A rotating equipment cross-working condition open set fault diagnosis method and device

Abstract

The application relates to the technical field of rotating equipment fault diagnosis, and discloses a rotating equipment cross-working-condition open-set fault diagnosis method and device, which comprises the following steps: constructing a source domain data set and a target domain data set; wherein the source domain data set contains rotating equipment vibration data with a fault state label under a first working condition, and the target domain data set contains rotating equipment vibration data under a second working condition; training a deep neural network by using the source domain data set and the target domain data set to obtain a trained deep neural network; the deep neural network comprises a feature extractor, a probabilistic neural network, a domain discriminator and an unknown class detector; and performing open-set classification on the target domain data set by using the trained deep neural network to obtain a rotating equipment cross-working-condition open-set fault class. The application realizes accurate identification of rotating equipment fault classes under different working conditions, and improves the accuracy of rotating equipment fault diagnosis.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of rotating equipment fault diagnosis, and particularly relates to a rotating equipment cross-condition open set fault diagnosis method and device. BACKGROUND

[0002] Rotating equipment is the core equipment of intelligent industry. Once it fails, it will affect production, cause unplanned maintenance and economic losses, or even cause personal injury accidents. The fault diagnosis technology of rotating equipment is the core problem of the state monitoring system.

[0003] Deep learning and vibration signals are used for fault diagnosis of rotating equipment. However, the rotating equipment is in different operating conditions for a long time, and the data distribution difference of vibration signals under different conditions increases the difficulty of fault recognition, and lacks the ability to identify fault categories, resulting in low accuracy of fault diagnosis of rotating equipment. SUMMARY

[0004] Therefore, the present application provides a rotating equipment cross-condition open set fault diagnosis method and device to solve the problems of high difficulty in fault recognition of rotating equipment under different conditions and lack of ability to identify fault categories.

[0005] In a first aspect, the present application provides a rotating equipment cross-condition open set fault diagnosis method, which comprises:

[0006] Constructing a source domain data set and a target domain data set; wherein the source domain data set contains rotating equipment vibration data with fault state labels under a first operating condition, and the target domain data set includes rotating equipment vibration data under a second operating condition;

[0007] Training a deep neural network using the source domain data set and the target domain data set to obtain a trained deep neural network; the deep neural network includes a feature extractor, a probabilistic neural network, a domain discriminator and an unknown class detector;

[0008] Using the trained deep neural network to perform open set classification on the target domain data set to obtain a rotating equipment cross-condition open set fault category.

[0009] The embodiment provides a rotating equipment cross-working condition open set fault diagnosis method, the deep neural network is trained through the rotating equipment vibration data with a fault state label and the rotating equipment vibration data without a fault state label under different working conditions, then the deep neural network after training is used for open set classification of a target domain data set, a rotating equipment cross-working condition open set fault class is obtained, the knowledge related to the fault is transferred from one working condition to another working condition in a transfer learning mode, and a fault class sample that does not exist in the other working condition exists, accurate identification of the rotating equipment fault class under different working conditions is realized, and the accuracy of the rotating equipment fault diagnosis is improved.

[0010] In an optional implementation, the deep neural network is trained by using the source domain data set and the target domain data set, and the deep neural network after training is obtained, including:

[0011] Based on the source domain data set, a classification training optimization objective function of the source domain data is constructed by using the feature extractor and the probabilistic neural network;

[0012] Based on the source domain data set and the target domain data set, a weighted domain adversarial training optimization objective function is constructed by using the feature extractor and the domain discriminator;

[0013] A weighted classification training optimization objective function is constructed by using the probabilistic neural network and the unknown class detector;

[0014] The total optimization objective function is constructed based on the classification training optimization objective function of the source domain data, the weighted domain adversarial training optimization objective function and the weighted classification training optimization objective function;

[0015] The deep neural network is trained by using the total optimization objective function until the deep neural network converges, and the deep neural network after training is obtained.

[0016] The rotating equipment cross-working condition open set fault diagnosis method provided by the embodiment is based on the source domain data set and the target domain data set, a total optimization objective function is constructed by using a feature extractor, a probabilistic neural network, a domain discriminator and an unknown class detector, and the deep neural network is trained by using the total optimization objective function, so that the training process of the deep neural network aims to accurately identify the fault class samples under different working conditions, and it is ensured that the deep neural network after training can accurately identify and accurately diagnose the rotating equipment fault class under different working conditions.

[0017] In an optional implementation, based on the source domain data set, a classification training optimization objective function of the source domain data is constructed by using the feature extractor and the probabilistic neural network, including:

[0018] mapping, by the feature extractor, the rotating equipment vibration data with the fault state label in the first operating condition in the source domain data set to a feature space to obtain source domain data features;

[0019] Based on the source domain data features, the probability neural network is used to obtain the probability distribution of the source domain data features;

[0020] Based on the probability distribution of the source domain data features, the classification loss of the source domain data set is calculated, and the classification training optimization objective function of the source domain data is constructed based on the classification loss of the source domain data set.

[0021] The rotating equipment cross-condition open set fault diagnosis method provided in this embodiment introduces a probability neural network to participate in the training of a deep neural network. The probability neural network is used to learn the probability distribution of the known class features in the source domain data set, and the classification loss of the source domain data set is calculated based on the probability distribution of the source domain data features. Then, the classification training optimization objective function of the source domain data is constructed based on the classification loss of the source domain data set, thereby realizing accurate classification of the source domain data.

[0022] In an optional implementation, based on the source domain data set and the target domain data set, a weighted domain adversarial training optimization objective function is constructed using the feature extractor and the domain discriminator, including:

[0023] mapping, by the feature extractor, the rotating equipment vibration data with the fault state label in the first operating condition in the source domain data set to a feature space to obtain source domain data features;

[0024] Based on the target domain data features, the probability neural network is used to obtain a plurality of probability likelihood values, and the maximum probability likelihood value is taken as a similarity indication value;

[0025] The similarity indication value is normalized to obtain a similarity value;

[0026] Based on the target domain data features and the similarity value, the domain discriminator is used to obtain a weighted domain adversarial training loss value;

[0027] The weighted domain adversarial training optimization objective function is constructed based on the weighted domain adversarial training loss value.

[0028] The rotating equipment cross-condition open set fault diagnosis method provided in this embodiment uses the output of the probability neural network to determine the similarity of the target domain sample, i.e., the similarity value. The strong feature distribution fitting capability of the probability neural network can more accurately calculate the similarity of the target domain sample, thereby improving the classification accuracy of the target domain sample.

[0029] In an optional implementation, a weighted classification training optimization objective function is constructed using the probability neural network and the unknown class detector, including:

[0030] Based on the source domain data characteristics, the unknown class detector is used to obtain a class classification loss value;

[0031] The target domain data characteristics are sorted based on the similarity values, and the class characteristics are divided based on the sorting results, and a weighted classification loss value is calculated based on the class characteristic division results;

[0032] The total classification loss value is calculated based on the class classification loss value and the weighted classification loss value, and a weighted classification training optimization objective function is constructed based on the total classification loss value.

[0033] The unknown class detector is used to identify known classes and unknown classes, the target domain data characteristics are sorted based on the similarity values, and then the class characteristics are divided based on the sorting results. In the process of deep neural network training, the source domain data and part of the target domain data are divided into known classes, part of the target domain data is divided into unknown classes, and the similarity values are used in the training of the unknown class detector, so that the training method is more reasonable, and the accuracy of the fault diagnosis of the rotating equipment is further improved.

[0034] In an optional implementation, the trained deep neural network is used to perform open set classification on the target domain data set to obtain a rotating equipment cross-condition open set fault class, including:

[0035] Based on the target domain samples in the target domain data set, the target domain sample characteristics are obtained through the feature extractor;

[0036] Based on the target domain sample characteristics, the output likelihood probability is obtained through the probabilistic neural network;

[0037] Based on the target domain sample characteristics, the unknown class detector is used to determine whether the target domain sample is a known class;

[0038] If the target domain sample is a known class, the output likelihood probability is used to determine the rotating equipment cross-condition open set fault class.

[0039] The rotating equipment cross-condition open set fault diagnosis method provided in this embodiment uses the unknown class detector to determine whether the target domain sample is a known class based on the target domain sample characteristics, without setting an empirical threshold value. The unknown class detector can directly output whether the sample is an unknown class, and the testing process is more direct and efficient. When the target domain sample is a known class, the output likelihood probability is used to determine the rotating equipment cross-condition open set fault class, thereby achieving accurate diagnosis of the rotating equipment cross-condition open set fault.

[0040] In a second aspect, the present application provides a rotating equipment cross-condition open set fault diagnosis device, which comprises:

[0041] The constructing module is configured to construct a source domain data set and a target domain data set, wherein the source domain data set contains rotating equipment vibration data with a fault state label under a first operating condition, and the target domain data set includes rotating equipment vibration data under a second operating condition;

[0042] The training module is configured to train the deep neural network by using the source domain data set and the target domain data set, so as to obtain a trained deep neural network; the deep neural network includes a feature extractor, a probabilistic neural network, a domain discriminator and an unknown class detector.

[0043] The classification module is configured to perform open set classification on the target domain data set by using the trained deep neural network, so as to obtain a rotating equipment cross-condition open set fault class.

[0044] In a third aspect, the present application provides a computer device, comprising a memory and a processor, the memory and the processor are communicatively connected with each other, the memory stores computer instructions, and the processor executes the computer instructions to perform the rotating equipment cross-condition open set fault diagnosis method of the first aspect or any of the corresponding embodiments thereof.

[0045] In a fourth aspect, the present application provides a computer readable storage medium, which stores computer instructions, and the computer instructions are used to make a computer execute the rotating equipment cross-condition open set fault diagnosis method of the first aspect or any of the corresponding embodiments thereof.

[0046] In a fifth aspect, the present application provides a computer program product, which comprises computer instructions, and the computer instructions are used to make a computer execute the rotating equipment cross-condition open set fault diagnosis method of the first aspect or any of the corresponding embodiments thereof. BRIEF DESCRIPTION OF DRAWINGS

[0047] In order to more clearly illustrate the specific embodiments of the present application or the technical solutions in the prior art, the drawings needed in the specific embodiments or the prior art description will be briefly introduced as follows. Obviously, the drawings in the following description are some embodiments of the present application, and those skilled in the art can also obtain other drawings according to these drawings without creative labor.

[0048] Figure 1 is a flowchart of a rotating equipment cross-condition open set fault diagnosis method according to an embodiment of the present application;

[0049] Figure 2 is a flowchart of another rotating equipment cross-condition open set fault diagnosis method according to an embodiment of the present application;

[0050] Figure 3is a flowchart of a deep neural network training process according to an embodiment of the present application;

[0051] Figure 4 is a flowchart of another rotating equipment cross-condition open set fault diagnosis method according to an embodiment of the present application;

[0052] Figure 5 is a flowchart of rotating equipment cross-condition open set fault class determination according to an embodiment of the present application;

[0053] Figure 6 is a structural block diagram of a rotating equipment cross-condition open set fault diagnosis device according to an embodiment of the present application;

[0054] Figure 7 is a hardware structure diagram of a computer device according to an embodiment of the present application. DETAILED DESCRIPTION

[0055] In order to make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0056] Vibration is an inherent characteristic of rotating equipment, and vibration signals contain a large amount of equipment operation state information. Analysis of the vibration signals of rotating equipment can identify the health status of the equipment. Deep learning technology can automatically extract fault-related features without manual feature engineering, and is a promising tool for fault diagnosis.

[0057] However, the rotating equipment fault diagnosis technology based on deep learning and vibration signals faces two major challenges. One is cross-condition fault diagnosis. Rotating equipment is in different operating conditions for a long time, such as different rotating speeds and different loads. The data distribution of vibration signals under different conditions (i.e. domain difference) is very different, which greatly increases the difficulty of fault identification of deep learning models. The other challenge is open set fault diagnosis. Training data cannot exhaust all fault categories, and deployed models often encounter fault categories that do not exist in the training phase, so the model should have both known class classification and unknown class identification functions, i.e. open set recognition ability. Training data and test data are collected under different conditions, and there are a large number of fault categories in the test data that do not exist in the training data. The goal of open set domain adaptation technology is to classify known categories in test data and identify unknown categories, which can well solve the above two challenges.

[0058] For the open set recognition technology without considering the domain influence, an edge probability distribution of data features can be fitted by using a flow model in the probabilistic neural network, and an unknown category sample is recognized by using an output likelihood probability of the flow model; or a variational autoencoder in the probabilistic neural network is improved into a conditional variational autoencoder to learn a conditional distribution of data features, and an unknown category sample is also recognized by using an output likelihood probability of the model.

[0059] The embodiment of the present application provides a rotating equipment cross-condition open set fault diagnosis method, which is applied to a mobile terminal, introduces a probabilistic neural network into open set domain adaptive fault diagnosis, proposes a unified probabilistic neural network training paradigm for learning a conditional distribution of data features, considers cross-domain fault diagnosis, proposes a similarity calculation method of target domain samples based on an output likelihood probability of the probabilistic neural network, adds an unknown category detector for directly identifying an abnormal category, and adopts a new method for training, so that the classification process no longer needs to depend on a traditional manual threshold setting method.

[0060] According to the embodiment of the present application, a rotating equipment cross-condition open set fault diagnosis method embodiment is provided, and it should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a group of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that here.

[0061] In the embodiment, a rotating equipment cross-condition open set fault diagnosis method is provided, which can be used in the mobile terminal described above, such as a fault diagnosis equipment connected with the rotating equipment, Figure 1 is a flowchart of a rotating equipment cross-condition open set fault diagnosis method according to the embodiment of the present application, as shown in Figure 1 The flowchart includes the following steps:

[0062] Step S101, constructing a source domain data set and a target domain data set; wherein the source domain data set contains rotating equipment vibration data with a fault state label under a first running condition, and the target domain data set includes rotating equipment vibration data under a second running condition.

[0063] Specifically, the rotating equipment vibration data of one or more vibration sensors on the rotating equipment under the first running condition is collected, and then the rotating equipment vibration data is preprocessed, and a fault state label is manually labeled, which is denoted as a source domain data set, and the source domain data set can be expressed as:

[0064]

[0065] Wherein, represents the preprocessed source domain sample data (i.e. the rotating equipment vibration data), represents the sample corresponding fault status label, n s denotes the number of source domain dataset samples, wherein the label Y s denotes the source domain label set, a total of |Y s | categories.

[0066] Further, the vibration data of the rotating equipment under another operating condition (i.e., the second operating condition) is collected and preprocessed, and is denoted as a target domain dataset, which can be expressed as:

[0067]

[0068] wherein, denotes the preprocessed target domain sample data, n t denotes the number of target domain dataset samples, sample corresponding unknown label It is assumed that the target domain label set is Y t , a total of |Y t | categories, that is, in addition to the known categories of the source domain, there are unknown categories in the target domain.

[0069] Further, the preprocessing method of the vibration data of the rotating equipment includes but is not limited to denoising methods such as low-pass filtering and wavelet transform, feature extraction methods such as fast Fourier transform and variational modal decomposition, data enhancement methods such as multi-channel data stacking and one-dimensional vibration to two-dimensional image, etc.

[0070] Step S102, using the source domain dataset and the target domain dataset to train the deep neural network to obtain the trained deep neural network; the deep neural network includes a feature extractor, a probabilistic neural network, a domain discriminator and an unknown class detector.

[0071] Specifically, the deep neural network includes a feature extractor E, |Y s | probabilistic neural networks a domain discriminator D, and an unknown class detector O; wherein the feature extractor maps the sample to a feature space; the probabilistic neural network is used to learn the probability distribution of each category feature; the domain discriminator is used for weighted adversarial training, aligning the known categories of the source domain and the target domain in the feature space, and the output dimension is 2, which is a probability vector of the sample belonging to the source domain and the target domain; the unknown class detector is used to classify the known categories and the unknown categories, and the output dimension is 2, which is a probability vector of the sample belonging to the known categories and the unknown categories.

[0072] Further, the probabilistic neural network includes but is not limited to variational autoencoder, probabilistic flow model and other neural networks that can be used for data probability density estimation.

[0073] Step S103: Use the trained deep neural network to perform open set classification on the target domain data set to obtain open set fault categories of rotating equipment across working conditions.

[0074] Specifically, the task of cross-operation open set fault diagnosis is to use the source domain dataset D s For the target domain dataset D t The samples can be classified correctly. t Known faults in and identified D t Unknown fault in .

[0075] Furthermore, corresponding maintenance actions are performed based on the open-set fault diagnosis output of the rotating equipment by the trained deep neural network; when the trained deep neural network determines that the rotating equipment is in a normal state, the maintenance strategy is not executed; when the trained deep neural network determines that the rotating equipment is a known fault, the corresponding maintenance strategy is executed; when the network determines that the rotating equipment is an unknown fault, the subsequent maintenance strategy should be comprehensively determined based on other data.

[0076] This embodiment provides a method for cross-operating-condition open-set fault diagnosis of rotating equipment. A deep neural network is trained using vibration data of rotating equipment with fault status labels and vibration data of rotating equipment without fault status labels under different operating conditions. The trained deep neural network is then used to perform open-set classification on a target domain data set to obtain cross-operating-condition open-set fault categories of rotating equipment. Transfer learning is used to transfer fault-related knowledge from one operating condition to another, and there are fault category samples that have never appeared in the other operating condition. This achieves accurate identification of rotating equipment fault categories under different operating conditions and improves the accuracy of fault diagnosis of rotating equipment.

[0077] In this embodiment, a method for cross-operating-condition open-set fault diagnosis of rotating equipment is provided, which can be used in the above-mentioned mobile terminal, such as a fault diagnosis device connected to the rotating equipment. Figure 2 FIG. 1 is a flow chart of a method for diagnosing faults of rotating equipment across operating conditions according to an embodiment of the present invention. Figure 2 As shown, the process includes the following steps:

[0078] Step S201: construct a source domain dataset and a target domain dataset; wherein the source domain dataset includes vibration data of rotating equipment with a fault state label under a first operating condition, and the target domain dataset includes vibration data of rotating equipment under a second operating condition. Figure 1 Step S101 of the illustrated embodiment will not be described in detail here.

[0079] Step S202, training the deep neural network using the source domain dataset and the target domain dataset to obtain a trained deep neural network; the deep neural network comprises a feature extractor, a probabilistic neural network, a domain discriminator, and an unknown class detector.

[0080] Specifically, as shown in Figure 3 the above step S202 comprises:

[0081] Step S2021, based on the source domain dataset, constructing a classification training optimization objective function of the source domain data using the feature extractor and the probabilistic neural network.

[0082] In some optional embodiments, the above step S2021 comprises:

[0083] Step a1, mapping the rotating equipment vibration data in the source domain dataset under the first operating condition with the fault state label to the feature space by the feature extractor to obtain source domain data features.

[0084] Specifically, for the source domain data sample The source domain data features extracted by the feature extractor E are:

[0085]

[0086] Step a2, based on the source domain data features, obtaining the probability distribution of the source domain data features using the probabilistic neural network.

[0087] Specifically, the source domain data features are input into the kth probabilistic neural network V k , and the log-likelihood probability is output as:

[0088]

[0089] where P Z represents the probability density function of the random variable Z.

[0090] Further, the method of calculating the log-likelihood probability by the kth probabilistic neural network is determined by the type of the selected probabilistic neural network, such as the variational autoencoder which calculates the log-likelihood probability by the encoding-decoding architecture and the variational model, and the flow model which calculates the log-likelihood probability by the Jacobian determinant derived by the reversible neural network.

[0091] Step a3, calculating the classification loss of the source domain dataset based on the probability distribution of the source domain data features, and constructing a classification training optimization objective function of the source domain data based on the classification loss of the source domain dataset.

[0092] Specifically, the classification loss Lclf The calculation formula is as follows:

[0093]

[0094] Among them, L ce Represents the cross entropy loss function with SoftMax function (normalized exponential function), γ and λ are hyperparameters, the first term Can ensure that the network is reliable for source domain data samples The classification is correct, the second Can guarantee the Probabilistic Neural Network Fully learned features The probability distribution of .

[0095] Furthermore, the expression of the classification training optimization objective function of source domain data is as follows:

[0096]

[0097] This embodiment provides a cross-operating condition open-set fault diagnosis method for rotating equipment, which introduces a probabilistic neural network to participate in the training of a deep neural network. The probabilistic neural network is used to learn the probability distribution of known category features in a source domain data set, and calculates the classification loss of the source domain data set based on the probability distribution of the source domain data features. Then, based on the classification loss of the source domain data set, a classification training optimization objective function of the source domain data is constructed, thereby achieving accurate classification of the source domain data.

[0098] Step S2022: Based on the source domain dataset and the target domain dataset, a feature extractor and a domain discriminator are used to construct a weighted domain adversarial training optimization objective function.

[0099] In some optional implementations, the above step S2022 includes:

[0100] Step b1: Mapping the vibration data of the rotating equipment under the second operating condition in the target domain data set to the feature space through a feature extractor to obtain target domain data features.

[0101] Specifically, for the target domain data sample The target domain data features extracted by feature extractor E for:

[0102]

[0103] In step b2, based on the target domain data characteristics, a probabilistic neural network is used to obtain multiple probability likelihood values, and the maximum probability likelihood value is used as the similarity indicator value.

[0104] Specifically, the target domain data features Input the kth probability neural network V k , and output a probability likelihood value is:

[0105]

[0106] Further, the maximum probability likelihood value of the output of the |Y s | probability neural networks on the feature is taken as a similarity indication value, and the similarity indication value ω j ' can be represented as:

[0107]

[0108] Step b3, normalize the similarity indication value to obtain a similarity value.

[0109] Specifically, within a training batch of the target domain dataset, the similarity indication value ω j ' is normalized to obtain a similarity value ω j , and the calculation formula of the similarity value ω

[0110]

[0111] where ω′ max and ω′ min are the maximum and minimum similarity indication values within the same training batch, and ε is a very small positive value. Usually, a positive value not greater than 10 -5 is selected.

[0112] Step b4, based on the target domain data features and the similarity value, a weighted domain adversarial training loss value is obtained using a domain discriminator.

[0113] Specifically, the domain discriminator is used to align the known classes in the source domain dataset and the target domain dataset in the feature space, and the calculation formula of the weighted domain adversarial training loss value is as follows:

[0114]

[0115] where d s and d t represent the domain labels of the source domain dataset and the target domain dataset, respectively, and ω j represents the similarity of the target domain sample to the source domain sample , i.e., the probability that the sample belongs to the known class.

[0116] Step b5, a weighted domain adversarial training optimization objective function is constructed based on the weighted domain adversarial training loss value.

[0117] Specifically, the expression of the weighted domain adversarial training optimization objective function is as follows:

[0118] min E -L adv (12)

[0119] min D L adv (13)

[0120] This embodiment provides a method for cross-operating condition open-set fault diagnosis of rotating equipment, which uses the output of a probabilistic neural network to determine the similarity of target domain samples, that is, the similarity value. The strong feature distribution fitting ability of the probabilistic neural network can more accurately calculate the similarity of the target domain samples, thereby improving the classification accuracy of the target domain samples.

[0121] Step S2023: construct a weighted classification training optimization objective function using a probabilistic neural network and an unknown class detector.

[0122] In some optional implementations, the above step S2022 includes:

[0123] In step c1, based on the source domain data features, the unknown class detector is used to obtain the category classification loss value.

[0124] Specifically, for the source domain data features The unknown class detector O classifies it as a known class, and the class classification loss value The calculation formula is as follows:

[0125]

[0126] Among them, d k Represents known category labels.

[0127] Step c2: sort the target domain data features based on the similarity value, divide the category features based on the sorting results, and calculate the weighted classification loss value based on the category feature division results.

[0128] Specifically, for the target domain data features According to ω j In order from small to large Sort the data and take the first ρ% of the target domain data features as the unknown class feature group, recorded as The number is [n t ρ%]; the target domain data features of the latter ρ% are taken as the known class feature group, recorded as The number is [n t ρ%]; the target domain data features in the middle part are taken as the undetermined group, where [*] represents the integer part of *; ρ is usually selected as a positive value not greater than 30.

[0129] Further, the calculation formula of the weighted classification loss value is as follows:

[0130]

[0131] wherein d uk represents an unknown class label.

[0132] Step c3, calculating a total classification loss value based on the class classification loss value and the weighted classification loss value, and constructing a weighted classification training optimization objective function based on the total classification loss value.

[0133] Specifically, the calculation formula of the total classification loss value L o is as follows:

[0134]

[0135] Further, the expression of the weighted classification training optimization objective function is as follows:

[0136] min E,O L o (17)

[0137] The method provided by the embodiment is used for identifying known classes and unknown classes by using an unknown class detector, sorting target domain data features based on similarity values, and then performing class feature division based on the sorting results. In the process of deep neural network training, the source domain data and part of the target domain data are divided into known classes, part of the target domain data is divided into unknown classes, and the similarity values are used for training the unknown class detector, so that the training method is more reasonable, and the accuracy of the fault diagnosis of the rotating equipment is further improved.

[0138] Step S2024, constructing a total optimization objective function based on the classification training optimization objective function of the source domain data, the weighted domain adversarial training optimization objective function, and the weighted classification training optimization objective function.

[0139] Specifically, the expression of the total optimization objective function is as follows:

[0140]

[0141] min D αL adv (19)

[0142] Wherein, α and β are adjustable balance coefficients.

[0143] Step S2025, the deep neural network is trained by using the total optimization objective function until the deep neural network converges, and a trained deep neural network is obtained.

[0144] Step S203, the open set classification is performed on the target domain data set by using the trained deep neural network, and a rotating equipment cross-condition open set fault class is obtained. For details, please refer to Figure 1 Step S103 of the embodiment shown in the figure will not be repeated here.

[0145] The rotating equipment cross-condition open set fault diagnosis method provided in the embodiment is based on the source domain data set and the target domain data set, and a total optimization objective function is constructed through the feature extractor, the probability neural network, the domain discriminator and the unknown class detector. The deep neural network is trained by using the total optimization objective function, so that the training process of the deep neural network aims to accurately identify fault class samples under different operating conditions, and ensures that the trained deep neural network can accurately identify and accurately diagnose rotating equipment fault classes under different operating conditions.

[0146] In the embodiment, a rotating equipment cross-condition open set fault diagnosis method is provided, which can be used in the mobile terminal described above, such as a fault diagnosis device connected to the rotating equipment, Figure 4 is a flowchart of a rotating equipment cross-condition open set fault diagnosis method according to an embodiment of the present application, as shown in the figure, the flowchart includes the following steps: Figure 4

[0147] Step S401, a source domain data set and a target domain data set are constructed; the source domain data set contains rotating equipment vibration data with fault state labels under a first operating condition, and the target domain data set includes rotating equipment vibration data under a second operating condition. For details, please refer to Figure 2 Step S201 of the embodiment shown in the figure will not be repeated here.

[0148] Step S402, the deep neural network is trained by using the source domain data set and the target domain data set, and a trained deep neural network is obtained; the deep neural network includes a feature extractor, a probability neural network, a domain discriminator and an unknown class detector. For details, please refer to Figure 2 Step S202 of the embodiment shown in the figure will not be repeated here.

[0149] Step S403, the open set classification is performed on the target domain data set by using the trained deep neural network, and a rotating equipment cross-condition open set fault class is obtained.

[0150] Specifically, as shown in the figure, the above step S403 includes: Figure 5

[0151] ​​Step S4031, based on the target domain sample in the target domain data set, the target domain sample feature is obtained through the feature extractor.

[0152] Specifically, the feature extractor E performs feature extraction on the target domain data sample to obtain the feature The calculation formula is the same as the above formula (7).

[0153] Step S4032, based on the target domain sample feature, the output likelihood probability is obtained through the probability neural network.

[0154] Specifically, the probability neural network performs classification on the feature The output likelihood probability of the feature

[0155] Step S4033, based on the target domain sample feature, the unknown class detector judges whether the target domain sample is a known class.

[0156] Step S4034, if the target domain sample is a known class, the output likelihood probability is used to determine the cross-condition open set fault class of the rotating equipment.

[0157] Specifically, when the unknown class detector O classifies the target domain sample feature as an unknown class, the sample is identified as an unknown fault; otherwise, when O classifies the target domain sample feature as a known class, the sample is identified as a known class The classification formula of the identification feature is the class .

[0158]

[0159] The rotating equipment cross-condition open set fault diagnosis method provided in the embodiment is based on the target domain sample feature, and the unknown class detector is used to judge whether the target domain sample is a known class. No experience threshold is needed, the unknown class detector can directly output whether the sample is an unknown class, the test process is more direct and efficient, and when the target domain sample is a known class, the output likelihood probability is used to determine the cross-condition open set fault class of the rotating equipment, thereby realizing accurate diagnosis of the cross-condition open set fault of the rotating equipment.

[0160] There is also provided in the embodiment a rotating equipment cross-condition open set fault diagnosis apparatus for implementing the above embodiment and preferred implementation, which has been described and will not be repeated. As used below, the term "module" can be a combination of software and / or hardware that implements a predetermined function. Although the apparatus described in the following embodiments is preferably implemented in software, hardware or a combination of software and hardware is also possible and contemplated.

[0161] The embodiment provides a rotating equipment cross-condition open set fault diagnosis apparatus, as shown in the accompanying drawings, comprising: Figure 6

[0162] The constructing module 601 is configured to construct a source domain data set and a target domain data set; the source domain data set contains rotating equipment vibration data with a fault state label under a first operating condition, and the target domain data set includes rotating equipment vibration data under a second operating condition.

[0163] The training module 602 is configured to train a deep neural network using the source domain data set and the target domain data set to obtain a trained deep neural network; the deep neural network includes a feature extractor, a probabilistic neural network, a domain discriminator, and an unknown class detector.

[0164] The classification module 603 is configured to perform open set classification on the target domain data set using the trained deep neural network to obtain a rotating equipment cross-condition open set fault class.

[0165] In some optional implementations, the training module 602 comprises:

[0166] The first constructing unit is configured to construct a classification training optimization objective function of the source domain data using the feature extractor and the probabilistic neural network based on the source domain data set.

[0167] The second constructing unit is configured to construct a weighted domain adversarial training optimization objective function using the feature extractor and the domain discriminator based on the source domain data set and the target domain data set.

[0168] The third constructing unit is configured to construct a weighted classification training optimization objective function using the probabilistic neural network and the unknown class detector.

[0169] The fourth constructing unit is configured to construct a total optimization objective function based on the classification training optimization objective function of the source domain data, the weighted domain adversarial training optimization objective function, and the weighted classification training optimization objective function.

[0170] The training unit is configured to train the deep neural network using the total optimization objective function until the deep neural network converges to obtain the trained deep neural network.

[0171] In some optional implementations, the first constructing unit comprises:​

[0172] The first mapping subunit is configured to map, by the feature extractor, the rotating equipment vibration data with the fault state label in the first operating condition in the source domain data set to a feature space to obtain source domain data features;

[0173] The first calculation subunit is configured to obtain a probability distribution of the source domain data features based on the source domain data features and by using a probabilistic neural network.

[0174] The first construction subunit is configured to calculate a classification loss of the source domain data set based on the probability distribution of the source domain data features, and construct a classification training optimization objective function of the source domain data based on the classification loss of the source domain data set.

[0175] In some optional embodiments, the second construction unit includes:

[0176] The second mapping subunit is configured to map, by the feature extractor, the rotating equipment vibration data in the second operating condition in the target domain data set to the feature space to obtain target domain data features.

[0177] The second calculation subunit is configured to obtain a plurality of probability likelihood values based on the target domain data features and by using the probabilistic neural network, and take a maximum probability likelihood value as a similarity indication value.

[0178] The normalization processing subunit is configured to perform normalization processing on the similarity indication value to obtain a similarity value.

[0179] The third calculation subunit is configured to obtain a weighted domain adversarial training loss value based on the target domain data features and the similarity value and by using a domain discriminator.

[0180] The second construction subunit is configured to construct a weighted domain adversarial training optimization objective function based on the weighted domain adversarial training loss value.

[0181] In some optional embodiments, the third construction unit includes:

[0182] The fourth calculation subunit is configured to obtain a class classification loss value based on the source domain data features and by using an unknown class detector.

[0183] The sorting subunit is configured to sort the target domain data features based on the similarity value, perform class feature division based on a sorting result, and calculate a weighted classification loss value based on the class feature division result.

[0184] The third construction subunit is configured to calculate a total classification loss value based on the class classification loss value and the weighted classification loss value, and construct a weighted classification training optimization objective function based on the total classification loss value.

[0185] In some optional embodiments, the classification module 603 includes:

[0186] An extraction unit, configured to obtain target domain sample features through a feature extractor based on target domain samples in a target domain dataset;

[0187] A computing unit, used to obtain the output likelihood probability through a probabilistic neural network based on the sample features of the target domain;

[0188] A judgment unit, configured to judge whether the target domain sample belongs to a known category through an unknown class detector based on the target domain sample features;

[0189] The determination unit is used to determine the cross-operating condition open set fault category of the rotating equipment based on the output likelihood probability if the target domain sample is a known category.

[0190] The further functional description of each of the above modules and units is the same as that of the above corresponding embodiments and will not be repeated here.

[0191] In this embodiment, a rotating equipment cross-operating condition open-set fault diagnosis device is presented in the form of a functional unit, where the unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and memory that executes one or more software or fixed programs, and / or other devices that can provide the above functions.

[0192] The embodiment of the present invention also provides a computer device having the above Figure 6 A rotating equipment cross-operating condition open-collection fault diagnosis device is shown.

[0193] See also Figure 7 , Figure 7 is a structural diagram of a computer device provided by an optional embodiment of the present invention, such as Figure 7 As shown, the computer device includes: one or more processors 10, memory 20, and interfaces for connecting various components, including high-speed interfaces and low-speed interfaces. Various components utilize different buses to communicate with each other and can be installed on a common mainboard or installed in other ways as needed. The processor can process the instructions executed in the computer device, including instructions stored in the memory or on the memory to display the graphical information of the GUI on an external input / output device (such as, a display device coupled to the interface). In some optional embodiments, if necessary, multiple processors and / or multiple buses can be used together with multiple memories and multiple memories. Equally, multiple computer devices can be connected, and each device provides part of the necessary operations (for example, as a server array, a group of blade servers, or a multi-processor system). Figure 7 A processor 10 is taken as an example.

[0194] The processor 10 can be a central processing unit, a network processing unit, or a combination thereof. The processor 10 can further include a hardware chip. The hardware chip can be an application specific integrated circuit, a programmable logic device, or a combination thereof. The programmable logic device can be a complex programmable logic device, a field programmable logic device, a generic array logic, or any combination thereof.

[0195] The memory 20 stores instructions executable by the at least one processor 10 to cause the at least one processor 10 to perform the methods illustrated in the above embodiments.

[0196] The memory 20 can include a program storage area and a data storage area. The program storage area can store an operating system and application programs required by at least one function. The data storage area can store data created according to the use of the computer device, and the like. In addition, the memory 20 can include a high-speed random access memory, and can further include a non-transitory memory, such as at least one magnetic disk storage device, a flash memory device, or other non-transitory solid-state memory device. In some alternative embodiments, the memory 20 can optionally include a memory disposed remotely from the processor 10, which can be connected to the computer device through a network. Examples of the network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and a combination thereof.

[0197] The memory 20 can include a volatile memory, such as a random access memory, and can also include a non-volatile memory, such as a flash memory, a hard disk, or a solid state disk. The memory 20 can further include a combination of the above-mentioned types of memories.

[0198] The computer device further includes an input device 30 and an output device 40. The processor 10, the memory 20, the input device 30, and the output device 40 can be connected by a bus or other means, Figure 7 For example, by a bus connection.

[0199] The input device 30 can receive input digital or character information, and generate key signal inputs related to the user settings and function controls of the computer device, such as a touch screen, a keypad, a mouse, a trackpad, a touchpad, a pointing stick, one or more mouse buttons, a trackball, a joystick, and the like. The output device 40 can include a display device, an auxiliary lighting device (e.g., an LED), a tactile feedback device (e.g., a vibration motor), and the like. The display device includes, but is not limited to, a liquid crystal display, a light emitting diode, a display, and a plasma display. In some alternative embodiments, the display device can be a touch screen.

[0200] The embodiments of the present application further provide a computer readable storage medium, and the method according to the embodiments of the present application can be implemented in hardware, firmware, or recorded in a storage medium, or stored in a remote storage medium or a non-transitory machine readable storage medium and downloaded to a local storage medium through network, so that the method described herein can be processed by such software on a storage medium using a general purpose computer, a special purpose processor, or programmable or special hardware. The storage medium can be a magnetic disk, an optical disk, a read-only memory, a random access memory, a flash memory, a hard disk, or a solid state disk, etc. Further, the storage medium can also include a combination of the above-mentioned memories. It can be understood that the computer, the processor, the microprocessor controller, or the programmable hardware includes a storage component that can store or receive software or computer code, when the software or computer code is accessed and executed by the computer, the processor, or the hardware, the method shown in the above embodiments is implemented.

[0201] Part of the present application can be applied as a computer program product, for example, computer program instructions, when executed by a computer, the operation of the computer can invoke or provide the method and / or technical solutions according to the present application. Those skilled in the art should understand that the form of computer program instructions in computer readable medium includes but is not limited to source file, executable file, installation package file, etc. Correspondingly, the way of computer program instructions executed by computer includes but is not limited to: the computer directly executes the instructions, or the computer compiles the instructions and then executes the corresponding compiled program, or the computer reads and executes the instructions, or the computer reads and installs the instructions and then executes the corresponding installed program. Here, the computer readable medium can be any available computer readable storage medium or communication medium accessible to the computer.

[0202] Although the embodiments of the present application are described in conjunction with the accompanying drawings, various modifications and changes can be made by those skilled in the art without departing from the spirit and scope of the present application, and such modifications and changes fall within the scope defined by the appended claims.

Claims

1. A method for cross-operating-condition open-set fault diagnosis of rotating equipment, characterized in that: The method comprises: Constructing a source domain dataset and a target domain dataset; wherein the source domain dataset includes vibration data of the rotating equipment with a fault state label under a first operating condition, and the target domain dataset includes vibration data of the rotating equipment under a second operating condition; Training a deep neural network using the source domain dataset and the target domain dataset to obtain a trained deep neural network; the deep neural network includes a feature extractor, a probabilistic neural network, a domain discriminator, and an unknown class detector; Performing open set classification on the target domain data set using the trained deep neural network to obtain open set fault categories of rotating equipment across working conditions; The method of training a deep neural network using the source domain dataset and the target domain dataset to obtain the trained deep neural network includes: Based on the source domain data set, using the feature extractor and the probabilistic neural network to construct a classification training optimization objective function for the source domain data; Based on the source domain dataset and the target domain dataset, constructing a weighted domain adversarial training optimization objective function using the feature extractor and the domain discriminator; Constructing a weighted classification training optimization objective function using the probabilistic neural network and the unknown class detector; Constructing an overall optimization objective function based on the classification training optimization objective function of the source domain data, the weighted domain adversarial training optimization objective function, and the weighted classification training optimization objective function; The deep neural network is trained using the overall optimization objective function until the deep neural network converges, thereby obtaining the trained deep neural network.

2. The method according to claim 1, characterized in that The step of constructing a classification training optimization objective function for source domain data based on the source domain data set by using the feature extractor and the probabilistic neural network includes: Mapping the vibration data of the rotating equipment with a fault state label under the first operating condition in the source domain data set to a feature space by a feature extractor to obtain source domain data features; Based on the source domain data features, using the probabilistic neural network to obtain a probability distribution of the source domain data features; The classification loss of the source domain data set is calculated based on the probability distribution of the source domain data features, and a classification training optimization objective function of the source domain data is constructed based on the classification loss of the source domain data set.

3. The method according to claim 2, characterized in that The step of constructing a weighted domain adversarial training optimization objective function based on the source domain dataset and the target domain dataset using the feature extractor and the domain discriminator includes: Mapping the vibration data of the rotating equipment under the second operating condition in the target domain data set to a feature space by a feature extractor to obtain target domain data features; Based on the target domain data features, using the probabilistic neural network to obtain multiple probability likelihood values, and using the maximum probability likelihood value as a similarity indicator value; Normalizing the similarity indicator value to obtain a similarity value; Based on the target domain data features and the similarity value, using the domain discriminator to obtain a weighted domain adversarial training loss value; A weighted domain adversarial training optimization objective function is constructed based on the weighted domain adversarial training loss value.

4. The method according to claim 3, characterized in that The method of constructing a weighted classification training optimization objective function using the probabilistic neural network and the unknown class detector includes: Based on the source domain data features, using the unknown class detector to obtain a category classification loss value; Sorting the target domain data features based on the similarity values, performing category feature division based on the sorting results, and calculating a weighted classification loss value based on the category feature division results; A total classification loss value is calculated based on the category classification loss value and the weighted classification loss value, and a weighted classification training optimization objective function is constructed based on the total classification loss value.

5. The method according to claim 1, wherein The open set classification of the target domain data set is performed using the trained deep neural network to obtain the cross-operating condition open set fault categories of the rotating equipment, including: Based on the target domain samples in the target domain dataset, obtaining target domain sample features through the feature extractor; Based on the target domain sample features, obtaining an output likelihood probability through the probabilistic neural network; Based on the target domain sample features, determining whether the target domain sample belongs to a known category by the unknown class detector; If the target domain sample is of a known category, the cross-operating condition open set fault category of the rotating equipment is determined based on the output likelihood probability.

6. A rotating equipment cross-operating condition open-set fault diagnosis device, characterized in that: The device comprises: A construction module is used to construct a source domain dataset and a target domain dataset; wherein the source domain dataset includes vibration data of the rotating equipment with a fault state label under a first operating condition, and the target domain dataset includes vibration data of the rotating equipment under a second operating condition; A training module, configured to train a deep neural network using the source domain dataset and the target domain dataset to obtain a trained deep neural network; the deep neural network includes a feature extractor, a probabilistic neural network, a domain discriminator, and an unknown class detector; A classification module, configured to perform open-set classification on the target domain dataset using the trained deep neural network to obtain open-set fault categories of rotating equipment across operating conditions; Training modules include: The first construction unit is used to construct a classification training optimization objective function for source domain data based on the source domain data set using a feature extractor and a probabilistic neural network; The second construction unit is used to construct a weighted domain adversarial training optimization objective function based on the source domain dataset and the target domain dataset using a feature extractor and a domain discriminator; A third construction unit is used to construct a weighted classification training optimization objective function using a probabilistic neural network and an unknown class detector; A fourth construction unit is used to construct an overall optimization objective function based on the classification training optimization objective function of the source domain data, the weighted domain adversarial training optimization objective function, and the weighted classification training optimization objective function; The training unit is used to train the deep neural network using the total optimization objective function until the deep neural network converges to obtain the trained deep neural network.

7. A computer device, characterized in that: include: A memory and a processor, wherein the memory and the processor are communicatively connected to each other, computer instructions are stored in the memory, and the processor executes the cross-operating condition open set fault diagnosis method for rotating equipment according to any one of claims 1 to 5 by executing the computer instructions.

8. 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 computer to execute the cross-operating-condition open-set fault diagnosis method for rotating equipment according to any one of claims 1 to 5.

9. A computer program product, characterized in that The method comprises computer instructions for causing a computer to execute the method for cross-operating-condition open-set fault diagnosis of rotating equipment according to any one of claims 1 to 5.

Images