Fault diagnosis method and device for equipment lubricating system, equipment and medium

Abstract

The embodiment of the invention discloses a fault diagnosis method and device for an equipment lubricating system, equipment and a medium, and relates to the technical field of fault detection of the equipment lubricating system. The method comprises the steps of performing iterative training on a to-be-trained sample expansion model through an original data set of an equipment lubrication system, and updating parameters of a generator for multiple times based on association information of generated data and real data in each iterative process; if it is determined that iteration of the sample expansion model is completed, determining a model obtained after iteration as a trained sample expansion model; the trained sample expansion model is used for expanding a training sample of the equipment lubrication system to obtain expansion data, training the model for identifying the lubrication system fault based on the expansion data, and processing the detection data of the to-be-diagnosed lubrication system based on the trained model for identifying the lubrication system fault to obtain the to-be-diagnosed lubrication system fault. And obtaining the fault type of the to-be-diagnosed lubricating system.

Description

Technical Field

[0001] The present invention relates to the technical field of equipment lubrication system fault detection, and in particular to a method, device, equipment and medium for fault diagnosis of an equipment lubrication system. Background Art

[0002] In the industrial production process, the normal operation of the equipment lubrication system is crucial to ensure the stability of the equipment and extend its service life. Due to the complex working environment of the equipment, the lubrication system is easily affected by factors such as temperature, pressure, and vibration, which leads to changes in the lubrication state, which may cause problems such as equipment wear and contamination, and even cause equipment failure.

[0003] Traditional equipment lubrication fault identification mainly relies on manual monitoring and experience-based judgment. This method is not only time-consuming and labor-intensive, but the accuracy of identification depends on the skills and experience of the operator, and there are problems of strong subjectivity and high misjudgment rate. Summary of the invention

[0004] The present invention provides a fault diagnosis method, device, equipment and medium for an equipment lubrication system, which can expand the data of the equipment lubrication system based on a generative adversarial network and determine an accurate fault diagnosis result based on the expanded data.

[0005] According to one aspect of the present invention, a fault diagnosis method for a lubrication system of an equipment is provided, the method comprising:

[0006] Obtain the original data set of the equipment lubrication system;

[0007] The training sample expansion model is iteratively trained through the original data set. In each iteration of the training sample expansion model, the parameters of the generator are updated multiple times based on the association information between the generated data and the real data; the association information between the generated data and the real data includes: the accuracy of the discriminator output, the cross entropy loss function, and the difference in features between the generated data and the real data; the accuracy of the discriminator output is the accuracy of the discriminator in classifying the real data and the generated data; the sample expansion model is constructed based on a generative adversarial network;

[0008] Enter the next iteration based on the sample expansion model after parameter update. If it is determined that the iteration of the sample expansion model is completed, the model obtained after the iteration is determined as the trained sample expansion model; the trained sample expansion model is used to expand the training samples of the equipment lubrication system to obtain expanded data, and the model for identifying lubrication system faults is trained based on the expanded data, and the detection data of the lubrication system to be diagnosed is processed based on the trained model for identifying lubrication system faults to obtain the fault type of the lubrication system to be diagnosed.

[0009] According to another aspect of the present invention, there is provided a fault diagnosis device for a lubrication system of an equipment, comprising:

[0010] The original data set acquisition module is used to acquire the original data set of the equipment lubrication system;

[0011] A sample expansion model training module, used for iteratively training the sample expansion model to be trained through the original data set, and in each iteration of the sample expansion model to be trained, the parameters of the generator are updated multiple times based on the association information between the generated data and the real data; the association information between the generated data and the real data includes: the accuracy of the discriminator output, the cross entropy loss function, and the difference in features between the generated data and the real data; the accuracy of the discriminator output is the accuracy of the discriminator in classifying the real data and the generated data; the sample expansion model is constructed based on a generative adversarial network;

[0012] The trained sample expansion model determination module is used to enter the next iteration based on the sample expansion model after parameter update. If it is determined that the sample expansion model iteration is completed, the model obtained after the iteration is determined as the trained sample expansion model; the trained sample expansion model is used to expand the training samples of the equipment lubrication system to obtain expansion data, and the model for identifying lubrication system faults is trained based on the expansion data, and the detection data of the lubrication system to be diagnosed is processed based on the trained model for identifying lubrication system faults to obtain the fault type of the lubrication system to be diagnosed.

[0013] According to another aspect of the present invention, an electronic device is provided, the electronic device comprising:

[0014] at least one processor; and

[0015] a memory communicatively connected to the at least one processor; wherein,

[0016] The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the fault diagnosis method for the equipment lubrication system described in any embodiment of the present invention.

[0017] According to another aspect of the present invention, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores computer instructions, and the computer instructions are used to implement the fault diagnosis method of the equipment lubrication system described in any embodiment of the present invention when executed by a processor.

[0018] The technical solution of the embodiment of the present application includes: obtaining an original data set of an equipment lubrication system; iteratively training a sample expansion model to be trained using the original data set, and in each iteration of the sample expansion model to be trained, updating the parameters of the generator multiple times based on the association information between generated data and real data; the association information between the generated data and the real data includes: the accuracy of the discriminator output, the cross entropy loss function, and the difference in features between the generated data and the real data; the accuracy of the discriminator output is the accuracy of the discriminator in classifying the real data and the generated data; the sample expansion model is constructed based on a generative adversarial network; entering the next iteration based on the sample expansion model after parameter update, if it is determined that the iteration of the sample expansion model is completed, the model obtained after the iteration is determined as the trained sample expansion model; the trained sample expansion model is used to expand the training samples of the equipment lubrication system to obtain expanded data, so as to train a model for identifying lubrication system faults based on the expanded data, and to process the detection data of the lubrication system to be diagnosed based on the trained model for identifying lubrication system faults to obtain the fault type of the lubrication system to be diagnosed. This technical solution updates the parameters of the sample expansion model multiple times by generating correlation information between data and real data, so that the parameters of the sample expansion model can quickly approach theoretically reasonable parameters, making the data obtained by sample expansion of the trained sample expansion model more reasonable, and then training the model for identifying lubrication system faults based on the expanded data to obtain the trained model, and then realizing equipment lubrication system fault detection based on the trained model.

[0019] It should be understood that the contents described in this section are not intended to identify the key or important features of the embodiments of the present invention, nor are they intended to limit the scope of the present invention. Other features of the present invention will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0021] Figure 1 is a flow chart of a fault diagnosis method for a lubrication system of an equipment provided in accordance with the first embodiment of the present application;

[0022] Figure 2 is a flow chart of a fault diagnosis method for a lubrication system of an equipment provided in accordance with the second embodiment of the present application;

[0023] Figure 3is a flow chart of a fault diagnosis method for a lubrication system of an equipment provided in accordance with Embodiment 3 of the present application;

[0024] Figure 4 is a flow chart of a fault diagnosis method for a lubrication system of an equipment provided in accordance with the fourth embodiment of the present application;

[0025] Figure 5 It is a structural schematic diagram of a fault diagnosis device for a lubrication system of an equipment provided in accordance with Embodiment 5 of the present application;

[0026] Figure 6 It is a structural schematic diagram of an electronic device for implementing a fault diagnosis method for a lubrication system of an equipment according to an embodiment of the present application. DETAILED DESCRIPTION

[0027] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only embodiments of a part of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0028] It should be noted that the terms "first", "second", "target", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0029] Embodiment 1

[0030] Figure 1 A flowchart of a method for diagnosing a fault in a lubricating system of an equipment is provided for the first embodiment of the present application. The embodiment of the present application can be applied to the case of diagnosing a fault in a lubricating system of an equipment. The method can be executed by a fault diagnosis device for a lubricating system of an equipment. The fault diagnosis device for a lubricating system of an equipment can be implemented in the form of hardware and / or software. The fault diagnosis device for a lubricating system of an equipment can be configured in an electronic device with data processing capabilities. Figure 1 As shown, the method includes:

[0031] S110, obtaining an original data set of a lubrication system of an equipment.

[0032] The original data set of the equipment lubrication system is a data set obtained by collecting data from the equipment lubrication system, and the original data set includes multiple training samples. The data in each training sample can be determined according to actual conditions, and the embodiment of the present application does not limit this. Exemplarily, the data in the training sample can be one or more of the following: temperature, vibration frequency, pressure, flow rate, conductivity, acoustic emission, oil color, pH value, water content, and viscosity. Among them, temperature (Ra) indicates the temperature of the lubricating part of the equipment and reflects the thermal stability of the lubricating oil; vibration frequency (Ba) indicates the degree of friction of mechanical parts; pressure (Ca) is the pressure in the lubrication system, and high pressure may indicate pipeline blockage; flow rate (Da) reflects the flow rate of the lubricating material, and a decrease in flow rate may be an indication of a pipeline leak; conductivity (Ea) is the conductivity of the lubricating oil, and changes may indicate that the lubricating oil is contaminated; acoustic emission (Fa) can reflect the fracture of the equipment, such as monitoring cracks and fractures of the equipment through acoustic emission; changes in oil color (Ga) reflect the aging or contamination degree of the oil; pH value (Ha) is the acidity and alkalinity of the lubricating oil, reflecting chemical stability; water content (Ia) reflects the water content in the lubricating oil; viscosity (Ja) is the viscosity of the lubricating oil, which is closely related to changes in temperature and pressure.

[0033] Specifically, since the number of training samples in the original data set is usually limited, when a model is trained based on existing training samples to identify equipment lubrication system faults, it will affect the recognition accuracy of the model. Therefore, the training samples (i.e., the original data set) can be expanded based on a generative adversarial network with feature decoupling.

[0034] Furthermore, the training samples in the original data set are labeled, and the labeled content reflects whether there is a fault in the equipment lubrication system corresponding to the training sample and the type of fault. For example, the labeled content is: "normal operation", "insufficient lubrication", "excessive lubrication", "contamination" or "mechanical damage".

[0035] S120, iteratively training the to-be-trained sample expansion model using the original data set, and in each iteration of the to-be-trained sample expansion model, updating the parameters of the generator multiple times based on the association information between the generated data and the real data.

[0036] Among them, the association information between the generated data and the real data includes: the accuracy of the discriminator output, the cross entropy loss function, and the difference in features between the generated data and the real data; the accuracy of the discriminator output is the accuracy of the discriminator in classifying the real data and the generated data; the sample expansion model is constructed based on a generative adversarial network, and the improved generative adversarial network in the embodiment of the present application can be called a generative adversarial network based on feature decoupling.

[0037] Specifically, when the to-be-trained sample expansion model is iteratively trained using the original data set, in each iteration of the to-be-trained sample expansion model, the parameters of the generator are updated three times based on the association information between the generated data and the real data, and the discriminator is updated once.

[0038] In the embodiment of the present application, for example, when the training sample expansion model is iteratively trained by the original data set, the generator and discriminator of the generative adversarial network are initialized, and the generator is G c , the discriminator is D c , the initialization method is expressed as:

[0039]

[0040]

[0041] In the formula, and Represent the parameter sets of the generator and the discriminator respectively; It means that 0 is the mean value. is a normal distribution with variance; Indicates that in [-β c ,β c ] is uniformly distributed within the range; σ c and β c are the standard deviation of the control parameter initialization and the boundary of the uniform distribution; ~ indicates that it obeys a specific distribution. For example, Set to 0.01, β c Set to 0.5.

[0042] In the embodiment of the present application, optionally, when the parameters of the generator are updated multiple times based on the association information between the generated data and the real data, one parameter update process includes steps A1-A3:

[0043] Step A1, adjust the current generator loss function weight according to the accuracy of the discriminator output to obtain the updated generator loss function weight; the accuracy of the discriminator output is proportional to the updated generator loss function weight.

[0044] In the embodiment of the present application, optionally, the updated generator loss function weight is determined according to the following formula:

[0045]

[0046] Among them, α c,t is the updated generator loss function weight, α c is the current generator loss function weight, κ cis the adjustment factor, is the accuracy of the discriminator output, τ c is the preset accuracy threshold.

[0047] Specifically, after subtracting the preset accuracy threshold from the accuracy of the discriminator output, it is possible to determine whether the performance of the discriminator is too good or insufficient, and then adjust the current generator loss function weight to obtain an updated generator loss function weight.

[0048] Exemplarily, since the loss functions of the generator and the discriminator in the traditional generative adversarial network are usually fixed, which may lead to instability in the early or late stages of training, the present application dynamically adjusts the weight of the loss function during the training process to adapt to the learning speed and ability of the generator and the discriminator. For example, in each iteration, the weight of the loss function will be automatically adjusted based on the classification accuracy of the discriminator on the generated data. If the accuracy of the discriminator is higher than a certain threshold, the penalty for the generator loss function will be increased, otherwise it will be reduced. This scheme is set up in this way so that the loss function is adjusted in real time according to the output of the discriminator, which is more in line with the actual situation during the training process, so that the data expanded by the sample expansion model obtained by the final training is more reasonable.

[0049] Step A2, constructing the generator loss function according to the updated generator loss function weight, the cross entropy loss function, and the difference in features between the generated data and the real data.

[0050] In the embodiment of the present application, optionally, the generator loss function is determined according to the following formula:

[0051]

[0052] in, is the generator loss function, α c,t is the updated generator loss function weight, BCE() represents the binary cross entropy loss function, label1 represents the false label (for the generator, it is hoped that the data generated by the generator is considered to be real), which can be the label considered by the generator, D c () represents the discriminator function, G c () represents the generator function, z c is the noise vector input to the generator, β c is the L1 norm weight, mc is the number of features of the generated samples, ω ic is the weight of the i-th feature dimension, ∥∥1 represents the L1 norm, ∥f ic (G c (z c ))-f ic (x c )∥1 reflects the difference in characteristics between the generated data and the real data, fic () is the independent activation function of the feature. For example, the independent activation function of the feature adopts the Sigmoid activation function, x c is a real data sample.

[0053] Step A3, determining the updated generator parameters based on the current generator parameters, the generator loss function, the learning rate of the generative adversarial network, and the gradient of the generator parameters.

[0054] In the embodiment of the present application, optionally, the parameter updating process is expressed as:

[0055]

[0056]

[0057] in, and Represent the parameter sets of the generator and the discriminator respectively, z c is the noise vector input to the generator; x c is a real data sample; η c is the learning rate of the generative adversarial network; represents the gradient of the generator parameters; represents the gradient of the discriminator parameters; G c () represents the generator function; D c () represents the discriminator function; Represents the loss function of the generator; ← represents the parameter update operation.

[0058] This scheme is set up in this way, so that the weights in the loss function can be adaptively adjusted according to the output of the discriminator during the training process, and the loss function of the generator includes the composite loss of cross entropy and feature differentiation, so that the parameter values ​​after the parameter update are more reasonable.

[0059] In the embodiment of the present application, optionally, when the parameters of the generator are updated multiple times based on the association information between the generated data and the real data, one of the parameter update processes includes: determining the updated generator parameters according to the following formula:

[0060]

[0061] in, is the updated generator parameter, ← indicates parameter update, is the current generator parameter, λ c is the learning rate of the feature decoupling process, is the gradient of the generator parameters, mc is the number of features of the generated samples, f ic is the independent activation function of the feature, G c() represents the generator function, z c is the noise vector input to the generator, x c is a real data sample, ∥∥2 is the L2 norm.

[0062] For example, in this parameter update, by decoupling different features, the generator can learn more detailed and independent feature expressions, thereby generating higher quality data.

[0063] In the embodiment of the present application, optionally, when the parameters of the generator are updated multiple times based on the association information between the generated data and the real data, one of the parameter update processes includes: determining the updated generator parameters according to the following formula:

[0064]

[0065] in, is the updated generator parameter, ← indicates parameter update, is the current generator parameter, γ c is the similarity adjustment rate, is the gradient of the generator parameters, mc is the number of features of the generated samples, h ic () represents the high-order similarity calculation function on the i-th dimension, G c () represents the generator function, z c is the noise vector input to the generator, x c is a real data sample.

[0066] Exemplarily, the third update of the generator parameters is performed, and the similarity feedback is used for adjustment. The update of the generator parameters is performed by comparing the intrinsic structure of the generated data with the real data. Specifically, according to the feedback provided by the discriminator, the intrinsic structure of the generated data is adjusted to make it closer to the internal characteristics of the real data.

[0067] S130, entering the next iteration based on the sample expansion model after parameter update, if it is determined that the iteration of the sample expansion model is completed, the model obtained after the iteration is determined as the trained sample expansion model; the trained sample expansion model is used to expand the training samples of the equipment lubrication system to obtain expanded data, and the model for identifying lubrication system faults is trained based on the expanded data, and the detection data of the lubrication system to be diagnosed is processed based on the trained model for identifying lubrication system faults to obtain the fault type of the lubrication system to be diagnosed.

[0068] Exemplarily, the iteration step S120 is repeated until a preset stop iteration condition is met, indicating that the model training is completed. In one embodiment, the preset stop iteration condition is reaching a preset maximum number of iterations. Exemplarily, the preset maximum number of iterations is set to 1000 times.

[0069] After the sample expansion model training is completed, the method further includes: increasing the number of samples through the sample expansion model. In one embodiment, assuming that the number of training samples in the original data set is 800, and the sample expansion model expands and generates 200 samples, the expanded data includes 1000 samples. Based on the expanded data, the model for identifying lubrication system faults is trained, and based on the trained model for identifying lubrication system faults, the detection data of the lubrication system to be diagnosed is processed to obtain the fault type of the lubrication system to be diagnosed.

[0070] The technical solution of the embodiment of the present application includes: obtaining an original data set of an equipment lubrication system; iteratively training a sample expansion model to be trained using the original data set, and in each iteration of the sample expansion model to be trained, updating the parameters of the generator multiple times based on the association information between generated data and real data; the association information between the generated data and the real data includes: the accuracy of the discriminator output, the cross entropy loss function, and the difference in features between the generated data and the real data; the accuracy of the discriminator output is the accuracy of the discriminator in classifying the real data and the generated data; the sample expansion model is constructed based on a generative adversarial network; entering the next iteration based on the sample expansion model after parameter update, if it is determined that the iteration of the sample expansion model is completed, the model obtained after the iteration is determined as the trained sample expansion model; the trained sample expansion model is used to expand the training samples of the equipment lubrication system to obtain expanded data, so as to train a model for identifying lubrication system faults based on the expanded data, and to process the detection data of the lubrication system to be diagnosed based on the trained model for identifying lubrication system faults to obtain the fault type of the lubrication system to be diagnosed. This technical solution updates the parameters of the sample expansion model multiple times by generating correlation information between data and real data, so that the parameters of the sample expansion model can quickly approach theoretically reasonable parameters, making the data obtained by sample expansion of the trained sample expansion model more reasonable, and then training the model for identifying lubrication system faults based on the expanded data to obtain the trained model, and then realizing equipment lubrication system fault detection based on the trained model.

[0071] Embodiment 2

[0072] Figure 2 This is a flow chart of a fault diagnosis method for an equipment lubrication system provided in Example 2 of the present application. The embodiment of the present application is optimized based on the above embodiment.

[0073] like Figure 2 As shown, the method of the embodiment of the present application specifically includes the following steps:

[0074] S210, obtaining expanded sample data; the expanded sample data includes: expanded data of the equipment lubrication system generated by the trained sample expansion model and training samples in the original data set.

[0075] S220, iteratively training the feature extraction model to be trained using the expanded sample data, and in each iteration of the feature extraction model to be trained, updating the parameters of the feature extraction model after the previous iteration according to the oscillation frequency of the current iteration, the phase of the oscillation of the current iteration, and the amplitude of the oscillation of the current iteration.

[0076] Wherein, the feature extraction model is constructed based on a fully connected neural network; exemplarily, the number of layers of the fully connected neural network is 6, and the number of nodes in each layer of the fully connected layer is 128, that is, the layer contains 128 neurons.

[0077] Specifically, the parameters of the neural network are initialized, and at the same time, the hyperparameters of the dynamic adaptive oscillation are initialized, including the initial phase of the oscillation and the initial amplitude of the oscillation. In one embodiment, the initialization method is expressed as:

[0078]

[0079]

[0080] φ 0p =0;

[0081] A 0p =1;

[0082] Among them, ~ indicates compliance with a specific distribution; It means the mean is 0 and the variance is Normal distribution of represents normal distribution; represents the initialization variance of the neural network; W 0p is the initial weight of the neural network, b 0p is the initial bias of the neural network; φ 0p is the initial phase of oscillation; A 0p is the initial amplitude of oscillation. Set to 0.01.

[0083] After the neural network is initialized, in each iteration process, the oscillation frequency of each parameter can be calculated based on the current loss function to obtain the oscillation frequency of the current iteration; and the phase of the oscillation of the current iteration and the amplitude of the oscillation of the current iteration are calculated; and then based on the oscillation frequency of the current iteration, the phase of the oscillation of the current iteration and the amplitude of the oscillation of the current iteration, the parameters of the feature extraction model after the previous iteration are updated to obtain updated parameters.

[0084] In the embodiment of the present application, optionally, the oscillation frequency of the current iteration is determined according to the following formula:

[0085]

[0086] Among them, ω p is the oscillation frequency, is the oscillation frequency of the tth iteration, ω0 is the basic oscillation frequency, α p is a hyperparameter for adjusting the oscillation responsiveness, Lp is the loss function of the neural network, is the symbol of partial derivative, is the weight of the neural network at the t-1th iteration, is the weight of the neural network at the kth iteration, β p is the historical gradient weight factor, which is used to control the influence of historical gradients in the current frequency adjustment. For example, α p Set to 5, β p Set to 0.3; the loss function of the neural network uses cross entropy loss, which is calculated by the preset Softmax on the output of the last layer of the neural network.

[0087] Exemplarily, when determining the oscillation frequency of the current iteration, the oscillation frequency of each parameter can be calculated based on the current loss function. The adjustment of the oscillation frequency can depend on the local curvature estimate of the loss surface and be adjusted using historical gradient information to make the parameter update smoother, thereby enabling more efficient search in the parameter space.

[0088] This technical solution adjusts the oscillation frequency of the current iteration through the historical gradient, so that the process of updating the oscillation frequency is smoother, thereby enabling more effective search in the parameter space.

[0089] In the embodiment of the present application, optionally, the phase of the oscillation of the current iteration is determined according to the following formula:

[0090]

[0091] Among them, φ p is the phase of the oscillation, is the phase of the oscillation at the tth iteration; is the phase of the oscillation at the t-1th iteration; δ p is the phase adjustment factor; tanh() is the hyperbolic tangent function, which can limit the adjustment range of the phase to avoid excessive adjustment leading to instability of the algorithm, λ p is a hyperparameter that adjusts the phase sensitivity. For example, δ p Set to 0.1, λ p Set to 0.3.

[0092] Specifically, since the update of each parameter is also affected by the phase difference, which is determined by the success rate of the previous update and the interaction strength between the parameters, the phase update depends on the effect of the previous parameter update. This setting of the present scheme can simulate the delayed effect of causality, making the parameter update process more reasonable.

[0093] The amplitude of the oscillation of the current iteration is determined according to the following formula:

[0094]

[0095] Among them, A p is the amplitude of the oscillation, is the amplitude of the oscillation at the tth iteration; is the amplitude of the oscillation at the t-1th iteration; γ p is the amplitude adjustment coefficient. For example, γ p Set to 0.95.

[0096] In the embodiment of the present application, the amplitude of the oscillation is automatically adjusted according to the effect of parameter updates in the past iterations. If the update of a certain parameter continues to reduce the loss, its amplitude is increased; otherwise, the amplitude is reduced. This arrangement of the present solution can appropriately adjust the amplitude according to the previous iterations, making the amplitude more reasonable.

[0097] In an embodiment of the present application, optionally, updating the parameters of the feature extraction model after the previous iteration according to the oscillation frequency of the current iteration, the phase of the oscillation of the current iteration, and the amplitude of the oscillation of the current iteration includes:

[0098] The weight of the neural network of the current iteration is determined according to the following formula:

[0099]

[0100] in, is the weight of the neural network at the tth iteration, is the weight of the neural network at the t-1th iteration, is the amplitude of the oscillation at the tth iteration, is the oscillation frequency of the tth iteration, t *is the current iteration number of the neural network, is the phase of the oscillation at the tth iteration;

[0101] The bias of the neural network for the current iteration is determined according to the following formula:

[0102]

[0103] in, is the bias of the neural network at the tth iteration, is the bias of the neural network at the t-1th iteration.

[0104] In an embodiment of the present application, when updating parameters (weights and biases), the update is performed in combination with the oscillation behavior, and sine and cosine functions are used to simulate the oscillation behavior, allowing the parameters to be periodically explored in the gradient direction of the loss function, so that the updated parameters are more reasonable.

[0105] S230, enter the next iteration based on the feature extraction model after parameter update. If it is determined that the iteration of the feature extraction model is completed, the model obtained after the iteration is determined as the trained feature extraction model; the trained feature extraction model is used to extract features from the detection data of the lubrication system to be diagnosed, and obtain a first feature, so as to determine the fault type of the lubrication system to be diagnosed based on the first feature.

[0106] Exemplarily, the iterative parameter updating step is repeated until a preset stop iteration condition is met, indicating that the feature extraction model training is completed. In one embodiment, the preset stop iteration condition is reaching a preset maximum number of iterations, preferably, the preset maximum number of iterations is set to 1000 times.

[0107] The method also includes: after obtaining the trained feature extraction model, acquiring the detection data of the lubrication system to be diagnosed; performing feature extraction on the detection data through the trained feature extraction model to obtain a first feature; processing the first feature through a feature dimension reduction model to obtain a second feature; processing the second feature through a lubrication system fault identification model to obtain a fault type of the lubrication system to be diagnosed.

[0108] The technical solution of the embodiment of the present application expands the original data set through the trained sample expansion model to obtain expanded sample data with a larger number of samples. The feature extraction model can be trained in large quantities based on the expanded sample data. When training the feature extraction model, the parameters of the feature extraction model after the previous iteration are updated according to the oscillation frequency of the current iteration, the phase of the oscillation of the current iteration, and the amplitude of the oscillation of the current iteration to obtain updated parameters. After the iteration, the trained feature extraction model is obtained. The detection data of the lubrication system to be diagnosed is processed based on feature extraction to obtain the first feature extracted therefrom, and the fault type of the lubrication system to be diagnosed can be determined based on the first feature.

[0109] Embodiment 3

[0110] Figure 3 This is a flow chart of a fault diagnosis method for an equipment lubrication system provided in Example 3 of the present application. The embodiment of the present application is optimized based on the above embodiment.

[0111] like Figure 3 As shown, the method of the embodiment of the present application specifically includes the following steps:

[0112] S310, acquiring feature extraction data; the feature extraction data is obtained by processing the expanded sample data with the trained feature extraction model; the expanded sample data includes: the expanded data of the equipment lubrication system generated by the trained sample expansion model and the training samples in the original data set.

[0113] In the embodiment of the present application, an autoencoding neural network based on feature refinement is used as a feature dimension reduction model. The autoencoding neural network based on feature refinement consists of three parts: an encoder, a decoder, and a feature adjustment module. The encoder is responsible for mapping high-dimensional input data to a low-dimensional feature space, and the decoder is used to reconstruct the reduced-dimensional features back to the original space to ensure the reversibility of the dimensionality reduction process. The feature adjustment module dynamically adjusts the reduced-dimensional feature space through recursive feature adaptive optimization, so that important features can be strengthened and secondary features are gradually weakened, so that the reduced-dimensional feature representation has better simplicity while retaining data information. The feature adjustment module is implemented through step S330 and related content.

[0114] S320, iteratively training the feature dimension reduction model to be trained through the feature extraction data, in each iteration process, processing the feature extraction data through the encoder of the feature dimension reduction model to obtain low-dimensional features; the feature dimension reduction model is constructed based on an autoencoder neural network.

[0115] Exemplarily, the feature extraction data is input into an encoder of a feature dimensionality reduction model, and the encoder of the feature dimensionality reduction model adopts a multi-layer nonlinear mapping structure to map the feature extraction data to an initial low-dimensional feature space, which is expressed as:

[0116] Z r =Sig enc (W r X r +b r )

[0117] In the formula, Z r represents low-dimensional features, W r is the weight matrix of the encoder, b r is the encoder bias vector, Sig enc () is the multi-layer Sigmoid activation function of the encoder.

[0118] Furthermore, the feature weights of the low-dimensional features are iteratively optimized through step S330.

[0119] S330, updating the feature weights of the previous iteration according to the gradient of the loss function to the feature weights and the learning rate of the autoencoder neural network to obtain the feature weights of the current iteration.

[0120] For example, A can be set r Corresponding to the feature weight of the entire low-dimensional feature, α r The feature weights of the individual features.

[0121] In the embodiment of the present application, optionally, the feature weight of the previous iteration is updated according to the gradient of the loss function to the feature weight and the learning rate of the autoencoder neural network to obtain the feature weight of the current iteration, including:

[0122] The feature weight of the current iteration is determined according to the following formula:

[0123]

[0124] in, represents the feature weight of the t+1th iteration, represents the feature weight of the tth iteration, η r is the learning rate of the autoencoder neural network, L r () is the loss function of the autoencoder neural network, Y r is the label data, which can be the annotation of the training samples in the original data set. Represents the gradient of the loss function with respect to the feature weight. Exemplarily, the loss function of the autoencoder neural network is the reconstruction error loss function, η r Set to 0.01.

[0125] For example, the feature weights can be initialized and then iteratively optimized. In each round of iteration, the weights of each feature are adjusted according to the performance of the features in the previous round, and those features with important influences are gradually enhanced, and redundant or noise features are gradually weakened, that is, the greater the importance of the feature, the greater its corresponding feature weight. This arrangement of the present solution can achieve the effect of highlighting important features.

[0126] S340, determining adjusted low-dimensional features according to the feature weights after the iteration and the low-dimensional features, and processing the adjusted low-dimensional features through a decoder of the feature dimensionality reduction model to obtain reconstructed data, so as to optimize the autoencoder neural network based on the reconstructed data.

[0127] For example, the adjusted low-dimensional features can be expressed as:

[0128] Z′ r =A r Z r ;

[0129] Among them, Z′ r is the adjusted low-dimensional feature, A r is the feature weight, Z r It is a low-dimensional feature.

[0130] Exemplarily, the reconstruction process of the decoder is expressed as:

[0131] X′ r =Sig dec (W′ r Z′ r +b′ r )

[0132] In the formula, X′ r is the reconstructed high-dimensional data, that is, the reconstructed data, W′ r is the weight matrix of the decoder, b′ r is the bias vector of the decoder, Sig dec () is the multi-layer Sigmoid activation function of the decoder.

[0133] S350, if it is determined that the iteration has ended, the model after the iteration is determined as the trained feature dimensionality reduction model; the trained feature dimensionality reduction model is used to perform feature dimensionality reduction on the first feature to obtain a second feature, so as to determine the fault type of the lubrication system to be diagnosed based on the second feature.

[0134] Exemplarily, if the feature dimension reduction model meets the preset stop iteration condition, it means that the model training is completed. In one embodiment, the preset stop iteration condition is reaching a preset maximum number of iterations. Exemplarily, the preset maximum number of iterations is set to 1000 times.

[0135] The method further includes: after obtaining the trained feature dimension reduction model, using the trained feature dimension reduction model to perform feature dimension reduction on the first feature to obtain a second feature, so as to determine the fault type of the lubrication system to be diagnosed based on the second feature. The first feature is obtained by extracting features from the detection data using the trained feature extraction model.

[0136] The technical solution of the embodiment of the present application sets feature weights for the low-dimensional features output by the encoder, and iterates the feature weights so that the feature weights are proportional to the degree of influence of the low-dimensional features, and then adjusts the low-dimensional features based on the feature weights to obtain adjusted low-dimensional features, and processes the adjusted low-dimensional features based on the encoder to complete the training of the feature dimensionality reduction model. Since the feature weights are proportional to the degree of influence of the low-dimensional features, the trained feature dimensionality reduction model can retain important features when performing feature dimensionality reduction, avoiding the problem of important feature loss.

[0137] Embodiment 4

[0138] Figure 4 This is a flow chart of a fault diagnosis method for an equipment lubrication system provided in Example 4 of the present application. The embodiment of the present application is optimized based on the above embodiment.

[0139] like Figure 4 As shown, the method of the embodiment of the present application specifically includes the following steps:

[0140] S410, obtaining feature dimension reduction data; the feature dimension reduction data is obtained by processing the feature extraction data with the trained feature dimension reduction model; the feature extraction data is obtained by processing the expanded sample data with the trained feature extraction model; the expanded sample data includes: the expanded data of the equipment lubrication system generated by the trained sample expansion model and the training samples in the original data set.

[0141] S420, iteratively train the lubrication system fault identification model to be trained through the feature dimension reduction data, and in each iteration process, update the parameters of the fractional-order neural network according to the loss function of the fractional-order neural network, the learning rate of the fractional-order neural network after annealing adjustment, the first-order moment estimate after bias correction, and the second-order moment estimate after bias correction.

[0142] The lubrication system fault identification model to be trained may be a classifier model, and the lubrication system fault identification model may be constructed based on a fractional-order neural network algorithm optimized by hierarchical annealing.

[0143] In an embodiment of the present application, the feature dimension reduction data is input into the lubrication system fault identification model to be trained, and the fractional-order neural network algorithm based on hierarchical annealing optimization optimizes the network training process through a hierarchical annealing strategy, thereby achieving more refined parameter adjustment, thereby improving classification accuracy and generalization ability.

[0144] Specifically, the feature dimension reduction data is passed through multiple neuron layers of the fractional-order neural network in the lubrication system fault recognition model to be trained, and each layer is nonlinearly transformed using a fractional-order activation function to finally obtain a classification prediction result. In the process of forward propagation of the feature dimension reduction data, the output of each layer is obtained by processing the output of the previous layer through a fractional-order activation function, which is expressed as:

[0145]

[0146] In the formula, is the output of the lth layer of the fractional-order neural network; is the weight matrix of the lth layer of the fractional-order neural network; is the bias vector of the lth layer of the fractional-order neural network; φ u () is a fractional-order activation function.

[0147] In one embodiment, assuming that the input of the function is x, the implementation of the fractional-order activation function is expressed as:

[0148]

[0149] Where u represents the fractional order parameter.

[0150] In an embodiment of the present application, optionally, the parameters of the fractional-order neural network are updated according to the loss function of the fractional-order neural network, the learning rate of the fractional-order neural network after annealing adjustment, the first-order moment estimate after bias correction, and the second-order moment estimate after bias correction, including:

[0151] The weights of the fractional neural network are determined according to the following formula:

[0152]

[0153] in, The weight of the updated fractional-order neural network in layer l, W u (l) is the weight of the fractional-order neural network at layer l, ← is the parameter update operation, η u (t) is the learning rate of the annealed fractional neural network at the tth iteration, and are the first-order moment estimate and the second-order moment estimate after bias correction respectively; ∈ yuis a constant, which is used to prevent division by zero. It usually takes a small positive value, such as 0.01. is the loss function of the fractional-order neural network;

[0154] The bias of the fractional-order neural network is determined according to the following formula:

[0155]

[0156] in, The bias of the updated fractional-order neural network in layer l, is the bias of the fractional-order neural network at layer l.

[0157] Exemplarily, the above parameter update formula uses the approximate gradient method combined with the hierarchical annealing strategy to optimize the parameters of each layer of the fractional-order neural network. The hierarchical annealing strategy is used to help the fractional-order neural network model balance between global search and local refinement, which can avoid falling into the local optimum too early.

[0158] In an embodiment of the present application, optionally, the loss function of the fractional-order neural network is determined according to the following formula:

[0159]

[0160] in, is the loss function of the fractional-order neural network; N is the number of samples input to the fractional-order neural network in the current batch; K is the total number of categories. For example, in the embodiment of the present application, the number of fault categories may be 5; t ik is the true label of the i-th sample belonging to the k-th one; s ik is the probability that the i-th sample obtained through the network belongs to the k-th class; λ u is the regularization coefficient of the fractional-order neural network, which is used to control the trade-off between intra-class compactness and inter-class separation; is the output of the last layer of the fractional neural network; Belongs to class c i The mean of the sample output; μ c , μ c′ They are different categories c and c ′ The sample mean of ;

[0161] Exemplarily, the loss function of the above-mentioned fractional-order neural network is a composite function, which not only calculates the error between the prediction result and the true label, but also considers the intra-class closeness and inter-class separation, which can enhance the discrimination ability of the model.

[0162] The learning rate of the annealed fractional-order neural network after the t-th iteration is determined according to the following formula:

[0163]

[0164] Among them, α u is the learning rate decay rate; η u0 is the learning rate of the initial fractional-order neural network; t us is the current iteration number of the fractional-order neural network; exemplarily, η u0 Set to 0.01, α u Set to 0.95.

[0165] The bias-corrected first-order moment estimate is determined according to the following formula:

[0166]

[0167] Among them, m u (t) is the first-order moment estimate of the t-th iteration, β1 is the attenuation factor of the first moment estimate,

[0168] The bias-corrected second-order moment estimate is determined according to the following formula:

[0169]

[0170] Among them, v u (t) is the second moment estimate of the tth iteration, and β2 is the decay factor of the second moment estimate.

[0171] The first-order moment estimate of the t-th iteration is determined according to the following formula:

[0172] m u (t) = β1m u (t us -1)+(1-β1)g u (t);

[0173] Among them, m u (t) is the first-order moment estimate of the t-th iteration, g u (t) is the gradient of the t-th iteration, β1 is the attenuation factor of the first moment estimate, t us is the current iteration number of the fractional-order neural network

[0174] The second-order moment estimate of the t-th iteration is determined according to the following formula:

[0175] v u (t) = β2v u (t us -1)+(1-β2)g u (t) 2 ;

[0176] Among them, v u (t) is the second-order moment estimate of the t-th iteration.

[0177] S430, entering the next iteration based on the lubrication system fault identification model after parameter update. If it is determined that the iteration of the lubrication system fault identification model is completed, the model after the iteration is determined as the trained lubrication system fault identification model; the trained lubrication system fault identification model is used to process the second feature to obtain the fault type of the lubrication system to be diagnosed.

[0178] Exemplarily, the lubrication system fault identification model is iterated until a preset stop iteration condition is met, indicating that the model training is completed. In one embodiment, the preset stop iteration condition is reaching a preset maximum number of iterations, and exemplary, the preset maximum number of iterations is set to 800 times.

[0179] In the embodiment of the present application: after the sample is expanded by the sample expansion model, the feature extraction model is trained based on the sample expansion data, and then the feature dimension reduction model is trained based on the feature extraction data, and then the lubrication system fault identification model is trained based on the feature dimension reduction data. After the above model training is completed, the method further includes step B1-B:

[0180] Step B1, obtaining detection data of the lubrication system to be diagnosed.

[0181] Step B2: extracting features from the detection data using the trained feature extraction model to obtain a first feature.

[0182] Step B, 3, processing the first feature through the trained feature dimension reduction model to obtain the second feature.

[0183] Step B4: Process the second feature through the trained lubrication system fault identification model to obtain the fault type of the lubrication system to be diagnosed.

[0184] Exemplarily, the fault type may be one or more of the following: “normal operation”, “insufficient lubrication”, “excessive lubrication”, “contamination” and “mechanical damage”.

[0185] Embodiment 5

[0186] Figure 5 This is a schematic diagram of the structure of a fault diagnosis device for a lubrication system of an equipment provided in Embodiment 5 of the present application. The device can execute the fault diagnosis method for the lubrication system of an equipment provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method. Figure 5 As shown, the device comprises:

[0187] The original data set acquisition module 510 is used to acquire the original data set of the equipment lubrication system;

[0188] The sample expansion model training module 520 is used to iteratively train the sample expansion model to be trained through the original data set. In each iteration of the sample expansion model to be trained, the parameters of the generator are updated multiple times based on the association information between the generated data and the real data; the association information between the generated data and the real data includes: the accuracy of the discriminator output, the cross entropy loss function, and the difference in features between the generated data and the real data; the accuracy of the discriminator output is the accuracy of the discriminator in classifying the real data and the generated data; the sample expansion model is constructed based on a generative adversarial network;

[0189] The trained sample expansion model determination module 530 is used to enter the next iteration based on the sample expansion model after parameter update. If it is determined that the iteration of the sample expansion model is completed, the model obtained after the iteration is determined as the trained sample expansion model; the trained sample expansion model is used to expand the training samples of the equipment lubrication system to obtain expanded data, so as to train the model for identifying lubrication system faults based on the expanded data, and to process the detection data of the lubrication system to be diagnosed based on the trained model for identifying lubrication system faults to obtain the fault type of the lubrication system to be diagnosed.

[0190] Optionally, the sample expansion model training module 520 includes: a first parameter updating unit, the parameter updating unit including:

[0191] A loss function weight updating subunit, used to adjust the current generator loss function weight according to the accuracy of the discriminator output to obtain an updated generator loss function weight; the accuracy of the discriminator output is proportional to the updated generator loss function weight;

[0192] The loss function updating subunit is used to construct the generator loss function according to the updated generator loss function weight, the cross entropy loss function, and the difference in features between the generated data and the real data;

[0193] The parameter updating subunit is used to determine the updated generator parameters based on the current generator parameters, the generator loss function, the learning rate of the generative adversarial network, and the gradient of the generator parameters.

[0194] Optionally, the loss function weight updating subunit is specifically used for:

[0195] The updated generator loss function weight is determined according to the following formula:

[0196]

[0197] Among them, α c,t is the updated generator loss function weight, α c is the current generator loss function weight, κc is the adjustment factor, is the accuracy of the discriminator output, τ c is the preset accuracy threshold.

[0198] Optionally, the loss function updates the subunit, specifically for:

[0199] The generator loss function is determined according to the following formula:

[0200]

[0201] in, is the generator loss function, α c,t is the updated generator loss function weight, BCE() represents the binary cross entropy loss function, label1 represents the false label, and D c () represents the discriminator function, G c () represents the generator function, z c is the noise vector input to the generator, β c is the L1 norm weight, mc is the number of features of the generated samples, ω ic is the weight of the i-th feature dimension, ∥∥1 represents the L1 norm, ∥f ic (G c (z c ))-f ic (x c )∥1 reflects the difference in characteristics between the generated data and the real data, f ic () is the independent activation function of the feature, x c is a real data sample.

[0202] Optionally, the sample expansion model training module 520 includes: a second parameter updating unit, which is specifically used to:

[0203] The updated generator parameters are determined according to the following formula:

[0204]

[0205] in, is the updated generator parameter, ← indicates parameter update, is the current generator parameter, λ c is the learning rate of the feature decoupling process, is the gradient of the generator parameters, mc is the number of features of the generated samples, f ic is the independent activation function of the feature, G c () represents the generator function, z c is the noise vector input to the generator, x c is a real data sample, ∥∥2 is the L2 norm.

[0206] Optionally, the sample expansion model training module 520 includes: a third parameter updating unit, which is specifically used to:

[0207] The updated generator parameters are determined according to the following formula:

[0208]

[0209] in, is the updated generator parameter, ← indicates parameter update, is the current generator parameter, γ c is the similarity adjustment rate, is the gradient of the generator parameters, mc is the number of features of the generated samples, h ic () represents the high-order similarity calculation function on the i-th dimension, G c () represents the generator function, z c is the noise vector input to the generator, x c is a real data sample.

[0210] Optionally, the device further comprises:

[0211] An expanded sample data acquisition module is used to acquire expanded sample data; the expanded sample data includes: expanded data of the equipment lubrication system generated by the trained sample expansion model and training samples in the original data set;

[0212] A feature extraction model training module, used for iteratively training the feature extraction model to be trained through the expanded sample data, and in each iteration process of the feature extraction model to be trained, updating the parameters of the feature extraction model after the previous iteration according to the oscillation frequency of the current iteration, the phase of the oscillation of the current iteration, and the amplitude of the oscillation of the current iteration; the feature extraction model is constructed based on a fully connected neural network;

[0213] The trained feature extraction model determination module is used to enter the next iteration based on the feature extraction model after parameter update. If it is determined that the iteration of the feature extraction model is completed, the model obtained after the iteration is determined as the trained feature extraction model; the trained feature extraction model is used to perform feature extraction on the detection data of the lubrication system to be diagnosed, and obtain a first feature, so as to determine the fault type of the lubrication system to be diagnosed based on the first feature.

[0214] Optional feature extraction model training module, specifically used for:

[0215] The weight of the neural network of the current iteration is determined according to the following formula:

[0216]

[0217] in, is the weight of the neural network at the tth iteration, is the weight of the neural network at the t-1th iteration, is the amplitude of the oscillation at the tth iteration, is the oscillation frequency of the tth iteration, t * is the current iteration number of the neural network, is the phase of the oscillation at the tth iteration;

[0218] The bias of the neural network for the current iteration is determined according to the following formula:

[0219]

[0220] in, is the bias of the neural network at the tth iteration, is the bias of the neural network at the t-1th iteration.

[0221] Optionally, the oscillation frequency of the current iteration is determined according to the following formula:

[0222]

[0223] Among them, ω p is the oscillation frequency, is the oscillation frequency of the tth iteration, ω0 is the basic oscillation frequency, α p is a hyperparameter for adjusting the oscillation responsiveness, Lp is the loss function of the neural network, is the symbol of partial derivative, is the weight of the neural network at the t-1th iteration, β p is the historical gradient weight factor;

[0224] The phase of the oscillation of the current iteration is determined according to the following formula:

[0225]

[0226] Among them, φ p is the phase of the oscillation, is the phase of the oscillation at the tth iteration; is the phase of the oscillation at the t-1th iteration; δ p is the phase adjustment factor; tanh() is the hyperbolic tangent function, λ p is a hyperparameter that adjusts phase sensitivity;

[0227] The amplitude of the oscillation of the current iteration is determined according to the following formula:

[0228]

[0229] Among them, A p is the amplitude of the oscillation, is the amplitude of the oscillation at the tth iteration; is the amplitude of the oscillation at the t-1th iteration; γ p is the amplitude adjustment factor.

[0230] Optionally, the device further comprises:

[0231] A feature extraction data acquisition module is used to acquire feature extraction data; the feature extraction data is obtained by processing the expanded sample data with the trained feature extraction model;

[0232] A low-dimensional feature determination module, used for iteratively training a feature dimension reduction model to be trained through the feature extraction data, and in each iteration, processing the feature extraction data through an encoder of the feature dimension reduction model to obtain low-dimensional features; the feature dimension reduction model is constructed based on an autoencoder neural network;

[0233] The feature weight update module is used to update the feature weight of the previous iteration according to the gradient of the loss function to the feature weight and the learning rate of the autoencoder neural network to obtain the feature weight of the current iteration;

[0234] A reconstructed data determination module, used to determine the adjusted weights according to the feature weights of the current iteration and the low-dimensional features, and process the adjusted weights through a decoder of the feature dimensionality reduction model to obtain reconstructed data, so as to optimize the autoencoder neural network based on the reconstructed data;

[0235] The module for determining the trained feature dimensionality reduction model is used to determine the model after the iteration as the trained feature dimensionality reduction model if it is determined that the iteration has ended; the trained feature dimensionality reduction model is used to perform feature dimensionality reduction on the first feature to obtain a second feature, so as to determine the fault type of the lubrication system to be diagnosed based on the second feature.

[0236] Optionally, the feature weight updating module is specifically used to:

[0237] The feature weight of the current iteration is determined according to the following formula:

[0238]

[0239] in, represents the feature weight of the t+1th iteration, represents the feature weight of the tth iteration, η r is the learning rate of the autoencoder neural network, L r () is the loss function of the autoencoder neural network, Y r is the label data, Represents the gradient of the loss function with respect to the feature weights.

[0240] Optionally, the device further comprises:

[0241] A feature dimension reduction data acquisition module is used to acquire feature dimension reduction data; the feature dimension reduction data is obtained by processing the feature extraction data with the trained feature dimension reduction model;

[0242] A lubrication system fault identification model training module is used to iteratively train the lubrication system fault identification model to be trained through the feature dimension reduction data, and in each iteration process, the parameters of the fractional-order neural network are updated according to the loss function of the fractional-order neural network, the learning rate of the fractional-order neural network after annealing adjustment, the first-order moment estimation after bias correction, and the second-order moment estimation after bias correction;

[0243] The trained lubrication system fault identification model determination module is used to enter the next iteration based on the lubrication system fault identification model after parameter update. If it is determined that the iteration of the lubrication system fault identification model is completed, the model after the iteration is determined as the trained lubrication system fault identification model; the trained lubrication system fault identification model is used to process the second feature to obtain the fault type of the lubrication system to be diagnosed.

[0244] Optionally, the lubrication system fault identification model training module includes:

[0245] The weight updating unit is used to determine the weight of the fractional-order neural network according to the following formula:

[0246]

[0247] in, The weights of the updated fractional-order neural network in layer l, is the weight of the fractional-order neural network at layer l, ← is the parameter update operation, η u (t) is the learning rate of the annealed fractional neural network at the tth iteration, and are the first-order moment estimate and the second-order moment estimate after bias correction respectively; ∈ yu is a constant, is the loss function of the fractional-order neural network;

[0248] The bias update unit is used to determine the bias of the fractional-order neural network according to the following formula:

[0249]

[0250] in, The bias of the updated fractional-order neural network in layer l, is the bias of the fractional-order neural network at layer l.

[0251] Optionally, the loss function of the fractional-order neural network is determined according to the following formula:

[0252]

[0253] in, is the loss function of the fractional neural network; N is the number of samples input to the fractional neural network in the current batch; K is the total number of categories; t ik is the true label of the i-th sample belonging to the k-th one; s ik is the probability that the i-th sample obtained through the network belongs to the k-th class; λ u is the regularization coefficient of the fractional-order neural network, which is used to control the trade-off between intra-class compactness and inter-class separation; is the output of the last layer of the fractional neural network; Belongs to class c i The mean of the sample output; μ c , μ c′ are the sample means of different categories c and c′ respectively;

[0254] The learning rate of the annealed fractional-order neural network after the t-th iteration is determined according to the following formula:

[0255]

[0256] Among them, α u is the learning rate decay rate; η u0 is the learning rate of the initial fractional-order neural network; t us is the current iteration number of the fractional-order neural network;

[0257] The bias-corrected first-order moment estimate is determined according to the following formula:

[0258]

[0259] Among them, m u (t) is the first-order moment estimate of the t-th iteration, β1 is the attenuation factor of the first moment estimate,

[0260] The bias-corrected second-order moment estimate is determined according to the following formula:

[0261]

[0262] Among them, v u (t) is the second moment estimate of the tth iteration, and β2 is the decay factor of the second moment estimate.

[0263] A fault diagnosis device for an equipment lubrication system provided in an embodiment of the present application can execute a fault diagnosis method for an equipment lubrication system provided in any embodiment of the present invention, and has functional modules and beneficial effects corresponding to the execution method.

[0264] Embodiment 6

[0265] Figure 6 A schematic diagram of the structure of an electronic device 10 that can be used to implement an embodiment of the present invention is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices (such as helmets, glasses, watches, etc.) and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present invention described and / or required herein.

[0266] like Figure 6 As shown, the electronic device 10 includes at least one processor 11, and a memory connected to the at least one processor 11, such as a read-only memory (ROM) 12, a random access memory (RAM) 13, etc., wherein the memory stores a computer program that can be executed by at least one processor, and the processor 11 can perform various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 12 or the computer program loaded from the storage unit 18 to the random access memory (RAM) 13. In the RAM 13, various programs and data required for the operation of the electronic device 10 can also be stored. The processor 11, the ROM 12, and the RAM 13 are connected to each other through a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0267] A number of components in the electronic device 10 are connected to the I / O interface 15, including: an input unit 16, such as a keyboard, a mouse, etc.; an output unit 17, such as various types of displays, speakers, etc.; a storage unit 18, such as a disk, an optical disk, etc.; and a communication unit 19, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 19 allows the electronic device 10 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0268] The processor 11 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, etc. The processor 11 executes the various methods and processes described above, such as a fault diagnosis method for a lubrication system of an equipment.

[0269] In some embodiments, the fault diagnosis method of the equipment lubrication system can be implemented as a computer program, which is tangibly contained in a computer-readable storage medium, such as a storage unit 18. In some embodiments, part or all of the computer program can be loaded and / or installed on the electronic device 10 via the ROM 12 and / or the communication unit 19. When the computer program is loaded into the RAM 13 and executed by the processor 11, one or more steps of the fault diagnosis method of the equipment lubrication system described above can be performed. Alternatively, in other embodiments, the processor 11 can be configured to execute the fault diagnosis method of the equipment lubrication system by any other appropriate means (for example, by means of firmware).

[0270] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0271] Computer programs for implementing the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that when the computer program is executed by the processor, the functions / operations specified in the flow chart and / or block diagram are implemented. The computer program may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0272] In the context of the present invention, a computer-readable storage medium may be a tangible medium that may contain or store a computer program for use by or in combination with an instruction execution system, device or equipment. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or equipment, or any suitable combination of the foregoing. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0273] To provide interaction with a user, the systems and techniques described herein may be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and a pointing device (e.g., a mouse or trackball) through which the user can provide input to the electronic device. Other types of devices may also be used to provide interaction with the user; for example, the feedback provided to the user may be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user may be received in any form (including acoustic input, voice input, or tactile input).

[0274] The systems and techniques described herein may be implemented in a computing system that includes backend components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes frontend components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such backend components, middleware components, or frontend components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), a blockchain network, and the Internet.

[0275] A computing system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The client and server relationship is generated by computer programs running on the corresponding computers and having a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system to solve the defects of difficult management and weak business scalability in traditional physical hosts and VPS services.

[0276] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps described in the present invention can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solution of the present invention can be achieved, and this document does not limit this.

[0277] The above specific implementations do not constitute a limitation on the protection scope of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A fault diagnosis method for a lubrication system of an equipment, characterized in that: include: Obtain the original data set of the equipment lubrication system; Iteratively training the training sample expansion model using the original data set, and updating the parameters of the generator multiple times based on the association information between the generated data and the real data during each iteration of the training sample expansion model; The association information between the generated data and the real data includes: the accuracy of the discriminator output, the cross entropy loss function, and the difference in features between the generated data and the real data; the accuracy of the discriminator output is the accuracy of the discriminator in classifying the real data and the generated data; the sample expansion model is constructed based on the generative adversarial network; Enter the next iteration based on the sample expansion model after parameter update. If it is determined that the iteration of the sample expansion model is completed, the model obtained after the iteration is determined as the trained sample expansion model; the trained sample expansion model is used to expand the training samples of the equipment lubrication system to obtain expanded data, and the model for identifying lubrication system faults is trained based on the expanded data, and the detection data of the lubrication system to be diagnosed is processed based on the trained model for identifying lubrication system faults to obtain the fault type of the lubrication system to be diagnosed.

2. The method according to claim 1, characterized in that: When the parameters of the generator are updated multiple times based on the association information between the generated data and the real data, one of the parameter update processes includes: The current generator loss function weight is adjusted according to the accuracy of the discriminator output to obtain an updated generator loss function weight; the accuracy of the discriminator output is proportional to the updated generator loss function weight; Construct the generator loss function based on the updated generator loss function weight, the cross entropy loss function, and the difference in features between the generated data and the real data; The updated generator parameters are determined based on the current generator parameters, the generator loss function, the learning rate of the generative adversarial network, and the gradient of the generator parameters.

3. The method according to claim 2, characterized in that The updated generator loss function weight is determined according to the following formula: Among them, α c,t is the updated generator loss function weight, α c is the current generator loss function weight, κ c is the adjustment factor, is the accuracy of the discriminator output, τ c is the preset accuracy threshold.

4. The method according to claim 2, characterized in that: The generator loss function is determined according to the following formula: in, is the generator loss function, α c,t is the updated generator loss function weight, BCE() represents the binary cross entropy loss function, label1 represents the false label, and D c () represents the discriminator function, G c () represents the generator function, z c is the noise vector input to the generator, β c is the L1 norm weight, mc is the number of features of the generated samples, ω ic is the weight of the i-th feature dimension, ∥∥1 represents the L1 norm, ∥f ic (G c (z c ))-f ic (x c )∥1 reflects the difference in characteristics between the generated data and the real data, f ic () is the independent activation function of the feature, x c is a real data sample.

5. The method according to claim 1, characterized in that When the parameters of the generator are updated multiple times based on the association information between the generated data and the real data, one of the parameter update processes includes: determining the updated generator parameters according to the following formula: in, is the updated generator parameter, ← indicates parameter update, is the current generator parameter, λ c is the learning rate of the feature decoupling process, is the gradient of the generator parameters, mc is the number of features of the generated samples, f ic is the independent activation function of the feature, G c () represents the generator function, z c is the noise vector input to the generator, x c is a real data sample, ∥∥2 is the L2 norm.

6. The method according to claim 1, characterized in that When the parameters of the generator are updated multiple times based on the association information between the generated data and the real data, one of the parameter update processes includes: determining the updated generator parameters according to the following formula: in, is the updated generator parameter, ← indicates parameter update, is the current generator parameter, γ c is the similarity adjustment rate, is the gradient of the generator parameters, mc is the number of features of the generated samples, h ic () represents the high-order similarity calculation function on the i-th dimension, G c () represents the generator function, z c is the noise vector input to the generator, x c is a real data sample.

7. The method according to claim 1, characterized in that The method further comprises: Acquire expanded sample data; the expanded sample data includes: expanded data of the equipment lubrication system generated by the trained sample expansion model and training samples in the original data set; Iteratively training the feature extraction model to be trained by using the expanded sample data, and in each iteration of the feature extraction model to be trained, updating the parameters of the feature extraction model after the previous iteration according to the oscillation frequency of the current iteration, the phase of the oscillation of the current iteration, and the amplitude of the oscillation of the current iteration; the feature extraction model is constructed based on a fully connected neural network; Enter the next iteration based on the feature extraction model after parameter update. If it is determined that the iteration of the feature extraction model is completed, the model obtained after the iteration is determined as the trained feature extraction model; the trained feature extraction model is used to perform feature extraction on the detection data of the lubrication system to be diagnosed, and obtain a first feature, so as to determine the fault type of the lubrication system to be diagnosed based on the first feature.

8. The method according to claim 7, characterized in that The parameters of the feature extraction model after the previous iteration are updated according to the oscillation frequency of the current iteration, the phase of the oscillation of the current iteration, and the amplitude of the oscillation of the current iteration, including: The weight of the neural network of the current iteration is determined according to the following formula: in, is the weight of the neural network at the tth iteration, is the weight of the neural network at the t-1th iteration, is the amplitude of the oscillation at the tth iteration, is the oscillation frequency of the tth iteration, t * is the current iteration number of the neural network, is the phase of the oscillation at the tth iteration; The bias of the neural network for the current iteration is determined according to the following formula: in, is the bias of the neural network at the tth iteration, is the bias of the neural network at the t-1th iteration.

9. The method according to claim 7, characterized in that: The oscillation frequency of the current iteration is determined according to the following formula: Among them, ω p is the oscillation frequency, is the oscillation frequency of the tth iteration, ω0 is the basic oscillation frequency, α p is a hyperparameter for adjusting the oscillation responsiveness, Lp is the loss function of the neural network, is the symbol of partial derivative, is the weight of the neural network at the t-1th iteration, β p is the historical gradient weight factor; The phase of the oscillation of the current iteration is determined according to the following formula: Among them, φ p is the phase of the oscillation, is the phase of the oscillation at the tth iteration; is the phase of the oscillation at the t-1th iteration; δ p is the phase adjustment factor; tanh() is the hyperbolic tangent function, λ p is a hyperparameter that adjusts phase sensitivity; The amplitude of the oscillation of the current iteration is determined according to the following formula: Among them, A p is the amplitude of the oscillation, is the amplitude of the oscillation at the tth iteration; is the amplitude of the oscillation at the t-1th iteration; γ p is the amplitude adjustment factor.

10. The method according to claim 7, characterized in that The method further comprises: Acquire feature extraction data; the feature extraction data is obtained by processing the expanded sample data with the trained feature extraction model; Iteratively training a feature dimension reduction model to be trained is performed using the feature extraction data. In each iteration, the feature extraction data is processed by an encoder of the feature dimension reduction model to obtain low-dimensional features; the feature dimension reduction model is constructed based on an autoencoder neural network; The feature weights of the previous iteration are updated according to the gradient of the loss function to the feature weights and the learning rate of the autoencoder neural network to obtain the feature weights of the current iteration; Determining adjusted weights according to feature weights of the current iteration and the low-dimensional features, and processing the adjusted weights through a decoder of the feature dimensionality reduction model to obtain reconstructed data, so as to optimize the autoencoder neural network based on the reconstructed data; If it is determined that the iteration has ended, the model after the iteration is determined as the trained feature dimensionality reduction model; the trained feature dimensionality reduction model is used to perform feature dimensionality reduction on the first feature to obtain a second feature, so as to determine the fault type of the lubrication system to be diagnosed based on the second feature.

11. The method according to claim 10, characterized in that The feature weights of the previous iteration are updated according to the gradient of the loss function to the feature weights and the learning rate of the autoencoder neural network to obtain the feature weights of the current iteration, including: The feature weight of the current iteration is determined according to the following formula: in, represents the feature weight of the t+1th iteration, represents the feature weight of the tth iteration, η r is the learning rate of the autoencoder neural network, L r () is the loss function of the autoencoder neural network, Y r is the label data, Represents the gradient of the loss function with respect to the feature weights.

12. The method according to claim 10, characterized in that The method further comprises: Acquire feature dimension reduction data; the feature dimension reduction data is obtained by processing the feature extraction data with the trained feature dimension reduction model; Iteratively training the lubrication system fault identification model to be trained is performed using the feature dimension reduction data, and in each iteration, the parameters of the fractional-order neural network are updated according to the loss function of the fractional-order neural network, the learning rate of the fractional-order neural network after annealing adjustment, the first-order moment estimation after bias correction, and the second-order moment estimation after bias correction; The next iteration is entered based on the lubrication system fault identification model after parameter update. If it is determined that the iteration of the lubrication system fault identification model is completed, the model after the iteration is determined as the trained lubrication system fault identification model; the trained lubrication system fault identification model is used to process the second feature to obtain the fault type of the lubrication system to be diagnosed.

13. The method according to claim 12, characterized in that The parameters of the fractional-order neural network are updated according to the loss function of the fractional-order neural network, the learning rate of the fractional-order neural network after annealing adjustment, the first-order moment estimate after bias correction, and the second-order moment estimate after bias correction, including: The weights of the fractional neural network are determined according to the following formula: in, The weights of the updated fractional-order neural network in layer l, is the weight of the fractional-order neural network at layer l, ← is the parameter update operation, η u (t) is the learning rate of the annealed fractional neural network at the tth iteration, and are the first-order moment estimate and the second-order moment estimate after bias correction respectively; ∈ yu is a constant, is the loss function of the fractional-order neural network; The bias of the fractional-order neural network is determined according to the following formula: in, The bias of the updated fractional-order neural network in layer l, is the bias of the fractional-order neural network at layer l.

14. The method according to claim 13, characterized in that The loss function of the fractional-order neural network is determined according to the following formula: in, is the loss function of the fractional neural network; N is the number of samples input to the fractional neural network in the current batch; K is the total number of categories; t ik is the true label of the i-th sample belonging to the k-th one; s ik is the probability that the i-th sample obtained through the network belongs to the k-th class; λ u is the regularization coefficient of the fractional-order neural network, which is used to control the trade-off between intra-class compactness and inter-class separation; is the output of the last layer of the fractional neural network; Belongs to class c i The mean of the sample output; μ c , μ c′ are the sample means of different categories c and c′ respectively; The learning rate of the annealed fractional-order neural network after the t-th iteration is determined according to the following formula: Among them, α u is the learning rate decay rate; η u0 is the learning rate of the initial fractional-order neural network; t us is the current iteration number of the fractional-order neural network; The bias-corrected first-order moment estimate is determined according to the following formula: Among them, m u (t) is the first-order moment estimate of the t-th iteration, β1 is the attenuation factor of the first moment estimate, The bias-corrected second-order moment estimate is determined according to the following formula: Among them, v u (t) is the second moment estimate of the tth iteration, and β2 is the decay factor of the second moment estimate.

15. A fault diagnosis device for a lubrication system of an equipment, characterized in that: include: The original data set acquisition module is used to acquire the original data set of the equipment lubrication system; A sample expansion model training module, used for iteratively training the sample expansion model to be trained using the original data set, and in each iteration of the sample expansion model to be trained, updating the parameters of the generator multiple times based on the association information between the generated data and the real data; The association information between the generated data and the real data includes: the accuracy of the discriminator output, the cross entropy loss function, and the difference in features between the generated data and the real data; the accuracy of the discriminator output is the accuracy of the discriminator in classifying the real data and the generated data; the sample expansion model is constructed based on the generative adversarial network; The trained sample expansion model determination module is used to enter the next iteration based on the sample expansion model after parameter update. If it is determined that the sample expansion model iteration is completed, the model obtained after the iteration is determined as the trained sample expansion model; the trained sample expansion model is used to expand the training samples of the equipment lubrication system to obtain expansion data, and the model for identifying lubrication system faults is trained based on the expansion data, and the detection data of the lubrication system to be diagnosed is processed based on the trained model for identifying lubrication system faults to obtain the fault type of the lubrication system to be diagnosed.

16. An electronic device, characterized in that: The electronic device comprises: at least one processor; and a memory communicatively connected to the at least one processor; wherein, The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the fault diagnosis method for the equipment lubrication system according to any one of claims 1 to 14.

17. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the fault diagnosis method for the equipment lubrication system according to any one of claims 1 to 14 when executed.

Images