Method and device for diagnosing faults of a moving equipment, client and server

Abstract

The application relates to the technical field of fault diagnosis, in particular to a dynamic equipment fault diagnosis method and device, a client and a server, and the method comprises the following steps: acquiring a first data set and a first public verification set; training a first model by using the first data set and the first public verification set, mining a first difficult sample of the first public verification set, uploading the first difficult sample and a first predicted value to a server, and receiving a second predicted value and a second difficult sample mined by the server in the process of training a second model; performing joint bidirectional knowledge distillation based on the predicted value and the difficult sample until the first model and the second model are converged, obtaining a trained joint learning model, and using the joint learning model to diagnose dynamic equipment faults of any participant. Therefore, the problems that in the related art, the fault samples of a single device are few, the local model performance is poor, and multi-device fault diagnosis cannot be realized are solved.

Description

Technical Field

[0001] This application relates to the field of fault diagnosis technology, and in particular to a method, apparatus, client and server for diagnosing faults in moving equipment. Background Technology

[0002] Industrial intelligence is a technology that deeply integrates artificial intelligence with industrial scenarios, mechanisms, and knowledge to achieve or surpass human capabilities in perception, analysis, and decision-making. Key industrial equipment, as the driving force of production, is increasingly intelligent. While the industrial sector has accumulated a vast amount of online monitoring data with the widespread adoption of online monitoring devices, the use of this data has largely remained at the threshold alarm level. Currently, based on the collected vibration data, many methods for rotating machinery fault diagnosis have been proposed, based on deep learning, knowledge graphs, and automated machine learning algorithms, such as convolutional neural networks and long short-term memory networks, to utilize this monitoring data for real-time intelligent fault diagnosis.

[0003] The purpose of fault diagnosis for rotating machinery is to detect, identify, and locate faults in a timely manner before they deteriorate. This is crucial for ensuring the safe and reliable operation of rotating machinery. However, most AI-based fault diagnosis methods for rotating machinery require training an independent model for each machine and have poor generalization ability, making it unable to handle the different data distributions caused by differences in structure and working conditions between equipment.

[0004] In related technologies, transfer learning and domain adaptation have been introduced into the field of fault diagnosis. However, existing transfer learning methods can only handle changes in operating conditions such as speed and load. When applied to the fault diagnosis of multiple machines, transfer learning and domain adaptation methods require a large amount of fault data and complete fault types to support the training process, which cannot meet the requirements in actual industrial applications. Summary of the Invention

[0005] This application provides a method, apparatus, client, and server for diagnosing faults in moving equipment, in order to solve the problems in related technologies such as the limited number of fault samples for single equipment, poor performance of local models, and the inability to diagnose faults in multiple equipment.

[0006] A first aspect of this application provides a method for diagnosing faults in moving equipment. The method is applied to a client, which includes a first model. The method includes the following steps: acquiring a first dataset and a first public validation set; training the first model using the first dataset and the first public validation set, mining a first hard sample from the first public validation set, uploading the first hard sample and a first predicted value to a server, and receiving a second predicted value and a mined second hard sample from the server during the training of a second model; performing joint bidirectional knowledge distillation based on the predicted value and the hard sample until both the first model and the second model converge to obtain a trained joint learning model, and using the joint learning model to diagnose faults in the moving equipment of any participating party.

[0007] Optionally, mining the first difficult sample in the first public validation set includes: obtaining the first predicted value and the first true label on the public validation set; traversing all samples in the public validation set, and if the difference between the first predicted value and the first true label is greater than a first threshold, adding the index of the corresponding sample to the difficult sample index set, until the traversal is completed, and obtaining the first difficult sample.

[0008] Optionally, the first difficult sample includes multiple feature maps, and uploading the first difficult sample and the first predicted value to the server includes: selecting feature maps that meet a first preset condition from the first difficult sample; quantizing and compressing the feature maps that meet the first preset condition to obtain a first compressed feature map; and uploading the first compressed feature map and the first predicted value to the server.

[0009] A second aspect of this application provides a method for diagnosing equipment faults. The method is applied to a server, which includes a second model. The method includes the following steps: acquiring a second dataset and a second public validation set; training the second model using the second dataset and the second public validation set; mining a second hard sample from the second public validation set; sending the second hard sample and a second predicted value to any client; and receiving a first predicted value and a mined first hard sample from the client during the training of the first model; performing joint bidirectional knowledge distillation based on the predicted value and the hard sample until both the first model and the second model converge, obtaining a trained joint learning model; and using the joint learning model to diagnose equipment faults of the arbitrary participant.

[0010] Optionally, mining the second hard sample in the second public validation set includes: obtaining the second predicted value and the second true label of the second sample in the public validation set; traversing all samples of the second sample in the public validation set, and if the difference between the second predicted value and the second true label is greater than a second threshold, then the index of the corresponding sample is added to the hard sample index set, until the traversal is completed, and the second hard sample is obtained.

[0011] Optionally, the second difficult sample includes multiple feature maps, and the step of sending the second difficult sample and the second predicted value to any client includes: selecting feature maps that satisfy a second preset condition from the second difficult sample; quantizing and compressing the feature maps that satisfy the second preset condition to obtain a second compressed feature map; and sending the second compressed feature map and the second predicted value to any client.

[0012] A third aspect of this application provides a fault diagnosis device for moving equipment. The device is applied to a client, which includes a first model. The device includes: a first acquisition module for acquiring a first dataset and a first public validation set; a first training module for training the first model using the first dataset and the first public validation set, mining a first hard sample from the first public validation set, uploading the first hard sample and a first predicted value to a server, and receiving a second predicted value and a mined second hard sample from the server during the training of a second model; and a first distillation module for performing joint bidirectional knowledge distillation based on the predicted value and the hard sample until both the first model and the second model converge to obtain a trained joint learning model, and using the joint learning model to diagnose faults in the moving equipment of any participating party.

[0013] A fourth aspect of this application provides a fault diagnosis device for moving equipment. The device is applied to a server, which includes a second model. The device includes: a second acquisition module for acquiring a second dataset and a second public validation set; a second training module for training the second model using the second dataset and the second public validation set, mining a second hard sample from the second public validation set, sending the second hard sample and a second predicted value to any client, and receiving a first predicted value and a mined first hard sample from the client during the training of the first model; and a second distillation module for performing joint bidirectional knowledge distillation based on the predicted value and the hard sample until both the first model and the second model converge to obtain a trained joint learning model, and using the joint learning model to diagnose the faults of the moving equipment of the arbitrary participant.

[0014] A fifth aspect of this application provides a client, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the equipment fault diagnosis method as described in the above embodiments.

[0015] A sixth aspect of this application provides a server, a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the equipment fault diagnosis method as described in the above embodiments.

[0016] Therefore, this application has at least the following beneficial effects:

[0017] This application's embodiments combine joint learning with knowledge distillation. Based on joint knowledge distillation, the bidirectional mining method for difficult samples in moving equipment fault diagnosis can greatly save communication overhead, improve communication efficiency, avoid the spread of "bad knowledge," and improve the overall performance of the model by first selecting difficult samples on the server / client side, determining the index of the difficult samples, and then compressing and uploading the corresponding feature maps during the joint bidirectional knowledge distillation process for moving equipment fault diagnosis. This not only protects data privacy and saves costs, but also supports model heterogeneity.

[0018] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0019] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein:

[0020] Figure 1 This is a flowchart of a fault diagnosis method for moving equipment provided according to an embodiment of this application;

[0021] Figure 2 Here is a flowchart of a fault diagnosis method for moving equipment according to another embodiment of this application;

[0022] Figure 3 This is a flowchart of a bidirectional mining method for difficult samples in dynamic equipment fault diagnosis based on joint knowledge distillation, according to an embodiment of this application.

[0023] Figure 4 This is a framework diagram for joint knowledge distillation in fault diagnosis of moving equipment provided according to embodiments of this application;

[0024] Figure 5 This is an example diagram of a fault diagnosis device for moving equipment provided according to an embodiment of this application;

[0025] Figure 6This is an example diagram of a fault diagnosis device for moving equipment provided according to another embodiment of this application;

[0026] Figure 7 This is a schematic diagram of the client structure according to an embodiment of this application;

[0027] Figure 8 This is a schematic diagram of the server structure according to an embodiment of this application. Detailed Implementation

[0028] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0029] Due to harsh working environments and prolonged operation, major accidents caused by the failure of critical equipment in industrial production are increasingly drawing widespread public attention, particularly in the petrochemical and energy sectors, where safety accidents due to equipment malfunctions occur frequently. Rotating machinery, such as steam turbines, fans, and pumps, are critical equipment in industrial production. Equipment failures can lead to unexpected unit shutdowns, resulting in high operating and maintenance costs, and potentially causing significant economic losses and personnel accidents. The purpose of rotating machinery fault diagnosis is to detect, identify, and locate faults before they deteriorate, which is crucial for ensuring the safe and reliable operation of rotating machinery.

[0030] However, most AI-based fault diagnosis methods for rotating machinery require training an independent model for each machine and suffer from poor generalization ability, failing to handle the varying data distributions caused by differences in structure and operating conditions between devices. To address this issue, transfer learning and domain adaptation are introduced into the field of fault diagnosis. However, most existing transfer learning methods can only handle variations in operating conditions such as speed and load; when applied to fault diagnosis of multiple machines, these methods still require a large amount of fault data and a complete range of fault types to support the training process, which is unacceptable in practical industrial applications.

[0031] To address the challenges of limited fault samples for single-device systems and poor local model performance, which hinders multi-device, multi-fault diagnosis, joint learning is considered. Joint learning can learn fault diagnosis models from dispersed data while ensuring privacy data remains within the local machine. However, in the scenario of rotating machinery fault diagnosis, traditional joint learning cannot meet the needs of heterogeneous equipment, data, and models. Training the same model for all participants does not yield ideal results. Furthermore, due to limited computing power among participants, training a single large model for each participant is impractical, and frequent uploading and downloading of model parameters incurs high communication overhead, impacting the efficiency of ordinary joint learning. Therefore, how to coordinate computing resources, better utilize server-side computing power and knowledge, and reduce communication overhead are urgent problems to be solved in rotating machinery fault diagnosis.

[0032] Joint knowledge distillation effectively addresses these issues by combining joint learning with knowledge distillation, protecting data privacy, saving communication costs, and supporting heterogeneous models. However, directly employing traditional knowledge distillation methods can easily lead to the propagation of "bad knowledge." How to accurately select knowledge and mine hard samples remains a pressing problem.

[0033] The following description, with reference to the accompanying drawings, outlines a method, apparatus, client, and server for diagnosing equipment faults according to embodiments of this application. Specifically, Figure 1 This is a flowchart illustrating a method for diagnosing faults in moving equipment provided in an embodiment of this application.

[0034] like Figure 1 As shown, the fault diagnosis method for moving equipment is applied to a client set up in any participating party. The client includes a first model. The method includes the following steps:

[0035] In step S101, the first dataset and the first public validation set are obtained.

[0036] The first dataset can be a local private dataset within each client, and the first public validation set is a dataset composed of a portion of data collected from each participating client. No specific restrictions are imposed here.

[0037] It is understood that, in this embodiment of the application, the first dataset and the first public validation set are obtained from the client to facilitate the subsequent training of the first model.

[0038] In step S102, the first model is trained using the first dataset and the first public validation set, the first hard sample in the first public validation set is mined, the first hard sample and the first predicted value are uploaded to the server, and the second predicted value and the mined second hard sample are received from the server during the training of the second model.

[0039] The first model can be a small model within each client, without any specific restrictions.

[0040] The second model can be a large model on the server side, without specific limitations.

[0041] The first difficult sample includes multiple feature maps, which refer to the feature maps transmitted from the client to the server, without specific limitations.

[0042] The second difficult sample includes multiple feature maps, which refer to the feature maps transmitted from the server to the client, without specific limitations.

[0043] The first predicted value can be the client's predicted value on the public validation set, and can be predicted according to the actual situation, without specific limitations here.

[0044] The second predicted value can be the predicted value of the server on the public validation set. It can be predicted according to the actual situation, and no specific limitation is made here.

[0045] It is understood that in this embodiment of the application, the first model is trained using the first dataset and the first public validation set, the first hard sample of the first public validation set is mined, and the first hard sample and the first predicted value are uploaded to the server. The client also receives the second predicted value and the mined second hard sample from the server in training the second model, so that the client and the server can transmit the predicted value and feature map, so as to facilitate subsequent training using joint bidirectional knowledge distillation.

[0046] In this embodiment of the application, mining the first hard sample in the first public validation set includes: obtaining the first predicted value and the first true label on the public validation set; traversing all samples in the public validation set, and if the difference between the first predicted value and the first true label is greater than a first threshold, the index of the corresponding sample is added to the hard sample index set until the traversal is completed, and the first hard sample is obtained.

[0047] The first real label can be the client's real label on the public verification set, without any specific restrictions.

[0048] The first threshold can be a threshold set by the user in advance, such as threshold δ, which can be modified or adjusted according to the actual situation, and is not specifically limited here.

[0049] It is understood that, in this embodiment of the application, by obtaining the first predicted value and the first true label on the public verification set, the sample of all public verification sets is traversed. If the difference between the first predicted value and the first true label is greater than a set threshold, the index of the sample is added to the hard sample index set until the traversal is completed and the first hard sample is obtained. Based on the condition that the difference between the predicted value and the label is greater than the set threshold, hard sample mining is performed to facilitate subsequent joint bidirectional knowledge distillation.

[0050] In this embodiment of the application, uploading the first difficult sample and the first predicted value to the server includes: selecting feature maps that meet the first preset conditions from the first difficult sample; quantizing and compressing the feature maps that meet the first preset conditions to obtain a first compressed feature map; and uploading the first compressed feature map and the first predicted value to the server.

[0051] The first preset condition can be that the client satisfies the feature map corresponding to the sample in the hard sample index, without being specifically limited here.

[0052] The first compressed feature map can be obtained by quantizing and compressing the feature map that meets the conditions of the client, and no specific limitation is made here.

[0053] It is understood that in this embodiment of the application, feature maps that meet the first preset conditions are selected from the first difficult samples, the feature maps that meet the conditions are quantized and compressed to obtain the first compressed feature map. While ensuring performance, the communication cost is reduced, and the first compressed feature map and the first predicted value are uploaded to the server so that they can be trained by joint bidirectional knowledge distillation in the future.

[0054] In step S103, joint bidirectional knowledge distillation is performed based on the predicted values ​​and hard samples until both the first model and the second model converge, resulting in a trained joint learning model. The joint learning model is then used to diagnose equipment failures of any participant.

[0055] In this application, the participating parties are all factories, and no specific limitations are made here.

[0056] It is understood that the embodiments of this application perform joint bidirectional knowledge distillation based on all predicted values ​​and hard samples until both the first model and the second model converge to obtain a trained joint learning model, which is then used to diagnose equipment faults of any participant. This not only enables accurate fault location and saves communication costs, but also supports model heterogeneity and improves model performance.

[0057] Specifically, the joint bidirectional knowledge distillation process based on predicted values ​​and hard samples is as follows: For the factory (i.e., the client): First, the local model is trained using a local private dataset. Then, the predicted values ​​and selected feature maps transmitted from the server are used to train the local model on a public validation set for knowledge distillation. Further, the local model is used to predict values ​​and feature maps on the public validation set. Based on the hard sample index required by the server, the obtained feature maps are selected and quantized to reduce the size of the selected feature maps. Then, based on the predicted values ​​of the local model on the public validation set, the hard sample index required by the factory is obtained. Finally, the compressed selected feature maps, predicted values, and the hard sample index required by the factory are transmitted to the server.

[0058] According to the fault diagnosis method for moving equipment proposed in this application, a first dataset and a first public validation set are obtained. A first model is trained using the first dataset and the first public validation set. First hard samples from the first public validation set are mined, and the first hard samples and first predicted values ​​are uploaded to the server. The client also receives second predicted values ​​and mined second hard samples from the server during the training of a second model. This allows for the transmission of predicted values ​​and feature maps between the client and the server. Joint bidirectional knowledge distillation is performed based on all predicted values ​​and hard samples until both the first and second models converge, resulting in a trained joint learning model. This model is then used to diagnose faults in moving equipment involving any participant. This method achieves accurate fault location, saves communication costs, supports model heterogeneity, and improves model performance. Therefore, it solves the problems in related technologies, such as the limited number of fault samples for single-device faults, poor local model performance, and the inability to diagnose faults in multiple devices.

[0059] Next, with reference to the accompanying drawings, a method for diagnosing faults in moving equipment proposed according to embodiments of this application is described.

[0060] Figure 2 This is a flowchart of a fault diagnosis method for moving equipment according to an embodiment of this application.

[0061] like Figure 2 As shown, this equipment fault diagnosis method is applied to a server, which includes a second model. The method includes the following steps:

[0062] In step S201, the second dataset and the second public validation set are obtained.

[0063] The second dataset includes a public dataset consisting of multiple types of faults (faults of shafts, bearings, gearboxes, etc.) and a large amount of publicly available data, such as fault simulation experimental data from universities and research institutions obtained from the internet. No specific limitations are made here.

[0064] The second public validation set includes a public validation set jointly constructed with the client factory, which is a dataset composed of data collected from each participating factory, and is not specifically limited here.

[0065] It is understood that the embodiments of this application obtain the second dataset and the second public validation set from the server in order to facilitate the subsequent training of the second model.

[0066] In step S202, the second model is trained using the second dataset and the second public validation set, the second hard sample of the second public validation set is mined, the second hard sample and the second predicted value are sent to any client, and the first predicted value and the mined first hard sample of any client are received during the training of the first model.

[0067] It is understood that the embodiments of this application use a second dataset and a second public validation set to train a second model, and mine a second hard sample from the second public validation set. The second hard sample and the second predicted value are sent to any client, and the first predicted value and the first hard sample mined from any client during the training of the first model are received. This enables the client and the server to transmit predicted values ​​and feature maps, so that they can be trained subsequently using joint bidirectional knowledge distillation.

[0068] In this embodiment of the application, mining the second hard sample in the second public validation set includes: obtaining the second predicted value and the second true label of the second sample in the public validation set; traversing all samples in the public validation set, if the difference between the second predicted value and the second true label is greater than a second threshold, then the index of the corresponding sample is added to the hard sample index set, until the traversal is completed, and the second hard sample is obtained.

[0069] The second real label can be the real label of the server on the public verification set, without being specifically limited here.

[0070] The second threshold can be a threshold set by the user in advance, such as threshold δ, which can be modified or adjusted according to the actual situation, and is not specifically limited here.

[0071] It is understood that, in this embodiment of the application, by obtaining the second predicted value and the second true label on the public validation set, all samples of the second on the public validation set are traversed. When the difference between the second predicted value and the second true label is greater than a set threshold, the index of the corresponding sample is added to the hard sample index set. This process continues until the traversal is completed, and the second hard sample is obtained. Based on the condition that the difference between the predicted value and the label is greater than the set threshold, hard sample mining is performed to facilitate subsequent joint bidirectional knowledge distillation.

[0072] In this embodiment of the application, the second difficult sample includes multiple feature maps. Sending the second difficult sample and the second predicted value to any client includes: selecting feature maps that meet the second preset conditions from the second difficult sample; quantizing and compressing the feature maps that meet the second preset conditions to obtain a second compressed feature map; and sending the second compressed feature map and the second predicted value to any client.

[0073] The second preset condition can be that the server satisfies the feature map corresponding to the sample in the hard sample index, without being specifically limited here.

[0074] The second compressed feature map can be obtained by quantizing and compressing the feature map that meets the conditions on the server side, and no specific limitation is made here.

[0075] It is understood that in this embodiment, feature maps that meet the second preset conditions are selected from the second difficult samples, the feature maps that meet the conditions are quantized and compressed to obtain the second compressed feature map. While ensuring performance, the communication cost is reduced, and the second compressed feature map and the second prediction value are sent to the server so that they can be trained by joint bidirectional knowledge distillation in the future.

[0076] In step S203, joint bidirectional knowledge distillation is performed based on the predicted values ​​and hard samples until both the first model and the second model converge, resulting in a trained joint learning model. The joint learning model is then used to diagnose equipment failures of any participant.

[0077] It is understood that the embodiments of this application perform joint bidirectional knowledge distillation based on all predicted values ​​and hard samples until both the first model and the second model converge to obtain a trained joint learning model, which is then used to diagnose equipment faults of any participant. This not only enables accurate fault location and saves communication costs, but also supports model heterogeneity and improves model performance.

[0078] Specifically, the process of joint bidirectional knowledge distillation based on predicted values ​​and hard samples is as follows:

[0079] For the server: First, obtain the predicted values ​​and selected feature maps transmitted from all factory clients, and then perform average aggregation on the predicted values ​​and selected feature maps. Next, use the predicted values ​​and selected feature maps to train the server-side model using knowledge distillation on the public validation set. Further, use the server-side model to predict values ​​and feature maps on the public validation set. Based on the hard sample index required by each factory, select the obtained feature maps and perform quantization compression to reduce the size of the selected feature maps. Then, based on the predicted values ​​of the server-side model on the public validation set, obtain the hard sample index required by the server. Finally, transmit the compressed selected feature maps, predicted values, and the hard sample index required by the server to each factory client respectively.

[0080] According to the fault diagnosis method for moving equipment proposed in this application, a second dataset and a second public validation set are obtained. A first model is trained using the second dataset and the second public validation set. First hard samples from the second public validation set are mined, and the second hard samples and second predicted values ​​are sent to the client. The server also receives the first predicted values ​​and the mined first hard samples from the client during the training of the first model, enabling the client and server to transmit predicted values ​​and feature maps. Joint bidirectional knowledge distillation is performed based on all predicted values ​​and hard samples until both the first and second models converge, resulting in a trained joint learning model. This model is then used to diagnose faults in moving equipment involving any participant. This method achieves accurate fault location, saves communication costs, supports model heterogeneity, and improves model performance. Therefore, it solves the problems in related technologies such as limited fault samples for single devices, poor local model performance, and inability to diagnose faults in multiple devices.

[0081] The following will combine Figure 3 and Figure 4 The methods for diagnosing faults in moving equipment are described in detail below:

[0082] (1) With the factory as the participant and a platform operator as the joint learning server, a joint learning framework for a large-scale equipment fault diagnosis model is constructed.

[0083] 1) The server possesses a large model, and the dataset includes a public dataset comprising multiple types of faults and a large amount of publicly available data, as well as a public validation set jointly constructed by all client factories. The server uses the public dataset to pre-train its large model. The multiple types of faults include those related to shaft systems, bearings, gearboxes, etc.

[0084] Among them, the public dataset refers to the fault simulation experiment data of universities and research institutions obtained from the Internet; the public validation set is a dataset composed of a portion of data collected from each participating factory.

[0085] 2) Each factory has its own small model, with datasets including local private datasets and public validation sets. Model heterogeneity is supported between factories, enabling the diagnosis of faults in one or two types of equipment. Factories train their models on their local private datasets.

[0086] 3) The server transmits predicted values ​​and feature maps to each factory, and trains the model using a bidirectional joint knowledge distillation method until the joint learning model converges. Specifically:

[0087] For the factory (i.e., the client): First, the local model is trained using a local private dataset. Then, the local model is trained on a public validation set using the predicted values ​​and selected feature maps transmitted from the server. Further, the local model is used to predict values ​​and feature maps on the public validation set. Based on the hard sample index required by the server, the obtained feature maps are selected and quantized to reduce their size. Then, based on the predicted values ​​of the local model on the public validation set, the hard sample index required by the factory is obtained. Finally, the compressed selected feature maps, predicted values, and the hard sample index required by the factory are transmitted to the server.

[0088] For the server: First, obtain the predicted values ​​and selected feature maps transmitted from all factory clients, and then perform average aggregation on the predicted values ​​and selected feature maps. Next, use the predicted values ​​and selected feature maps to train the server-side model using knowledge distillation on the public validation set. Further, use the server-side model to predict values ​​and feature maps on the public validation set. Based on the hard sample index required by each factory, select the obtained feature maps and perform quantization compression to reduce the size of the selected feature maps. Then, based on the predicted values ​​of the server-side model on the public validation set, obtain the hard sample index required by the server. Finally, transmit the compressed selected feature maps, predicted values, and the hard sample index required by the server to each factory client respectively.

[0089] (2) In terms of communication, since transmitting feature maps requires a lot of communication overhead, feature maps are selected based on difficult samples for uploading, and the selected feature maps are further quantized and compressed to ensure performance while reducing communication costs by 7 to 10 times.

[0090] The process for obtaining the hard sample indexes required by the factory / server is as follows:

[0091] 1) Obtain the predicted values ​​and true labels of the factory / server on the public validation set;

[0092] 2) Iterate through all samples in the public validation set. If the difference between the predicted value and the true label is greater than the threshold δ, add the index of the sample to the hard sample index set.

[0093] The transmission of predicted values ​​and feature maps between the server and each factory refers to the transmission of feature maps corresponding to samples indexed by hard samples, while the predicted values ​​are uploaded directly.

[0094] 3) After completing the sample traversal, the resulting set of hard sample indexes is the hard sample index required by the factory / server.

[0095] The selection of feature maps refers to: retaining the feature maps corresponding to samples whose index is in the hard sample index, and discarding the feature maps corresponding to samples whose index is not in the hard sample index.

[0096] In summary, the bidirectional mining method for difficult samples in moving equipment fault diagnosis based on joint knowledge distillation provided in this application can greatly save communication overhead, avoid the spread of "bad knowledge", and improve the overall performance of the model by first selecting difficult samples on the server / client side, determining the index of the difficult samples, and then compressing and uploading the corresponding feature maps during the joint bidirectional knowledge distillation process of moving equipment fault diagnosis.

[0097] Next, the fault diagnosis device for moving equipment proposed according to the embodiments of this application is described with reference to the accompanying drawings.

[0098] Figure 5 This is a block diagram of a fault diagnosis device for moving equipment according to an embodiment of this application.

[0099] like Figure 5 As shown, the equipment fault diagnosis device 10 is applied to a client set up in any participating party. The client includes a first model. The device includes a first acquisition module 110, a first training module 120 and a first distillation module 130.

[0100] The first acquisition module 110 is used to acquire a first dataset and a first public validation set; the first training module 120 is used to train a first model using the first dataset and the first public validation set, mine the first hard sample in the first public validation set, upload the first hard sample and the first predicted value to the server, and receive the second predicted value and the mined second hard sample from the server during the training of the second model; the first distillation module 130 is used to perform joint bidirectional knowledge distillation based on the predicted value and the hard sample until both the first model and the second model converge, to obtain a trained joint learning model, and to use the joint learning model to diagnose the faults of the moving equipment of any participant.

[0101] It should be noted that the foregoing explanation of the embodiments of the moving equipment fault diagnosis method also applies to the moving equipment fault diagnosis device of this embodiment, and will not be repeated here.

[0102] According to the equipment fault diagnosis device proposed in this application, a first dataset and a first public validation set are acquired. A first model is trained using the first dataset and the first public validation set. First hard samples from the first public validation set are mined, and the first hard samples and first predicted values ​​are uploaded to the server. The client also receives second predicted values ​​and mined second hard samples from the server during the training of a second model. This allows for the transmission of predicted values ​​and feature maps between the client and the server. Joint bidirectional knowledge distillation is performed based on all predicted values ​​and hard samples until both the first and second models converge, resulting in a trained joint learning model. This model is then used to diagnose equipment faults involving any participant. This achieves accurate fault location, saves communication costs, supports model heterogeneity, and improves model performance. Therefore, it solves the problems in related technologies such as limited fault samples for single devices, poor local model performance, and inability to diagnose faults in multiple devices.

[0103] Next, the fault diagnosis device for moving equipment proposed according to the embodiments of this application is described with reference to the accompanying drawings.

[0104] Figure 6 This is a block diagram of a fault diagnosis device for moving equipment according to an embodiment of this application.

[0105] like Figure 6 As shown, the fault diagnosis device 20 for the moving equipment is applied to the server. The server includes a second model. The device includes a second acquisition module 210, a second training module 220, and a second distillation module 230.

[0106] The second acquisition module 210 is used to acquire the second dataset and the second public validation set; the second training module 220 is used to train the second model using the second dataset and the second public validation set, mine the second hard sample in the second public validation set, send the second hard sample and the second predicted value to any client, and receive the first predicted value and the first hard sample mined by any client during the training of the first model; the second distillation module 230 is used to perform joint bidirectional knowledge distillation based on the predicted value and the hard sample until both the first model and the second model converge, to obtain the trained joint learning model, and to use the joint learning model to diagnose the faults of the moving equipment of any participant.

[0107] It should be noted that the foregoing explanation of the embodiments of the moving equipment fault diagnosis method also applies to the moving equipment fault diagnosis device of this embodiment, and will not be repeated here.

[0108] According to the equipment fault diagnosis device proposed in this application, a second dataset and a second public validation set are acquired. A first model is trained using the second dataset and the second public validation set. First hard samples from the second public validation set are mined, and the second hard samples and second predicted values ​​are sent to the client. The server also receives the first predicted values ​​and the mined first hard samples from the client during the training of the first model, enabling the client and server to transmit predicted values ​​and feature maps. Joint bidirectional knowledge distillation is performed based on all predicted values ​​and hard samples until both the first and second models converge, resulting in a trained joint learning model. This model is then used to diagnose equipment faults involving any participant, achieving accurate fault location, saving communication costs, supporting model heterogeneity, and improving model performance. This solves the problems in related technologies, such as limited fault samples for single devices, poor local model performance, and the inability to diagnose faults in multiple devices.

[0109] Figure 7 A schematic diagram of the client provided in an embodiment of this application. The client may include:

[0110] The memory 701, the processor 702, and the computer program stored on the memory 701 and executable on the processor 702.

[0111] When the processor 702 executes the program, it implements the equipment fault diagnosis method provided in the above embodiments.

[0112] Furthermore, the client also includes:

[0113] Communication interface 703 is used for communication between memory 701 and processor 702.

[0114] The memory 701 is used to store computer programs that can run on the processor 702.

[0115] The memory 701 may include high-speed RAM (Random Access Memory) memory, and may also include non-volatile memory, such as at least one disk storage.

[0116] If the memory 701, processor 702, and communication interface 703 are implemented independently, then the communication interface 703, memory 701, and processor 702 can be interconnected via a bus to complete communication between them. The bus can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus, or an EISA (Extended Industry Standard Architecture) bus, etc. The bus can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 7 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0117] Optionally, in a specific implementation, if the memory 701, processor 702, and communication interface 703 are integrated on a single chip, then the memory 701, processor 702, and communication interface 703 can communicate with each other through an internal interface.

[0118] The processor 702 may be a CPU (Central Processing Unit), an ASIC (Application Specific Integrated Circuit), or one or more integrated circuits configured to implement the embodiments of this application.

[0119] Figure 8 A schematic diagram of the server structure provided in an embodiment of this application. The server may include:

[0120] The memory 801, the processor 802, and the computer program stored on the memory 801 and capable of running on the processor 802.

[0121] When the processor 802 executes the program, it implements the equipment fault diagnosis method provided in the above embodiments.

[0122] Furthermore, the server-side also includes:

[0123] Communication interface 803 is used for communication between memory 801 and processor 802.

[0124] The memory 801 is used to store computer programs that can run on the processor 802.

[0125] The memory 801 may include high-speed RAM (Random Access Memory) memory, and may also include non-volatile memory, such as at least one disk storage.

[0126] If the memory 801, processor 802, and communication interface 803 are implemented independently, then the communication interface 803, memory 801, and processor 802 can be interconnected via a bus to complete communication between them. The bus can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus, or an EISA (Extended Industry Standard Architecture) bus, etc. The bus can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 8 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0127] Optionally, in a specific implementation, if the memory 801, processor 802, and communication interface 803 are integrated on a single chip, then the memory 801, processor 802, and communication interface 803 can communicate with each other through an internal interface.

[0128] The processor 802 may be a CPU (Central Processing Unit), an ASIC (Application Specific Integrated Circuit), or one or more integrated circuits configured to implement the embodiments of this application.

[0129] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0130] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "N" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0131] Any process or method described in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more N executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0132] It should be understood that the various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, the N steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (FPGAs), field-programmable gate arrays (FPGAs), etc.

[0133] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

[0134] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. A method for diagnosing faults in moving equipment, characterized in that, The method is applied to a client, the client including a first model, wherein the method includes the following steps: Obtain the first dataset and the first public validation set; The first model is trained using the first dataset and the first public validation set. The first hard sample in the first public validation set is mined. The first hard sample and the first predicted value are uploaded to the server. The second predicted value and the mined second hard sample are received from the server during the training of the second model. Based on the predicted values ​​and hard samples, joint bidirectional knowledge distillation is performed until both the first model and the second model converge, resulting in a trained joint learning model. The joint learning model is then used to diagnose equipment failures of any participant. For the client, the joint bidirectional knowledge distillation based on predicted values ​​and hard samples includes: using the predicted values ​​and selected feature maps transmitted from the server to train the first model on a public validation set for knowledge distillation. The second model is the large model on the server side; Obtain the predicted values ​​and true labels of the first / second model on the public validation set; iterate through all samples in the public validation set, and if the difference between the predicted value and the true label is greater than a threshold, add the index of the corresponding sample to the hard sample index set; complete the sample traversal to obtain the hard samples. The collected data is vibration data.

2. The method according to claim 1, characterized in that, The mining of the first hard sample in the first public validation set includes: Obtain the first predicted value and the first true label of the first model on the public validation set; Iterate through all samples of the first model in the public validation set. If the difference between the first predicted value and the first true label is greater than the first threshold, the index of the corresponding sample is added to the hard sample index set. Continue until the iteration is complete to obtain the first hard sample.

3. The method according to claim 1 or 2, characterized in that, The first difficult sample includes multiple feature maps, and uploading the first difficult sample and the first predicted value to the server includes: Select feature maps that satisfy the first preset conditions from the first difficult samples; The feature map that meets the first preset condition is quantized and compressed to obtain the first compressed feature map. The first compressed feature map and the first predicted value are then uploaded to the server.

4. A method for diagnosing faults in moving equipment, characterized in that, The method is applied to a server, which includes a second model, wherein the method includes the following steps: Obtain the second dataset and the second public validation set; The second model is trained using the second dataset and the second public validation set. The second hard sample in the second public validation set is mined. The second hard sample and the second predicted value are sent to any client. The first predicted value and the mined first hard sample are received from the client during the training of the first model. Based on the predicted values ​​and hard samples, joint bidirectional knowledge distillation is performed until both the first model and the second model converge, resulting in a trained joint learning model. The joint learning model is then used to diagnose the equipment failure of any participant. For the server, the joint bidirectional knowledge distillation based on predicted values ​​and hard samples includes: obtaining the predicted values ​​and selected feature maps transmitted by the first model, and averaging and aggregating the predicted values ​​and selected feature maps; and training the server model with knowledge distillation using the predicted values ​​and selected feature maps on a public validation set. The first model is a small model within each client; Obtain the predicted values ​​and true labels of the first / second model on the public validation set; iterate through all samples in the public validation set, and if the difference between the predicted value and the true label is greater than a threshold, add the index of the corresponding sample to the hard sample index set; complete the sample traversal to obtain the hard samples. The collected data is vibration data.

5. The method according to claim 4, characterized in that, The mining of the second hard sample in the second public validation set includes: Obtain the second predicted value and the second true label of the second model on the public validation set; Iterate through all samples of the second model in the public validation set. If the difference between the second predicted value and the second true label is greater than the second threshold, the index of the corresponding sample is added to the hard sample index set. Continue until the iteration is complete to obtain the second hard sample.

6. The method according to claim 4 or 5, characterized in that, The second difficult sample includes multiple feature maps. Sending the second difficult sample and the second predicted value to any client includes: Select feature maps that satisfy the second preset conditions from the second difficult samples; The feature map that meets the second preset condition is quantized and compressed to obtain the second compressed feature map. The second compressed feature map and the second predicted value are then sent to any client.

7. A fault diagnosis device for moving equipment, characterized in that, The device is applied to a client, the client including a first model, wherein the device includes: The first acquisition module is used to acquire the first dataset and the first public validation set; The first training module is used to train the first model using the first dataset and the first public validation set, mine the first hard sample in the first public validation set, upload the first hard sample and the first predicted value to the server, and receive the second predicted value and the mined second hard sample from the server during the training of the second model. The first distillation module is used to perform joint bidirectional knowledge distillation based on predicted values ​​and hard samples until both the first model and the second model converge, thereby obtaining a trained joint learning model. The joint learning model is then used to diagnose equipment failures of any participant. For the client, the joint bidirectional knowledge distillation based on predicted values ​​and hard samples includes: using the predicted values ​​and selected feature maps transmitted from the server to train the first model on a public validation set for knowledge distillation. The second model is the large model on the server side; Obtain the predicted values ​​and true labels of the first / second model on the public validation set; iterate through all samples in the public validation set, and if the difference between the predicted value and the true label is greater than a threshold, add the index of the corresponding sample to the hard sample index set; complete the sample traversal to obtain the hard samples. The collected data is vibration data.

8. A fault diagnosis device for moving equipment, characterized in that, The device is applied to a server, the server including a second model, wherein the device includes: The second acquisition module is used to acquire the second dataset and the second public validation set; The second training module is used to train the second model using the second dataset and the second public validation set, mine the second hard sample in the second public validation set, send the second hard sample and the second predicted value to any client, and receive the first predicted value and the first hard sample mined by the arbitrary client during the training of the first model. The second distillation module is used to perform joint bidirectional knowledge distillation based on predicted values ​​and hard samples until both the first model and the second model converge, thereby obtaining a trained joint learning model. The joint learning model is then used to diagnose the equipment failure of any participant. For the server, the joint bidirectional knowledge distillation based on predicted values ​​and hard samples includes: obtaining the predicted values ​​and selected feature maps transmitted by the first model, and averaging and aggregating the predicted values ​​and selected feature maps; and training the server model with knowledge distillation using the predicted values ​​and selected feature maps on a public validation set. The first model is a small model within each client; Obtain the predicted values ​​and true labels of the first / second model on the public validation set; iterate through all samples in the public validation set, and if the difference between the predicted value and the true label is greater than a threshold, add the index of the corresponding sample to the hard sample index set; complete the sample traversal to obtain the hard samples. The collected data is vibration data.

9. A client, characterized in that, include: A memory, a processor, and a computer program stored in the memory and executable on the processor, the processor executing the program to implement the fault diagnosis method for moving equipment as described in any one of claims 1-3.

10. A server-side component, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor, the processor executing the program to implement the fault diagnosis method for moving equipment as described in any one of claims 4-6.

Images