Equipment fault determination method and device

Abstract

The invention provides an equipment fault determination method and device, and relates to the technical field of fault detection. According to the method, open set fault identification is carried out based on rotor position information of a magnetic suspension bearing in equipment, fault types occurring in a training process can be identified, fault types which do not occur in the training process are marked as unknown faults, the similarity between fault samples of the unknown types and known fault types is mined, and fault identification is carried out. And the accuracy and practicability of the magnetic suspension bearing fault diagnosis system are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of fault detection, in particular to a device fault determination method and device. BACKGROUND

[0002] In the fault diagnosis mechanism of large equipment, the traditional method usually focuses on key components (such as bearings, gearboxes, etc.), and the monitoring sensors are arranged near the components to be diagnosed. Such faults are often caused by long-term operation, component wear and fatigue damage.

[0003] For large equipment equipped with magnetic suspension bearings, due to its inherent non-friction and non-contact characteristics, the failure probability of the magnetic bearing itself is extremely low, and most of the failures are caused by the failure of the large equipment system equipped with magnetic suspension bearings or other components. The failure types are various. The existing fault recognition scheme is difficult to collect all kinds of fault samples for training, and it is difficult to effectively identify the faults of large equipment equipped with magnetic suspension bearings. SUMMARY

[0004] Therefore, the present application provides a device determination method and device to effectively identify the faults of large equipment equipped with magnetic suspension bearings.

[0005] Specifically, the present application is realized by the following technical solutions:

[0006] In a first aspect, the present application provides a device fault determination method, comprising:

[0007] Obtaining rotor position information of a magnetic suspension bearing in a target device within a target time period;

[0008] Performing open set fault recognition based on the rotor position information to determine a first target fault type matching the rotor position information from a plurality of fault types; the plurality of fault types include a plurality of preset fault types and an unknown fault type representing a type other than the preset fault types;

[0009] In the case where the first target fault type is an unknown fault type, determining a second target fault type from the plurality of preset fault types based on the attribute features of the rotor position information; the second target fault type is used to provide a reference for determining the actual type corresponding to the unknown fault type.

[0010] Optionally, the open set fault recognition based on the rotor position information to determine a first target fault type matching the rotor position information from a plurality of fault types comprises:

[0011] input the rotor position information into the trained fault classification model to obtain a first target fault type output by the fault classification model; the fault classification model is a neural network with an OpenMax layer.

[0012] Optionally, the fault classification model is determined by the following steps:

[0013] An initial neural network is constructed.

[0014] The initial neural network is trained by using a first rotor position information sample and a true value label corresponding to the first rotor position information sample to obtain a trained neural network.

[0015] The normalization layer of the trained neural network is replaced by the OpenMax layer to obtain the fault classification model.

[0016] Optionally, the second target fault type is determined from the plurality of preset fault types based on the attribute feature of the rotor position information, and the second target fault type comprises:

[0017] A feature distance between the attribute feature of the rotor position information and attribute features of a plurality of second rotor position information samples is determined.

[0018] A target rotor position information sample is determined from the plurality of second rotor position information samples based on the feature distance and a preset approximate threshold.

[0019] A true value label corresponding to the target rotor position information sample is determined as the second target fault type.

[0020] Optionally, the attribute feature of the rotor position information comprises a statistical attribute feature.

[0021] The statistical attribute feature is determined by the following steps:

[0022] Statistical values of the rotor position information in a plurality of statistical dimensions are determined.

[0023] For any statistical dimension, a statistical attribute feature in the statistical dimension is determined based on a statistical value in the statistical dimension and a judgment threshold corresponding to the statistical dimension.

[0024] Optionally, the attribute feature comprises a self-learning attribute feature.

[0025] The self-learning attribute feature is determined by the following steps:

[0026] The rotor position information is input into a trained feature extraction model to obtain a feature coefficient vector output by the feature extraction model, and the feature coefficient vector is taken as the self-learning attribute feature.

[0027] Optionally, the feature extraction model is trained by the following steps:

[0028] An initial feature extraction model is constructed, the initial feature extraction model is used to encode rotor position information to obtain a first feature vector, the first feature vector is reconstructed by using a feature matrix and a feature coefficient vector to obtain a second feature vector, and the second feature vector is decoded to obtain reconstruction information corresponding to the rotor position information;

[0029] The initial feature extraction model is trained based on a first loss function, a second loss function and a third loss function to obtain a trained feature extraction model;

[0030] The trained feature extraction model has a feature matrix that makes the trained feature extraction model satisfy the first loss function, the second loss function and the third loss function.

[0031] The first loss function is used to minimize the difference between the first feature vector and the second feature vector, the second loss function is used to minimize the L0 norm of the feature coefficient vector, and the third loss function is used to minimize the difference between the rotor position information and the reconstruction information.

[0032] Optionally, in the case where the attribute features include statistical attribute features and self-learning attribute features, the feature distance between the attribute features of the rotor position information and the attribute features of a plurality of second rotor position information samples is determined, including:

[0033] For any second rotor position information sample, first difference information between self-learning attribute features in the rotor position information and self-learning attribute features in the attribute features of the second rotor position information sample is determined, and second difference information between statistical attribute features in the rotor position information and statistical attribute features in the attribute features of the second rotor position information sample is determined.

[0034] Based on the first difference information, the second difference information and the weight corresponding to the second difference information, the feature distance between the attribute features of the rotor position information and the attribute features of the second rotor position information sample is determined.

[0035] In a second aspect, the embodiments of the present application also provide a device fault determination apparatus, including:

[0036] The acquisition module is configured to acquire rotor position information of a magnetic suspension bearing in a target device in a target time period.

[0037] The first identification module is configured to perform open set fault identification based on the rotor position information, and determine a first target fault type matching the rotor position information from a plurality of fault types; the plurality of fault types include a plurality of preset fault types and an unknown fault type representing a type other than the preset fault types.

[0038] The second identification module is configured to, in a case where the first target fault type is the unknown fault type, determine a second target fault type from the plurality of preset fault types based on an attribute feature of the rotor position information; the second target fault type is used to provide a reference for determining an actual type corresponding to the unknown fault type.

[0039] Optionally, the first identification module is specifically configured to:

[0040] input the rotor position information into a trained fault classification model to obtain the first target fault type output by the fault classification model; the fault classification model is a neural network having an OpenMax layer.

[0041] Optionally, the apparatus further includes a first training module configured to:

[0042] construct an initial neural network;

[0043] train the initial neural network using a first rotor position information sample and a true value label corresponding to the first rotor position information sample to obtain a trained neural network;

[0044] replace a normalization layer of the trained neural network with the OpenMax layer to obtain the fault classification model.

[0045] Optionally, the second identification module is specifically configured to:

[0046] determine a feature distance between the attribute feature of the rotor position information and attribute features of a plurality of second rotor position information samples, respectively;

[0047] determine a target rotor position information sample from the plurality of second rotor position information samples based on the feature distance and a preset approximation threshold;

[0048] determine a true value label corresponding to the target rotor position information sample as the second target fault type.

[0049] Optionally, the attribute feature of the rotor position information includes a statistical attribute feature; and the second identification module is specifically configured to:

[0050] determine statistical values of the rotor position information in a plurality of statistical dimensions;

[0051] For any statistical dimension, based on a statistical value under the statistical dimension and a judgment threshold corresponding to the statistical dimension, a statistical property feature under the statistical dimension is determined.

[0052] Optionally, the property feature includes a self-learning property feature; and the second identification module is specifically configured to:

[0053] input the rotor position information into the trained feature extraction model to obtain a feature coefficient vector output by the feature extraction model, and take the feature coefficient vector as the self-learning property feature.

[0054] Optionally, the device further includes a second training module configured to:

[0055] construct an initial feature extraction model, wherein the initial feature extraction model is configured to encode the rotor position information to obtain a first feature vector, reconstruct the first feature vector by using a feature matrix and a feature coefficient vector to obtain a second feature vector, and decode the second feature vector to obtain reconstruction information corresponding to the rotor position information;

[0056] train the initial feature extraction model based on a first loss function, a second loss function and a third loss function to obtain a trained feature extraction model;

[0057] the trained feature extraction model has a feature matrix that makes the trained feature extraction model satisfy the first loss function, the second loss function and the third loss function;

[0058] the first loss function is configured to minimize the difference between the first feature vector and the second feature vector, the second loss function is configured to minimize the L0 norm of the feature coefficient vector, and the third loss function is configured to minimize the difference between the rotor position information and the reconstruction information.

[0059] Optionally, in the case where the property feature includes a statistical property feature and a self-learning property feature, the second identification module is specifically configured to:

[0060] for any second rotor position information sample, determine first difference information between the self-learning property feature in the rotor position information and the self-learning property feature in the property feature of the second rotor position information sample, and determine second difference information between the statistical property feature in the rotor position information and the statistical property feature in the property feature of the second rotor position information sample;

[0061] Determine a feature distance between the attribute feature of the rotor position information and the attribute feature of the second rotor position information sample based on the first difference information, the second difference information, and a weight corresponding to the second difference information.

[0062] In a third aspect, the optional implementation of the present application further provides a computer device, a processor and a memory, the memory stores machine readable instructions executable by the processor, and the processor is configured to execute the machine readable instructions stored in the memory, and the machine readable instructions are executed by the processor to perform the steps of the first aspect or any possible implementation of the first aspect.

[0063] In a fourth aspect, the optional implementation of the present application further provides a computer readable storage medium, and the computer readable storage medium stores a computer program, and the computer program is executed to perform the steps of the first aspect or any possible implementation of the first aspect.

[0064] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, but not limiting the technical solutions of the present application.

[0065] The device fault determination method and device provided by the embodiments of the present application can identify open set faults based on rotor position information of a magnetic suspension bearing in a device, can identify fault types that occur in a training process, mark fault types that do not occur in the training process as unknown faults, and mine the similarity between unknown fault samples and known fault types, thereby improving the accuracy and practicability of a magnetic suspension bearing fault diagnosis system.

[0066] In order to make the above objectives, characteristics and advantages of the present application more apparent and easy to understand, the following preferred embodiments are specifically described below, and the accompanying drawings are referred to, and the detailed description is as follows. BRIEF DESCRIPTION OF DRAWINGS

[0067] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiments, the drawings herein are incorporated into the description and form a part of the description, the drawings show the embodiments consistent with the present application, and are used to illustrate the technical solutions of the present application together with the description. It should be understood that the following drawings only show some embodiments of the present application, and therefore should not be regarded as a limitation on the scope, and for those skilled in the art, other related drawings can also be obtained without creative labor on the basis of these drawings.

[0068] Figure 1 A flowchart of a device fault determination method provided by the embodiments of the present application;

[0069] Figure 2A schematic diagram of a feature extraction model provided by an embodiment of the present application;

[0070] Figure 3 A schematic diagram of a device fault determination apparatus provided by an embodiment of the present application;

[0071] Figure 4 A schematic diagram of a computer device structure provided by an embodiment of the present application. DETAILED DESCRIPTION

[0072] The exemplary embodiments will be described in detail herein with reference to the attached drawings. In the following description, the same or similar components are designated by the same or similar reference numerals, and the description thereof will not be repeated. The following exemplary embodiments described in the following description are not presented to cause ambiguity with respect to the scope of the present application. Instead, they are presented merely to most clearly and completely describe the exemplary embodiments consistent with some aspects of the present application as set forth in the appended claims.

[0073] The terminology used in the present application is for the purpose of describing particular embodiments only and is not intended to be limiting of the present application. As used in the description of the application and the appended claims, the singular forms "a", "an" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It also will be understood that the term "and / or" as used herein refers to and encompasses any and all possible combinations of one or more of the associated listed items.

[0074] It is to be understood that the singular forms "a", "an", and "the" include plural referents unless the context clearly dictates otherwise. It is to be understood that the term "and / or" as used herein refers to and encompasses any and all possible combinations of one or more of the associated listed items.

[0075] The term "and / or", as used herein, merely describes one of the associated associated terms, and can represent three relationships, for example, A and / or B, can represent three cases of A alone, A and B together, and B alone. In addition, the term "at least one" herein means any one of the plurality or any combination of at least two of the plurality, for example, including at least one of A, B and C, can mean including any one or more elements selected from the set consisting of A, B and C.

[0076] It is found through research that the prior art usually focuses on key components such as bearings in the fault diagnosis mechanism of large equipment, and the monitoring sensors are arranged near the components to be diagnosed. Such faults are often caused by component wear and fatigue damage caused by long-term operation. However, for large equipment equipped with magnetic suspension bearings, due to the inherent non-friction and non-contact characteristics of magnetic bearings, the failure probability of magnetic bearings is extremely low, and most of the failures are caused by system-level failures of large equipment or failures of other components. The failure recognition scheme in the prior art is difficult to collect all kinds of fault samples for training, and it is difficult to effectively identify the faults of large equipment equipped with magnetic suspension bearings.

[0077] Therefore, the present application provides a device fault determination method and device, which performs open set fault recognition based on the rotor position information of the magnetic suspension bearing in the device, can identify the fault types that occur in the training process, mark the fault types that do not occur in the training process as unknown faults, and mine the similarity between the unknown fault samples and the known fault types, thereby improving the accuracy and practicability of the magnetic suspension bearing fault diagnosis system.

[0078] The above-mentioned defects of the prior art are the results of the inventors' practice and careful research, and therefore, the discovery process of the above-mentioned problems and the solutions proposed by the present application to solve the above-mentioned problems are the contributions of the inventors to the present application.

[0079] In order to facilitate the understanding of the present embodiment, first, a device fault determination method disclosed by the present embodiment is introduced in detail. The execution subject of the device fault determination method provided by the present embodiment is generally a computer device with certain computing power, which includes, for example, a terminal device or a server or other processing device. In some possible implementation manners, the device fault determination method can be realized by a processor calling computer readable instructions stored in a memory.

[0080] Referring to Figure 1 The device fault determination method provided by the present embodiment is a flowchart of a device fault determination method, which includes the following steps.

[0081] S101, obtaining rotor position information of a magnetic suspension bearing in a target device in a target time period.

[0082] In this step, the target device can be a device equipped with a magnetic bearing. Magnetic bearing is an advanced bearing technology that uses magnetic force to suspend the rotor (rotating part) in the stator (stationary part) without contact. It completely eliminates the physical contact and mechanical friction in traditional mechanical bearings (such as ball bearings, sliding bearings). Magnetic bearing is widely used in situations that require high speed, high precision, long life, no pollution, low maintenance or extreme environment, such as high-speed rotating machinery, energy field, medical equipment, aerospace, semiconductor manufacturing, etc.

[0083] In the magnetic bearing-rotor system, the magnetic bearing is usually equipped with displacement sensors to monitor the running state of the bearing in real time. This step can use the displacement offset information collected by the displacement sensor within a certain period of time to identify faults.

[0084] For example, the magnetic bearing is equipped with multiple displacement sensors, such as horizontal and vertical displacement sensors and axial displacement sensors at both ends of the magnetic bearing. These sensors can capture the displacement changes of the magnetic bearing in different directions, forming multi-dimensional time series data.

[0085] The collected raw displacement data is preprocessed, including denoising, filtering, and normalization, to eliminate interference signals and unify data formats. The preprocessed displacement data is segmented according to a predefined time window (target time period) to obtain a series of rotor position information samples.

[0086] In practical applications, the collection frequency and time window length of rotor position information should be reasonably set according to the running characteristics and fault characteristics of the device to ensure that the collected data can fully reflect the running state and potential fault characteristics of the magnetic bearing.

[0087] S102, based on the rotor position information, open set fault identification is performed to determine a first target fault type matching the rotor position information from a plurality of fault types; the plurality of fault types include a plurality of preset fault types and an unknown fault type representing a type other than the preset fault types.

[0088] In this step, the rotor position information obtained is identified by open set identification technology. Unlike traditional closed set identification, open set fault identification can identify fault types that have not been seen in the training process and mark them as "unknown faults", which is of great significance for fault diagnosis in actual engineering applications.

[0089] In some possible implementations, the rotor position information can be input into a trained fault classification model to obtain a first target fault type output by the fault classification model; the fault classification model is a neural network with an OpenMax layer for open set identification.

[0090] The fault classification model can be used to identify the fault of the magnetic suspension bearing. The fault classification model can be trained according to the following steps:

[0091] An initial neural network is constructed;

[0092] The initial neural network is trained using the first rotor position information samples and the true value labels corresponding to the first rotor position information samples, to obtain a trained neural network;

[0093] The normalization layer of the trained neural network is replaced with an OpenMax layer to obtain the fault classification model.

[0094] The initial neural network can be a convolutional neural network (CNN) or a multilayer perceptron (MLP) structure, and a suitable network structure can be selected according to the characteristics of the rotor position information. The initial neural network can be used as a basic classification model, and after the basic classification model is trained, it can be modified into a model that can perform open set fault identification.

[0095] The first rotor position information samples are samples of known fault types, which are used to train the neural network to identify known fault types. During the training process, the network parameters can be optimized through a backpropagation algorithm to enable the network to accurately identify known fault types. The first rotor position information samples can be used as training samples for the initial neural network, and the first rotor position information samples can be input into the initial neural network to obtain the identification results of the initial neural network for the first rotor position information.

[0096] In this step, the first rotor position information samples can be represented as X={x1,x2,...,x n}, and the corresponding true value labels can be represented as Y∈[1,2,3,...,C], where n represents the number of dimensions of the displacement information in the first rotor position information samples, and C represents the number of known fault types. Then a trained initial neural network f:X→Y can be obtained.

[0097] Exemplarily, the initial neural network described above can include multiple convolutional layers, pooling layers, and fully connected layers. The convolutional layers are responsible for extracting local features in the rotor position information, such as frequency domain features and time domain features; the pooling layers are used for dimensionality reduction and extracting main features; and the fully connected layers map the extracted features to the fault type space. In the network design process, factors such as the dimensionality of the data, the complexity of the features, and the limitations of the computing resources need to be considered, and parameters such as the depth, width, and activation function of the network are set reasonably. In addition, a normalized SoftMax layer needs to be added to the output layer of the network to convert the output of the network into a probability distribution of each fault type. This initial neural network will serve as the basis for the open set fault recognition model, which will be trained and modified in subsequent steps.

[0098] Training the neural network is a key step to achieve accurate fault recognition. In this stage, a large number of rotor position information samples of known fault types need to be prepared as training data. These samples can cover as many known fault types and working conditions as possible to improve the generalization ability of the model. Each sample has a corresponding true value label indicating the fault type to which the sample belongs. The training process uses a supervised learning method to optimize the network parameters through the backpropagation algorithm and gradient descent method.

[0099] Exemplarily, first, the rotor position information samples are input into the initial neural network to obtain the predicted output of the network. Then, the loss between the predicted output and the true value label is calculated, and the commonly used loss functions include the cross-entropy loss function. Next, the gradient of the loss function with respect to the network parameters is calculated through the backpropagation algorithm, and the network parameters are updated using the gradient descent method to reduce the loss. This process is repeated multiple times (multiple training rounds) until the network converges or reaches the preset training rounds. To prevent overfitting, some regularization techniques such as L1 / L2 regularization, Dropout, etc. are usually used. In addition, techniques such as cross-validation can be used to evaluate the performance of the model and adjust the network structure and hyperparameters. After sufficient training, the neural network can accurately recognize the known fault types in the training set, preparing for subsequent open set fault recognition.

[0100] The trained neural network usually uses a SoftMax layer as the output layer to map the features to the probability distribution of the fault types. Replacing the SoftMax layer with an open set recognition OpenMax layer can achieve open set fault recognition function, i.e. it can recognize fault types that have not been seen in the training process. The OpenMax layer calculates the distance between the input sample and the known fault types to determine whether the sample belongs to a known fault type. If it does not belong to any known fault type, it will be labeled as an unknown fault type.

[0101] Traditional neural networks typically use a SoftMax layer as the final normalization layer, which transforms the network's output into probability distributions for each class, with the sum of these probabilities being 1. This design performs well in closed-set recognition problems, but has limitations in open-set recognition problems because it always classifies input samples into a known class, even if the sample does not actually belong to any known class. This makes it difficult to perform effective fault identification in the specific scenario of magnetic levitation bearings. To address this issue, this application replaces the SoftMax layer with an OpenMax layer.

[0102] The OpenMax layer works by calculating the distance between an input sample and each known category based on the statistical properties of known categories, and introducing an additional "unknown" category. In its implementation, a statistical model is first built for each known fault category, typically based on the distribution of training samples in the feature space. Then, for a new input sample, its distance or similarity to the statistical models of each known category is calculated. If the distance between a sample and all known categories exceeds a certain threshold, it is identified as the "unknown" category (output 0); otherwise, it is identified as the closest known category. This method effectively distinguishes between known and unknown fault types, improving the accuracy and reliability of the fault diagnosis system in practical applications. Through this modification, we obtain a fault classification model with open-set recognition capabilities, which can accurately identify not only known fault types but also unknown fault types not seen during training.

[0103] Thus, we can obtain the trained fault classification model f1: X→Y′, where Y′∈[0,1,2,3,...,C]. When the output is 0, it indicates that the sample does not belong to any known fault type, that is, it is an unknown fault.

[0104] S103. When the first target fault type is an unknown fault type, a second target fault type is determined from the multiple preset fault types based on the attribute characteristics of the rotor position information; the second target fault type is used to provide a reference for determining the actual type corresponding to the unknown fault type.

[0105] In this step, when the first target fault type is identified as an unknown fault type (i.e., the output is 0), the known fault type most similar to the unknown fault can be found by analyzing the attribute characteristics of the rotor position information, providing a reference for fault cause analysis.

[0106] In some possible implementation, a property feature of the rotor position information can be determined, and a feature distance between the property feature and property features of a plurality of second rotor position information samples can be determined. Then, based on the feature distance and a preset approximation threshold, a target rotor position information sample can be determined from the plurality of second rotor position information samples, and a true value label corresponding to the target rotor position information sample is determined as the second target fault type.

[0107] In the formula, the second rotor position information sample is a sample of a known fault type, and is used for feature comparison with the unknown fault sample. The feature distance represents a distance between two samples in a feature space, and can be measured by using a Euclidean distance, a cosine similarity, or the like. The preset approximation threshold is used for judging whether two samples are similar. If the feature distance is less than the threshold, the two samples are considered to be similar.

[0108] In some possible implementation, the property feature of the rotor position information includes a statistical property feature, and the statistical property feature is determined by the following steps.

[0109] A statistical value of the rotor position information in a plurality of statistical dimensions is determined. For any statistical dimension, a statistical property feature in the statistical dimension is determined based on a statistical value in the statistical dimension and a judgment threshold corresponding to the statistical dimension.

[0110] The statistical property feature can be a statistical feature extracted from the rotor position information, and includes a statistical quantity such as a mean value, a variance, a kurtosis, and a skewness. The plurality of statistical dimensions can include a time domain statistical dimension and a frequency domain statistical dimension. The time domain statistical dimension includes a mean value, a variance, a kurtosis, and a skewness. The frequency domain statistical dimension includes a typical component in a frequency spectrum, such as a 2-fold frequency, a 3-fold frequency, or the like. In one possible implementation, the statistical property feature can include a feature obtained based on a rule such as expert experience. One rule can be regarded as one statistical dimension.

[0111] For each statistical dimension, a statistical property feature in the dimension is determined according to a statistical value in the dimension and a corresponding judgment threshold. For example, if the statistical value in the statistical dimension exceeds the judgment threshold, the feature value in the dimension is 1, otherwise, the feature value in the dimension is 0. In this way, a binary feature vector representing the statistical property feature of the rotor position information can be obtained.

[0112] In the fault diagnosis of magnetic suspension bearings, statistical characteristics are important indicators of system operation status. After obtaining the rotor position information, we need to extract statistical values in multiple statistical dimensions from it. These statistical values can describe the characteristics of rotor position information from different angles. Specifically, statistical dimensions can be divided into two categories: time domain statistical dimensions and frequency domain statistical dimensions. Time domain statistical dimensions mainly focus on the statistical characteristics of rotor position information in time series, including mean, standard deviation, variance, kurtosis, skewness, peak-to-peak value, waveform factor, pulse factor, and margin factor. The mean reflects the average level of position deviation, the standard deviation and variance reflect the fluctuation degree of position deviation, the kurtosis reflects the sharpness of the distribution, and the skewness reflects the asymmetry of the distribution. These indicators are important for identifying different types of faults. For example, bearing looseness faults usually cause an increase in the standard deviation of position deviation, while imbalance faults may cause an increase in the amplitude of position deviation at specific frequencies. Frequency domain statistical dimensions focus on the characteristics of rotor position information in the frequency domain. Typically, the Fourier transform is performed on the time domain signal to obtain the frequency spectrum, and then features such as the amplitude of specific frequency components (e.g., 1x, 2x, etc. of the rotor speed), the energy distribution of the frequency spectrum, and the spectral kurtosis are extracted. These frequency domain features are particularly important for identifying periodic faults (such as imbalance, misalignment, etc.). By calculating these multi-dimensional statistical values, we can fully capture the characteristics of rotor position information and provide strong support for subsequent fault identification.

[0113] After obtaining the statistical values in each statistical dimension, these continuous statistical values can be converted into discrete attribute features to facilitate subsequent feature comparison and fault identification. This conversion process can be based on pre-set judgment thresholds. For each statistical dimension, there is a corresponding judgment threshold, which is usually determined based on a large amount of historical data and expert experience. When the statistical value in this dimension exceeds or is lower than the judgment threshold, it is considered that there is an anomaly in this dimension, and the corresponding feature value is set to 1; otherwise, the feature value is set to 0.

[0114] For example, if the kurtosis value of the pose offset signal exceeds the preset kurtosis threshold, the feature value under the statistical dimension of "kurtosis" is 1, indicating that there is a kurtosis anomaly. Similarly, if the 2x frequency amplitude of the pose offset signal exceeds the preset 2x frequency threshold, the feature value under the statistical dimension of "2x frequency" is also 1, indicating that there is a 2x frequency anomaly. In this way, continuous statistical values can be converted into a binary feature vector, and each element represents whether there is an anomaly under a specific statistical dimension. This binary representation simplifies the feature comparison process while retaining the key information required for fault diagnosis. It should be noted that the setting of the judgment threshold has an important influence on the effect of feature extraction. If the threshold is too high, some fault features may be ignored, and if the threshold is too low, normal fluctuations may be misjudged as fault features. Therefore, in actual application, the judgment thresholds of various statistical dimensions need to be reasonably set according to the specific situation and operating environment of the device to obtain appropriate fault diagnosis effect.

[0115] In some possible implementations, the attribute feature includes a self-learning attribute feature; and the self-learning attribute feature is determined by the following steps:

[0116] The rotor position information is input into the trained feature extraction model to obtain a feature coefficient vector output by the feature extraction model, and the feature coefficient vector is taken as the self-learning attribute feature.

[0117] In this step, deep features of the rotor position information can be extracted in an automatic learning manner. These features can be difficult to identify or define under normal circumstances. To this end, a feature extraction model can be designed, which adopts the structure of an autoencoder and can automatically learn the feature representation of the rotor position information.

[0118] Referring to Figure 2 FIG. 1 is a schematic diagram of a feature extraction model provided by an embodiment of the present application. The model includes an encoder, a feature reconstruction module, and a decoder. The encoder can encode the input rotor position information X into a feature vector Z, which captures the key features of the rotor position information. The feature reconstruction module reconstructs the feature vector using a pre-learned feature matrix A and a feature coefficient vector a to obtain a reconstructed feature vector Z'. The feature matrix A = [A1, A2,..., A M ] contains a series of basic features, and the feature coefficient vector a = [a1, a2,..., a M ] represents the weights of these basic features. The reconstructed feature vector can be represented as z i ′ = a1A1 + a2A2 +... + a M A M , that is, the weighted sum of the basic features. Finally, the decoder decodes the reconstructed feature vector into reconstructed information This reconstructed information is as close to the original input X as possible.

[0119] During the model training process, three optimization objectives are adopted: (1) minimizing the difference between the feature vector Z and the reconstructed feature vector Z', ensuring the accuracy of feature reconstruction; (2) minimizing the L0 norm of the feature coefficient vector a, making the feature representation as sparse as possible and improving the interpretability of the features; (3) minimizing the difference between the original input X and the reconstructed information , ensuring the accuracy of the entire auto-encoding process.

[0120] The feature extraction model trained in this way can map the rotor position information into a sparse feature coefficient vector, which captures the key features of the rotor position information and can be used as a self-learning attribute feature for subsequent fault identification. Compared with statistical features, self-learning features can capture more complex and abstract fault features, improving the accuracy of fault identification.

[0121] In some possible implementations, the feature extraction model can be trained by the following steps:

[0122] An initial feature extraction model is constructed; the initial feature extraction model is used to encode the rotor position information to obtain a first feature vector; the first feature vector is reconstructed using a feature matrix and a feature coefficient vector to obtain a second feature vector; the second feature vector is decoded to obtain reconstructed information corresponding to the rotor position information; based on a first loss function, a second loss function and a third loss function, the initial feature extraction model is trained to obtain a trained feature extraction model; the trained feature extraction model has a feature matrix that makes the trained feature extraction model satisfy the first loss function, the second loss function and the third loss function; the first loss function is used to minimize the difference between the first feature vector and the second feature vector; the second loss function is used to minimize the L0 norm of the feature coefficient vector; and the third loss function is used to minimize the difference between the rotor position information and the reconstructed information.

[0123] The initial feature extraction model described above can be an auto-encoder structure, including an encoder and a decoder. The encoder encodes the rotor position information into a feature vector, and the decoder decodes the feature vector into reconstructed information of the rotor position information. A feature reconstruction module is located between the encoder and the decoder, and is used to reconstruct the feature vector using the feature matrix and the feature coefficient vector.

[0124] During the training process, the feature extraction model can be trained by optimizing three loss functions to achieve the above three optimization objectives: the first loss function is used to minimize the difference between the first feature vector (the feature vector output by the encoder) and the second feature vector (the reconstructed feature vector), to ensure the accuracy of feature reconstruction; the second loss function is used to minimize the L0 norm of the feature coefficient vector, so that the feature coefficient vector is as sparse as possible, and the interpretability of the feature is improved; the third loss function is used to minimize the difference between the rotor position information and the reconstructed information, to ensure the reconstruction accuracy of the entire autoencoder.

[0125] wherein the first loss function can be represented as min||z′ i -z i ||, the second loss function can be represented as min||a||0, and the third loss function can be represented as

[0126] The feature extraction model trained in this way can automatically learn the features of the rotor position information and output a sparse feature coefficient vector as a self-learning attribute feature.

[0127] In some possible implementations, when the attribute feature includes a statistical attribute feature and a self-learning attribute feature, the feature distance can be determined by the following steps:

[0128] For any second rotor position information sample, determine the first difference information between the self-learning attribute feature in the rotor position information and the self-learning attribute feature in the attribute feature of the second rotor position information sample, and determine the second difference information between the statistical attribute feature in the rotor position information and the statistical attribute feature in the attribute feature of the second rotor position information sample; based on the first difference information, the second difference information, and the weight corresponding to the second difference information, determine the feature distance between the attribute feature of the rotor position information and the attribute feature of the second rotor position information sample.

[0129] wherein the first difference information represents the difference between the two samples in the self-learning feature space, which can be calculated using the Euclidean distance, cosine similarity, etc. The second difference information represents the difference between the two samples in the statistical feature space, and since the statistical feature is a binary vector, the Hamming distance can be used to calculate it.

[0130] The weight is used to balance the importance of the self-learning feature and the statistical feature in the calculation of the feature distance. By adjusting the weight, the contribution proportion of the two features in the calculation of the feature distance can be controlled. The feature distance can be represented as: feature distance = first difference information + weight * second difference information.

[0131] For example, the feature distance can be represented as di is a feature distance between the rotor position information and the i-th second rotor position information sample; the first difference information can be represented as ‖a i -a test ‖, where a i is a self-learning attribute feature in the attribute features of the i-th second rotor position information sample, and a test is a self-learning attribute feature in the rotor position information; the second difference information can be represented as ‖t i -t test ‖, where t i is a statistical attribute feature in the attribute features of the i-th second rotor position information sample, and t test is a statistical attribute feature in the rotor position information.

[0132] Those skilled in the art can understand that, in the above method of the specific implementation, the writing order of each step does not mean a strict execution order and does not constitute any limitation on the implementation process. The specific execution order of each step should be determined according to its function and possible internal logic.

[0133] Corresponding to the above-mentioned embodiments of the device fault determination method, the present application also provides embodiments of a device fault determination apparatus.

[0134] The embodiments of the device fault determination apparatus of the present application can be applied to a device equipped with a magnetic bearing. The apparatus embodiments can be realized by software, or by hardware or a combination of software and hardware. Taking software realization as an example, as a logically meaningful apparatus, it is formed by reading the corresponding computer program instructions in the non-volatile memory into the memory for running by the processor of the device in which it is located.

[0135] Referring to Figure 3 , a schematic diagram of a device fault determination apparatus provided by an embodiment of the present application is shown.

[0136] The apparatus includes:

[0137] The acquisition module 310 is configured to acquire rotor position information of a magnetic bearing in a target device within a target time period;

[0138] The first identification module 320 is configured to perform open set fault identification based on the rotor position information, and determine a first target fault type matching the rotor position information from a plurality of fault types; the plurality of fault types include a plurality of preset fault types, and an unknown fault type representing a type other than the preset fault types;

[0139] The second identification module 330 is configured to, when the first target fault type is an unknown fault type, determine a second target fault type from the plurality of preset fault types based on an attribute feature of the rotor position information, and the second target fault type is used to provide a reference for determining an actual type corresponding to the unknown fault type.

[0140] Optionally, the first identification module 320 is specifically configured to:

[0141] input the rotor position information into a trained fault classification model to obtain a first target fault type output by the fault classification model, and the fault classification model is a neural network with an OpenMax layer.

[0142] Optionally, the apparatus further includes a first training module 340 configured to:

[0143] construct an initial neural network;

[0144] train the initial neural network by using a first rotor position information sample and a true value label corresponding to the first rotor position information sample to obtain a trained neural network;

[0145] replace a normalization layer of the trained neural network with the OpenMax layer to obtain the fault classification model.

[0146] Optionally, the second identification module 330 is specifically configured to:

[0147] determine a feature distance between the attribute feature of the rotor position information and an attribute feature of each of a plurality of second rotor position information samples;

[0148] determine a target rotor position information sample from the plurality of second rotor position information samples based on the feature distance and a preset approximate threshold;

[0149] determine a true value label corresponding to the target rotor position information sample as the second target fault type.

[0150] Optionally, the attribute feature of the rotor position information includes a statistical attribute feature, and the second identification module 330 is specifically configured to:

[0151] determine a statistical value of the rotor position information in a plurality of statistical dimensions;

[0152] for any statistical dimension, determine a statistical attribute feature in the statistical dimension based on a statistical value in the statistical dimension and a judgment threshold corresponding to the statistical dimension.

[0153] Optionally, the attribute feature comprises a self-learning attribute feature; and the second identification module 330 is specifically configured to:

[0154] inputting the rotor position information into the trained feature extraction model to obtain a feature coefficient vector output by the feature extraction model, and taking the feature coefficient vector as the self-learning attribute feature.

[0155] Optionally, the device further comprises a second training module 350 configured to:

[0156] construct an initial feature extraction model, wherein the initial feature extraction model is configured to encode the rotor position information to obtain a first feature vector, reconstruct the first feature vector by using a feature matrix and a feature coefficient vector to obtain a second feature vector, and decode the second feature vector to obtain reconstruction information corresponding to the rotor position information;

[0157] train the initial feature extraction model based on a first loss function, a second loss function and a third loss function to obtain a trained feature extraction model;

[0158] the trained feature extraction model has a feature matrix that makes the trained feature extraction model satisfy the first loss function, the second loss function and the third loss function;

[0159] the first loss function is configured to minimize the difference between the first feature vector and the second feature vector; the second loss function is configured to minimize the L0 norm of the feature coefficient vector; and the third loss function is configured to minimize the difference between the rotor position information and the reconstruction information.

[0160] Optionally, in the case where the attribute feature comprises a statistical attribute feature and a self-learning attribute feature, the second identification module 330 is specifically configured to:

[0161] for any second rotor position information sample, determine first difference information between the self-learning attribute feature in the rotor position information and the self-learning attribute feature in the attribute feature of the second rotor position information sample, and determine second difference information between the statistical attribute feature in the rotor position information and the statistical attribute feature in the attribute feature of the second rotor position information sample;

[0162] determine a feature distance between the attribute feature of the rotor position information and the attribute feature of the second rotor position information sample based on the first difference information, the second difference information and a weight corresponding to the second difference information.

[0163] The implementation process of the functions and roles of each unit in the above device is specifically described in the implementation process of the corresponding steps in the above method, which will not be repeated here.

[0164] The embodiment of the present application also provides a computer device, as shown in the figure, which is a structural schematic diagram of the computer device provided by the embodiment of the present application, comprising: Figure 4 The embodiment of the present application also provides a computer device, as shown in the figure, which is a structural schematic diagram of the computer device provided by the embodiment of the present application, comprising:

[0165] The processor 41 and the memory 42; the memory 42 stores machine readable instructions executable by the processor 41, and the processor 41 is used for executing the machine readable instructions stored in the memory 42, and when the machine readable instructions are executed by the processor 41, the processor 41 executes the following steps:

[0166] Obtaining rotor position information of the magnetic suspension bearing in the target device in a target time period;

[0167] Based on the rotor position information, open set fault identification is performed, and a first target fault type matching the rotor position information is determined from a plurality of fault types; the plurality of fault types include a plurality of preset fault types and an unknown fault type representing a type other than the preset fault type;

[0168] In the case that the first target fault type is the unknown fault type, a second target fault type is determined from the plurality of preset fault types based on an attribute feature of the rotor position information; the second target fault type is used to provide a reference for determining an actual type corresponding to the unknown fault type.

[0169] The above memory 42 includes a memory 421 and an external memory 422; the memory 421 here is also called an internal memory, and is used for temporarily storing operation data in the processor 41 and exchanging data with the external memory 422 such as a hard disk, and the processor 41 exchanges data with the external memory 422 through the memory 421.

[0170] The specific execution process of the above instructions can refer to the steps of the device fault determination method described in the embodiment of the present application, which will not be repeated here.

[0171] For the device embodiment, since it basically corresponds to the method embodiment, the related parts can be referred to the part of the method embodiment. The device embodiment described above is only schematic, and the units described as separate components can or can not be physically separated, and the components displayed as units can or can not be physical units, that is, they can be located in one place, or distributed on multiple network units. According to actual needs, part or all of the modules can be selected to achieve the purpose of the present application. Those skilled in the art can understand and implement without creative labor.

[0172] The embodiment of the present application further provides a computer readable storage medium, and the computer readable storage medium stores a computer program. When the computer program is executed by a processor, the steps of the device fault determination method described in the method embodiment are executed. The storage medium can be a volatile or non-volatile computer readable storage medium.

[0173] The embodiment of the present application further provides a computer program product, which includes computer programs / instructions, and when the computer programs / instructions are executed by a processor, the device fault determination method provided by the embodiments of the present application is implemented.

[0174] The computer program product can be specifically implemented by means of hardware, software or a combination thereof. In an optional embodiment, the computer program product is specifically embodied as a computer storage medium, and in another optional embodiment, the computer program product is specifically embodied as a software product, such as a software development kit (SDK) and the like.

[0175] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the system and device described above can refer to the corresponding process in the foregoing method embodiments, and will not be described here. In the several embodiments provided by the present application, it can be understood that the disclosed system, device and method can be implemented by other ways. The device embodiments described above are only schematic, for example, the division of the units is only a logical function division, and there can be another division manner in actual implementation, for example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the units shown or discussed can be indirect coupling or communication connection through some communication interfaces, devices or units, and can be electrical, mechanical or other forms.

[0176] The units described as separate components can or can not be physically separate, and the components shown as units can or can not be physical units, that is, they can be located in one place, or can be distributed on a plurality of network units. According to actual needs, part or all of the units can be selected to achieve the purpose of the embodiment.

[0177] In addition, each functional unit in each embodiment of the present application can be integrated in one processing unit, or each unit can exist physically, or two or more units can be integrated in one unit.

[0178] If the functions are implemented in the form of software function units and sold or used as independent products, they can be stored in a nonvolatile computer readable storage medium executable by a processor. Based on this understanding, the technical solutions of the present application essentially or the parts that contribute to the prior art or parts of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a number of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (Read-Only Memory, ROM), a random access memory (Random Access Memory, RAM), a magnetic disk or an optical disk, and various media that can store program codes.

[0179] Finally, it should be noted that: the above-described embodiments are merely specific embodiments of the present application, used to illustrate the technical solutions of the present application, and not to limit them. The protection scope of the present application is not limited thereto, although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art within the technical range disclosed by the present application can still modify or easily think of changes to the technical solutions described in the foregoing embodiments, or make equivalent replacements to some technical features; and these modifications, changes or replacements do not make the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

[0180] The above-described is only the preferred embodiment of the present application, and does not limit the present application. Any modification, equivalent replacement, improvement, etc. within the spirit and principles of the present application should be included in the protection scope of the present application.

Claims

1. A method of determining a device failure, the method comprising: The method comprises: acquiring rotor position information of a magnetic suspension bearing in a target device in a target time period; performing open-set fault recognition based on the rotor position information to determine a first target fault type matching the rotor position information from a plurality of fault types; the plurality of fault types comprise a plurality of preset fault types and an unknown fault type representing a type other than the preset fault types; in a case where the first target fault type is the unknown fault type, determining a second target fault type from the plurality of preset fault types based on an attribute feature of the rotor position information; the second target fault type is used to provide a reference for determining an actual type corresponding to the unknown fault type.

2. The method of claim 1, wherein, The open-set fault recognition based on the rotor position information to determine a first target fault type matching the rotor position information from a plurality of fault types comprises: inputting the rotor position information into a trained fault classification model to obtain a first target fault type output by the fault classification model; the fault classification model is a neural network with an OpenMax layer.

3. The method of claim 2, wherein, The fault classification model is determined by the following steps: constructing an initial neural network; training the initial neural network using first rotor position information samples and true value labels corresponding to the first rotor position information samples to obtain a trained neural network; replacing a normalization layer of the trained neural network with the OpenMax layer to obtain the fault classification model.

4. The method of claim 1, wherein, The second target fault type determined from the plurality of preset fault types based on the attribute feature of the rotor position information comprises: determining a feature distance between the attribute feature of the rotor position information and attribute features of a plurality of second rotor position information samples; determining a target rotor position information sample from the plurality of second rotor position information samples based on the feature distance and a preset approximation threshold; determining a true value label corresponding to the target rotor position information sample as the second target fault type.

5. The method of claim 4, wherein, The attribute feature of the rotor position information comprises a statistical attribute feature; The statistical attribute feature is determined by the following steps: determining statistical values of the rotor position information in a plurality of statistical dimensions; for any statistical dimension, determining a statistical attribute feature in the statistical dimension based on a statistical value in the statistical dimension and a judgment threshold corresponding to the statistical dimension.

6. The method according to claim 4 or 5, characterized in that, The attribute feature comprises a self-learning attribute feature; The self-learning attribute feature is determined by the following steps: inputting the rotor position information into a trained feature extraction model to obtain a feature coefficient vector output by the feature extraction model, and taking the feature coefficient vector as the self-learning attribute feature.

7. The method of claim 6, wherein, The feature extraction model is trained by the following steps: constructing an initial feature extraction model; the initial feature extraction model is used to encode rotor position information to obtain a first feature vector; a second feature vector is obtained by reconstructing the first feature vector using a feature matrix and a feature coefficient vector; the second feature vector is decoded to obtain reconstruction information corresponding to the rotor position information; The initial feature extraction model is trained based on the first loss function, the second loss function, and the third loss function to obtain a trained feature extraction model; The trained feature extraction model has a feature matrix that makes the trained feature extraction model satisfy the first loss function, the second loss function, and the third loss function; The first loss function is used to minimize the difference between the first feature vector and the second feature vector; the second loss function is used to minimize the L0 norm of the feature coefficient vector; and the third loss function is used to minimize the difference between the rotor position information and the reconstructed information.

8. The method of claim 7, wherein, In the case where the attribute features include statistical attribute features and self-learning attribute features, the feature distance between the attribute features of the rotor position information and the attribute features of a plurality of second rotor position information samples is determined as follows: For any second rotor position information sample, first difference information between the self-learning attribute features in the rotor position information and the self-learning attribute features in the attribute features of the second rotor position information sample is determined, and second difference information between the statistical attribute features in the rotor position information and the statistical attribute features in the attribute features of the second rotor position information sample is determined; Based on the first difference information, the second difference information, and the weight corresponding to the second difference information, the feature distance between the attribute features of the rotor position information and the attribute features of the second rotor position information sample is determined.

9. An apparatus failure determination device characterized by comprising: The device comprises: An acquisition module configured to acquire rotor position information of a magnetic bearing in a target device within a target time period; A first identification module configured to perform open set fault identification based on the rotor position information, and determine a first target fault type matching the rotor position information from a plurality of fault types; the plurality of fault types include a plurality of preset fault types and an unknown fault type representing a type other than the preset fault types; A second identification module configured to, in the case where the first target fault type is the unknown fault type, determine a second target fault type from the plurality of preset fault types based on attribute features of the rotor position information; the second target fault type is used to provide a reference for determining an actual type corresponding to the unknown fault type.

10. A computer-readable storage medium having stored thereon a computer program, characterized in that, The program is executed by the processor to implement the steps of the method of any one of claims 1 to 8.

11. A computer device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor executes the program to implement the steps of the method of any one of claims 1 to 8.

Images