A method and system for detecting abnormal operation of rotating mechanical equipment in multiple working conditions

Abstract

The present invention discloses a method for detecting abnormal operation of rotating mechanical equipment under multiple working conditions, including: collecting and selecting vibration signals of rotating mechanical equipment under normal operation under multiple working conditions, learning description models, collecting vibration signals of rotating mechanical equipment under real-time monitoring and describing models for abnormal detection; learning an improved support vector data description model from vibration signals of rotating mechanical equipment under normal operation under multiple working conditions as a training process, and performing abnormal detection on real-time monitoring signals of rotating mechanical equipment as a testing process; training process: extracting time domain features and frequency domain features of subsequences and standardizing the features, learning this improved support vector data description model; testing process: extracting time domain features and frequency domain features of subsequences and using feature standardization parameters in the training process to standardize the extracted features, and obtaining abnormal detection results. The present invention can detect the operation status of the equipment and give an alarm for abnormal operation status.

Description

Technical Field

[0001] The present invention relates to the technical field of abnormal operation status detection of rotating mechanical equipment, and in particular to a method and system for detecting abnormal operation status of rotating mechanical equipment in multiple working conditions. Background Art

[0002] Rotating machinery and equipment refers to mechanical equipment that can complete specific functions through the rotation of rotating parts, and is widely used in the industrial field. However, due to possible harsh working conditions and long-term operation, rotating machinery and equipment will inevitably produce various forms of failure, thus affecting the smooth implementation of production tasks. Therefore, it is of great significance to detect abnormal operation status of rotating machinery and equipment, which will provide technical support for condition-based maintenance.

[0003] The operation of rotating mechanical equipment will generate vibration signals containing information representing the equipment status in real time. Most of the current anomaly detection technologies at home and abroad regard the anomaly detection task as a binary supervised learning problem or an unsupervised learning problem in terms of methods. Through methods such as support vector machines, neural networks, clustering, and isolation forests, a model is trained in a training set that contains both normal samples and abnormal samples, and then the model is used to test the data to be detected. Although the above unsupervised learning methods have overcome the problem of requiring experts to label sample labels, which is costly and difficult to implement, a more common and practical problem is that in the task of detecting the abnormal operation status of many equipment, the equipment is in normal operation most of the time, and the abnormal operation status has been terminated in time. Fault injection is difficult and costly, so there is very little abnormal sample data or it is not available at all, which limits the application of traditional anomaly detection methods. Summary of the invention

[0004] The purpose of the present invention is to provide a method and system for detecting abnormal operation of rotating mechanical equipment in multiple operating conditions, so as to solve the above-mentioned problems.

[0005] The present invention solves the technical problem by adopting the following technical solution:

[0006] A method for detecting abnormal operation of a rotating mechanical device under multiple operating conditions, comprising:

[0007] Collect and select vibration signals of rotating machinery equipment under normal operation in multiple working conditions, learn an improved support vector data description model, and then collect vibration signals of rotating machinery equipment under real-time monitoring and perform anomaly detection through the established improved support vector data description model;

[0008] The training process is to learn and improve the support vector data description model from the vibration signals of the rotating machinery equipment in normal operation under multiple working conditions, and the testing process is to detect anomalies in the real-time monitoring signals of the rotating machinery equipment.

[0009] Training process: The normal state historical data of multiple working conditions are truncated with a sliding window and divided into subsequences of equal length, and the time domain features and frequency domain features of the subsequences are extracted and standardized; the standardized features are used as input data of the improved support vector data description model to learn the improved support vector data description model;

[0010] Testing process: The real-time collected vibration signal is truncated with a sliding window to obtain the latest subsequence, the time domain features and frequency domain features of the subsequence are extracted, and the extracted features are standardized using the feature standardization parameters in the training process; the standardized features are used as input data for the improved support vector data description model to obtain the anomaly detection results of the model.

[0011] Furthermore, the specific steps of the model training process for improving support vector data description are as follows:

[0012] S1, establishing the optimal Gaussian mixture model for the training data X, and dividing the sample into a certain data subset according to the maximum response of the sample; wherein the optimal number of Gaussian mixture model components K for the training data X is determined based on the optimal model selection criterion;

[0013] S2. For each data subset divided in step S1, based on the multi-core learning method of center core alignment, learn the combination coefficients of the base core kernel function of the data subset, and construct the multi-core kernel function of the data subset;

[0014] S3. For each data subset divided in step S1, the multi-core kernel function constructed in step S2 is used, and a fuzzy weight is assigned to each sample according to the probability density information of the Gaussian mixture model in step S1, and the fuzzy multi-core support vector data descriptor model is learned.

[0015] Furthermore, in S2, for data subset A k ,remember y i is a sample x in the data subset i The corresponding label, because the method only needs samples in normal state for model training, so y i =1, i=1,2,...,|A k |; Let K * =yy T , for the N×N kernel matrix K, the corresponding centralized kernel matrix Where I is the N×N identity matrix, l is the N×1 vector whose elements are all 1; the combination coefficient of the base kernel function of the data subset is obtained by solving the following formula:

[0016]

[0017]

[0018] Among them, η m is the combination coefficient of the mth base kernel function, M is the number of base kernel functions, K m is the kernel matrix corresponding to the mth base kernel function, K mc is the centralized kernel matrix corresponding to the mth base kernel function, It is a centralized K * ,<·,·> F is the Frobenius inner product, ||·|| F is the Frobenius norm;

[0019] The multi-core kernel function constructed for this data subset is Among them, k m (·,·) is the mth basis kernel function, x i and x j are two samples of the kernel function to be determined in the data subset.

[0020] Furthermore, in S2, for data subset A k , the fuzzy multi-core support vector data descriptor model learning is solved by the following formula:

[0021]

[0022]

[0023] Among them, k m (·,·) is the mth basis kernel function, C k is the penalty factor, is the sample x i The fuzzy weight of and is the sample x i and x j The corresponding Lagrange multiplier;

[0024] Obtain the optimal solution α for the above problem k Finally, an important parameter calculation formula of the fuzzy multi-core support vector data descriptor model is as follows:

[0025]

[0026] Among them, the sample x l Corresponding is a non-zero value.

[0027] Furthermore, the specific steps of the model testing process for improving support vector data description are as follows:

[0028] S1, using each learned fuzzy multi-core support vector data descriptor model in turn to detect whether the test data z is in a normal operating state;

[0029] S2. When any fuzzy multi-core support vector data description sub-model detects that the test data is in a normal operating state, the test data operating state is determined to be ultimately normal; when all fuzzy multi-core support vector data description sub-models detect that the test data is in an abnormal operating state, the test data operating state is determined to be ultimately abnormal.

[0030] Furthermore, the detection of whether the test data z is in a normal operating state is obtained by the following formula:

[0031]

[0032] Among them, k m (·,·) is the mth basis kernel function, η m is the combination coefficient of the base kernel function, M is the number of base kernel functions; if f(z)≤R k , then in the kth sub-model, the test data is detected as being in normal operation; if f(z)>R k , then in the kth sub-model, the test data is detected as being in an abnormal operating state.

[0033] Furthermore, the optimal model selection criterion adopted in step S1 during the model training process of the improved support vector data description is the Bayesian information criterion. The optimal number of Gaussian mixture model components K for the training data X is determined by the following formula:

[0034]

[0035] Where N is the number of samples of training data X, P(x i |θ) is the sample x in the training data X i The global probability density estimate in the Gaussian mixture model, θ is the parameter of the Gaussian mixture model.

[0036] A multi-operating state abnormality detection system for rotating mechanical equipment, comprising: an equipment data management module, a data preprocessing and feature extraction module, an abnormality detection core module, a database module and a human-computer interaction module;

[0037] The equipment data management module uses the acceleration sensor to realize the real-time collection of vibration signal data; the equipment data management module realizes the storage management operation of real-time data and the reading management operation of historical data;

[0038] The data preprocessing and feature extraction module divides the original vibration signal data into subsequences of equal length using a sliding window; the data preprocessing and feature extraction module extracts the time domain features and frequency domain features of the subsequences; the data preprocessing and feature extraction module realizes the standardization of the extracted time domain features and frequency domain features;

[0039] The anomaly detection core module uses the standardized time domain features and frequency domain features of the training set to implement model training; the anomaly detection core module performs model testing and detection result judgment on the standardized time domain features and frequency domain features of the real-time acquisition data to achieve anomaly detection of the current operating status of the equipment;

[0040] The database module implements the storage of collected data; the database module implements the storage of time domain feature and frequency domain feature extraction algorithms; the database module implements parameter storage for feature standardization after the training set data is truncated by the sliding window and feature extraction; the database management module implements model storage after the anomaly detection core module completes model training;

[0041] The human-computer interaction module visualizes the monitoring data collected in real time; the human-computer interaction module operates and controls the selection of training data, the selection of feature extraction algorithms, the selection of feature standardization parameters, the model training process, and the selection of detection models; the human-computer interaction module displays the results of anomaly detection in real time.

[0042] Furthermore, the feature extraction submodule in the data preprocessing and feature extraction module can select feature extraction algorithms in the database module feature extraction algorithm library including: mean, root mean square value, peak value, kurtosis, skewness, peak factor, pulse factor, waveform factor and margin factor.

[0043] Beneficial effects of the present invention:

[0044] (1) The present invention makes full use of the vibration signal data naturally generated by the normal operation of the rotating mechanical equipment. The model training in the method does not require abnormal state data or data labeling. The data acquisition method is simple and the data acquisition cost is low.

[0045] (2) The method of the present invention can maintain a high anomaly detection accuracy rate and a low false alarm rate and missed alarm rate when the normal state data actually has multiple working conditions and each working condition is not distinguished and labeled;

[0046] (3) The method of the present invention can maintain a high anomaly detection accuracy rate and a low false alarm rate and missed alarm rate when the normal state data is actually mixed with the abnormal state data in an unknown manner. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 is a flow chart of the method of the present invention;

[0048] Figure 2 It is a schematic diagram of the structure of the system of the present invention. DETAILED DESCRIPTION

[0049] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0050] refer to Figure 1 The present invention discloses a method for detecting abnormal operation of a rotating mechanical device under multiple operating conditions, comprising:

[0051] Collect and select vibration signals of rotating machinery equipment under normal operation in multiple working conditions, learn an improved support vector data description model, and then collect vibration signals of rotating machinery equipment under real-time monitoring and perform anomaly detection through the established improved support vector data description model;

[0052] The training process is to learn and improve the support vector data description model from the vibration signals of the rotating machinery equipment in normal operation under multiple working conditions, and the testing process is to detect anomalies in the real-time monitoring signals of the rotating machinery equipment.

[0053] Training process: The normal state historical data of multiple working conditions are truncated with a sliding window and divided into subsequences of equal length, and the time domain features and frequency domain features of the subsequences are extracted and standardized; the standardized features are used as input data of the improved support vector data description model to learn the improved support vector data description model;

[0054] Testing process: The real-time collected vibration signal is truncated with a sliding window to obtain the latest subsequence, the time domain features and frequency domain features of the subsequence are extracted, and the extracted features are standardized using the feature standardization parameters in the training process; the standardized features are used as input data for the improved support vector data description model to obtain the anomaly detection results of the model.

[0055] A further improvement of the present invention is that the specific steps of the model training process of the improved support vector data description are:

[0056] S1, establishing the optimal Gaussian mixture model for the training data X, and dividing the sample into a certain data subset according to the maximum response of the sample; wherein the optimal number of Gaussian mixture model components K for the training data X is determined based on the optimal model selection criterion;

[0057] S2. For each data subset divided in step S1, based on the multi-core learning method of center core alignment, learn the combination coefficients of the base core kernel function of the data subset, and construct the multi-core kernel function of the data subset;

[0058] For the data set A k ,remember y i is a sample x in the data subset i The corresponding label, because the method only needs samples in normal state for model training, so y i =1, i=1,2,...,|A k |. Let K * =yy T , for the N×N kernel matrix K, the corresponding centralized kernel matrix Where I is the N×N identity matrix, and l is an N×1 vector whose elements are all 1. The combination coefficients of the base kernel function of this data subset are obtained by solving the following problem:

[0059]

[0060]

[0061] Among them, η m is the combination coefficient of the mth base kernel function, M is the number of base kernel functions, K m is the kernel matrix corresponding to the mth base kernel function, K mc is the centralized kernel matrix corresponding to the mth base kernel function, It is a centralized K * ,<·,·> F is the Frobenius inner product, ||·|| F is the Frobenius norm;

[0062] The multi-core kernel function constructed for this data subset is Among them, k m (·,·) is the mth basis kernel function, η m is the combination coefficient of the base kernel function, M is the number of base kernel functions; x i and x j are two samples of the kernel function to be determined in the data subset.

[0063] S3, for each data subset divided in step S1, the multi-core kernel function constructed in step S2 is used, and a fuzzy weight is assigned to each sample according to the probability density information of the Gaussian mixture model in step S1, and a fuzzy multi-core support vector data description sub-model is learned;

[0064] For the data set A k, the fuzzy multi-core support vector data descriptor model learning is completed by solving the following problems:

[0065]

[0066]

[0067] Among them, k m (·,·) is the mth basis kernel function, η m is the combination coefficient of the base kernel function, M is the number of base kernel functions, C k is the penalty factor, is the sample x i The fuzzy weight of and is the sample x i and x j The corresponding Lagrange multiplier;

[0068] Obtain the optimal solution α for the above problem k Finally, an important parameter calculation formula of the fuzzy multi-core support vector data descriptor model is as follows:

[0069]

[0070] Among them, k m (·,·) is the mth basis kernel function, η m is the combination coefficient of the base kernel function, M is the number of base kernel functions, and sample x l Corresponding is a non-zero value;

[0071] A further improvement of the present invention is that the specific steps of the model testing process of the improved support vector data description are:

[0072] S1. Use each learned fuzzy multi-core support vector data descriptor model in turn to detect whether the test data z is in a normal operating state; more specifically, calculate the following formula:

[0073]

[0074] Among them, k m (·,·) is the mth basis kernel function, η m is the combination coefficient of the base kernel function, and M is the number of base kernel functions. k , then in the kth sub-model, the test data is detected as being in normal operation; if f(z)>R k , then in the kth sub-model, the test data is detected as being in an abnormal operating state;

[0075] S2. When any fuzzy multi-core support vector data description sub-model detects that the test data is in a normal operating state, the test data operating state is determined to be ultimately normal; when all fuzzy multi-core support vector data description sub-models detect that the test data is in an abnormal operating state, the test data operating state is determined to be ultimately abnormal.

[0076] To further optimize the above technical solution, the optimal model selection criterion adopted in step S1 during the model training process of the improved support vector data description is the Bayesian information criterion. The optimal number of Gaussian mixture model components K for the training data X is determined by the following formula:

[0077]

[0078] Where N is the number of samples of training data X, P(x i |θ) is the sample x in the training data X i The global probability density estimate in the Gaussian mixture model, θ is the parameter of the Gaussian mixture model.

[0079] like Figure 2 As shown, the embodiment of the present invention discloses a multi-operating state abnormality detection system for rotating mechanical equipment, including: an equipment data management module, a data preprocessing and feature extraction module, an abnormality detection core module, a database module, and a human-computer interaction module;

[0080] The equipment data management module includes a real-time data acquisition submodule and a data management operation submodule; the real-time data acquisition submodule uses an acceleration sensor to transmit the real-time collected vibration signal data to the data preprocessing and feature extraction module for preprocessing and feature extraction for detection, and at the same time transmits the real-time collected vibration signal data to the data management operation submodule for data storage, and at the same time transmits the real-time collected vibration signal data to the human-computer interaction module; the data management operation submodule receives the data transmitted by the real-time data acquisition submodule and stores it in the historical database of the database module, and on the other hand, reads the historical data from the historical database of the database module and transmits it to the data preprocessing and feature extraction module for preprocessing and feature extraction for model training; the data management operation submodule receives the historical data reading selection control from the human-computer interaction module;

[0081] The data preprocessing and feature extraction module includes a sliding window truncation submodule, a feature extraction submodule and a feature standardization submodule; the sliding window truncation submodule receives data from the device data management module, divides the data into subsequences of equal length using the sliding window, and transmits the data to the feature extraction submodule; the feature extraction submodule receives the subsequence truncated from the sliding window, reads the feature extraction algorithm library in the database module, extracts the time domain features and frequency domain features of the subsequence, and transmits them to the feature standardization submodule; the feature extraction algorithm library receives the feature extraction algorithm selection control from the human-computer interaction module; the feature standardization submodule receives the features extracted from the feature extraction submodule, standardizes the features during the training process, and stores the relevant parameters used for feature standardization in the feature standardization parameter library of the database module; the feature standardization submodule receives the features extracted from the feature extraction submodule, reads the parameters in the feature standardization parameter library of the database module during the test process, and standardizes the received features using the read parameters; the feature standardization submodule transmits the standardized features to the anomaly detection core module; the feature standardization parameter library receives the parameter selection control from the human-computer interaction module;

[0082] The anomaly detection core module includes a model training submodule, a model testing submodule and a detection result determination submodule; during the training process, the model training submodule receives the standardized features from the data preprocessing and feature extraction module, learns to improve the support vector data description model, and stores the learned model in the model library of the database module; during the testing process, the model testing submodule receives the standardized features from the data preprocessing and feature extraction module, reads the model library in the database module, receives the model selection control from the human-computer interaction module, uses the selected model to perform preliminary anomaly detection, and transmits the preliminary detection results to the detection result determination submodule; the detection result determination submodule receives the preliminary detection results from the model testing submodule, and gives the final anomaly detection results according to the comprehensive determination logic and transmits them to the human-computer interaction module; wherein the comprehensive determination logic is that if all the preliminary detection results are abnormal, the final anomaly detection result is abnormal, and if any of the preliminary detection results is normal, the final anomaly detection result is normal;

[0083] The database module includes a historical database, a feature standardization parameter library, a feature extraction algorithm library and a model library; the historical database receives data from the device data management module to realize the storage of real-time collected data; the historical database transmits historical data to the device data management module for model training; the feature extraction algorithm library transmits feature extraction algorithms to the data preprocessing and feature extraction module; during the training process, the feature standardization parameter library receives the storage of feature standardization parameters transmitted by the data preprocessing and feature extraction module; during the testing process, the feature standardization parameter library transmits feature standardization parameters to the data preprocessing and feature extraction module; the feature extraction algorithm library realizes the storage of time domain features and frequency domain feature extraction algorithms; the model library receives the storage of learned models from the anomaly detection core module; the model library transmits the learned models to the anomaly detection core module for model testing;

[0084] The human-computer interaction module includes a monitoring signal visualization submodule, an operation control interface submodule and an anomaly detection result submodule; the monitoring signal visualization submodule receives data from the equipment data management module to realize the visualization of real-time collected vibration signal data; the operation control interface submodule operates the equipment data management module to perform reading and selection control on the historical database in the database module; the operation control interface submodule operates the data preprocessing and feature extraction module to perform parameter selection control and algorithm selection control on the feature standardization parameter library and feature extraction algorithm library in the database module; the operation control interface submodule operates the anomaly detection core module to perform model selection control on the model library in the database module; the operation control interface submodule operates the training process and the testing process; the anomaly detection submodule receives the final anomaly detection result from the anomaly detection core module and displays the final anomaly detection result.

[0085] To further optimize the above technical solution, the feature extraction submodule in the data preprocessing and feature extraction module can select the feature extraction algorithm in the database module feature extraction algorithm library, including: calculation of mean, root mean square value, peak value, kurtosis, skewness, peak factor, pulse factor, waveform factor and margin factor, and its formula is as follows:

[0086] (1) Mean

[0087] (2) Root mean square value

[0088] (3) Peak value X p =max(|x(N)|)

[0089] (4) Kurtosis

[0090] (5) Skewness

[0091] (6) Peak factor

[0092] (7) Pulse Factor

[0093] (8) Form factor

[0094] (9) Margin factor

[0095] where x(N) = {x1,x2,x3,…,x N} is the vibration signal subsequence, N is the subsequence length, μ is the mean of all values ​​in the subsequence, and x(N)-μ means subtracting the mean of all values ​​in the subsequence from each value in the vibration signal subsequence.

[0096] The present invention makes full use of vibration signal data naturally generated by the normal operation of rotating mechanical equipment. The model training in the method does not require abnormal state data, nor does it require data labeling. The data acquisition method is simple and the data acquisition cost is low. The method of the present invention can maintain a high anomaly detection accuracy rate and a low false alarm rate and missed alarm rate when the normal state data actually has multiple working conditions and the working conditions are not distinguished and labeled. The method of the present invention can maintain a high anomaly detection accuracy rate and a low false alarm rate and missed alarm rate when the normal state data is actually mixed with abnormal state data in an unknown manner.

[0097] In order to solve the practical problem of the complete lack of abnormal data, the present invention provides a method and system for detecting the abnormal operation state of rotating mechanical equipment that does not rely on any abnormal operation data. The method and system can adapt to the situation where normal state data used for model training have multiple working conditions, and can adapt to the situation where some abnormal state data are unknownly mixed into the normal state data used for model training. The method and system have high accuracy, low false alarm rate, low missed alarm rate, and low model training time complexity.

[0098] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for detecting abnormal operation of rotating mechanical equipment in multiple operating conditions, characterized in that: include: Collect and select vibration signals of rotating machinery equipment under normal operation in multiple working conditions, learn an improved support vector data description model, and then collect vibration signals of rotating machinery equipment under real-time monitoring and perform anomaly detection through the established improved support vector data description model; The training process is to learn and improve the support vector data description model from the vibration signals of the rotating machinery equipment in normal operation under multiple working conditions, and the testing process is to detect anomalies in the real-time monitoring signals of the rotating machinery equipment. Training process: The normal state historical data of multiple working conditions are truncated with a sliding window and divided into subsequences of equal length, and the time domain features and frequency domain features of the subsequences are extracted and standardized; the standardized features are used as input data of the improved support vector data description model to learn the improved support vector data description model; Testing process: The real-time collected vibration signal is truncated with a sliding window to obtain the latest subsequence, the time domain features and frequency domain features of the subsequence are extracted, and the extracted features are standardized using the feature standardization parameters in the training process; the standardized features are used as input data for the improved support vector data description model to obtain the anomaly detection results of the model; The specific steps of the model training process to improve the support vector data description are: S1, establishing the optimal Gaussian mixture model for the training data X, and dividing the sample into a certain data subset according to the maximum response of the sample; wherein the optimal number of Gaussian mixture model components K for the training data X is determined based on the optimal model selection criterion; S2. For each data subset divided in step S1, based on the multi-core learning method of center core alignment, learn the combination coefficients of the base core kernel function of the data subset, and construct the multi-core kernel function of the data subset; S3, for each data subset divided in step S1, the multi-core kernel function constructed in step S2 is used, and a fuzzy weight is assigned to each sample according to the probability density information of the Gaussian mixture model in step S1, and a fuzzy multi-core support vector data description sub-model is learned; In S2, for data subset A k ,remember y i is a sample x in the data subset i The corresponding label, because the method only needs samples in normal state for model training, so y i =1, i=1,2,...,|A k |; Let K * =yy T , for the N×N kernel matrix K, the corresponding centralized kernel matrix Where I is the N×N identity matrix, l is the N×1 vector whose elements are all 1; the combination coefficient of the base kernel function of the data subset is obtained by solving the following formula: Among them, η m is the combination coefficient of the mth base kernel function, M is the number of base kernel functions, K m is the kernel matrix corresponding to the mth base kernel function, K mc is the centralized kernel matrix corresponding to the mth base kernel function, It is a centralized K * ,<·,·> F is the Frobenius inner product, ||·|| F is the Frobenius norm; The multi-core kernel function constructed for this data subset is Among them, k m (·,·) is the mth basis kernel function, x i and x j are two samples of the kernel function to be found in the data subset; In S2, for data subset A k , the fuzzy multi-core support vector data descriptor model learning is solved by the following formula: Among them, k m (·,·) is the mth basis kernel function, C k is the penalty factor, is the sample x i The fuzzy weight of and is the sample x i and x j The corresponding Lagrange multiplier; Find the optimal solution α k Finally, an important parameter calculation formula of the fuzzy multi-core support vector data descriptor model is as follows: Among them, the sample x l Corresponding is a non-zero value; The specific steps of the model testing process for improving support vector data description are as follows: S1, using each learned fuzzy multi-core support vector data descriptor model in turn to detect whether the test data z is in a normal operating state; S2. When any fuzzy multi-core support vector data description sub-model detects that the test data is in a normal operating state, the test data operating state is determined to be finally normal; when all fuzzy multi-core support vector data description sub-models detect that the test data is in an abnormal operating state, the test data operating state is determined to be finally abnormal; Whether the test data z is in normal operation is determined by the following formula: Among them, k m (·,·) is the mth basis kernel function, η m is the combination coefficient of the base kernel function, M is the number of base kernel functions; if f(z)≤R k , then in the kth sub-model, the test data is detected as being in normal operation; if f(z)>R k , then in the kth sub-model, the test data is detected as being in an abnormal operating state.

2. According to claim 1, a method for detecting abnormal operation of rotating machinery in multiple operating conditions, characterized in that: The optimal model selection criterion adopted in step S1 during the model training process of the improved support vector data description is the Bayesian information criterion. The optimal number of Gaussian mixture model components K for the training data X is determined by the following formula: Where N is the number of samples of training data X, P(x i |θ) is the sample x in the training data X i The global probability density estimate in the Gaussian mixture model, θ is the parameter of the Gaussian mixture model.

3. A multi-operating state abnormality detection system for rotating mechanical equipment, characterized in that: The method for detecting abnormal operation of a rotating mechanical device in multiple operating conditions according to any one of claims 1 to 2 comprises: an equipment data management module, a data preprocessing and feature extraction module, an abnormality detection core module, a database module and a human-computer interaction module; The equipment data management module uses the acceleration sensor to realize the real-time collection of vibration signal data; the equipment data management module realizes the storage management operation of real-time data and the reading management operation of historical data; The data preprocessing and feature extraction module divides the original vibration signal data into subsequences of equal length using a sliding window; the data preprocessing and feature extraction module extracts the time domain features and frequency domain features of the subsequences; the data preprocessing and feature extraction module realizes the standardization of the extracted time domain features and frequency domain features; The anomaly detection core module uses the standardized time domain features and frequency domain features of the training set to implement model training; the anomaly detection core module performs model testing and detection result judgment on the standardized time domain features and frequency domain features of the real-time collected data to achieve anomaly detection of the current operating status of the equipment; The database module implements the storage of collected data; the database module implements the storage of time domain feature and frequency domain feature extraction algorithms; the database module implements parameter storage for feature standardization after the training set data is truncated by the sliding window and feature extraction; the database management module implements model storage after the anomaly detection core module completes model training; The human-computer interaction module visualizes the monitoring data collected in real time; the human-computer interaction module operates and controls the selection of training data, the selection of feature extraction algorithms, the selection of feature standardization parameters, the model training process, and the selection of detection models; the human-computer interaction module displays the results of anomaly detection in real time.

4. According to claim 3, a rotating mechanical equipment multi-operating state abnormality detection system is characterized in that: The feature extraction submodule in the data preprocessing and feature extraction module can select feature extraction algorithms in the database module feature extraction algorithm library including: mean, root mean square value, peak value, kurtosis, skewness, peak factor, pulse factor, waveform factor and margin factor.

Images