Fault prediction method and system for rotor bearing of unmanned aerial vehicle, and program product

Abstract

The invention discloses an unmanned aerial vehicle rotor bearing fault prediction method, system and program product, and belongs to the technical field of artificial intelligence. The method comprises the following steps: acquiring a to-be-diagnosed sample; extracting a hidden code feature vector through a feature extractor; inputting the hidden code feature vector into a fault diagnosis learner to obtain a fault data vector of the diagnosed bearing; the training process comprises the following steps: acquiring a small batch of unlabeled training samples; ranking the unlabeled training samples by using a ranker, labeling the top K unlabeled training samples by experts, and converting the top K unlabeled training samples into labeled training samples; according to the method, the non-label samples are fully utilized, part of the non-label samples serve as characterization of diagnosis uncertainty and then are embedded into the feature hidden codes, the overall potential subspace of the data is rebuilt through adversarial training, information related to diagnosis is introduced, and therefore the diagnosis perception capacity is enhanced.

Description

Technical Field

[0001] The present invention relates to a method, a system and a program product for predicting faults of a drone rotor bearing, belonging to the technical field of artificial intelligence. Background Art

[0002] In the prior art, a variety of methods for predicting faults of bearings are provided. For example, the Chinese patent application with the publication number CN119337204A discloses a method for predicting faults of a bearing based on physical information constrained VMD-PIML. This method improves the FNGO algorithm. By introducing a chaotic map to initialize the distribution of the eagle group, introducing a sine function and a refraction reverse learning strategy, the calculation accuracy and search ability of the algorithm are enhanced; the improved FNGO algorithm is used to iterate the important parameters of VMD to form [K, ɑ] to extract signal features; finally, the improved FNGO algorithm is used to optimize the hyperparameters in the PIML prediction model; the PIML prediction model can combine physical constraints and a machine learning model, can comprehensively consider the bearing signal features, and improve the accuracy and robustness of the model prediction. This invention constrains the PIML prediction model through physical knowledge, uses the improved NGO algorithm to optimize the VMD denoising model and the PIML prediction model, and improves the generalization and robustness of bearing fault prediction.

[0003] The Chinese patent application with the publication number CN119246068A discloses a method for predicting early faults of a rolling bearing based on Bayesian optimization of DCAE-RNN. The steps are as follows: constructing a DCAE-RNN model, where the DCAE-RNN model includes a DCAE model and an RNN model, and the DCAE model is composed of an SADE and a CAE, and using a Bayesian optimization function to find the optimal hyperparameters of the SDAE and the CAE; training the DCAE-RNN model with a training set to obtain a trained DCAE-RNN model; inputting the real-time vibration signal of the rolling bearing operation into the trained DCAE-RNN model, and calculating the residual between the predicted value and the real value output by the model to detect early faults of the rolling bearing. This invention can well describe the operation state of the rolling bearing, predict the early faults of the rolling bearing earlier and more stably, and has strong accuracy and robustness. Moreover, the model is convenient for deployment, which is very important for the degradation monitoring of the operation state of the rolling bearing.

[0004] However, these patent applications do not disclose how to utilize the uncertain information of unlabeled samples. Summary of the Invention

[0005] To overcome the shortcomings of the prior art, the object of the present invention is to provide a fault prediction method, system and program product for an unmanned aerial vehicle rotor bearing, which makes full use of unlabeled samples, uses some unlabeled samples as a characterization of diagnostic uncertainty, and then embeds them into the feature latent code. By adversarial training, the overall potential subspace of the data is reshaped, thereby introducing diagnostic-related information and enhancing the diagnostic perception ability.

[0006] To achieve the above object, the present invention provides a fault prediction method for an unmanned aerial vehicle rotor bearing, which includes the following steps: Step 1: Obtain the sample to be diagnosed ; Step 2: Extract the latent code feature vector according to the sample to be diagnosed through the feature extractor ; Step 3: Input the latent code feature vector into the fault diagnosis learner to obtain the fault data vector of the bearing to be diagnosed , is the optimal parameter of the fault diagnosis learner, and the training process of the fault diagnosis learner includes: S01: Obtain a small batch of unlabeled training samples and labeled training samples , where N and M are positive integers greater than or equal to 2; S02: Use the ranker to rank the unlabeled training samples to obtain the ranking value , sort the ranking values from largest to smallest, and submit the top K unlabeled training samples with the highest ranking values to an expert for labeling and convert them into labeled training samples, is the parameter of the ranker; calculate the ranking loss ; S03: Extract the latent code feature vector from the labeled training samples through the feature extractor ; the fault diagnosis learner obtains the fault data vector of the bearing to be diagnosed according to the latent code feature vector , , is the parameter of the fault diagnosis learner.

[0007] The present invention also provides a fault prediction system for an unmanned aerial vehicle rotor bearing, which includes: a memory and a processor, and a computer program that can run on the processor is stored on the memory. The processor executes the computer program to implement the above-mentioned fault prediction method for an unmanned aerial vehicle rotor bearing.

[0008] To achieve the above-mentioned invention objective, the present invention also provides a computer program product, which includes computer program code that can be called by a processor to execute the above-mentioned fault prediction method for a drone rotor bearing.

[0009] The present invention has the following beneficial effects: The fault prediction method, system and program product for a drone rotor bearing provided by the present invention make full use of unlabeled samples, use some unlabeled samples as the representation of diagnostic uncertainty, and then embed them into the feature latent code. By adversarial training, the overall potential subspace of the data is reshaped, thereby introducing diagnostic-related information and enhancing the diagnostic perception ability. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 is a flowchart of the fault prediction method for a drone rotor bearing provided by the first embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0011] In order to make the technical problems, technical solutions and beneficial effects to be solved by the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0012] First Embodiment

[0013] Figure 1 is a flowchart of the fault prediction method for a drone rotor bearing provided by the first embodiment of the present invention. As Figure 1 shown, the first embodiment of the present invention provides a fault prediction method for a drone rotor bearing, which includes the following steps: Step 1: Obtain a sample to be diagnosed ; Step 2: Extract a latent code feature vector from the sample to be diagnosed through a feature extractor ; ; Step 3: Input the latent code feature vector into a fault diagnosis learner to obtain a fault data vector of the bearing to be diagnosed , being the optimal parameter of the fault diagnosis learner.

[0014] In the first embodiment, the training process of the fault diagnosis learner includes: S01: Obtain a small batch of unlabeled training samples and labeled training samples , where N and M are positive integers greater than or equal to 2.

[0015] In the first embodiment, the training process of the fault diagnosis learner further includes: S02: Using a ranker Rank the unlabeled training samples to obtain ranking values , sort the ranking values from largest to smallest, and submit the top K unlabeled training samples with the largest ranking values to an expert for labeling and convert them into labeled training samples, which are the parameters of the ranker; calculate the ranking loss ; In the present invention, the ranking loss is calculated by the following formula : , wherein, is the ranking value obtained by the ranker for ranking ; is the ranking test data; are the parameters of the ranker, is the activation function.

[0016] In the first embodiment, the training process of the fault diagnosis learner further includes: S03: Extract the hidden code feature vector from the labeled training samples by a feature extractor; the fault diagnosis learner obtains the fault data vector of the bearing to be diagnosed according to the hidden code feature vector , , are the parameters of the fault diagnosis learner.

[0017] The training process of the fault diagnosis learner provided by the present invention further includes: S04: Calculate the loss of the fault diagnosis learner : , wherein, is the measured fault data vector of the bearing to be diagnosed, is the fault data vector of the bearing to be diagnosed output by the fault diagnosis learner.

[0018] The training process of the fault diagnosis learner provided by the present invention further includes: S05: Update and calculate the parameters of the fault diagnosis learner and the parameters of the ranker: ; , is the ranking loss Scale adjustment hyperparameter; is the first learning coefficient; is the second learning coefficient; is the gradient of the parameter ; is the gradient of the parameter ;

[0019] The training process of the fault diagnosis learner provided by the present invention further includes: S06: Calculate the reconstruction loss and adversarial loss of the VAE network (Variational Autoencoder network): , wherein, is the expectation; is the encoder of the VAE network, is the parameter of the encoder, is the hyperparameter; is 's predicted ranking; is the divergence; is the decoder of the VAE network, is the parameter of the decoder, is the latent code feature vector extracted from ; is 's predicted ranking, is the distribution of the latent code feature vector z, i = 1, …, M + K, j = 1, … N - K; , wherein, is the expectation; represents the discriminator, is the parameter of the discriminator; Update the parameters of the encoder and parameters of the decoder of the VAE network through the following formula: , wherein, is the adversarial loss 's scale adjustment hyperparameter; is the third learning coefficient; is the gradient of the parameter ; is the fourth learning coefficient; is the gradient of the parameter ;

[0020] The training process of the fault diagnosis learner provided by the present invention further includes: S07: Calculate the loss of the discriminator : , wherein, is the expectation; represents the discriminator, are the parameters of the discriminator; is the decoder of the VAE network, are the parameters of the decoder of the VAE network; i = 1, …, M + K, j = 1, …, N - K.

[0021] Update the parameters of the discriminator through the following formula : , wherein, is the fifth learning coefficient; is the gradient of the parameter .

[0022] The training process of the fault diagnosis learner provided by the present invention further includes: S08: Repeat steps S02 to S07 until the stop condition is reached to obtain the optimal , , , , .

[0023] In the first embodiment of the present invention, by fully utilizing the unlabeled samples, part of the unlabeled samples are used as the characterization of diagnostic uncertainty, and then embedded into the latent code features. The overall potential subspace of the data is reshaped through adversarial training, thereby introducing diagnostic-related information, and thus enhancing the diagnostic perception ability.

[0024] Second Embodiment

[0025] Only the content different from the first embodiment in the second embodiment of the present invention is described, and the same content will not be repeated.

[0026] The second embodiment of the present invention provides a fault prediction system for a drone rotor bearing, which includes a memory and a processor. A computer program that can run on the processor is stored on the memory, and the processor executes the computer program to implement the fault prediction method for the drone rotor bearing described in the first embodiment.

[0027] The beneficial effects of the second embodiment of the present invention are the same as those of the first embodiment and will not be repeated here.

[0028] Third Embodiment

[0029] Only the content different from the first embodiment of the present invention will be described in the second embodiment of the present invention, and the same content will not be repeated.

[0030] A third embodiment of the present invention provides a program product, which includes computer program code that can be called by a processor to execute the method described in the first embodiment.

[0031] The beneficial effects of the third embodiment of the present invention are the same as those of the first embodiment, and will not be repeated here.

[0032] It should be noted that the terms "first" and "second" are only used for descriptive purposes and cannot be construed as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, features defined with "first" and "second" may explicitly or implicitly include one or more of such features. In the description of the present invention, "a plurality" means two or more, unless otherwise specifically defined.

[0033] The foregoing has shown and described the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments, and what is described in the above embodiments and the specification is only to illustrate the principle of the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and all these changes and improvements fall within the scope of the present invention claimed.

Claims

1. A fault prediction method for a drone rotor bearing, characterized in that It includes the following steps: Step 1: Obtain the sample to be diagnosed ; Step 2: According to the sample to be diagnosed, the feature extractor extracts the hidden code feature vector ; Step 3: Input the hidden code feature vector into the fault diagnosis learner to obtain the fault data vector of the bearing to be diagnosed , being the optimal parameters of the fault diagnosis learner, and the training process of the fault diagnosis learner includes: S01: Obtain a small batch of unlabeled training samples and labeled training samples , where N and M are positive integers greater than or equal to 2; S02: Using a ranker For unlabeled training samples Rank them to obtain ranking values , sort the ranking values from largest to smallest, and submit the top K unlabeled training samples with higher ranking values to an expert for labeling and convert them into labeled training samples Are parameters of the ranker; calculate the ranking loss ; S03: Extract the latent code feature vectors from the labeled training samples Extract the latent code feature vectors ; Fault diagnosis learner According to the latent code feature vectors Obtain the fault data vectors of the bearings to be diagnosed , , are the parameters of the fault diagnosis learner.

2. The fault prediction method for the drone rotor bearing according to claim 1, wherein The training process of the fault diagnosis learner further includes: S04: Calculate the loss of the fault diagnosis learner ;​ S05: Update the parameters of the fault diagnosis learner and the parameters of the ranker : ; , is the ranking loss is the scale adjustment hyperparameter; is the first learning coefficient; is the second learning coefficient; is the gradient of the parameter ; is the gradient of the parameter ; S06: Calculate the reconstruction loss of the VAE network and the adversarial loss , and update the parameters of the encoder of the VAE network and the parameters of the decoder : ； ， In the formula, is the scale adjustment hyperparameter of the adversarial loss ; is the third learning coefficient; is the gradient of the parameter ; is the fourth learning coefficient; is the gradient of the parameter ; S07: Calculate the discriminator loss , and update the discriminator's parameters by the following formula : ， In the formula, is the fifth learning coefficient; is the gradient of the parameter ; S08: Repeat steps S02 to S07 until the stop condition is reached to obtain the optimal , , , , .

3. The fault prediction method for the drone rotor bearing according to claim 2, wherein ， In the formula, is the measured fault data vector of the bearing to be diagnosed, is the fault data vector of the bearing to be diagnosed output by the fault diagnosis learner.

4. The fault prediction method for the drone rotor bearing according to claim 3, wherein ， In the formula, is the ranking value obtained by the ranker for ranking ; is the ranking test data; is the parameter of the ranker, is the activation function.

5. The fault prediction method for the drone rotor bearing according to claim 4, wherein ， is the expectation; is the encoder of the VAE network, are the parameters of the encoder, are hyperparameters; is the predicted ranking of; is the divergence; is the decoder of the VAE network, are the parameters of the decoder, is from the latent code feature vector extracted, is the predicted ranking of, is the distribution of the latent code feature vector z, i = 1, …, M + K, j = 1, …, N - K.

6. The fault prediction method for the drone rotor bearing according to claim 5, wherein ， In the formula, is the expectation; represents a decision maker, is the parameter of the decision maker.

7. The fault prediction method for the drone rotor bearing according to claim 6, wherein ， In the formula, is the expectation; represents a decision device, is the parameter of the decision device; i = 1, …, M + K, j = 1, …, N - K.

8. A fault prediction system for a drone rotor bearing, comprising: A memory and a processor, wherein a computer program is stored on the memory and can run on the processor, and the processor executes the computer program to implement the fault prediction method for the drone rotor bearing according to any one of claims 1-7.

9. A computer program product, which includes computer program code, and the computer program code can be called by a processor to execute the fault prediction method for the drone rotor bearing according to any one of claims 1-7.

Images