A method for processing electromechanical device fault diagnostic data

Abstract

The present application relates to the technical field of information communication, and aims to provide a method for processing fault diagnosis data of electromechanical equipment, which comprises the following steps: collecting a local data set, calculating a stochastic gradient, and obtaining a momentum buffer function required for gradient tracking; calculating an adaptive learning rate according to the stochastic gradient; repeating the above steps to collect the momentum buffer function and the adaptive learning rate of the client, and uploading them to the server side when a preset period is reached; aggregating the momentum buffer function and the adaptive learning rate to obtain a global adaptive matrix, generating and outputting a model, and providing it to the client. The data from the client is optimized, and the information of all clients is comprehensively optimized to optimize the model, thereby protecting the data privacy and improving the accuracy of the equipment health assessment and fault prediction.

Description

Technical Field

[0001] This invention relates to the field of information and communication technology, and more specifically, to a method for processing fault diagnosis data of electromechanical equipment. Background Technology

[0002] Various types of electromechanical equipment have become highly information-based and networked, generating and collecting massive amounts of equipment status data. This vast amount of data, after being processed by machine learning algorithms, can enable intelligent diagnosis and predictive maintenance of electromechanical equipment. However, equipment status data often contains core business information of enterprises, raising concerns about data privacy. Furthermore, the distribution of data collected by different enterprises may vary, increasing the difficulty of data utilization. Therefore, how to achieve collaborative analysis of massive amounts of heterogeneous equipment status data while protecting data privacy, accurately assessing equipment health status, and predicting faults is a pressing problem that needs to be solved. Currently, centralized data analysis methods used for equipment maintenance require enterprises to upload local data to third-party cloud services, posing a risk of data leakage. Using encryption or differential privacy methods to process data before uploading reduces the accuracy of the model.

[0003] Federated learning is a machine learning paradigm that has emerged in recent years, enabling the collaboration of data from multiple enterprises without transmitting raw data. Currently, there are two main algorithms in the field of federated learning: FedAvg and FedAdam. 1) FedAvg (Federated Averaging) is a basic federated learning algorithm. It allows each client to train the model locally multiple times, and then periodically sends the parameters of the local model to the server for averaging. Specifically, the client uses local data to run SGD to train the model, and the server collects the model parameters from all clients, averages them, and broadcasts them to all clients as a global model. 2) FedAdam (Federated Adaptive Moment Estimation) is an improved algorithm of FedAvg. Based on FedAvg, the gradients calculated locally by the devices are also aggregated in a central server, and the model parameters are adaptively adjusted on the server side. However, existing algorithms have the following drawbacks: 1) They require a large number of training iterations, resulting in slow convergence; 2) They lack adaptability and rely on manual adjustment of the learning rate; 3) They depend on data homogeneity, which is not compatible with the growth of model and data scale, and they perform poorly in handling heterogeneous data, making them difficult to apply to practical prediction and maintenance.

[0004] Therefore, there is an urgent need to develop a faster and more efficient federated learning algorithm to improve the accuracy of device health assessment and fault prediction while protecting data privacy. Summary of the Invention

[0005] The present invention aims to provide a method for processing fault diagnosis data of electromechanical equipment to solve the technical problems of low accuracy in equipment fault prediction and inability to adequately protect data privacy.

[0006] The technical solution adopted by this invention to solve its technical problem is: a method for processing fault diagnosis data of electromechanical equipment, comprising:

[0007] S1 Client Information Collection

[0008] S11 collects local dataset D i Calculate the stochastic gradient And obtain the momentum buffer function m required for gradient tracking. t,i ;

[0009] S12 calculates the adaptive learning rate v based on the stochastic gradient in step S11. t,i ;

[0010] Repeat steps S11 and S12 to collect the momentum buffer function m from t clients. t,i With an adaptive learning rate, the data is uploaded to the server after the preset period t is reached;

[0011] S2 will collect m from t clients. t,i and v t,i Perform aggregation to obtain the global result. and Obtain the global adaptive matrix, generate the output model, and provide it to the client.

[0012] This invention optimizes the adaptive learning rate using historical gradient information from client terminals, and uploads the optimized adaptive learning rate and gradient information to a server. The server then optimizes the model based on the adaptive learning rate and gradient information from different clients, and finally provides the optimized model information to the clients. This invention optimizes from client data and integrates information from all clients for model optimization, improving the accuracy of device health assessment and fault prediction while protecting data privacy.

[0013] In this invention, it can be understood that in each iteration cycle, the server terminal is connected to t clients, each client contains b motor devices, and each motor device is numbered i. Based on this, the following calculations are performed.

[0014] In this invention, the initial information of the local dataset mentioned in step S11 includes at least one of the current, voltage, and vibration of the motor equipment; the initial information is converted into a time-frequency domain image through continuous wavelet transform.

[0015] Preferably, the calculation of the stochastic gradient in step S11 includes:

[0016] Local dataset D i This includes collecting data from b local motors, where the data from b local motors is represented as follows: Where i represents the electromechanical equipment number; D i This represents the local dataset of the i-th electromechanical device;

[0017] Using the current model parameters x t,i Using the collected local motor data B t,i Calculate the loss function f i (x t,i B y,i ):

[0018]

[0019] Calculate local motor data B in the client t,i Corresponding stochastic gradient

[0020]

[0021] The stochastic gradient of the t-th iteration is denoted as... Calculate the parameters x for each client i in the current model. t,i stochastic gradient on

[0022]

[0023] Calculating local stochastic gradients allows for immediate feedback on global model parameters from a local dataset, improving the model's adaptability and accuracy on heterogeneous data. Furthermore, local processing reduces reliance on a central server, accelerating the overall learning process.

[0024] Preferably, the momentum buffer function m required for gradient tracking in step S11 is... t,i The calculations include:

[0025] Calculate the stochastic gradient of the current step. This represents the stochastic gradient calculated for the i-th client point in round t:

[0026]

[0027] Calculate the model parameters x from the previous step t-1,i B on the current dataset t,i stochastic gradient This represents the stochastic gradient calculated for the i-th client point in round t-1:

[0028]

[0029] Momentum buffer function m t,i :

[0030]

[0031] Where, α t ∈(0,1) represents the momentum parameter.

[0032] m t,i Gradient information is tracked by accumulating a linear combination of the current gradient and the contents of the previous buffer.

[0033] The main function of gradient tracking is to smooth the learning process and accelerate convergence by utilizing historical gradient information. Momentum helps the model descend steadily during training by adjusting the update direction of the current step by considering previous gradients, thereby overcoming possible training fluctuations and quickly approaching the optimal solution.

[0034] Preferably, the calculation of the adaptive learning rate in step S12 includes:

[0035] Calculate the stochastic gradient for the current step from the previous step. Further, obtain the square of the current stochastic gradient:

[0036]

[0037] Calculate the adaptive learning rate v based on the square of the current stochastic gradient. t,i :

[0038]

[0039] Where β∈(0,1) represents the momentum parameter.

[0040] v t,i The adaptive learning rate is estimated by accumulating the exponential moving average of the squared historical gradients. By taking into account the amount of gradient change for each client, the adaptive learning rate helps maintain stability when facing different data distributions, avoiding slow or unstable learning caused by learning rates that are too large or too small.

[0041] Preferably, the global process in step S2 and The calculations include:

[0042] Collect the m uploaded by each client i t,i and v t,i ;

[0043] m for all clients t,i Calculate the average to obtain the global result.

[0044]

[0045] v for all clients t,i Calculate the average to obtain the global result.

[0046]

[0047] Repeat the above collection and averaging operations every q steps to obtain the global result. and

[0048] By aggregating all client information, a global result was obtained. and This prepares for generating an adaptive matrix. Gradient and learning rate information from all clients is synchronized and aggregated globally to achieve coordinated model updates. This reduces model bias caused by uneven data distribution while improving the overall model's robustness and performance through collective intelligence.

[0049] An adaptive matrix is ​​generated on the server. The main purpose of generating the adaptive matrix is ​​to create a global update strategy that reflects the shared contributions of different clients to model updates. By aggregating adaptive learning rate information from different clients, the adaptive matrix ensures that each global update takes into account the learning characteristics of all clients, thereby optimizing model performance and ensuring the consistency and effectiveness of updates.

[0050] Preferably, the generation of the global adaptive matrix in step S2 includes:

[0051]

[0052] Here, ρ is a constant greater than 0, used to smooth adaptive information and prevent overfitting.

[0053] It should be noted that the adaptive matrix A t It is a diagonal matrix, obtained through global stochastic gradient descent. This provides adaptive information for updating the client's local model.

[0054] Finally, the output model is updated in a personalized manner based on the client's data to improve the model's accuracy and adaptability to specific client data. This ensures the effective execution of the global update strategy in the local environment and enhances the model's responsiveness to local data characteristics.

[0055] Preferably, the generation of the output model in step S2 includes:

[0056] Each client i performs the following operations:

[0057] Get the global adaptive matrix A sent by the server.t Calculate its inverse matrix A t -1 ;

[0058] Use m t,i The global adaptive matrix is ​​updated as follows:

[0059]

[0060] Where, η t This represents the learning rate.

[0061] The beneficial effects of this invention are:

[0062] Compared to existing technologies, this application provides a method for processing fault diagnosis data of electromechanical equipment. It optimizes the adaptive learning rate using historical gradient information from client machines, and uploads the optimized adaptive learning rate and gradient information to a server. The server then optimizes the model based on the adaptive learning rate and gradient information from different clients, ultimately providing the optimized model information to the client. Optimizing from client data and integrating information from all clients improves the accuracy of equipment health assessment and fault prediction while protecting data privacy. Attached Figure Description

[0063] Figure 1 This is a schematic flowchart of a method for processing fault diagnosis data of electromechanical equipment according to the present invention.

[0064] The accompanying drawings are for illustrative purposes only and should not be construed as limiting the invention. To better illustrate the embodiments, some parts in the drawings may be omitted, enlarged, or reduced, and do not represent the actual product dimensions. It is understandable to those skilled in the art that some well-known structures and their descriptions may be omitted in the drawings. Detailed Implementation

[0065] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0066] like Figure 1 As shown, N electromechanical devices are treated as N clients, and each device is assigned a number i. Time-series data collected from electromechanical device i, such as motor current, voltage, and vibration signals, are converted into a time-frequency domain image through continuous wavelet transform. This image is used as the dataset D. i The input algorithm is used. Simultaneously, the server distributes initial model parameters x0 to each client for local learning.

[0067] A method for processing fault diagnosis data of electromechanical equipment according to the present invention includes the following steps:

[0068] 1. Calculate the local stochastic gradient

[0069] From local dataset D i Randomly sampled small batches of data Where b is the size of the mini-batch data. The mini-batch data contains motor operation data such as current, voltage, and vibration, which have been converted into time-frequency domain images through continuous wavelet transform;

[0070] Using the current model parameters x t,i Using the collected local motor data B t,i Calculate the loss function f i (x t,i B t,i ):

[0071]

[0072] Calculate local motor data B in the client t,i Corresponding stochastic gradient

[0073]

[0074] The stochastic gradient of the t-th iteration is denoted as... Calculate the parameters x for each client i in the current model. t,i stochastic gradient on

[0075]

[0076] 2. Gradient tracking

[0077] Calculate the stochastic gradient of the current step. This represents the stochastic gradient calculated for the i-th client point in round t:

[0078]

[0079] Calculate the model parameters x from the previous step t-1,i B on the current dataset t,i stochastic gradient This represents the stochastic gradient calculated for the i-th client point in round t-1:

[0080]

[0081] Momentum buffer function m t,i :

[0082]

[0083] Where, α t∈(0,1) represents the momentum parameter.

[0084] m t,i Gradient information is tracked by accumulating a linear combination of the current gradient and the contents of the previous buffer.

[0085] 3. Estimating the adaptive learning rate

[0086] Perform the following operations on each client i:

[0087] Calculate the stochastic gradient for the current step from the previous step. Further, obtain the square of the current stochastic gradient:

[0088]

[0089] Calculate the adaptive learning rate v based on the square of the current stochastic gradient. t,i :

[0090]

[0091] Where β∈(0,1) represents the momentum parameter.

[0092] v t,i The adaptive learning rate is estimated by accumulating the exponential moving average of the squared historical gradients.

[0093] 4. Server-side average adaptive information

[0094] The server performs the following operations:

[0095] Collect the m uploaded by each client i t,i and v t,i ;

[0096] m for all clients t,i Calculate the average to obtain the global result.

[0097]

[0098] v for all clients t,i Calculate the average to obtain the global result.

[0099]

[0100] Repeat the above collection and averaging operations every q steps to obtain the global result. and

[0101] By aggregating all client information, a global result was obtained. and This prepares for generating an adaptive matrix.

[0102] 5. Generate adaptive matrix

[0103] The server performs the following operations to generate the global adaptive matrix:

[0104]

[0105] Where ρ is a constant greater than 0 (e.g., it can be 1, 2, 3, etc.), used to smooth adaptive information and prevent overfitting.

[0106] It should be noted that the adaptive matrix A t It is a diagonal matrix, obtained through global stochastic gradient descent. This provides adaptive information for updating the client's local model.

[0107] 6. Local client adaptive updates

[0108] Each client i performs the following operations:

[0109] Get the global adaptive matrix A sent by the server. t Calculate its inverse matrix A t -1 ;

[0110] Use m t,i The global adaptive matrix is ​​updated as follows:

[0111]

[0112] Where, η t This represents the learning rate.

[0113] Compared to existing federated learning algorithms, this invention introduces a momentum mechanism to track and utilize historical gradient information, effectively accelerating model convergence. Furthermore, by collaboratively generating an adaptive matrix among clients, the number of communication rounds required for training is reduced, decreasing sample complexity to O(ε-3) and communication complexity to O(ε-2). Specifically, the introduction of the adaptive matrix significantly improves the ability to handle heterogeneous data, enabling each client to quickly adapt to the characteristics of its own dataset, thereby accelerating the entire model training process. Existing federated learning algorithms typically require more communication and have slower convergence speeds. Through these innovations, this invention achieves a significant improvement in communication and computational efficiency while ensuring data security and privacy, meeting the high standards of intelligent manufacturing and equipment maintenance technologies required in the Industry 4.0 era. This invention is easy to promote and apply, and can be continuously upgraded. This provides an effective way to improve fault diagnosis using massive distributed electromechanical equipment data.

[0114] The model for motor fault diagnosis obtained through the above process has the following three main advantages compared to traditional fault detection methods.

[0115] (1) Data privacy and utilization efficiency: Traditional centralized data processing methods require centralized processing of sensitive data, which poses a risk of leakage. This patent uses a federated learning method to update the fault diagnosis model, which can perform data analysis through local computation and only share model parameters without sharing the original data, greatly enhancing data privacy protection. This method enables all parties involved to jointly optimize and update the model used for fault diagnosis while ensuring data security.

[0116] (2) Adaptability to heterogeneous data: Different companies' electromechanical equipment may have different data acquisition standards and formats, and traditional methods are inefficient when processing such heterogeneous data. This patent uses adaptive learning rate and dynamic model adjustment to adjust the fault diagnosis model, allowing the model to better adapt to data from different sources, thereby improving the accuracy and reliability of the model in multi-source data environments.

[0117] (3) Real-time performance and accuracy: Using the federated learning algorithm, the model can receive and process updated data from various clients in real time, quickly provide feedback on fault diagnosis results, accelerate the speed of fault detection, give the model the ability to integrate information from multiple sources, and improve the accuracy of fault judgment. In terms of predictive maintenance, the model can predict potential faults based on the real-time status data of the equipment, providing a scientific basis for maintenance decisions.

[0118] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the implementation of the present invention. Those skilled in the art can make other variations or modifications based on the above description. It is neither necessary nor possible to exhaustively describe all embodiments here. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the claims of the present invention.

Claims

1. A method for processing fault diagnosis data of electromechanical equipment, characterized in that, include: S1 Client Information Collection S11 collects local dataset D i Calculate the stochastic gradient And obtain the momentum buffer function m required for gradient tracking. t,i ; S12 calculates the adaptive learning rate v based on the stochastic gradient. t,i ; Repeat steps S11 and S12 to collect the momentum buffer function m from t clients. t,i and adaptive learning rate v t,i Once t reaches the preset period, it is uploaded to the server. S2 will collect m from t clients. t,i and v t,i Perform aggregation to obtain the global result. and Obtain the global adaptive matrix, generate and output the model, and provide it to the client; The momentum buffer function required for gradient tracking mentioned in step S11 includes: Calculate the stochastic gradient of the current step. This represents the stochastic gradient calculated for the i-th client point in round t: Calculate the model parameters x from the previous step t-1,i B on the current dataset t,i stochastic gradient This represents the stochastic gradient calculated for the i-th client point in round t-1: Momentum buffer function m t,i : Where, α t ∈(0,1) represents the momentum parameter; This represents data for b local motors.

2. The method for processing fault diagnosis data of electromechanical equipment according to claim 1, characterized in that, The initial information of the local dataset mentioned in step S11 includes at least one of the current, voltage, and vibration of the motor equipment; The initial information is converted into a time-frequency domain image through continuous wavelet transform.

3. The method for processing fault diagnosis data of electromechanical equipment according to claim 1, characterized in that, The calculation of stochastic gradients in step S11 includes: Local dataset D i This includes collecting data from b local motors, where the data from b local motors is represented as follows: Where i represents the electromechanical equipment number; D i This represents the local dataset of the i-th electromechanical device; Using the current model parameters x t,i Using the collected local motor data B t,i Calculate the loss function Calculate local motor data B in the client t,i Corresponding stochastic gradient The stochastic gradient of the t-th iteration is denoted as... Calculate the parameters x for each client i in the current model. t,i stochastic gradient on 4. The method for processing fault diagnosis data of electromechanical equipment according to claim 1, characterized in that, The calculation of the adaptive learning rate in step S12 includes: Calculate the stochastic gradient for the current step from the previous step. Further, obtain the square of the current stochastic gradient: Calculate the adaptive learning rate v based on the square of the current stochastic gradient. t,i : Where β∈(0,1) represents the momentum parameter.

5. The method for processing fault diagnosis data of electromechanical equipment according to claim 1, characterized in that, The global result obtained in step S2 and include: Collect the m uploaded by each client i t,i and v t,i ; m for all clients t,i Calculate the average to obtain the global result. v for all clients t,i Calculate the average to obtain the global result. Every q steps, repeat the above collection and averaging operations to obtain the global result. and 6. The method for processing fault diagnosis data of electromechanical equipment according to claim 1, characterized in that, The step S2 of obtaining the global adaptive matrix includes: Where ρ is a constant greater than 0.

7. The method for processing fault diagnosis data of electromechanical equipment according to claim 1, characterized in that, The generation and output of the model in step S2 includes: Each client i performs the following operations: Get the global adaptive matrix A sent by the server. t Calculate its inverse matrix A t -1 ; Use m t,i The global adaptive matrix is ​​updated as follows: Where, η t This represents the learning rate.

Images