A real-time fault diagnosis method and device for a turbine cooler

Abstract

The present application provides a method and device for real-time fault diagnosis of a turbine cooler. The method provided in the present application includes: establishing a turbine cooler simulation model, the turbine cooler simulation model includes a one-dimensional simulation model and a three-dimensional simulation model; simulating various fault modes based on the turbine cooler simulation model to obtain fault data under various fault modes; determining the state monitoring parameters of the turbine cooler according to the fault data; establishing a probabilistic neural network alternative model set based on the state monitoring parameters; forming a fault diagnosis model set based on the probabilistic neural network alternative model set and operating condition parameters, and calculating a matching matrix based on the operating condition parameters and the fault diagnosis model set; determining a target fault diagnosis model from the fault diagnosis model set based on the matching matrix, and performing fault diagnosis based on the target fault diagnosis model and the real-time fault operating condition. The method and device for real-time fault diagnosis of a turbine cooler provided in the present application can realize real-time fault diagnosis under variable operating conditions.

Description

Technical Field

[0001] The present application relates to the technical field of turbine cooler fault diagnosis, and in particular to a turbine cooler real-time fault diagnosis method and device. Background Art

[0002] The turbine cooler is a core component of the aircraft environmental control system. It is installed in the environmental control accessory compartment of the mid-fuselage and is used to expand the high-temperature air drawn from the engine to do work, reduce temperature and pressure, and provide cooling air for the aircraft cabin and electronic equipment compartment. Its working condition directly affects the cooling of pilots and electronic equipment.

[0003] Existing health status diagnosis methods for aviation electromechanical products are divided into expert knowledge-based and model-based fault diagnosis methods. Expert knowledge-based methods rely on knowledge and experience, and are characterized by high randomness and low accuracy; model-based methods need to combine prior knowledge obtained from historical experience and establish accurate mathematical models for the object of study based on its physical, mathematical and other principles, but this is often difficult to do in reality.

[0004] For turbine coolers, due to their large nonlinearity and strong coupling, it is difficult to carry out modeling mechanism and fault diagnosis research. In addition, since the working environment and operating task conditions of turbine coolers often change, they often operate alternately between various working conditions. The working state differences under different task conditions are quite obvious, and the signals generated when the working conditions change will also have a disturbance effect on them. Therefore, fault diagnosis of turbine coolers under variable working conditions has become an urgent problem to be solved. Summary of the invention

[0005] In view of this, the present application provides a real-time fault diagnosis method and device for a turbine cooler, which can realize real-time fault diagnosis under variable operating conditions.

[0006] Specifically, the present application is implemented through the following technical solutions:

[0007] The first aspect of the present application provides a real-time fault diagnosis method for a turbine cooler, the method comprising:

[0008] Establishing a turbine cooler simulation model, wherein the turbine cooler simulation model includes a one-dimensional simulation model and a three-dimensional simulation model;

[0009] Simulating various fault modes based on the turbine cooler simulation model to obtain fault data under various fault modes;

[0010] determining a condition monitoring parameter of the turbine cooler according to the fault data;

[0011] Establishing a probabilistic neural network candidate model set based on the condition monitoring parameters;

[0012] Forming a fault diagnosis model set based on the probabilistic neural network candidate model set and operating condition parameters, and calculating a matching matrix based on the operating condition parameters and the fault diagnosis model set;

[0013] A target fault diagnosis model is determined from the fault diagnosis model set based on the matching matrix, and fault diagnosis is performed based on the target fault diagnosis model and the real-time fault condition.

[0014] The second aspect of the present application provides a real-time fault diagnosis device for a turbine cooler, the device comprising a building module, a simulation module, a determination module, a calculation module and a diagnosis module; wherein:

[0015] The establishing module is used to establish a turbine cooler simulation model, wherein the turbine cooler simulation model includes a one-dimensional simulation model and a three-dimensional simulation model;

[0016] The simulation module is used to simulate various fault modes based on the turbine cooler simulation model to obtain fault data under various fault modes;

[0017] The determination module is used to determine the state monitoring parameters of the turbine cooler according to the fault data;

[0018] The establishment module is further used to establish a probabilistic neural network candidate model set based on the state monitoring parameters;

[0019] The calculation module is used to form a fault diagnosis model set based on the probabilistic neural network candidate model set and operating parameters, and calculate a matching matrix based on the operating parameters and the fault diagnosis model set;

[0020] The diagnosis module is used to determine a target fault diagnosis model from the fault diagnosis model set based on the matching matrix, and perform fault diagnosis based on the target fault diagnosis model and the real-time fault condition.

[0021] The real-time fault diagnosis method and device for a turbine cooler provided in the present application establish a turbine cooler simulation model including a one-dimensional simulation model and a three-dimensional simulation model, and use the two models to simulate the best links respectively. The output signal obtained by the one-dimensional model can be used as the input of the three-dimensional model, and the output of the three-dimensional model can optimize the one-dimensional model. The output signals of the two models are complementary, and the model with the most accurate simulation simulates the signal of a specific environment and a specific part. When the two have the same accuracy and consistent output, the corresponding model can be selected in a manner with the shortest calculation time and the lowest calculation amount, which can more comprehensively reflect the behavior of the turbine cooler under various operating conditions and fault modes, obtain more comprehensive data, and simulate the behavior of the turbine cooler in a high-precision, low-computation manner, providing a basis for subsequent analysis; by simulating various fault modes, rich fault data can be obtained, reducing the cost and risk of fault testing, and more sensor information can be obtained, which is helpful to deeply understand the performance of the turbine cooler under different fault modes and provide support for subsequent fault analysis modeling; a probabilistic neural network alternative model set and a fault diagnosis model set are established, and considering the differences in fault distribution under different operating conditions, multiple diagnosis models are established for each operating condition and fault. Each time, the matching matrix is ​​calculated based on the operating parameters, monitoring parameters and fault characteristics, and the correlation between each model to screen out the optimal fault diagnosis model, rather than just judging by the accuracy of a single fault identification model. By matching the current fault characteristics with the best identification model in real time, it can better adapt to different operating conditions, determine the target fault diagnosis model based on the matching matrix, and select the most suitable model from multiple fault diagnosis models. It can perform fault diagnosis according to real-time operating conditions under variable operating conditions, flexibly adjust fault handling strategies, and improve maintenance efficiency and accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 A flowchart of a first embodiment of a real-time fault diagnosis method for a turbine cooler provided in the present application;

[0023] Figure 2 A schematic diagram of the structure of the probabilistic neural network model provided for this application;

[0024] Figure 3 This is a flowchart of state detection parameter special diagnosis extraction and information fusion shown in an exemplary embodiment of the present application;

[0025] Figure 4 This is a schematic structural diagram of Embodiment 1 of the turbine cooler real-time fault diagnosis device provided in the present application. DETAILED DESCRIPTION

[0026] Here, exemplary embodiments are described in detail, and examples thereof are shown in the accompanying drawings. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application.

[0027] The terms used in this application are only for the purpose of describing specific embodiments and are not intended to limit this application. The singular forms of "a", "said" and "the" used in this application are also intended to include plural forms, unless the context clearly indicates other meanings. It should also be understood that the term "and / or" used in this article refers to and includes any or all possible combinations of one or more associated listed items.

[0028] It should be understood that although the terms first, second, third, etc. may be used in the present application to describe various information, these information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of the present application, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. Depending on the context, the word "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining".

[0029] Specific embodiments are given below to introduce the technical solution of the present application in detail.

[0030] Embodiment 1

[0031] Figure 1 This is a flow chart of the first embodiment of the real-time fault diagnosis method for a turbine cooler provided in this application. Figure 1 , the method provided in this embodiment may include:

[0032] S101. Establish a turbine cooler simulation model, wherein the turbine cooler simulation model includes a one-dimensional simulation model and a three-dimensional simulation model.

[0033] It should be noted that the establishment of a turbine cooler simulation model includes:

[0034] (1) A one-dimensional simulation model is established based on the working principle and thermodynamic formula of the turbine cooler.

[0035] It should be noted that the turbine cooler mainly expands the high-temperature air drawn from the engine to do work, cools down the temperature and pressure, and provides cooling air for the aircraft cabin and electronic equipment compartment. Its main components include turbines, compressors / fans, etc. The working process generally involves compression, expansion and heat exchange, and the energy conversion of the flowing working fluid in the whole system conforms to the laws of thermodynamics.

[0036] Specifically, without considering the circuit loss, a one-dimensional simulation model can be established based on the relationship between flow balance, pressure balance, and energy balance:

[0037] ;

[0038] in, are the outlet temperature and outlet pressure of the fan / compressor end, are the outlet temperature and outlet pressure of the turbine end, is the rotation speed, are the inlet temperature, inlet pressure, flow rate and compression ratio at the fan / compressor end, is the expansion ratio at the turbine end.

[0039] (2) Determine the mathematical function form of the fan / compressor end and the turbine end based on the one-dimensional simulation model.

[0040] Specifically, the mathematical function form of the fan / compressor end is determined based on the one-dimensional simulation model:

[0041] ;

[0042] in, are the inlet temperature, inlet pressure, flow rate and compression ratio at the fan / compressor end, is the rotation speed, They are the outlet temperature and outlet pressure of the fan / compressor end respectively.

[0043] The mathematical function form of the turbine end determined based on the one-dimensional simulation model is:

[0044] ;

[0045] in, are the inlet temperature, inlet pressure, flow rate and expansion ratio of the turbine end, is the rotation speed, are the outlet temperature and outlet pressure of the turbine end respectively.

[0046] (3) Determine the basic performance parameters of the turbine according to the mathematical function form.

[0047] Combined with the above formula, basic performance parameters such as outlet temperature and pressure of the turbine under various conditions can be obtained. Please refer to the previous description for details.

[0048] (4) A three-dimensional simulation model is established based on thermodynamic finite element and multi-body dynamics finite element; the three-dimensional simulation model is used to obtain flow field parameters and vibration parameters of the turbine.

[0049] It should be noted that thermodynamic finite element is a numerical calculation method that applies finite element analysis to thermodynamic problems. It is mainly used to analyze the temperature field, heat flux, stress and deformation of solid materials or structures under thermal effects; multi-body dynamics finite element combines finite element analysis with multi-body dynamics to study dynamic systems in which multiple rigid or flexible bodies interact with each other in motion.

[0050] Specifically, a three-dimensional simulation model is established to obtain the flow field parameters and vibration parameters of the turbine, wherein the flow field parameter is the temperature of the physical field inside the turbine cooler.

[0051] (5) A turbine cooler simulation model is established based on the one-dimensional simulation model and the three-dimensional simulation model.

[0052] It should be noted that the jointly established turbine cooler simulation model can use the one-dimensional simulation model to obtain the basic performance parameters of the turbine, and can also use the three-dimensional simulation model to obtain the flow field parameters and vibration parameters of the turbine, so as to achieve a complementary effect and obtain more complete monitoring parameter information during the operation of the turbine cooler. In addition, after the turbine cooler simulation model is jointly established according to the one-dimensional simulation model and the three-dimensional simulation model, it includes: using the output of the one-dimensional simulation model as the input of the three-dimensional simulation model to determine the boundary conditions of the three-dimensional simulation model; using the output of the three-dimensional simulation model as the verification and calibration signal of the one-dimensional simulation model to correct the calculation results of the one-dimensional simulation model; and performing simulation operations under various working conditions based on the iteratively trained turbine cooler simulation model to determine the monitoring parameters.

[0053] Specifically, the one-dimensional simulation model can quickly calculate the basic performance parameters of the turbine cooler, such as pressure change, airflow velocity, etc., based on the thermodynamic formula of the turbine cooler and the simplified aerodynamic theory. The model is mainly used to predict the overall trend of the turbine cooler and the large-scale operating response. The three-dimensional simulation model obtains the complex flow field and vibration response inside the turbine through detailed analysis of thermodynamic finite elements and multi-body dynamics finite elements, covering the microscopic distribution of airflow, flow field disturbances and vibration characteristics. By using the output results of the one-dimensional simulation model, such as macro parameters such as pressure and airflow velocity, as the boundary conditions or input parameters of the three-dimensional simulation model, the flow field and vibration response inside the turbine can be refined; and the detailed flow field and vibration analysis results of the three-dimensional simulation model, such as the local temperature field, airflow velocity field, vibration parameters, etc., as the input of the one-dimensional simulation model, can be used to optimize the system characteristic curve and correct the simplified assumptions of the one-dimensional simulation model, so that the results of the one-dimensional simulation model are more accurate. Therefore, the one-dimensional simulation model and the three-dimensional simulation model are jointly operated to form a simulation system for the entire turbine cooler. The one-dimensional model is used to quickly predict the macroscopic working conditions, and the three-dimensional model is used to refine the local analysis. Through the iterative optimization of the two, accurate simulation can be performed under different working conditions. And the joint model can call the one-dimensional and three-dimensional simulation modules respectively according to different operating conditions to achieve a comprehensive analysis of the performance of the turbine cooler.

[0054] S102: Simulate each failure mode based on the turbine cooler simulation model to obtain failure data under each failure mode.

[0055] Specifically, simulating various fault modes based on the turbine cooler simulation model to obtain fault data under various fault modes includes:

[0056] (1) Based on FMMEA analysis of turbine cooler, typical failure modes are determined.

[0057] It should be noted that FMMEA (Failure Modes, Mechanisms, and Effects Analysis) is a systematic and structured analysis method, which is mainly used to identify potential failure modes, failure mechanisms and their impact on system functions in equipment or systems. For the fault analysis of turbine coolers, FMMEA can help determine the types of faults, causes of faults and their impacts that may occur in turbine coolers under different operating conditions, thereby providing a basis for real-time monitoring and fault diagnosis. For example, in one embodiment, through FMMEA analysis, it is determined that the typical failure modes of turbine coolers may be blade wear, bearing wear, fatigue damage, fracture and jamming, etc. Specifically, the causes of these failure modes are mostly fatigue failures caused by installation accuracy, material organization, sudden operating conditions, etc.

[0058] (2) Determining monitoring parameters based on relevant factors affecting turbine cooler fault diagnosis; the relevant factors at least include fault mechanism and sensor layout.

[0059] It should be noted that, in order to determine the monitoring parameters according to the relevant factors affecting the fault diagnosis of the turbine cooler, it is necessary to comprehensively consider the fault mechanism, sensor layout, and actual installation constraints (such as space limitations, environmental tolerance, and maintenance convenience, etc.), and select reasonable and effective monitoring parameters and corresponding sensor types. For example, in one embodiment, the determined monitoring parameters may be fan / compressor inlet pressure, fan / compressor inlet temperature, fan / compressor outlet pressure, fan / compressor outlet temperature, turbine inlet pressure, turbine inlet temperature, turbine outlet pressure, turbine outlet temperature, shaft speed, bearing housing vibration, turbine end bearing temperature, and fan / compressor bearing temperature. Among them, in this embodiment, the inlet parameters are mainly used to control and adjust the working state of the turbine cooler, and the remaining parameters (such as turbine end outlet temperature, bearing vibration and temperature, etc.) are used for real-time status monitoring, so as to detect faults or abnormal conditions in a timely manner.

[0060] (3) Modifying the monitoring parameters or characteristic curves of the turbine cooler simulation model according to the typical failure modes and the monitoring parameters, and calculating the failure data under each typical failure mode by changing the operating input conditions of the modified simulation model.

[0061] First, typical fault modes are identified, and then, according to each fault mode, the monitoring parameters related thereto are analyzed, and then the monitoring parameters or characteristic curves in the turbine cooler simulation model are adjusted according to the analysis results. Specifically, for example, in one embodiment, the typical fault mode identified is turbine bearing wear. At this time, it can be known through analysis that turbine bearing wear will cause vibration and heat, and it is necessary to monitor the bearing vibration and temperature, and then replace or add the bearing vibration and temperature to the turbine cooler simulation model to ensure that the model can reflect the changes in these parameters under the fault mode, or, according to the influence of the fault mode, modify the characteristic curve to more accurately describe the performance changes of the turbine cooler under the fault state. By simulating with the turbine cooler simulation model after modifying the parameters, the fault data under each typical fault mode can be obtained.

[0062] S103. Determine a status monitoring parameter of the turbine cooler according to the fault data.

[0063] It should be noted that since there are a lot of noise, abnormalities, and inconsistent data in the original fault data of the turbine cooler, which seriously affects the accuracy of fault diagnosis, it is very necessary to preprocess the data. Specifically, preprocessing starts from the perspectives of specific tasks, system boundaries, parameter characteristics, noise types, etc., to filter out redundant data that is not related to data mining, improve data quality, and improve the efficiency of data feature extraction.

[0064] After preprocessing, the condition monitoring parameters of the turbine cooler can be determined by performing sensitivity analysis on all parameters in the fault data and deleting insensitive parameters to reduce the parameter dimension. Furthermore, when performing parameter sensitivity analysis, a rough analysis of fault sensitive parameters or fault mode discrimination can be performed when new measurement points are added or new faults occur, so as to find out which faults are easier to distinguish in which parameter spaces.

[0065] Specifically, determining the state monitoring parameters of the turbine cooler according to the fault data includes:

[0066] (1) Construct a CART decision tree model based on the fault data and the extreme random tree algorithm.

[0067] It should be noted that the CART (Classification and Regression Trees) decision tree model is a decision tree algorithm for classification and regression. It models data by creating a tree structure to help identify the relationship between input variables and target variables.

[0068] It should be noted that the extreme random tree is composed of multiple decision trees, and the decision tree is composed of nodes and directed edges. Among them, the leaf node is the fault mode category label, and the non-leaf node is the evaluation condition. During classification, a turbine cooler input data with an unknown category label is matched with the nodes of the tree structure to finally find the only leaf node, and the category label of the leaf node is the fault category corresponding to the data. Specifically, the construction of the extreme random tree is divided into three steps: parameter selection, decision tree generation, and decision tree pruning.

[0069] Specifically, the process of building a decision tree is to summarize a set of classification rules from the training set so that it has a small conflict with the training data and has a strong generalization ability. It is basically divided into the following steps:

[0070] ① Calculate the information entropy of the fault data set before partitioning;

[0071] ② Traverse all parameters used as partitioning conditions and calculate the information entropy after the fault data set is partitioned according to each parameter;

[0072] ③ Select the parameter with the largest information gain and use this feature as the data partitioning node to partition the data;

[0073] ④ Recursively process all the divided sub-data sets, repeat the above steps from the parameters that have not been selected, and select the optimal data partitioning parameters to divide the sub-data sets.

[0074] It should be noted that there are generally two conditions for the recursion to end: one is that all parameters have been used up, and the other is that the information entropy gain after partitioning is small enough. For this stop condition, it is necessary to implement the threshold value of the information gain as the recursive end condition.

[0075] When generating a decision tree, too much consideration is given to how to improve the correct fault mode classification of the training data, thus constructing an overly complex decision tree, which often accurately classifies the training data, but the classification accuracy of unknown test data decreases, resulting in overfitting. The solution to overfitting is to simplify the decision tree, that is, to cut off some subtrees or leaf nodes in the generated tree, so that the parent node of the subtree or leaf node becomes a new leaf node, thereby simplifying the classification tree model.

[0076] It should also be noted that decision tree pruning is often achieved by minimizing the overall loss function or cost function of the decision tree. Minimizing the loss function during pruning is equivalent to the maximum likelihood estimation of the regularization. Select the best subtree T.

[0077] In addition, before constructing the CART decision tree model according to the fault data and the extreme random tree algorithm, it also includes standardization processing for the fault sample data.

[0078] (2) Based on the CART decision tree model and using the out-of-bag data, a first out-of-bag data error and a second out-of-bag data error are calculated; the second out-of-bag data error is an error calculated after randomly adding noise interference to the out-of-bag data.

[0079] Specifically, for each decision tree in the extreme random tree, the corresponding OOB (out-of-bag data) data is used to calculate its out-of-bag data error, recorded as errOOB1 (first out-of-bag data error), and noise interference is randomly added to the parameter X of all samples of the out-of-bag data OOB (which can randomly change the value of the sample at the parameter X), and its out-of-bag data error is calculated again, recorded as errOOB2 (second out-of-bag data error).

[0080] (3) Determine the sensitivity of each parameter in the fault data based on the calculation results.

[0081] Combined with the above example, if there are Ntree trees in the extreme random tree, then the importance of parameter X is ∑(errOOB2-errOOB1) / Ntree, which means that if noise is randomly added to a parameter, the out-of-bag accuracy is greatly reduced, which means that this parameter has a great influence on the classification result of the sample, that is, its importance is relatively high, and then based on the importance ranking, the sensitivity of each parameter in the fault data is determined. Among them, the higher the importance ranking, the more sensitive it is.

[0082] (4) Calculate the Spearman rank correlation coefficient between every two fault-sensitive parameters and output the results.

[0083] It should be noted that, based on the parameter sensitivity analysis, correlation analysis is generally used to eliminate some redundant information with low sensitivity and strong correlation to prevent model overfitting and increase the model training speed.

[0084] Since aviation electromechanical components belong to complex coupled systems, there are correlations between many parameters. In particular, the high correlation between some parameters presents redundancy at the data level. In addition, due to the ambiguity of the fault mechanism, the feature extraction method will generate a large number of fault-irrelevant features. By analyzing the correlation between features and faults, characteristic parameters related to the fault can also be screened out.

[0085] Specifically, the Spearman rank correlation coefficient can be used to calculate the correlation between fault-sensitive parameters, where the Spearman rank correlation coefficient is used to estimate the correlation between two variables X and Y, where the correlation between variables can be described by a monotonic function. If there are no identical elements in the two sets of values ​​of the two variables, then when one of the variables can be expressed as a good monotonic function of the other variable (that is, the change trend of the two variables is the same), the ρ between the two variables can reach +1 or -1. Assume that there are two random variables X and Y, and their number of elements is n. The two random variables take the i-th (1<=i<=n) value and are represented by Xi and Yi respectively. Sort X and Y (in ascending or descending order at the same time) to obtain two element ranking sets x and y, where the elements xi and yi are the ranking of Xi in X and the ranking of Yi in Y respectively. Subtract the corresponding elements in the sets x and y to obtain a ranking difference set d, where di=xi-yi, 1<=i<=n. The Spearman rank correlation coefficient between random variables X and Y can be calculated using x, y or d, and is calculated as follows:

[0086] ;

[0087] (5) According to the output result, retain a set of features with the highest correlation coefficient from the multiple sets of features, delete other less relevant features, and obtain the turbine cooler state monitoring parameters.

[0088] For example, in one embodiment, the turbine cooler status monitoring parameters finally obtained may be turbine outlet pressure, turbine outlet temperature, fan / compressor outlet pressure, fan / compressor outlet temperature, shaft speed, bearing housing vibration, turbine end bearing temperature, and fan / compressor bearing temperature.

[0089] S104: Establish a probabilistic neural network candidate model set based on the state monitoring parameters.

[0090] It should be noted that the process of establishing a probabilistic neural network candidate model set based on condition monitoring parameters can be as follows: normalizing the condition monitoring parameters to ensure the consistency of dimensions between different features and improve the model training effect; selecting key features from the condition monitoring parameters according to the needs of fault diagnosis to reduce model complexity and improve diagnostic accuracy; designing the structures of multiple probabilistic neural network models, including the number of nodes in the input layer, pattern layer, summation layer and output layer and the selection of activation functions; using historical data (including monitoring parameters under normal operation and fault conditions) to train each model to learn the relationship between input and output; evaluating the performance of each model through methods such as cross-validation, and selecting the best performing model as part of the candidate model set; and combining multiple verified models together to form the final probabilistic neural network candidate model set.

[0091] Specifically, the probabilistic neural network candidate model set includes multiple probabilistic neural network models. Figure 2 For the structural diagram of the probabilistic neural network model provided in this application, please refer to Figure 2 The probabilistic neural network model includes an input layer, a pattern layer, a summation layer and an output layer; wherein the workflow of the probabilistic neural network model is as follows:

[0092] The input layer receives the characteristic input vector of each state monitoring parameter and performs normalization processing;

[0093] The pattern layer receives the normalized learning samples, calculates the distance between each sample in the sample matrix to be identified and various patterns in the learning matrix, and activates radial basis function neurons; the learning samples are obtained based on the historical operation data and fault data of the turbine cooler, and the sample matrix to be identified is the monitoring parameter sample of the current turbine cooler;

[0094] The summation layer obtains the initial probability sum of each sample in the sample matrix to be identified being classified into various categories; the categories include different fault types or normal operating states;

[0095] The output layer calculates and outputs the initial probability of the current sample according to different samples to be identified.

[0096] In combination with the above content, it should also be noted that the status monitoring data of the turbine cooler is a typical time series data. When the turbine cooler has faults such as wear, leakage, and jamming, its fault information will be reflected in certain feature quantities in the time series data. If the feature quantities related to the fault can be extracted, it will play an important role in improving the accuracy of fault diagnosis. The method of fault feature extraction generally adopts knowledge-based methods and data-driven methods. The knowledge-based method is based on a clear understanding of the fault mechanism and clearly understands which features will be clearly manifested after the fault occurs; the data-driven method is based on a large amount of fault sample data, analyzing the sensitivity of various data features and faults, and screening out sensitive features of the fault.

[0097] The research on condition monitoring parameter feature extraction focuses on extracting the fault information contained in the original multivariate time series condition monitoring parameters through various time domain, frequency domain, and time-frequency domain feature extraction methods, and maximizing the differences between different faults from the multi-fault feature space. As a typical rotating machinery, the vibration signal of the turbine cooler can timely reflect its internal state changes. Once the turbine cooler fails, it will cause changes in the signal energy in different frequency bands. Therefore, the components of each frequency band are obtained through wavelet packet decomposition, and the energy of each component signal is monitored and calculated to obtain the entropy value as a fault judgment feature, which can significantly improve the accuracy of the subsequent diagnosis model.

[0098] Figure 3 This is a flowchart showing the extraction of state detection parameter characteristics and information fusion according to an exemplary embodiment of the present application. Figure 3 Specifically, the method for obtaining the characteristic input vector can be as follows: performing recursive average filtering calculation on the temperature and pressure signals in the state monitoring parameters to obtain the average value and effective value after denoising; performing wavelet packet decomposition on the vibration signal in the state monitoring parameter, and calculating the kurtosis value and normalized energy of each decomposition component, and calculating the overall energy entropy of the kurtosis value and normalized energy; fusing the average value and effective value of the temperature and pressure signals after denoising with the kurtosis value, normalized energy and overall energy entropy of the wavelet packet decomposition component of the vibration signal to form a characteristic input vector.

[0099] The energy entropy is calculated as follows: Assuming that the decomposed components are , contains different frequency components and has different energies , the energy entropy is:

[0100] ;

[0101] in, Indicates the energy of the i-th wavelet packet decomposition component in the total energy % of.

[0102] S105, forming a fault diagnosis model set based on the probabilistic neural network candidate model set and operating condition parameters, and calculating a matching matrix based on the operating condition parameters and the fault diagnosis model set.

[0103] It should be noted that forming a fault diagnosis model set based on the probabilistic neural network candidate model set and the operating parameters, and calculating a matching matrix based on the operating parameters and the fault diagnosis model set include:

[0104] (1) Analyze the performance and functional characteristics of the turbine cooler to obtain operating condition information, and determine operating condition parameters based on the operating condition information.

[0105] It should be noted that the performance and functional characteristics of the turbine cooler can be analyzed through experiments, simulations and related data analysis, the key factors affecting the operation of the turbine cooler can be identified, the various operating conditions of the system can be determined, and then the operating parameters such as temperature, pressure, flow rate, etc. can be determined.

[0106] (2) Training each model in the probabilistic neural network candidate model set based on the operating condition parameters to form a fault diagnosis model set.

[0107] The collected monitoring data and corresponding working condition parameters are input into the probabilistic neural network candidate model for training. Through repeated training and verification of the model, a fault diagnosis model set for specific working conditions is formed.

[0108] (3) The operating condition parameters are formed into an operating condition parameter vector, which is input into the fault diagnosis model set, and the correlation between the operating condition parameter vector and the fault diagnosis model set is calculated to obtain a matching matrix.

[0109] After extracting the characteristics of the operating parameters and converting them into vector format, they are input into the fault diagnosis model set. By calculating the correlation between these parameter vectors and the output of each fault diagnosis model, the corresponding relationship between the operating parameters and the fault diagnosis model is obtained to form a matching matrix.

[0110] S106 , determining a target fault diagnosis model from the fault diagnosis model set based on the matching matrix, and performing fault diagnosis based on the target fault diagnosis model and the real-time fault condition.

[0111] It should be noted that determining a target fault diagnosis model from the fault diagnosis model set based on the matching matrix, and performing fault diagnosis based on the target fault diagnosis model and the real-time fault condition include:

[0112] (1) Determine the current real-time operating condition information according to the current turbine inlet parameters and the turbine cooler simulation model.

[0113] It should be noted that by real-time monitoring of the inlet parameters of the turbine cooler (such as pressure, temperature, flow rate, etc.) and combining the simulation model of the turbine cooler (according to the above formula), the current operating condition information can be calculated.

[0114] (2) Based on the real-time operating condition information, obtain the real-time fault operating condition parameters that best match the operating condition.

[0115] By analyzing the real-time operating condition information and combining it with the previously generated fault diagnosis model and matching matrix, the fault condition parameters that best match the current operating condition are extracted.

[0116] (3) Determine a target fault diagnosis model according to the corresponding relationship between the real-time fault condition parameters and the fault diagnosis model in the matching matrix, and perform fault diagnosis based on the determined fault diagnosis model.

[0117] Using the corresponding relationship recorded in the matching matrix, the real-time fault condition parameters are compared with the fault diagnosis model, and the optimal target fault diagnosis model is selected. Then, the model is used to analyze the current monitoring data to determine whether there is a fault in the turbine cooler and the type of fault.

[0118] The real-time fault diagnosis method for a turbine cooler provided in this embodiment establishes a turbine cooler simulation model including a one-dimensional simulation model and a three-dimensional simulation model, and uses the two models to simulate the links they are best at respectively. The output signal obtained by the one-dimensional model can be used as the input of the three-dimensional model, and the output of the three-dimensional model can optimize the one-dimensional model. The output signals of the two models are complementary, and the model with the most accurate simulation simulates the signal of a specific environment and a specific part. When the accuracy of the two is the same and the output is consistent, the corresponding model can be selected in a manner with the shortest calculation time and the lowest calculation amount, which can more comprehensively reflect the behavior of the turbine cooler under various operating conditions and fault modes, obtain more comprehensive data, and simulate the behavior of the turbine cooler in a high-precision and low-computation manner. Provide a basis for subsequent analysis; by simulating various fault modes, rich fault data can be obtained, reducing the cost and risk of fault testing, and more sensor information can be obtained, which is helpful to deeply understand the performance of the turbine cooler in different fault modes and provide support for subsequent fault analysis modeling; then, based on the fault data, sensitivity analysis and correlation analysis are performed to identify the key parameters of the turbine cooler that can be used to characterize the health state, support sensor layout optimization, and reduce environmental interference through signal feature extraction methods, and adopt multi-source information fusion methods to improve diagnostic accuracy; establish a probabilistic neural network candidate model set and a fault diagnosis model set, and consider the differences in fault distribution under different working conditions, and set up multiple diagnostic models for each working condition and fault. Each time, according to the working condition parameters, monitoring parameters and fault characteristics, the matching matrix is ​​calculated with the correlation between each model to screen out the optimal fault diagnosis model, rather than just judging by the accuracy of a single fault identification model. By matching the current fault characteristics with the best identification model in real time, it can better adapt to different working conditions, determine the target fault diagnosis model based on the matching matrix, select the most suitable model from multiple fault diagnosis models, and perform fault diagnosis according to real-time working conditions under variable working conditions, flexibly adjust the fault handling strategy, and improve maintenance efficiency and accuracy. In addition, based on the probabilistic neural network model combined with feature extraction, accurate fault identification and location can be achieved in the case of small sample data, and maintenance or replacement strategies can be formulated according to the health status of each component to ensure the reliable operation of the turbine cooler. This application implements the fault diagnosis modeling process of the turbine cooler from fault mode analysis, joint fault simulation modeling, state parameter identification, deep feature extraction and time-varying working conditions, and establishes a real-time fault diagnosis method with high diagnostic accuracy. It can be used for real-time monitoring and fault diagnosis of the health status of the turbine cooler during use and operation and maintenance, and supports rapid fault location and maintenance guarantee of the turbine cooler.

[0119] Corresponding to the aforementioned embodiment of a turbine cooler real-time fault diagnosis method, the present application also provides an embodiment of a turbine cooler real-time fault diagnosis device.

[0120] Figure 4This is a schematic diagram of the structure of the first embodiment of the turbine cooler real-time fault diagnosis device provided by this application. Figure 4 The device provided in this embodiment includes an establishment module 410, a simulation module 420, a determination module 430, a calculation module 440 and a diagnosis module 450; wherein,

[0121] The establishing module 410 is used to establish a turbine cooler simulation model, and the turbine cooler simulation model includes a one-dimensional simulation model and a three-dimensional simulation model;

[0122] The simulation module 420 is used to simulate various fault modes based on the turbine cooler simulation model to obtain fault data under various fault modes;

[0123] The determination module 430 is used to determine the state monitoring parameters of the turbine cooler according to the fault data;

[0124] The establishment module 410 is also used to establish a probabilistic neural network candidate model set based on the state monitoring parameters;

[0125] The calculation module 440 is used to form a fault diagnosis model set based on the probabilistic neural network candidate model set and the operating parameters, and calculate a matching matrix based on the operating parameters and the fault diagnosis model set;

[0126] The diagnosis module 450 is used to determine a target fault diagnosis model from the fault diagnosis model set based on the matching matrix, and perform fault diagnosis based on the target fault diagnosis model and the real-time fault condition.

[0127] The device of this embodiment can be used to perform Figure 1 The steps, specific implementation principles and implementation processes of the method embodiment shown are similar and will not be repeated here.

[0128] The implementation process of the functions and effects of each unit in the above-mentioned device is specifically described in the implementation process of the corresponding steps in the above-mentioned method, and will not be repeated here.

[0129] For the device embodiment, since it basically corresponds to the method embodiment, the relevant parts can refer to the partial description of the method embodiment. The device embodiment described above is only schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the present application scheme. A person of ordinary skill in the art can understand and implement it without paying any creative work.

[0130] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application shall be included in the scope of protection of the present application.

Claims

1. A real-time fault diagnosis method for a turbine cooler, characterized in that: The method comprises: Establishing a turbine cooler simulation model, wherein the turbine cooler simulation model includes a one-dimensional simulation model and a three-dimensional simulation model; Simulating various fault modes based on the turbine cooler simulation model to obtain fault data under various fault modes; determining a condition monitoring parameter of the turbine cooler according to the fault data; Establishing a probabilistic neural network candidate model set based on the condition monitoring parameters; Forming a fault diagnosis model set based on the probabilistic neural network candidate model set and operating condition parameters, and calculating a matching matrix based on the operating condition parameters and the fault diagnosis model set; Determine a target fault diagnosis model from the fault diagnosis model set based on the matching matrix, and perform fault diagnosis based on the target fault diagnosis model and the real-time fault condition; The step of establishing a turbine cooler simulation model comprises: A one-dimensional simulation model is established based on the working principle and thermodynamic formula of the turbo cooler; Determining mathematical function forms of the fan / compressor end and the turbine end based on the one-dimensional simulation model; Determine basic turbine performance parameters according to the mathematical function form; A three-dimensional simulation model is established based on thermodynamic finite elements and multi-body dynamics finite elements; the three-dimensional simulation model is used to obtain flow field parameters and vibration parameters of the turbine; Establishing a turbine cooler simulation model based on the one-dimensional simulation model and the three-dimensional simulation model; After the turbine cooler simulation model is jointly established according to the one-dimensional simulation model and the three-dimensional simulation model, the method includes: Using the output of the one-dimensional simulation model as the input of the three-dimensional simulation model to determine the boundary conditions of the three-dimensional simulation model; Using the output of the three-dimensional simulation model as a verification and calibration signal of the one-dimensional simulation model to correct the calculation results of the one-dimensional simulation model; Based on the iteratively trained turbine cooler simulation model, simulated operation under various working conditions is carried out to determine the monitoring parameters.

2. The method according to claim 1, characterized in that: The forming of a fault diagnosis model set based on the probabilistic neural network candidate model set and operating condition parameters, and calculating a matching matrix based on the operating condition parameters and the fault diagnosis model set, includes: Analyze the performance and functional characteristics of the turbine cooler to obtain operating condition information, and determine operating condition parameters based on the operating condition information; Training each model in the probabilistic neural network candidate model set based on the operating condition parameters to form a fault diagnosis model set; The operating condition parameters are formed into an operating condition parameter vector, which is input into the fault diagnosis model set, and the correlation between the operating condition parameter vector and the fault diagnosis model set is calculated to obtain a matching matrix.

3. The method according to claim 1, characterized in that The simulating each fault mode based on the turbine cooler simulation model to obtain fault data under each fault mode includes: Determine typical failure modes based on FMMEA analysis of turbine coolers; Determining monitoring parameters according to relevant factors affecting fault diagnosis of the turbine cooler; the relevant factors at least include fault mechanism and sensor layout; The monitoring parameters or characteristic curves of the turbine cooler simulation model are modified according to the typical failure modes and the monitoring parameters, and the failure data under each typical failure mode is calculated by changing the operating input conditions of the modified simulation model.

4. The method according to claim 1, characterized in that: Determining the state monitoring parameter of the turbine cooler according to the fault data includes: Constructing a CART decision tree model based on the fault data and the extreme random tree algorithm; Based on the CART decision tree model and using the out-of-bag data, a first out-of-bag data error and a second out-of-bag data error are calculated; the second out-of-bag data error is an error calculated after randomly adding noise interference to the out-of-bag data; Determine the sensitivity of each parameter in the fault data based on the calculation results; Calculate the Spearman rank correlation coefficient between every two fault-sensitive parameters and output the result; According to the output results, a set of features with the highest correlation coefficient is retained from the multiple sets of features, and other less relevant features are deleted to obtain turbine cooler state monitoring parameters.

5. The method according to claim 1, characterized in that: The probabilistic neural network model includes an input layer, a pattern layer, a summation layer and an output layer; wherein the workflow of the probabilistic neural network model is as follows: The input layer receives the characteristic input vector of each state monitoring parameter and performs normalization processing; The pattern layer receives the normalized learning samples, calculates the distance between each sample in the sample matrix to be identified and various patterns in the learning matrix, and activates radial basis function neurons; the learning samples are obtained based on the historical operation data and fault data of the turbine cooler, and the sample matrix to be identified is the monitoring parameter sample of the current turbine cooler; The summation layer obtains the initial probability sum of each sample in the sample matrix to be identified being classified into various categories; the categories include different fault types or normal operating states; The output layer calculates and outputs the initial probability of the current sample according to different samples to be identified.

6. The method according to claim 1, characterized in that The step of determining a target fault diagnosis model from the fault diagnosis model set based on the matching matrix, and performing fault diagnosis based on the target fault diagnosis model and the real-time fault condition includes: Determine the current real-time operating condition information according to the current turbine inlet parameters and the turbine cooler simulation model; Based on the real-time operating condition information, obtaining a real-time fault operating condition parameter that best matches the operating condition; According to the corresponding relationship between the real-time fault condition parameters and the fault diagnosis model in the matching matrix, a target fault diagnosis model is determined, and fault diagnosis is performed based on the determined fault diagnosis model.

7. The method according to claim 5, characterized in that The method for obtaining the feature input vector is: Perform recursive average filtering calculation on the temperature and pressure signals in the state monitoring parameters to obtain the average value and effective value after denoising; Performing wavelet packet decomposition on the vibration signal in the state monitoring parameter, calculating the kurtosis value and normalized energy of each decomposition component, and obtaining the overall energy entropy of the kurtosis value and normalized energy; The denoised average and effective values ​​of the temperature and pressure signals are fused with the kurtosis value, normalized energy and overall energy entropy of the wavelet packet decomposition components of the vibration signal. Constitute the feature input vector.

8. A real-time fault diagnosis device for a turbine cooler, characterized in that: The device includes a building module, a simulation module, a determination module, a calculation module and a diagnosis module; wherein, The establishing module is used to establish a turbine cooler simulation model, wherein the turbine cooler simulation model includes a one-dimensional simulation model and a three-dimensional simulation model; The simulation module is used to simulate various fault modes based on the turbine cooler simulation model to obtain fault data under various fault modes; The determination module is used to determine the state monitoring parameters of the turbine cooler according to the fault data; The establishment module is further used to establish a probabilistic neural network candidate model set based on the state monitoring parameters; The calculation module is used to form a fault diagnosis model set based on the probabilistic neural network candidate model set and operating parameters, and calculate a matching matrix based on the operating parameters and the fault diagnosis model set; The diagnosis module is used to determine a target fault diagnosis model from the fault diagnosis model set based on the matching matrix, and perform fault diagnosis based on the target fault diagnosis model and the real-time fault condition; The step of establishing a turbine cooler simulation model comprises: A one-dimensional simulation model is established based on the working principle and thermodynamic formula of the turbo cooler; Determining mathematical function forms of the fan / compressor end and the turbine end based on the one-dimensional simulation model; Determine basic turbine performance parameters according to the mathematical function form; A three-dimensional simulation model is established based on thermodynamic finite elements and multi-body dynamics finite elements; the three-dimensional simulation model is used to obtain flow field parameters and vibration parameters of the turbine; Establishing a turbine cooler simulation model based on the one-dimensional simulation model and the three-dimensional simulation model; After the turbine cooler simulation model is jointly established according to the one-dimensional simulation model and the three-dimensional simulation model, the method includes: Using the output of the one-dimensional simulation model as the input of the three-dimensional simulation model to determine the boundary conditions of the three-dimensional simulation model; Using the output of the three-dimensional simulation model as a verification and calibration signal of the one-dimensional simulation model to correct the calculation results of the one-dimensional simulation model; Based on the iteratively trained turbine cooler simulation model, simulated operation under various working conditions is carried out to determine the monitoring parameters.

Images