型聚合方法，对本地模型执行聚合操作，得到全局模型，并利用全局模型对本地模型进行更新，基于更新后得到的本地模型对设备进行故障诊断。
本公开实施例采用的上述至少一个技术方案能够达到以下有益效果：
通过基于预先创建的联合学习架构确定多个参与方，并获取参与方的设备在运行过程中产生的振动信号，其中参与方中包含至少一个设备；利用预设的信号分析算法，对振动信号执行预处理操作，得到振动信号对应的信号特征数据，并基于信号特征数据，对设备执行聚类操作，得到多个相似设备组；将每个参与方中同一相似设备组内的设备所对应的信号特征数据进行整合，得到训练数据，利用训练数据对多故障诊断模型进行训练得到本地模型；基于每个相似设备组对应的本地模型，以及预设的模型聚合方法，对本地模型执行聚合操作，得到全局模型，并利用全局模型对本地模型进行更新，基于更新后得到的本地模型对设备进行故障诊断。本公开无需为每台设备训练单独模型，适用于复杂场景下的设备故障诊断，提升设备故障诊断结果的精确度以及设备故障诊断的效率。
附图说明
为了更清楚地说明本公开实施例中的技术方案，下面将对实施例或现有技术描述中所需要使用的附图作简单地介绍，显而易见地，下面描述中的附图仅仅是本公开的一些实施例，对于本领域普通技术人员来讲，在不付出创造性劳动的前提下，还可以根据这些附图获得其它的附图。
图1是本公开实施例提供的一种联合学习的架构示意图；
图2是本公开实施例提供的设备故障诊断方法的流程示意图；
图3是本公开实施例提供的实际场景中设备故障诊断方法的整体流程示意图；
图4是本公开实施例提供的实际场景中设备振动信号的波形示意图；
图5是本公开实施例提供的实际场景中对设备振动信号进行预处理得到的频谱图；
图6是本公开实施例提供的实际场景中对设备振动信号进行预处理得到的连续小波系数矩阵的示意图；
图7是本公开实施例提供的隐私聚类算法的原理示意图；
图8是本公开实施例提供的基于卷积神经网络的多故障诊断模型的结构示意图；
图9是本公开实施例提供的个性化联合学习模型的分层聚合及更新操作的示意图；
图10是本公开实施例提供的设备故障诊断装置的结构示意图；
图11是本公开实施例提供的电子设备的结构示意图。
具体实施方式
以下描述中，为了说明而不是为了限定，提出了诸如特定系统结构、技术之类的具体细节，以便透彻理解本公开实施例。然而，本领域的技术人员应当清楚，在没有这些具体细节的其它实施例中也可以实现本公开。在其它情况中，省略对众所周知的系统、装置、电路以及方法的详细说明，以免不必要的细节妨碍本公开的描述。
在社会整体安全水平日益提升的大背景下，因工业设备的故障导致的重大事故愈发受到社会的深切关注。特别是在石化、能源等领域，因为设备故障造成的安全事故频繁发生。以旋转机械设备为例，如汽轮机、风机、泵机等是工业生产中的关键设备，如果发生设备故障将可能导致机组意外停机，从而导致较高的运行和维护成本，还有可能造成巨大的经济损失和安全事故。旋转机械设备故障诊断的目的在于故障劣化前对其及时进行检测、识别和定位，对于确保旋转机械设备的安全可靠运行至关重要。
随着工业4.0和智慧能源的提出，工业领域对生产设备的智能控制和管理的要求越来越高，生产管理者和操作人员需要实时掌握设备的运行状态和健康程度，因此对旋转机械设备进行实时在线监测和智能诊断的要求也越来越高。近些年来，随着在线监测设备的普及，工业领域已经积累了大量的在线监测数据，但对这些数据的使用只停留于阈值报警层面；如何更加充分利用这些监测数据进行实时智能的故障诊断已成为一个紧迫而极具挑战性的问题。
近些年来，基于所采集的振动数据，已经提出了许多基于人工智能算法的旋转机械故障诊断方法，如卷积神经网络、长短时记忆网络等。但大多数基于人工智能的旋转机械故障诊断方法需要为每台机器训练一个独立的模型，并且泛化能力较差，无法处理设备间因结构和工作条件的不同引起的数据分布不同。针对这一问题，将转移学习和领域自适应引入故障诊断领域。然而，现有的大多数迁移学习方法只能处理转速和负载等工况的变化。当应用于多台机器设备的故障诊断时，这些方法仍然需要大量的故障数据和完整的故障类型来支持训练 过程，这在实际的工业应用中是无法满足的。
为解决大量机械设备的故障诊断问题，本公开提出一种基于个性化联合学习的机械设备故障诊断方法，联合学习相较于上述人工智能算法以及迁移学习算法，具有能够从多个参与方获取知识，同时保护数据隐私的优点。但是在旋转机械设备故障诊断的场景下，由于数据的异构性，为所有参与者维护相同模型的联合学习框架可能不会对所有机器都有很好的效果，因此需要根据旋转机械设备故障诊断的实际情况对联合学习进行改进。
联合学习是指在确保数据安全及用户隐私的前提下，综合利用多种AI(Artificial Intelligence，人工智能)技术，联合多方合作共同挖掘数据价值，催生基于联合建模的新的智能业态和模式。联合学习至少具有以下特点：
(1)参与节点控制自有数据的弱中心化联合训练模式，确保共创智能过程中的数据隐私安全。
(2)在不同应用场景下，利用筛选和/或组合AI算法、隐私保护计算，建立多种模型聚合优化策略，以获取高层次、高质量的模型。
(3)在确保数据安全及用户隐私的前提下，基于多种模型聚合优化策略，获取提升联合学习引擎的效能方法，其中效能方法可以是通过解决包括计算架构并行、大规模跨域网络下的信息交互、智能感知、异常处理机制等，提升联合学习引擎的整体效能。
(4)获取各场景下多方用户的需求，通过互信机制，确定合理评估各联合参与方的真实贡献度，进行分配激励。
基于上述方式，可以建立基于联合学习的AI技术生态，充分发挥行业数据价值，推动垂直领域的场景落地。
图1是本公开实施例提供的一种联合学习的架构示意图。如图1所示，联合学习的架构可以包括服务器(中心节点)101以及参与方102、参与方103和参与方104。
在联合学习过程中，基本模型可以通过服务器101建立，服务器101将该模型发送至与其建立通信连接的参与方102、参与方103和参与方104。基本模型还可以是任一参与方建立后上传至服务器101，服务器101将该模型发送至与其建立通信连接的其他参与方。参与方102、参与方103和参与方104根据下载的基本结构和模型参数构建模型，利用本地数据进行模型训练，获得更新的模型参数，并将更新的模型参数加密上传至服务器101。服务器101对参与方102、参与方103和参与方104发送的模型参数进行聚合，获得全局模型参数，并将全局模型参数传回至参与方102、参与方103和参与方104。参与方102、参与方103和参与方104根据接收的全局模型参数对各自的模型进行迭代，直到模型最终收敛，从而实现对模型的训练。在联合学习过程中，参与方102、参与方103和参与方104上传的数据为 模型参数，本地数据并不会上传至服务器101，且所有参与方可以共享最终的模型参数，因此可以在保证数据隐私的基础上实现共同建模。
需要说明的是，参与方的数量不限于如上的三个，而是可以根据需要进行设置，本公开实施例对此不作限制。
下面将结合附图详细说明根据本公开实施例的一种设备故障诊断方法和装置。
图2是本公开实施例提供的设备故障诊断方法的流程示意图。图2的设备故障诊断方法可以由联合学习的服务器执行。如图2所示，该设备故障诊断方法具体可以包括：
S201，基于预先创建的联合学习架构确定多个参与方，并获取参与方的设备在运行过程中产生的振动信号，其中参与方中包含至少一个设备；
S202，利用预设的信号分析算法，对振动信号执行预处理操作，得到振动信号对应的信号特征数据，并基于信号特征数据，对设备执行聚类操作，得到多个相似设备组；
S203，将每个参与方中同一相似设备组内的设备所对应的信号特征数据进行整合，得到训练数据，利用训练数据对多故障诊断模型进行训练得到本地模型；
S204，基于每个相似设备组对应的本地模型，以及预设的模型聚合方法，对本地模型执行聚合操作，得到全局模型，并利用全局模型对本地模型进行更新，基于更新后得到的本地模型对设备进行故障诊断。
具体地，本公开实施例的设备可以认为是目前工业生产中的机械设备，例如旋转机械设备(汽轮机、风机、泵机等)。机械设备通常安装在工厂中，因此，本公开的联合学习架构中的参与方可以认为是工厂，即本公开是对工厂中的机械设备的故障进行诊断。需要说明的是，本公开以下实施例是以旋转机械设备为例进行描述的，但是，应当理解的是，本公开实施例的应用场景不限于旋转机械设备的故障诊断，任何可能发生故障的机械设备，都适用于本公开技术方案，以下实施例的应用场景不构成对本公开技术方案的限定。
进一步地，每一个参与方对应联合学习框架中的一个节点，参与方通过安装在PC、平板电脑、智能手机等智能终端上的客户端与联合学习服务端进行数据交互。联合学习框架中还具有为客户端提供服务的节点(即服务端)，服务端可以是用于执行聚合操作的服务器，服务端可以协调多个客户端进行联合学习以得到联合学习模型。服务器可以是独立的物理服务器，也可以是多个物理服务器构成的服务器集群或者云计算服务器。
根据本公开实施例提供的技术方案，基于预先创建的联合学习架构确定多个参与方，并获取参与方的设备在运行过程中产生的振动信号，其中参与方中包含至少一个设备；利用预设的信号分析算法，对振动信号执行预处理操作，得到振动信号对应的信号特征数据，并基于信号特征数据，对设备执行聚类操作，得到多个相似设备组；将每个参与方中同一相似 设备组内的设备所对应的信号特征数据进行整合，得到训练数据，利用训练数据对多故障诊断模型进行训练得到本地模型；基于每个相似设备组对应的本地模型，以及预设的模型聚合方法，对本地模型执行聚合操作，得到全局模型，并利用全局模型对本地模型进行更新，基于更新后得到的本地模型对设备进行故障诊断。本公开无需为每台设备训练单独模型，适用于复杂场景下的设备故障诊断，提升设备故障诊断结果的精确度以及设备故障诊断的效率。
下面结合附图，对本公开基于联合学习的设备故障诊断方法的实现流程进行详细说明，图3是本公开实施例提供的实际场景中设备故障诊断方法的整体流程示意图。如图3所示，该基于联合学习的设备故障诊断方法的实现流程具体可以包括以下内容：
本公开的联合学习框架中，将工厂作为参与方，将工厂中的机械设备作为数据采集的对象，通过在每台设备上加装振动传感器，采集设备运行过程中产生的振动信号，并根据对振动信号预处理得到的时域指标和频域指标，对来自不同参与方(即工厂)的设备进行聚类，划分成多个相似设备组。并且，根据对设备的振动信号进行预处理得到的频谱图和连续小波系数矩阵，将其对应的特征向量作为多任务故障诊断模型(即多故障诊断模型)的输入，利用多任务故障诊断模型中的多任务故障分类模块(即多故障分类模块)，判断该设备对应每一类型下的故障是否会发生。
在一些实施例中，基于预先创建的联合学习架构确定多个参与方，并获取所述参与方的设备在运行过程中产生的振动信号，包括：将预设的目标对象作为参与方，利用所述参与方构建用于设备故障诊断的联合学习架构，并利用安装在所述设备上的振动传感器，使用不同的采样频率，采集所述设备在不同采样频率下产生的振动信号；其中，所述预设的目标对象包括工厂对象。
具体地，对于多个工厂的多台旋转机械设备，在每台设备上加装振动传感器，采集设备运行过程中产生的振动信号。在一个具体实施例中，例如对于一个由3个工厂参与的设备故障诊断联合学习场景而言，假设3个工厂分别有N 1、N 2和N 3台旋转机械设备。通过在每个设备上安装振动速度和加速度传感器来获取设备的振动信号。加速度振动传感器可以安装在设备的驱动端或者非驱动端。
进一步地，在振动信号的采集过程中，为了关注不同频段的振动信号特征，因此可以分别以32k、16k和8k的采样频率，采集不同频率的振动信号，将采集到的不同频率的振动信号分别记为x 1、x 2和x 3。
在一些实施例中，信号特征数据包括时域特征、频域特征、频谱图、以及连续小波系数矩阵；利用预设的信号分析算法，对振动信号执行预处理操作，得到振动信号对应的信号特征数据，包括：利用预设的特征提取方法对振动信号进行特征提取，得到振动信号对应的 时域特征和频域特征，对振动信号进行傅里叶变换，得到振动信号对应的频谱图，并利用小波变化方法获取振动信号的连续小波系数矩阵。
具体地，对于采集到的设备振动信号进行数据的预处理，使用信号分析方法提取振动信号的时域指标(即时域特征)和频域指标(即频域特征)，然后对振动信号进行傅里叶变换，获取与其对应的频谱图；最后使用小波变化方法获取振动信号的连续小波系数矩阵。下面结合附图以及具体实施例，对利用信号分析算法进行数据的预处理过程以及预处理后得到的信号特征数据进行详细说明。图4是本公开实施例提供的实际场景中设备振动信号的波形示意图；图5是本公开实施例提供的实际场景中对设备振动信号进行预处理得到的频谱图；图6是本公开实施例提供的实际场景中对设备振动信号进行预处理得到的连续小波系数矩阵的示意图。
进一步地，从振动信号中提取的振动时域特征包括峰值、峰峰值、有效值、峭度、偏度、裕度等，频域特征包括设备转速的0.5倍频幅值、1倍频幅值、2倍频幅值、3倍频幅值、4倍频幅值、5倍频幅值等，进而可以将上述三种不同采样频率的振动信号所对应的时域特征和频域特征组合成为一个30维的特征向量a。
进一步地，对上述三种不同采样频率的振动信号进行傅里叶变换，获取三种振动信号的频谱图，并将频谱图截取为三个一维特征向量f 1、f 2和f 3。在实际应用中，f 1、f 2和f 3长度可以分别为128、128和256。然后，对上述三种不同采样频率的振动信号进行连续小波变换，获取三种不同频率振动信号的连续小波系数矩阵，并将连续小波系数矩阵进行截取，作为三个二维特征方阵C 1、C 2和C 3。在实际应用中，连续小波变换的小波函数可以选用Morlet小波函数，三种不同频率的振动信号的分解尺度可以分别为128、256和384，此时C 1、C 2和C 3的尺寸分别对应为128×128、256×256和384×384。
在一些实施例中，基于信号特征数据，对设备执行聚类操作，得到多个相似设备组，包括：将不同采样频率下的振动信号对应的时域特征和频域特征进行组合，得到与设备相对应的特征向量，将特征向量作为隐私聚类算法的输入，利用隐私聚类算法对设备进行聚类，以便将具有相似特征向量的设备划分为同一相似设备组。
具体地，使用隐私聚类算法，以设备的时域指标和频域指标作为输入，对不同工厂的多台旋转机械设备进行聚类，将设备划分到多个相似设备组。下面结合附图以及具体实施例，对本公开隐私聚类算法的原理及实现过程进行详细说明。图7是本公开实施例提供的隐私聚类算法的原理示意图，如图7所示，基于隐私聚类算法对设备进行划分，得到相似设备组的过程可以包括以下内容：
每台设备的特征向量a用于隐私聚类算法的输入。为了避免故障引起的设备数据分布的 变化，只使用每台设备的正常数据对相似的设备组进行聚类，设备所属的类别由其大部分数据所属的类别决定。
进一步地，本公开选用联邦k-means作为隐私聚类算法，其基本原理和训练流程如图7所示。每个客户端代表一个参与联合学习的工厂，在联邦k-means的训练过程中，客户端在本地执行k-means训练，并将获得的聚类中心上载到服务器。然后，服务端从客户端接收聚类中心，并将聚类中心进一步聚合以获得图中正方形表示的全局聚类中心。接下来，客户端从服务端下载全局聚类中心并继续本地训练。最后，通过多轮迭代以达到联邦k-means聚类算法的收敛，直到所有客户端的本地聚类中心和相应的全局聚类中心之间的距离足够小。
在一些实施例中，利用训练数据对多故障诊断模型进行训练得到本地模型，包括：针对每个参与方内的相似设备组，获取相似设备组内各个设备所对应振动信号的频谱图和连续小波系数矩阵，基于振动信号的采样频率，将频谱图和连续小波系数矩阵分别截取为一维特征向量和二维特征方阵；将不同采样频率的振动信号、一维特征向量和二维特征方阵作为卷积神经网络模型的输入，根据预设的设备故障类型，构建基于卷积神经网络的多故障诊断模型；利用训练数据对多故障诊断模型进行训练，得到与每个相似设备组相对应的本地模型，其中，多故障诊断模型中包含特征提取模块以及多故障分类模块。
具体地，在利用每一相似设备组的训练数据对多故障诊断模型进行训练得到本地模型之前，需要先对多故障诊断模型的结构进行构造。下面结合附图以及具体实施例，对本公开基于卷积神经网络的多故障诊断模型的结构进行详细说明。图8是本公开实施例提供的基于卷积神经网络的多故障诊断模型的结构示意图，如图8所示，该基于卷积神经网络的多故障诊断模型具体可以包括以下内容：
本公开实施例根据振动信号、频谱图和连续小波系数矩阵的形式，以及需要诊断的旋转机械设备的故障种类，构建基于卷积神经网络的旋转机械设备的多任务故障诊断模型，并将模型分为通用振动特征提取模块(即特征提取模块)和多任务故障诊断模块(即多故障分类模块)。
在基于卷积神经网络的旋转机械设备的多任务故障诊断模型的构造过程中，根据振动信号、频谱图和连续小波系数矩阵的尺寸确定输入特征的数量和尺寸，根据需要诊断的故障种类确定多任务故障诊断模型的输出数量。如图8所示为本公开在实际场景中所构造的旋转机械设备的多任务故障诊断模型的结构，输入1、输入2和输入3分别对应上述实施例的x 1、x 2和x 3，输入4、输入5和输入6分别对应上述实施例的f 1、f 2和f 3，输入7、输入8和输入9分别对应上述实施例的C 1、C 2和C 3。每种类型的输入经过不同参数的卷积运算后，合并为三维特征矩阵，将三维特征矩阵作为通用振动特征提取模块的输出。多任务故障诊断模 块以通用振动特征为输入，分别为每种故障构造单独的诊断子网络，最终由sigmoid函数给出该类故障是否发生的判断结果。
需要说明的是，本公开实施例的多任务故障诊断模型是针对每一个工厂(参与方)内的不同相似设备组所构建的本地模型，即每一个工厂内的每一个相似设备组对应一个本地模型。
在一些实施例中，基于每个相似设备组对应的本地模型，以及预设的模型聚合方法，对本地模型执行聚合操作，得到全局模型，并利用全局模型对本地模型进行更新，包括：获取全部相似设备组所对应的本地模型，将全部相似设备组的本地模型中的特征提取模块进行聚合，得到全局特征提取模块，并获取同一相似设备组所对应的本地模型，将同一相似设备组的本地模型中的多故障分类模块进行聚合，得到同一相似设备组对应的全局多故障分类模块；针对每个参与方内的相似设备组，从联合学习架构的服务端下载相似设备组标识对应的全局特征提取模块以及全局多故障分类模块，并利用全局特征提取模块以及全局多故障分类模块，分别对相似设备组的本地模型中的特征提取模块及多故障分类模块进行更新；在相似设备组的本地模型更新之后，多次重复执行聚合及更新操作，直至联合学习模型达到收敛。
具体地，本公开实施例是以工厂作为参与方，将平台运营方作为联合学习框架中的服务端，从而构建用于设备故障诊断的个性化联合学习框架。在每个工厂内部(即参与方内部)，按照相似设备组将预处理后的设备振动数据进行汇总和本地训练，工厂间则按照相似设备组进行联合学习的分层模型聚合和更新，直至联合学习模型收敛。下面结合附图以及具体实施例，对本公开个性化联合学习模型的分层聚合及更新过程进行详细说明。图9是本公开实施例提供的个性化联合学习模型的分层聚合及更新操作的示意图，如图9所示，该个性化联合学习模型的分层聚合及更新操作具体可以包括以下内容：
在实际应用中，假设工厂A和工厂B是参与联合学习的两个工厂，工厂的所有设备被分成两个相似设备组，即相似设备组1和相似设备组2。因此，对于每个工厂建立两个客户端，每个工厂对应两个相似设备组。对于每个工厂的每个客户端，首先将工厂中同一相似设备组对应设备的训练数据进行汇总，并训练获得该相似设备组的本地模型。其次，对相似设备组的本地模型进行分层模型聚合和更新，分层模型聚合和更新操作可以包括以下步骤：
(1)将所有工厂的所有相似设备组所训练获得的本地模型中的通用特征提取模块进行聚合，获得全局的通用特征提取模块；
(2)对于每个相似设备组，将所有工厂训练获得的对应该相似设备组的本地模型中的多任务故障分类模块进行聚合，获取该相似设备组的全局多任务故障分类模块；
(3)针对每个工厂对应的每个相似设备组的本地模型，从服务端下载全局的通用特征提取模块，以及该相似设备组对应的全局多任务故障分类模块，完成本地模型的更新。
最后，在经过多次的模型聚合和本地模型的更新操作后，个性化联合学习模型将达到收敛，最终将会获得一个全局的通用特征提取模块以及每个相似设备组对应的多任务故障分类模块。在本公开实施例中，本地模型聚合可以采用平均聚合的方式。
在一些实施例中，基于更新后得到的本地模型对设备进行故障诊断，包括：在对设备进行故障诊断时，根据设备产生的振动信号所对应的时域特征和频域特征，对设备进行分类，以确定设备对应的相似设备组；基于相似设备组标识，从联合学习架构的服务端下载与相似设备组标识对应的全局特征提取模块以及全局多故障分类模块，并将全局特征提取模块以及全局多故障分类模块组合成完整的多故障诊断模型，将多故障诊断模型作为设备的本地模型；根据对设备的振动信号进行预处理得到的频谱图和连续小波系数矩阵，将频谱图对应的一维特征向量、连续小波系数矩阵对应的二维特征方阵、以及不同采样频率的振动信号作为本地模型的输入，以使本地模型对设备对应每种故障的发生做出判断。
具体地，基于上述训练后的个性化联合学习模型，在对原参与方的设备或者新增参与方的设备进行故障诊断时，可以根据设备振动信号的时域和频域特征，将该设备划分到某个相似设备组，然后从服务端下载该相似设备组对应的模型，对这台设备进行故障诊断。
进一步地，对于一台需要进行故障诊断的旋转机械设备，如果该设备已经参与过联合学习模型的训练，则可以直接获取其相似设备组信息；如果该设备未参与过联合学习模型的训练，则可以通过对其振动信号进行时域和频域特征的提取，进而基于隐私聚类算法所获得的聚类中心，将其划分到某个相似设备组中去。然后，根据该设备的相似设备组信息，从服务端下载全局的通用特征提取模块和对应相似设备组的多任务故障分类模块，并将两者组合为完整的旋转机械设备的多任务故障诊断模型。
进一步地，对该设备的振动波形进行预处理，获取与故障诊断模型输入尺寸相符合的频谱图和连续小波系数矩阵。最后，将振动波形、频谱图和连续小波系数矩阵输入到旋转机械多任务故障诊断模型，得到每种故障的发生情况。
根据本公开实施例提供的技术方案，通过采用隐私设备聚类的方式对多个工厂的设备进行相似设备组的划分，并构建基于卷积神经网络的旋转机械设备的多任务故障诊断模型，最终基于分层模型聚合，实现了对每个相似设备组的个性化模型训练和聚合，解决了旋转机械设备故障诊断中不同设备的数据异构问题。本公开通过利用联合聚类算法对多个工厂的大量旋转机械设备进行相似设备聚类，并使用基于分层模型聚合的个性化联合学习算法实现多个工厂多类型旋转设备的精准的多类型故障诊断，实现设备故障诊断方法的应用场景的全覆盖，提升了设备故障诊断结果的精确度以及设备故障诊断的效率。
下述为本公开装置实施例，可以用于执行本公开方法实施例。对于本公开装置实施例 中未披露的细节，请参照本公开方法实施例。
图10是本公开实施例提供的设备故障诊断装置的结构示意图。如图10所示，该设备故障诊断装置包括：
获取模块1001，被配置为基于预先创建的联合学习架构确定多个参与方，并获取参与方的设备在运行过程中产生的振动信号，其中参与方中包含至少一个设备；
聚类模块1002，被配置为利用预设的信号分析算法，对振动信号执行预处理操作，得到振动信号对应的信号特征数据，并基于信号特征数据，对设备执行聚类操作，得到多个相似设备组；
训练模块1003，被配置为将每个参与方中同一相似设备组内的设备所对应的信号特征数据进行整合，得到训练数据，利用训练数据对多故障诊断模型进行训练得到本地模型；
诊断模块1004，被配置为基于每个相似设备组对应的本地模型，以及预设的模型聚合方法，对本地模型执行聚合操作，得到全局模型，并利用全局模型对本地模型进行更新，基于更新后得到的本地模型对设备进行故障诊断。
在一些实施例中，图10的获取模块1001将预设的目标对象作为参与方，利用参与方构建用于设备故障诊断的联合学习架构，并利用安装在设备上的振动传感器，使用不同的采样频率，采集设备在不同采样频率下产生的振动信号；其中，预设的目标对象包括工厂对象。
在一些实施例中，信号特征数据包括时域特征、频域特征、频谱图、以及连续小波系数矩阵；图10的聚类模块1002利用预设的特征提取方法对振动信号进行特征提取，得到振动信号对应的时域特征和频域特征，对振动信号进行傅里叶变换，得到振动信号对应的频谱图，并利用小波变化方法获取振动信号的连续小波系数矩阵。
在一些实施例中，图10的聚类模块1002将不同采样频率下的振动信号对应的时域特征和频域特征进行组合，得到与设备相对应的特征向量，将特征向量作为隐私聚类算法的输入，利用隐私聚类算法对设备进行聚类，以便将具有相似特征向量的设备划分为同一相似设备组。
在一些实施例中，图10的训练模块1003针对每个参与方内的相似设备组，获取相似设备组内各个设备所对应振动信号的频谱图和连续小波系数矩阵，基于振动信号的采样频率，将频谱图和连续小波系数矩阵分别截取为一维特征向量和二维特征方阵；将不同采样频率的振动信号、一维特征向量和二维特征方阵作为卷积神经网络模型的输入，根据预设的设备故障类型，构建基于卷积神经网络的多故障诊断模型；利用训练数据对多故障诊断模型进行训练，得到与每个相似设备组相对应的本地模型，其中，多故障诊断模型中包含特征提取模块以及多故障分类模块。
在一些实施例中，图10的诊断模块1004获取全部相似设备组所对应的本地模型，将全部相似设备组的本地模型中的特征提取模块进行聚合，得到全局特征提取模块，并获取同一相似设备组所对应的本地模型，将同一相似设备组的本地模型中的多故障分类模块进行聚合，得到同一相似设备组对应的全局多故障分类模块；针对每个参与方内的相似设备组，从联合学习架构的服务端下载相似设备组标识对应的全局特征提取模块以及全局多故障分类模块，并利用全局特征提取模块以及全局多故障分类模块，分别对相似设备组的本地模型中的特征提取模块及多故障分类模块进行更新；在相似设备组的本地模型更新之后，多次重复执行聚合及更新操作，直至联合学习模型达到收敛。
在一些实施例中，图10的诊断模块1004在对设备进行故障诊断时，根据设备产生的振动信号所对应的时域特征和频域特征，对设备进行分类，以确定设备对应的相似设备组；基于相似设备组标识，从联合学习架构的服务端下载与相似设备组标识对应的全局特征提取模块以及全局多故障分类模块，并将全局特征提取模块以及全局多故障分类模块组合成完整的多故障诊断模型，将多故障诊断模型作为设备的本地模型；根据对设备的振动信号进行预处理得到的频谱图和连续小波系数矩阵，将频谱图对应的一维特征向量、连续小波系数矩阵对应的二维特征方阵、以及不同采样频率的振动信号作为本地模型的输入，以使本地模型对设备对应每种故障的发生做出判断。
应理解，上述实施例中各步骤的序号的大小并不意味着执行顺序的先后，各过程的执行顺序应以其功能和内在逻辑确定，而不应对本公开实施例的实施过程构成任何限定。
图11是本公开实施例提供的电子设备11的结构示意图。如图11所示，该实施例的电子设备11包括：处理器1101、存储器1102以及存储在该存储器1102中并且可以在处理器1101上运行的计算机程序1103。处理器1101执行计算机程序1103时实现上述各个方法实施例中的步骤。或者，处理器1101执行计算机程序1103时实现上述各装置实施例中各模块/单元的功能。
示例性地，计算机程序1103可以被分割成一个或多个模块/单元，一个或多个模块/单元被存储在存储器1102中，并由处理器1101执行，以完成本公开。一个或多个模块/单元可以是能够完成特定功能的一系列计算机程序指令段，该指令段用于描述计算机程序1103在电子设备11中的执行过程。
电子设备11可以是桌上型计算机、笔记本、掌上电脑及云端服务器等电子设备。电子设备11可以包括但不仅限于处理器1101和存储器1102。本领域技术人员可以理解，图11仅仅是电子设备11的示例，并不构成对电子设备11的限定，可以包括比图示更多或更少的部件，或者组合某些部件，或者不同的部件，例如，电子设备还可以包括输入输出设备、网 络接入设备、总线等。
处理器1101可以是中央处理单元(Central Processing Unit，CPU)，也可以是其它通用处理器、数字信号处理器(Digital Signal Processor，DSP)、专用集成电路(Application Specific Integrated Circuit，ASIC)、现场可编程门阵列(Field-Programmable Gate Array，FPGA)或者其它可编程逻辑器件、分立门或者晶体管逻辑器件、分立硬件组件等。通用处理器可以是微处理器或者该处理器也可以是任何常规的处理器等。
存储器1102可以是电子设备11的内部存储单元，例如，电子设备11的硬盘或内存。存储器1102也可以是电子设备11的外部存储设备，例如，电子设备11上配备的插接式硬盘，智能存储卡(Smart Media Card，SMC)，安全数字(Secure Digital，SD)卡，闪存卡(Flash Card)等。进一步地，存储器1102还可以既包括电子设备11的内部存储单元也包括外部存储设备。存储器1102用于存储计算机程序以及电子设备所需的其它程序和数据。存储器1102还可以用于暂时地存储已经输出或者将要输出的数据。
所属领域的技术人员可以清楚地了解到，为了描述的方便和简洁，仅以上述各功能单元、模块的划分进行举例说明，实际应用中，可以根据需要而将上述功能分配由不同的功能单元、模块完成，即将装置的内部结构划分成不同的功能单元或模块，以完成以上描述的全部或者部分功能。实施例中的各功能单元、模块可以集成在一个处理单元中，也可以是各个单元单独物理存在，也可以两个或两个以上单元集成在一个单元中，上述集成的单元既可以采用硬件的形式实现，也可以采用软件功能单元的形式实现。另外，各功能单元、模块的具体名称也只是为了便于相互区分，并不用于限制本申请的保护范围。上述系统中单元、模块的具体工作过程，可以参考前述方法实施例中的对应过程，在此不再赘述。
在上述实施例中，对各个实施例的描述都各有侧重，某个实施例中没有详述或记载的部分，可以参见其它实施例的相关描述。
本领域普通技术人员可以意识到，结合本文中所公开的实施例描述的各示例的单元及算法步骤，能够以电子硬件、或者计算机软件和电子硬件的结合来实现。这些功能究竟以硬件还是软件方式来执行，取决于技术方案的特定应用和设计约束条件。专业技术人员可以对每一个特定的应用来使用不同方法来实现所描述的功能，但是这种实现不应认为超出本公开的范围。
在本公开所提供的实施例中，应该理解到，所揭露的装置/计算机设备和方法，可以通过其它的方式实现。例如，以上所描述的装置/计算机设备实施例仅仅是示意性的，例如，模块或单元的划分，仅仅为一种逻辑功能划分，实际实现时可以有另外的划分方式，多个单元或组件可以结合或者可以集成到另一个系统，或一些特征可以忽略，或不执行。另一点， 所显示或讨论的相互之间的耦合或直接耦合或通讯连接可以是通过一些接口，装置或单元的间接耦合或通讯连接，可以是电性，机械或其它的形式。
作为分离部件说明的单元可以是或者也可以不是物理上分开的，作为单元显示的部件可以是或者也可以不是物理单元，即可以位于一个地方，或者也可以分布到多个网络单元上。可以根据实际的需要选择其中的部分或者全部单元来实现本实施例方案的目的。
另外，在本公开各个实施例中的各功能单元可以集成在一个处理单元中，也可以是各个单元单独物理存在，也可以两个或两个以上单元集成在一个单元中。上述集成的单元既可以采用硬件的形式实现，也可以采用软件功能单元的形式实现。
集成的模块/单元如果以软件功能单元的形式实现并作为独立的产品销售或使用时，可以存储在一个计算机可读存储介质中。基于这样的理解，本公开实现上述实施例方法中的全部或部分流程，也可以通过计算机程序来指令相关的硬件来完成，计算机程序可以存储在计算机可读存储介质中，该计算机程序在被处理器执行时，可以实现上述各个方法实施例的步骤。计算机程序可以包括计算机程序代码，计算机程序代码可以为源代码形式、对象代码形式、可执行文件或某些中间形式等。计算机可读介质可以包括：能够携带计算机程序代码的任何实体或装置、记录介质、U盘、移动硬盘、磁碟、光盘、计算机存储器、只读存储器(Read-Only Memory，ROM)、随机存取存储器(Random Access Memory，RAM)、电载波信号、电信信号以及软件分发介质等。需要说明的是，计算机可读介质包含的内容可以根据司法管辖区内立法和专利实践的要求进行适当的增减，例如，在某些司法管辖区，根据立法和专利实践，计算机可读介质不包括电载波信号和电信信号。
以上实施例仅用以说明本公开的技术方案，而非对其限制；尽管参照前述实施例对本公开进行了详细的说明，本领域的普通技术人员应当理解：其依然可以对前述各实施例所记载的技术方案进行修改，或者对其中部分技术特征进行等同替换；而这些修改或者替换，并不使相应技术方案的本质脱离本公开各实施例技术方案的精神和范围，均应包含在本公开的保护范围之内。

Claims (10)

  1. 一种设备故障诊断方法，其特征在于，包括：
    基于预先创建的联合学习架构确定多个参与方，并获取所述参与方的设备在运行过程中产生的振动信号，其中所述参与方中包含至少一个所述设备；
    利用预设的信号分析算法，对所述振动信号执行预处理操作，得到所述振动信号对应的信号特征数据，并基于所述信号特征数据，对所述设备执行聚类操作，得到多个相似设备组；
    将每个所述参与方中同一所述相似设备组内的设备所对应的信号特征数据进行整合，得到训练数据，利用所述训练数据对多故障诊断模型进行训练得到本地模型；
    基于每个所述相似设备组对应的本地模型，以及预设的模型聚合方法，对所述本地模型执行聚合操作，得到全局模型，并利用所述全局模型对所述本地模型进行更新，基于所述更新后得到的本地模型对设备进行故障诊断。
  2. 根据权利要求1所述的方法，其特征在于，所述基于预先创建的联合学习架构确定多个参与方，并获取所述参与方的设备在运行过程中产生的振动信号，包括：
    将预设的目标对象作为参与方，利用所述参与方构建用于设备故障诊断的联合学习架构，并利用安装在所述设备上的振动传感器，使用不同的采样频率，采集所述设备在不同采样频率下产生的振动信号；其中，所述预设的目标对象包括工厂对象。
  3. 根据权利要求1所述的方法，其特征在于，所述信号特征数据包括时域特征、频域特征、频谱图、以及连续小波系数矩阵；
    所述利用预设的信号分析算法，对所述振动信号执行预处理操作，得到所述振动信号对应的信号特征数据，包括：
    利用预设的特征提取方法对所述振动信号进行特征提取，得到所述振动信号对应的时域特征和频域特征，对所述振动信号进行傅里叶变换，得到所述振动信号对应的频谱图，并利用小波变化方法获取所述振动信号的连续小波系数矩阵。
  4. 根据权利要求3所述的方法，其特征在于，所述基于所述信号特征数据，对所述设备执行聚类操作，得到多个相似设备组，包括：
    将不同采样频率下的所述振动信号对应的时域特征和频域特征进行组合，得到与所述设备相对应的特征向量，将所述特征向量作为隐私聚类算法的输入，利用所述隐私聚类算法对所述设备进行聚类，以便将具有相似特征向量的设备划分为同一所述相似设备组。
  5. 根据权利要求3所述的方法，其特征在于，所述利用所述训练数据对多故障诊断模 型进行训练得到本地模型，包括：
    针对每个所述参与方内的所述相似设备组，获取所述相似设备组内各个设备所对应振动信号的频谱图和连续小波系数矩阵，基于所述振动信号的采样频率，将所述频谱图和所述连续小波系数矩阵分别截取为一维特征向量和二维特征方阵；
    将所述不同采样频率的振动信号、所述一维特征向量和所述二维特征方阵作为卷积神经网络模型的输入，根据预设的设备故障类型，构建基于卷积神经网络的多故障诊断模型；
    利用所述训练数据对所述多故障诊断模型进行训练，得到与每个所述相似设备组相对应的本地模型，其中，所述多故障诊断模型中包含特征提取模块以及多故障分类模块。
  6. 根据权利要求5所述的方法，其特征在于，所述基于每个所述相似设备组对应的本地模型，以及预设的模型聚合方法，对所述本地模型执行聚合操作，得到全局模型，并利用所述全局模型对所述本地模型进行更新，包括：
    获取全部相似设备组所对应的本地模型，将所述全部相似设备组的所述本地模型中的特征提取模块进行聚合，得到全局特征提取模块，并获取同一相似设备组所对应的本地模型，将所述同一相似设备组的所述本地模型中的多故障分类模块进行聚合，得到所述同一相似设备组对应的全局多故障分类模块；
    针对每个所述参与方内的所述相似设备组，从所述联合学习架构的服务端下载相似设备组标识对应的全局特征提取模块以及全局多故障分类模块，并利用所述全局特征提取模块以及所述全局多故障分类模块，分别对所述相似设备组的本地模型中的特征提取模块及多故障分类模块进行更新；
    在所述相似设备组的本地模型更新之后，多次重复执行所述聚合及更新操作，直至联合学习模型达到收敛。
  7. 根据权利要求6所述的方法，其特征在于，所述基于所述更新后得到的本地模型对设备进行故障诊断，包括：
    在对所述设备进行故障诊断时，根据所述设备产生的振动信号所对应的时域特征和频域特征，对所述设备进行分类，以确定所述设备对应的相似设备组；
    基于相似设备组标识，从所述联合学习架构的服务端下载与所述相似设备组标识对应的全局特征提取模块以及全局多故障分类模块，并将所述全局特征提取模块以及所述全局多故障分类模块组合成完整的多故障诊断模型，将所述多故障诊断模型作为所述设备的本地模型；
    根据对所述设备的振动信号进行预处理得到的频谱图和连续小波系数矩阵，将所述频谱图对应的一维特征向量、所述连续小波系数矩阵对应的二维特征方阵、以及所述不同采样 频率的振动信号作为所述本地模型的输入，以使所述本地模型对所述设备对应每种故障的发生做出判断。
  8. 一种设备故障诊断装置，其特征在于，包括：
    获取模块，被配置为基于预先创建的联合学习架构确定多个参与方，并获取所述参与方的设备在运行过程中产生的振动信号，其中所述参与方中包含至少一个所述设备；
    聚类模块，被配置为利用预设的信号分析算法，对所述振动信号执行预处理操作，得到所述振动信号对应的信号特征数据，并基于所述信号特征数据，对所述设备执行聚类操作，得到多个相似设备组；
    训练模块，被配置为将每个所述参与方中同一所述相似设备组内的设备所对应的信号特征数据进行整合，得到训练数据，利用所述训练数据对多故障诊断模型进行训练得到本地模型；
    诊断模块，被配置为基于每个所述相似设备组对应的本地模型，以及预设的模型聚合方法，对所述本地模型执行聚合操作，得到全局模型，并利用所述全局模型对所述本地模型进行更新，基于所述更新后得到的本地模型对设备进行故障诊断。
  9. 一种电子设备，包括存储器，处理器及存储在存储器上并可在处理器上运行的计算机程序，所述处理器执行所述程序时实现如权利要求1所述的方法。
  10. 一种计算机可读存储介质，所述计算机可读存储介质存储有计算机程序，其特征在于，所述计算机程序被处理器执行时实现如权利要求1所述的方法。
PCT/CN2022/115738 2021-11-29 2022-08-30 设备故障诊断方法、装置、电子设备及存储介质 WO2023093177A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202111434704.6A CN114118156B (zh) 2021-11-29 2021-11-29 设备故障诊断方法、装置、电子设备及存储介质
CN202111434704.6 2021-11-29

Publications (1)

Publication Number Publication Date
WO2023093177A1 true WO2023093177A1 (zh) 2023-06-01

Family

ID=80371287

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/115738 WO2023093177A1 (zh) 2021-11-29 2022-08-30 设备故障诊断方法、装置、电子设备及存储介质

Country Status (2)

Country Link
CN (1) CN114118156B (zh)
WO (1) WO2023093177A1 (zh)

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117113200A (zh) * 2023-10-24 2023-11-24 中海石油气电集团有限责任公司 转子故障诊断方法、装置、电子设备及介质
CN117668622A (zh) * 2024-02-01 2024-03-08 山东能源数智云科技有限公司 设备故障诊断模型的训练方法、故障诊断方法及装置
CN117763621A (zh) * 2024-02-22 2024-03-26 青岛他坦科技服务有限公司 一种基于联邦学习的能源大数据安全保护方法
CN117932230A (zh) * 2024-03-21 2024-04-26 北京航空航天大学 小样本条件下基于特征解耦与样本重组的故障诊断方法
CN117932230B (zh) * 2024-03-21 2024-05-28 北京航空航天大学 小样本条件下基于特征解耦与样本重组的故障诊断方法

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114118156B (zh) * 2021-11-29 2023-05-02 新智我来网络科技有限公司 设备故障诊断方法、装置、电子设备及存储介质
CN117589444B (zh) * 2024-01-18 2024-04-02 湖南科技大学 一种基于联邦学习的风力发电机齿轮箱故障诊断方法

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20210048809A1 (en) * 2019-08-14 2021-02-18 Hitachi, Ltd. Multi task learning with incomplete labels for predictive maintenance
CN112906859A (zh) * 2021-01-27 2021-06-04 重庆邮电大学 一种用于轴承故障诊断的联邦学习算法
CN114118156A (zh) * 2021-11-29 2022-03-01 新智我来网络科技有限公司 设备故障诊断方法、装置、电子设备及存储介质

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20210048809A1 (en) * 2019-08-14 2021-02-18 Hitachi, Ltd. Multi task learning with incomplete labels for predictive maintenance
CN112906859A (zh) * 2021-01-27 2021-06-04 重庆邮电大学 一种用于轴承故障诊断的联邦学习算法
CN114118156A (zh) * 2021-11-29 2022-03-01 新智我来网络科技有限公司 设备故障诊断方法、装置、电子设备及存储介质

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117113200A (zh) * 2023-10-24 2023-11-24 中海石油气电集团有限责任公司 转子故障诊断方法、装置、电子设备及介质
CN117113200B (zh) * 2023-10-24 2024-02-02 中海石油气电集团有限责任公司 转子故障诊断方法、装置、电子设备及介质
CN117668622A (zh) * 2024-02-01 2024-03-08 山东能源数智云科技有限公司 设备故障诊断模型的训练方法、故障诊断方法及装置
CN117668622B (zh) * 2024-02-01 2024-05-10 山东能源数智云科技有限公司 设备故障诊断模型的训练方法、故障诊断方法及装置
CN117763621A (zh) * 2024-02-22 2024-03-26 青岛他坦科技服务有限公司 一种基于联邦学习的能源大数据安全保护方法
CN117932230A (zh) * 2024-03-21 2024-04-26 北京航空航天大学 小样本条件下基于特征解耦与样本重组的故障诊断方法
CN117932230B (zh) * 2024-03-21 2024-05-28 北京航空航天大学 小样本条件下基于特征解耦与样本重组的故障诊断方法

Also Published As

Publication number Publication date
CN114118156A (zh) 2022-03-01
CN114118156B (zh) 2023-05-02

Similar Documents

Publication Publication Date Title
WO2023093177A1 (zh) 设备故障诊断方法、装置、电子设备及存储介质
Yang et al. A time efficient approach for detecting errors in big sensor data on cloud
CN111597247A (zh) 一种数据异常分析方法、装置及存储介质
CN112468347A (zh) 一种云平台的安全管理方法、装置、电子设备及存储介质
CN110430224A (zh) 一种基于随机块模型的通信网络异常行为检测方法
Yin et al. Parallel implementing improved k-means applied for image retrieval and anomaly detection
CN111626360A (zh) 用于检测锅炉故障类型的方法、装置、设备和存储介质
US20190289480A1 (en) Smart Building Sensor Network Fault Diagnostics Platform
CN117634905A (zh) 一种电网业务数据特征提取方法
CN113572639B (zh) 一种载波网络故障的诊断方法、系统、设备和介质
CN116150692A (zh) 基于云平台和改进孤立森林的电网故障诊断方法和系统
CN113821418B (zh) 故障根因分析方法及装置、存储介质和电子设备
CN115293236A (zh) 一种基于混合聚类的电力设备并行故障诊断方法及装置
Dai et al. K-modes clustering algorithm based on weighted overlap distance and its application in intrusion detection
Naukudkar et al. Enhancing performance of security log analysis using correlation-prediction technique
CN114422324B (zh) 一种告警信息的处理方法、装置、电子设备及存储介质
CN116186018B (zh) 一种基于安全控制的电力数据标识和解析方法
Xia et al. Histogram-Based Gradient Boosting Tree: A Federated Learning Approach for Collaborative Fault Diagnosis
CN116823233B (zh) 一种基于全周期运维的用户数据处理方法及系统
CN118018567A (zh) 基于数据智能的工业互联网云平台
CN114826867B (zh) 处理数据的方法、装置、系统及存储介质
Yang Research on Traffic Detection Method of Secure Transmission Industrial Internet of Things Based on Computer Vision
He et al. Data privacy protection health status assessment for rotating machinery with dual-task feature fusion framework
CN117113266A (zh) 基于图同构网络的无人工厂异常检测方法、装置
Boubouh et al. Democratizing Machine Learning: Resilient Distributed Learning with Heterogeneous Participants