CN112814854B

CN112814854B - Joint learning-based turbine fan maintenance method and device

Info

Publication number: CN112814854B
Application number: CN202011630307.1A
Authority: CN
Inventors: 张敏; 高庆
Original assignee: Ennew Digital Technology Co Ltd
Current assignee: Ennew Digital Technology Co Ltd
Priority date: 2020-12-31
Filing date: 2020-12-31
Publication date: 2022-04-29
Anticipated expiration: 2040-12-31
Also published as: CN112814854A

Abstract

The invention is suitable for the technical field of operation and maintenance of a turbine fan, and provides a method and a device for maintaining the turbine fan based on joint learning, wherein the method comprises the following steps: receiving a global model issued by a server under a joint learning architecture; acquiring historical operating data of the turbine fan, and generating sample data for training the global model; training the global model by using the sample data, and updating a local model; and if the local model reaches a preset convergence condition, creating a model for the predictive maintenance of the turbine fan, and performing the predictive maintenance on the turbine fan by using the model. The model obtained by training is higher in quality, so that the accuracy of operation and maintenance prediction is improved; in addition, the aggregation mode of the server side is improved, so that the training efficiency of the prediction model is further improved.

Description

Joint learning-based turbine fan maintenance method and device

Technical Field

The invention belongs to the technical field of energy, and particularly relates to a method and a device for maintaining a turbofan based on joint learning.

Background

The turbine fan is maintained at present and is mostly maintained with "passive form fortune, mainly relies on-the-spot staff to carry out periodic maintenance and troubleshooting, and this kind of simple dependence workman squats the point and maintains, and fortune dimension cost is high, also appears easily because the staff level is inconsistent leads to the power generation loss even operation safety problem.

In the prior art, a machine learning algorithm is also used for establishing a prediction model to perform prediction maintenance on the turbine fan. However, it has been found experimentally that the quality of the predictive model built using the operational characteristic data of the turbine fan is not high. Therefore, how to further improve the accuracy of the maintenance model of the turbine fan to ensure the accuracy and efficiency of the operation and maintenance of the fan is a current technical problem.

Disclosure of Invention

In view of this, embodiments of the present invention provide a method and an apparatus for maintaining a turbofan based on joint learning, so as to solve the problem that the operation and maintenance efficiency and accuracy of the turbofan are not high.

In a first aspect of the present invention, a method for maintaining a turbofan based on joint learning is provided, including: receiving a global model issued by a server under a joint learning architecture; acquiring historical operating data of the turbine fan, and generating sample data for training the global model; training the global model by using the sample data, and updating a local model; and if the local model reaches a preset convergence condition, creating a model for the predictive maintenance of the turbine fan, and performing the predictive maintenance on the turbine fan by using the model.

In some alternatives, the historical operating data includes: a temperature of the turbo fan, a pressure of the turbo fan, and/or a rotational speed of the turbo fan.

In a second aspect of the present invention, a method for maintaining a turbofan based on joint learning is provided, which includes: receiving a local model uploaded by a client; aggregating the local models and updating the global model; and determining the next global model updating time according to the updating result of the updated global model, wherein the updating result comprises information about whether the updating is successful and the updated attribute information.

In some alternatives, the receiving the local model uploaded by the client comprises: detecting the local model catalog and obtaining a detection result; and receiving the local model according to the detection result.

In a third aspect of the present invention, a combined learning based maintenance apparatus for a turbofan is provided, which includes: the global model receiving module is used for receiving a global model issued by a server under the joint learning architecture; the sample data generating module is used for acquiring historical operating data of the turbine fan and generating sample data used for training the global model; the global model training module is used for training the global model by using the sample data and updating a local model; and the fan prediction maintenance module is used for creating a model for the turbine fan prediction maintenance if the local model reaches a preset convergence condition, and performing prediction maintenance on the turbine fan by using the model.

In a fourth aspect of the present invention, a combined learning-based maintenance apparatus for a turbofan is provided, which includes: the model receiving module is used for receiving the local model uploaded by the client; the model aggregation module is used for aggregating the local models and updating the global model; and the iteration updating module is used for determining the next updating time of the global model according to the updating result of the updated global model, wherein the updating result comprises information about whether the updating is successful and the updated attribute information.

In some alternatives, the model receiving module specifically includes: the directory checking unit is used for detecting the local model directory and obtaining a detection result; and the model receiving unit is used for receiving the local model according to the detection result.

In a fifth aspect of the present invention, there is provided an electronic device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, the processor implementing the steps of the method according to any one of the first and second aspects when executing the computer program.

A sixth aspect of the invention provides a storage medium storing a computer program, wherein the computer program, when executed by a processor, implements the steps of the method according to any one of the first and second aspects.

Compared with the prior art, the embodiment of the invention has the following beneficial effects: the method realizes the training of the operation and maintenance prediction model of the turbofan through the joint learning architecture, improves sample data of local training, and improves the quality of the model obtained by training, so that the accuracy of the operation and maintenance prediction is improved; in addition, the aggregation mode of the server side is improved, so that the training efficiency of the prediction model is further improved.

Drawings

In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings needed for the embodiments or the prior art descriptions will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings without creative efforts.

FIG. 1 is a joint learning architecture to which the following embodiments of the present invention may be applied;

FIG. 2 is a flow chart of a method for joint learning based maintenance of a turbofan according to one embodiment of the invention;

fig. 3 is a structural diagram of a joint learning-based maintenance apparatus for a turbofan according to a third embodiment of the present invention.

Detailed Description

In the following description, for purposes of explanation and not limitation, specific details are set forth, such as particular system structures, techniques, etc. in order to provide a thorough understanding of the embodiments of the invention. It will be apparent, however, to one skilled in the art that the present invention may be practiced in other embodiments that depart from these specific details. In other instances, detailed descriptions of well-known systems, devices, circuits, and methods are omitted so as not to obscure the description of the present invention with unnecessary detail.

Fig. 1 is a joint learning architecture to which the following embodiments of the present invention can be applied, as shown in fig. 1, in which the joint learning architecture includes: the system comprises a server 101 and a plurality of

clients

102, 103 and 104, wherein the plurality of

clients

102, 103 and 104 are respectively connected with the server in a communication mode.

The server 101 is a trust center for joint learning, which is also called a central server; the

clients

102, 103, and 104 are clients of the joint learning. Generally, the number of clients can be K, K ≧ 2 and a positive integer, such as client-1 and client-2 … in FIG. 1.

Specifically, the client may specifically include, but is not limited to, an electronic device such as a computer, a server, a workstation, and the like. Illustratively, the electronic device may specifically include: the system comprises a memory, a processor and a computer program stored in the memory and capable of running on the processor, wherein when the computer program is executed by the processor, the steps of the complete method are realized as in any one of the following embodiments.

The joint learning allows a plurality of users to cooperatively train the shared global model without sharing data in the local equipment, and the central server coordinates and completes a plurality of rounds of joint learning to obtain the final global model. Wherein, at the beginning of each round, the central server sends the current global model to the clients participating in the joint learning; each client trains the received global model according to local data of the client, and returns the updated model to the central server after the training is finished; and after the central server collects the updates returned by all the clients, the global model is updated once, and the updating of the current round is finished.

Therefore, in the following embodiments of the present invention, the operation and maintenance data of the turbine fan is used to perform training and improvement on the maintenance model through a joint learning method, so as to improve the operation and maintenance efficiency and accuracy of fan maintenance.

Example one

The embodiment will describe the joint learning based turbine maintenance method provided by the invention from the client side in the joint learning architecture, that is, the execution subject of the embodiment is the client of the joint learning, for example, the method provided by the embodiment is executed by the

clients

102, 103, and 104 shown in fig. 1.

FIG. 2 is a flowchart illustrating a method for jointly learning based maintenance of a turbofan according to an embodiment of the present invention, and as shown in FIG. 2, the method for jointly learning based maintenance of a turbofan includes the following steps S01-S04.

S101: and receiving a global model issued by a server under the joint learning architecture.

Besides the passive receiving of the global model issued by the server, the step also includes actively downloading the global model from the server. Specifically, the global model includes a server-side initialized global model, for example, a global model issued at a place where the joint learning task starts. In addition, the global model also comprises a local model uploaded by the data nodes where the various turbo fans participating in the joint learning are aggregated by the server.

S102: and acquiring historical operating data of the turbine fan, and generating sample data for training the global model.

The historical operating data comprises measuring point data such as the temperature of the turbine fan, the pressure of the turbine fan, the rotating speed of the turbine fan and the like.

Specifically, after the operational data of the turbofan is acquired, the operational data can be input into the NASA simulation software to generate sample data, so that the sample data can be consistent with the actual operational condition, and the trained model has higher precision. It should be appreciated that the operational conditions of the turbine fan can be obtained by using the data input simulation software, so that high-quality sample data is formed.

S103: and training the global model by using the sample data, and updating a local model.

Under the joint learning architecture, the global model obtained from the server side is an intermediate parameter of the global model, for example, a gradient of a training model. Therefore, the parameters corresponding to the global model are utilized, the sample data is utilized to carry out training locally, and the parameters of the model are further updated in the training process, so that the local model is obtained.

S104: and if the local model reaches a preset convergence condition, creating a model for the predictive maintenance of the turbine fan, and performing the predictive maintenance on the turbine fan by using the model.

Wherein, the judgment whether the convergence condition is reached can be updated by a preset iteration number, that is, the number of times of repeating the above steps S101 to S103. Alternatively, the accuracy of the model may be preset, for example, after the local model is updated, whether the local model meets the preset accuracy is determined, and if so, it is determined that the local model reaches the convergence condition. Therefore, a model for the predictive maintenance of the turbine fan can be created locally, and the model is used for the predictive maintenance of the turbine fan.

Example two

The embodiment will describe the joint learning based turbofan maintenance method provided by the invention from the server side in the joint learning architecture, that is, the execution subject of the embodiment is the server side of the joint learning, for example, the method provided by the embodiment is executed by the server 101 shown in fig. 1.

Specifically, the method for maintaining the turbofan based on the joint learning provided by the embodiment specifically includes the following steps S201 to S203.

Step S201: and receiving the local model uploaded by the client.

With reference to the first embodiment, the local model is a local model generated by the client based on a global model issued by the server and updated after being trained by using local data.

In an example, the step S201 may specifically include:

step S211: detecting the local model catalog and obtaining a detection result;

step S212: and receiving the local model according to the detection result.

Wherein, the above results specifically include: judging whether a new local model is generated: if yes, receiving the local model; if not, returning to detect the local model catalog.

Wherein, the local model directory is client information currently participating in joint learning. Specifically, the client can access the server for joint learning before starting the joint learning task; or adding the joint learning midway after the joint learning task is started. Therefore, by checking the local model directory, the situation of the client currently participating in the joint learning can be determined, so as to judge whether a new local model is uploaded.

Step S202: and aggregating the local models and updating the global model.

Wherein the aggregating may specifically include performing a mean aggregation on all uploaded local models.

In addition, in the case of combining the above examples, in order to deal with the problem of system heterogeneity or data heterogeneity, when the local model is aggregated, an asynchronous joint aggregation method may also be adopted. Illustratively, the step S202 may include: and according to the received newly uploaded local model, combining the global model after the previous aggregation to perform the aggregation, and updating the global model according to the result after the aggregation. That is to say, in this example, it is not necessary to wait for all clients to upload the local models before performing aggregation, so that the training efficiency is improved.

Step S203: and determining the next global model updating time according to the updating result of the updated global model, wherein the updating result comprises information about whether the updating is successful and the updated attribute information.

And according to the global model updating time, performing aggregate updating on the received local models when the updating time is reached. Meanwhile, the updated attribute information is intermediate parameter information of the updated global model, for example, gradient information of the global model. And when the updating result is determined to be successful, the attribute information can be issued to each client to complete the joint learning task.

In summary, the first embodiment and the second embodiment implement the training of the operation and maintenance prediction model of the turbofan through the joint learning architecture, and improve sample data of local training, so that the quality of the model obtained by training is higher, and the accuracy of the operation and maintenance prediction is improved; in addition, the aggregation mode of the server side is improved, so that the training efficiency of the prediction model is further improved.

The prediction model obtained by the above embodiment of the present invention based on the joint learning is compared with the model obtained by the deep learning in the prior art, that is, the model is tested by using the same test data, and the accuracy of the test results of the two models is shown. The specific experimental results are shown in the following table 1:

TABLE 1

The user 1, the user 2, and the user 3 represent data processing nodes local to the turbine fan, and may be specifically edge data processing nodes, for example. As is clear from table 1 above, the accuracy of the present invention is significantly improved compared to the prior art.

It should be understood that, the sequence numbers of the steps in the foregoing embodiments do not imply an execution sequence, and the execution sequence of each process should be determined by its function and inherent logic, and should not constitute any limitation to the implementation process of the embodiments of the present invention.

EXAMPLE III

Under the same inventive concept, the third embodiment further provides a device corresponding to each step in the method embodiment shown in fig. 2.

Fig. 3 is a structural diagram of a joint learning-based turbofan maintenance apparatus according to a third embodiment of the present invention.

Referring to fig. 3, the joint learning based maintenance apparatus 300 for a turbine fan provided in this embodiment is generally disposed at a client, and specifically includes: a global model receiving module 301, configured to receive a global model delivered by a server under a joint learning architecture; a sample data generating module 302, configured to obtain historical operating data of the turbofan, and generate sample data used for training the global model; a global model training module 303, configured to train the global model using the sample data, and update a local model; and the fan prediction maintenance module 304 is configured to create a model for the turbine fan prediction maintenance if the local model reaches a preset convergence condition, and perform the prediction maintenance on the turbine fan by using the model.

In some alternatives, the historical operating data includes: at least one of a temperature of the turbo fan, a pressure of the turbo fan, and/or a rotational speed of the turbo fan.

In addition, this embodiment still provides a turbo fan maintains device based on joint learning, sets up in the server end, and it includes: the model receiving module is used for receiving the local model uploaded by the client; the model aggregation module is used for aggregating the local models and updating the global model; and the iteration updating module is used for determining the next updating time of the global model according to the updating result of the updated global model, wherein the updating result comprises information about whether the updating is successful and the updated attribute information.

In some alternatives, the model aggregation module is specifically configured to perform aggregation this time by combining a global model after previous aggregation according to a received newly uploaded local model, and update the global model according to a result after aggregation this time.

The above-mentioned embodiments are only used for illustrating the technical solutions of the present invention, and not for limiting the same; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; such modifications and substitutions do not substantially depart from the spirit and scope of the embodiments of the present invention, and are intended to be included within the scope of the present invention.

Claims

1. A method for maintaining a turbine fan based on joint learning is characterized in that an execution subject is a client side of the joint learning, and comprises the following steps:

receiving a global model issued by a server under a joint learning architecture;

acquiring historical operating data of the turbine fan, and inputting the historical operating data into NASA simulation software to generate sample data for training the global model;

training the global model by using the sample data, and updating a local model;

if the local model reaches a preset convergence condition, creating a model for the predictive maintenance of the turbine fan, and performing the predictive maintenance on the turbine fan by using the model;

the execution subject is a server side of the joint learning, comprising:

receiving a local model uploaded by a client;

aggregating the local models, and updating a global model, so that the client trains the updated global model to create a model for the predictive maintenance of the turbine fan;

determining the next global model updating time according to the updating result of the updated global model, wherein the updating result comprises information about whether the updating is successful and attribute information of the updating;

wherein, the receiving of the local model uploaded by the client comprises:

detecting the local model catalog and obtaining a detection result;

judging whether a new local model is generated or not according to the detection result; if yes, receiving the new local model; if not, returning to detect the local model catalog.

2. The joint learning based turbofan maintenance method of claim 1 wherein the historical operating data comprises: a temperature of the turbo fan, a pressure of the turbo fan, and/or a rotational speed of the turbo fan.

3. A joint learning based turbine fan maintenance device for performing the method of claim 1 or 2, comprising:

the global model receiving module is used for receiving a global model issued by a server under the joint learning architecture;

the sample data generating module is used for acquiring historical operating data of the turbine fan and generating sample data used for training the global model;

the global model training module is used for training the global model by using the sample data and updating a local model;

and the fan prediction maintenance module is used for creating a model for the turbine fan prediction maintenance if the local model reaches a preset convergence condition, and performing prediction maintenance on the turbine fan by using the model, wherein the preset convergence condition is a preset iteration updating time or a preset model accuracy.

4. A joint learning based turbine fan maintenance device for performing the method of claim 1 or 2, comprising:

the model receiving module is used for receiving the local model uploaded by the client;

the model aggregation module is used for aggregating the local models and updating the global model so that the client trains the updated global model to create a model for the predictive maintenance of the turbine fan;

the iteration updating module is used for determining the next updating time of the global model according to the updating result of the updated global model, wherein the updating result comprises information about whether the updating is successful and attribute information of the updating;

the model receiving module specifically includes:

the directory checking unit is used for detecting the local model directory and obtaining a detection result;

the model receiving unit is used for judging whether a new local model is generated or not according to the detection result; if yes, receiving the new local model; if not, returning to detect the local model catalog.

5. An electronic device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that the steps of the method according to claim 1 or 2 are implemented when the processor executes the computer program.

6. A storage medium storing a computer program, characterized in that the computer program realizes the steps of the method according to claim 1 or 2 when executed by a processor.