Method and device for detecting gas path faults in a combined power unit

Abstract

The present application provides a method and device for detecting gas path faults in a combined power device. The method provided by the present application includes: obtaining the measurement parameters at the current moment and the state parameters at the previous moment; inputting the state parameters at the previous moment into the simulation model corresponding to the combined power device to obtain the predicted measurement parameters at the previous moment; determining the Kalman gain matrix of the Kalman filter at the current moment according to the pre-constructed state covariance matrix of the Kalman filter at the previous moment, the measurement noise matrix at the current moment, the system noise matrix, and the combined Jacobian matrix corresponding to the simulation model; the combined Jacobian matrix is ​​obtained based on the initial Jacobian matrix of the simulation model and the initial Jacobian matrix of the prediction model; based on the Kalman filter, the health parameters are calculated according to the health parameters at the previous moment, the measurement parameters at the current moment, the predicted measurement parameters at the previous moment, and the Kalman gain matrix at the current moment.

Description

Technical Field

[0001] The present application relates to the technical field of fault detection, and in particular to a method and device for detecting gas path faults in a combined power device. Background Art

[0002] Under the development trend of realizing comprehensive integration of aircraft energy, function, physics and control, lightweight and integration of aircraft are very critical. As the most complex system on the aircraft, the advancement of electromechanical system directly affects the overall performance of the aircraft.

[0003] As a device that integrates aircraft auxiliary power, emergency power, and thermal management systems, the combined power unit (thermal management combined power unit) is increasingly used in aircraft because it can reduce the overall cost of the aircraft, improve hardware utilization, and improve the overall performance of the aircraft. However, during the use and maintenance phases of the combined power unit, it is very important to monitor the health status and diagnose faults of the device. Among them, the diagnosis of gas path faults inside the device is an important part of its health management.

[0004] Compared with traditional power component circuits, combined power devices have high system integration, high nonlinearity, strong coupling, a large number of circuit components, and many related physical quantities. Therefore, it is more difficult to accurately locate a fault when it occurs. Summary of the invention

[0005] In view of this, the present application provides a method and device for performing gas path fault detection on a combined power device, so as to accurately perform gas path fault detection on the combined power device.

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

[0007] In a first aspect, the present application provides a method for detecting a gas path fault in a combined power device, the method comprising:

[0008] Acquire the measurement parameters at the current moment and the state parameters at the previous moment; wherein the state parameters include the specified parameters and the health parameters related to the measurement parameters; the measurement parameters are the parameters related to the gas path performance among the parameters of the compressor, combustion chamber, power turbine, and cooling turbine in the combined power device;

[0009] Inputting the state parameters of the previous moment into the simulation model corresponding to the combined power device, so that the simulation model outputs the predicted measurement parameters of the previous moment according to the state parameters of the previous moment; wherein the simulation model is used to characterize the correlation between the state parameters and the measurement parameters;

[0010] The Kalman gain matrix of the Kalman filter at the current moment is determined according to the state covariance matrix of the Kalman filter at the previous moment, the measurement noise matrix of the Kalman filter at the current moment, the system noise matrix of the Kalman filter at the current moment, and the combined Jacobian matrix corresponding to the simulation model; wherein the combined Jacobian matrix corresponding to the simulation model is obtained based on the first initial Jacobian matrix of the simulation model and the second initial Jacobian matrix of the pre-trained prediction model; the pre-trained prediction model is used to predict the corresponding health parameters based on the input measurement parameters; the prediction model is trained based on a preset number of groups of sample data, and each group of sample data includes actual measurement parameters and actual health parameters at any moment;

[0011] Based on the Kalman filter, the health parameter at the current moment is calculated according to the health parameter at the previous moment, the measurement parameter at the current moment, the predicted measurement parameter at the previous moment and the Kalman gain matrix at the current moment, so as to perform gas path fault detection on the combined power device according to the health parameter at the current moment;

[0012] The state covariance matrix at the current moment is updated according to the Kalman gain matrix at the current moment, the combined Jacobian matrix, and the state covariance matrix at the previous moment.

[0013] A second aspect of the present application provides a device for detecting gas path faults in a combined power device, the device comprising an acquisition module, a processing module, a determination module, a calculation module and an update module, wherein:

[0014] The acquisition module is used to acquire the measurement parameters at the current moment and the state parameters at the previous moment; wherein the state parameters include specified parameters and health parameters related to the measurement parameters; the measurement parameters are parameters related to gas path performance among various parameters of the compressor, combustion chamber, power turbine, and cooling turbine in the combined power device;

[0015] The processing module is used to input the state parameters of the previous moment into the simulation model corresponding to the combined power device, so that the simulation model outputs the predicted measurement parameters of the previous moment according to the state parameters of the previous moment; wherein the simulation model is used to characterize the correlation between the state parameters and the measurement parameters;

[0016] The determination module is used to determine the Kalman gain matrix of the Kalman filter at the current moment according to the pre-constructed state covariance matrix of the Kalman filter at the previous moment, the measurement noise matrix of the Kalman filter at the current moment, the system noise matrix of the Kalman filter at the current moment, and the combined Jacobian matrix corresponding to the simulation model; wherein the combined Jacobian matrix corresponding to the simulation model is obtained based on the first initial Jacobian matrix of the simulation model and the second initial Jacobian matrix of the pre-trained prediction model; the pre-trained prediction model is used to predict the corresponding health parameters based on the input measurement parameters; the prediction model is trained based on a preset number of groups of sample data, and each group of sample data includes actual measurement parameters and actual health parameters at any moment;

[0017] The calculation module is used to calculate the health parameter at the current moment based on the Kalman filter according to the health parameter at the previous moment, the measurement parameter at the current moment, the predicted measurement parameter at the previous moment and the Kalman gain matrix at the current moment, so as to perform gas path fault detection on the combined power device according to the health parameter at the current moment;

[0018] The updating module is used to update the state covariance matrix at the current moment according to the Kalman gain matrix at the current moment, the combined Jacobian matrix, and the state covariance matrix at the previous moment.

[0019] The present application provides a method and device for detecting gas path faults in a combined power unit. On the one hand, the first initial Jacobian matrix of the joint simulation model and the second initial Jacobian matrix of the pre-trained prediction model are used to determine the combined Jacobian matrix of the simulation model, and then the Kalman gain matrix of the Kalman filter at the current moment is determined based on the combined Jacobian matrix. In this way, the advantages of the simulation model and the prediction model are maximized, the description ability of each model on the system state is optimized, and the dynamic changes in the complex system are captured more accurately, thereby improving the prediction accuracy and more accurately estimating the health parameters of the combined power unit. On the second hand, by jointly using the simulation model and the prediction model, accurate detection of gas path faults can be maintained in a complex and changeable operating environment, and the robustness to external disturbances can be increased. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 A flow chart of a method for detecting gas path faults in a combined power device according to a first embodiment of the present application;

[0021] Figure 2 A schematic diagram of a combined power device shown as an exemplary embodiment of the present application;

[0022] Figure 3A flow chart of a second embodiment of a method for detecting gas path faults in a combined power device provided by the present application;

[0023] Figure 4 This is a schematic diagram of a third embodiment of a method for detecting gas path faults in a combined power device according to an exemplary embodiment of the present application;

[0024] Figure 5 This is a diagram showing experimental results of a verification experiment according to an exemplary embodiment of the present application;

[0025] Figure 6 This is a structural schematic diagram of embodiment 1 of the device for detecting gas path faults in a combined power 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] Figure 1 This is a flow chart of the first embodiment of the method for detecting gas path faults in a combined power device provided by the present application. Figure 1 , the method provided in this embodiment may include:

[0031] S101. Obtain the measurement parameters at the current moment and the state parameters at the previous moment; the state parameters include specified parameters and health parameters related to the measurement parameters; the measurement parameters are parameters related to the gas path performance among various parameters of the compressor, combustion chamber, power turbine, and cooling turbine in the combined power device.

[0032] The method and device provided in the present application are specifically used for performing gas path fault detection on a combined power device, which is a thermal management type combined power device. Figure 2 This is a schematic diagram of a combined power device shown in an exemplary embodiment of the present application, please refer to Figure 2 The combined power unit is a thermal management combined power unit that integrates aircraft auxiliary power, emergency power and thermal management systems. It mainly includes a compressor, a combustion chamber, a power turbine, a cooling turbine, etc. Among them, the compressor is used to inhale and compress air to increase the air pressure, and provide high-pressure air for the combustion process in the combustion chamber during the operation stage to enhance the combustion efficiency; the combustion chamber is a chamber where fuel and compressed air are mixed and burned, and the chemical energy contained in the fuel is converted into thermal energy to produce high-temperature and high-pressure combustion gas; the power turbine is driven by the high-temperature and high-pressure combustion gas generated in the combustion chamber, and the thermal energy contained in the combustion gas is converted into mechanical energy, thereby driving the output shaft of the combined power unit, and other connected equipment is driven by the output shaft; the cooling turbine is used to absorb part of the combustion gas from the combined power unit or inhale external air to drive the blades of the cooling turbine to rotate, reduce the temperature of the combined power unit, and play a cooling role.

[0033] It should be noted that the state parameters and measurement parameters related to gas circuit fault detection are predetermined. It is understandable that in gas circuit fault detection, the main research targets are pipeline flow, component efficiency, and total pressure recovery coefficient. Therefore, in one possible implementation, eight health parameters of the combined power unit are finally determined, namely, compressor flow, compressor efficiency, combustion chamber combustion efficiency, combustion chamber total pressure recovery coefficient, power turbine flow, power turbine efficiency, cooling turbine flow, and cooling turbine efficiency. Furthermore, considering that the measurement parameters should be quantities that are easily measured by the sensor, eight measurement parameters were finally selected, namely, compressor outlet pressure, compressor outlet temperature, combustion chamber outlet pressure, combustion chamber outlet temperature, fuel flow, power turbine exhaust temperature, cooling turbine outlet flow, and cooling turbine outlet temperature.

[0034] Referring to the above description, it should be noted that the state parameters include specified parameters and health parameters related to the measurement parameters, wherein the specified parameters include shaft speed, outlet flow, outlet temperature, outlet pressure, and output work.

[0035] In this step, the measurement parameters at the current moment and the state parameters at the previous moment can be obtained, and the measurement parameters at the current moment are recorded as , the state parameter at the previous moment is recorded as .

[0036] S102. Input the state parameters of the previous moment into the simulation model corresponding to the combined power device, so that the simulation model outputs the predicted measurement parameters of the previous moment according to the state parameters of the previous moment; wherein the simulation model is used to characterize the correlation between the state parameters and the measurement parameters.

[0037] Specifically, the simulation model is pre-established, and the simulation model is specifically shown in the following formula:

[0038] ;

[0039] in, There are 8 measurement parameters, specifically, is the compressor outlet pressure, is the compressor outlet temperature, is the combustion chamber outlet pressure, is the combustion chamber outlet temperature, is the fuel flow rate, is the power turbine exhaust temperature, To cool the turbine outlet flow, To cool the turbine outlet temperature; There are 8 health parameters. To specify parameters, specifically, is the compressor flow, is the compressor efficiency, Combustion efficiency of the combustion chamber, is the total pressure recovery coefficient, is the power turbine flow, The power turbine efficiency, To cool the turbine flow, To cool the turbine efficiency, is the shaft speed, For environmental pressure, is the ambient temperature, is the overall output shaft power of the combined power unit, Heat exchange for environmental control.

[0040] Referring to the above formula, it can be understood that the simulation model is used to characterize the correlation between the state parameters and the measurement parameters.

[0041] Optionally, in a possible implementation, the process of establishing the simulation model may include the following steps:

[0042] (1) According to the characteristics of the compressor, the combustion chamber, the power turbine and the cooling turbine, a first mathematical model corresponding to the compressor, a second mathematical model corresponding to the combustion chamber, a third mathematical model corresponding to the power turbine and a fourth mathematical model corresponding to the cooling turbine are respectively established; wherein each mathematical model is used to characterize the correlation between some parameters of the measured parameters and some parameters of the health parameters.

[0043] Specifically, according to the working principle of the combined power device, the input and output interfaces corresponding to the health parameters and measurement parameters are added to the compressor, combustion chamber, power turbine model, and cooling turbine, and the corresponding mathematical models are established.

[0044] The first mathematical model corresponding to the compressor can be expressed in the following mathematical function form:

[0045] ;

[0046] in, is the compressor flow rate, is the compressor efficiency, For environmental pressure, is the ambient temperature, is the compressor pressure ratio, is the shaft speed, is the compressor outlet flow rate, is the compressor outlet temperature, is the outlet pressure of the compressor, is the work consumed by the compressor.

[0047] The second mathematical model corresponding to the combustion chamber can be expressed in the following mathematical function form:

[0048] ;

[0049] in, is the combustion efficiency of the combustion chamber, is the total pressure recovery coefficient, is the inlet pressure of the combustion chamber, is the combustion chamber inlet temperature, is the inlet flow rate of the combustion chamber, is the fuel quantity, is the outlet flow rate of the combustion chamber, is the combustion chamber outlet temperature, is the outlet pressure of the combustion chamber.

[0050] The third mathematical model corresponding to the power turbine model can be expressed in the following mathematical function form:

[0051] ;

[0052] in, is the power turbine flow, is the power turbine efficiency, is the inlet pressure of the power turbine, is the inlet temperature of the power turbine, is the expansion ratio, is the power turbine outlet flow, is the outlet temperature of the power turbine, is the outlet pressure of the power turbine, is the output work of the power turbine, is the shaft speed.

[0053] The fourth mathematical model corresponding to the cooling turbine model can be expressed in the following mathematical function form:

[0054] ;

[0055] in, To cool the turbine flow, To cool the turbine efficiency, The inlet pressure of the cooling turbine, The inlet temperature of the cooling turbine is is the expansion ratio, To cool the turbine outlet flow, The outlet temperature of the cooling turbine is To cool the turbine outlet pressure, To cool the turbine output, is the shaft speed.

[0056] (2) According to the flow balance relationship, pressure balance relationship and energy balance relationship of the combined power device, the first mathematical model, the second mathematical model, the third mathematical model and the fourth mathematical model are integrated to obtain the simulation model.

[0057] Specifically, the flow balance relationship of the combined power unit is as follows:

[0058] Combustion chamber outlet flow = power turbine inlet flow;

[0059] Compressor outlet flow = cooling turbine inlet flow + combustion chamber inlet flow;

[0060] Furthermore, the pressure balance relationship is as follows:

[0061] Compressor outlet pressure = combustion chamber inlet pressure;

[0062] Compressor outlet pressure = cooling turbine inlet pressure + pressure loss;

[0063] Furthermore, the energy balance relationship is as follows:

[0064] Power generation = work done by power turbine + work done by cooling turbine - work done by compressor.

[0065] In this step, according to the above flow balance relationship, pressure balance relationship, and energy balance relationship, the first mathematical model, the second mathematical model, the third mathematical model, and the fourth mathematical model are integrated to establish a simulation model of the combined power device. The specific mathematical expression of the simulation model is shown above and will not be repeated here.

[0066] Referring to the foregoing introduction, it can be understood that the simulation model is used to characterize the correlation between the state parameters and the measurement parameters, and it can predict the measurement parameters according to the state parameters.

[0067] In this step, the state parameters of the previous moment are input into the simulation model, and the simulation model can output the predicted measurement parameters of the previous moment according to the state parameters of the previous moment. The details are as follows:

[0068] = ;

[0069] in, is the predicted measurement parameter for the previous moment;

[0070] is the health parameter in the state parameter of the previous moment;

[0071] It is the specified parameter in the state parameter of the previous moment.

[0072] S103. Determine the Kalman gain matrix of the Kalman filter at the current moment based on the pre-constructed state covariance matrix of the Kalman filter at the previous moment, the measurement noise matrix of the Kalman filter at the current moment, the system noise matrix of the Kalman filter at the current moment, and the combined Jacobian matrix corresponding to the simulation model; wherein the combined Jacobian matrix corresponding to the simulation model is obtained based on the first initial Jacobian matrix of the simulation model and the second initial Jacobian matrix of the pre-trained prediction model; the pre-trained prediction model is used to predict the corresponding health parameters based on the input measurement parameters; the prediction model is trained based on a preset number of groups of sample data, and each group of sample data includes actual measurement parameters and actual health parameters at any moment.

[0073] Specifically, in a possible implementation, the Kalman gain matrix of the Kalman filter at the current moment may be determined according to the following formula:

[0074] ;

[0075] ;

[0076] Among them, the is the Kalman gain matrix of the Kalman filter at the current moment;

[0077] Said is the combined Jacobian matrix corresponding to the simulation model;

[0078] Said is the intermediate amount;

[0079] Said is the state covariance matrix of the Kalman filter at the previous moment;

[0080] Said is the measurement noise matrix of the Kalman filter at the current moment;

[0081] Said is the system noise matrix of the Kalman filter at the current moment.

[0082] It should be noted that the Kalman filter is pre-built and is used to predict the health parameters of the combined power unit.

[0083] Specifically, when determining the Kalman gain matrix, the state covariance matrix of the Kalman filter, the measurement noise matrix, the system noise matrix and the combined Jacobian matrix corresponding to the simulation model are comprehensively considered, so that accurate estimation of the state of the combined power unit can be achieved.

[0084] It should be noted that the measurement noise matrix and the system noise matrix are fixed diagonal matrices determined based on the characteristics of the combined power device and the measurement accuracy.

[0085] Further, in this embodiment, the combined Jacobian matrix corresponding to the simulation model is obtained based on the first initial Jacobian matrix of the simulation model and the second initial Jacobian matrix of the pre-trained prediction model. For example, in one possible implementation, the combined Jacobian matrix corresponding to the simulation model is the weighted value of the first initial Jacobian matrix of the simulation model and the second initial Jacobian matrix of the pre-trained prediction model.

[0086] It should be noted that the pre-trained prediction model is used to predict the corresponding health parameters based on the input measurement parameters. The prediction model is trained based on a preset number of groups of sample data, and each group of sample data includes actual measurement parameters and actual health parameters at any time.

[0087] Optionally, in a possible implementation, the process of constructing the prediction model includes:

[0088] Step 1: Build the initial model.

[0089] Specifically, the initial model can be a multilayer perceptron neural network (Multilayer Perceptron, MLP), which, as a feedforward artificial neural network model, can realize a nonlinear mapping relationship from multidimensional input to multidimensional output. It can be understood that MLP includes an input layer, a hidden layer, and an output layer, and the nodes of each layer are connected to the next layer. Each node is a neuron with a nonlinear activation function, which is trained through a supervised learning method of back propagation for continuous input and output.

[0090] In this embodiment, the input layer of the MLP is used to receive the input measurement parameters and pass them to the hidden layer. The number of nodes in the input layer is the same as the number of features of the input data. After that, the hidden layer performs nonlinear transformation and feature extraction on the measurement parameters received by the input layer, and passes the obtained results to the output layer, which outputs the final predicted health parameters.

[0091] Combined with the above introduction, it can be seen that the input to the input layer should be an 8-dimensional variable: compressor flow, compressor efficiency, combustion chamber combustion efficiency, combustion chamber total pressure recovery coefficient, power turbine flow, power turbine efficiency, cooling turbine flow and cooling turbine efficiency; the output is also an 8-dimensional variable: compressor outlet pressure, compressor outlet temperature, combustion chamber outlet pressure, combustion chamber outlet temperature, fuel flow, power turbine exhaust temperature, cooling turbine outlet flow and cooling turbine outlet temperature. Therefore, in the specific implementation, the number of input layer nodes of the MLP network is set to 8, the number of output layer nodes is set to 8, the initial number of hidden layers is set to 1, and some layers are reserved for subsequent improvement of the processing space of input data. For the 8 input measurement parameters, the initial number of nodes on the hidden layer is set to 128 (where 2^7=128).

[0092] It should be noted that the number of hidden layers in the MLP can be one or more layers, which can be set according to actual needs and is not limited here. For example, in one embodiment, the number of hidden layers can be 2 layers; for another example, in another embodiment, the number of hidden layers can be 3 layers. Furthermore, the nonlinear activation function on each node is also selected according to actual needs and is not limited here. For example, in one embodiment, the nonlinear activation function on the node can be a sigmoid function or a ReLU function.

[0093] Step 2: Get a preset number of sample data sets.

[0094] It should be noted that the preset number is set according to actual needs and is not limited in this embodiment.

[0095] Optionally, in a possible implementation, a process of obtaining a set of sample data may include:

[0096] (1) Collecting whole-machine data of the combined power device at a first preset sampling interval to collect actual measurement parameters.

[0097] (2) performing component-level data collection on the compressor, the combustion chamber, the power turbine, and the cooling turbine respectively at a second preset sampling interval, and calculating actual health parameters based on the collected raw data; wherein the second preset sampling interval is greater than the first preset sampling interval.

[0098] (3) The actual measured parameters and actual health parameters at the same time are taken as a set of sample data.

[0099] It should be noted that the first preset sampling interval is relatively small, and the second preset sampling interval is relatively large. For example, in one embodiment, the first preset sampling interval is 1 hour, and the second preset sampling interval is 15 hours.

[0100] Based on the above introduction, when collecting whole machine data and component-level data for a combined power unit, it is planned to obtain M groups of data between 0 and T1 hours of working of the whole machine and each component, and N (N is greater than or equal to zero) groups of data after T1 hours.

[0101] In the specific implementation, the test data of the whole machine at each moment is recorded starting from the working time of each component being 0 hours. The specific content of the record includes the input conditions corresponding to this moment: speed, power generation power, ambient temperature and pressure, heat exchanger heat transfer capacity, etc., as well as the aforementioned 8 measurement parameters: compressor outlet pressure, compressor outlet temperature, combustion chamber outlet pressure, combustion chamber outlet temperature, fuel flow rate, power turbine exhaust temperature, cooling turbine outlet flow rate and cooling turbine outlet temperature.

[0102] Furthermore, the whole machine is systemically disassembled every t hours (for example, in one embodiment, t is set to 15). After the disassembly, the compressor, combustion chamber, power turbine and cooling turbine are respectively subjected to component-level tests to obtain original data (such as external input conditions (such as rotation speed, power generation power, ambient temperature and pressure, etc.)), and then the health parameters at each collection time (compressor flow, compressor efficiency, combustion chamber combustion efficiency, combustion chamber total pressure recovery coefficient, power turbine flow, power turbine efficiency, cooling turbine flow and cooling turbine efficiency) are calculated based on the original data.

[0103] It should be noted that when calculating health parameters based on raw data, the calculation can be performed based on traditional methods. For example, in one possible implementation, the test output results of each component when it is used for the first time can be used as a standard. For each component, while ensuring that the external input conditions are the same, the current test output results are compared with the test output results of the first use to obtain the current health parameter values ​​of each component.

[0104] Furthermore, the actual measurement parameters and actual health parameters at the same time are taken as a set of sample data, so that M sets of data between 0-T1 hours and N sets of data after T1 hours can be finally obtained. Each set of data contains 8 health parameters and 8 measurement parameters.

[0105] It can be understood that the first preset sampling interval is smaller and the second preset sampling interval is larger. Therefore, the whole machine test data is far more than the component-level test data. The 8 measurement parameters obtained from the whole machine test at the same time and the 8 health parameters obtained from the component-level test are used as a set of valid data for training the prediction model.

[0106] Combined with the above example, for example, in a possible implementation, starting from the working time of each component is 0 hours, the whole machine test data at each moment is recorded, and the system is split every 15 hours, and the compressor, combustion chamber, power turbine and cooling turbine are tested at the component level. The test is carried out until the working time is 915 hours, and 60 groups of valid data are obtained, each of which contains 8 health parameters and 8 measurement parameters.

[0107] It should be noted that after the prediction model is corrected with a large number of sample data, the obtained prediction model has a higher model diagnosis accuracy. At this time, the number of experiments can be reduced.

[0108] Step 3: Use the preset number of sample data to train the initial model to obtain a trained prediction model.

[0109] Specifically, the initial model can be trained by the supervised learning method of back propagation. Back propagation refers to the process of training MLP, firstly passing the input sample data to the input layer of MLP, each node in the input layer receives the data of a parameter and passes it to the hidden layer of the next layer, starting from the first hidden layer in the hidden layer, each node of the next hidden layer receives the input signal from the previous hidden layer, and performs weighted summation on the transmitted signal, and then processes it through the nonlinear activation function in the hidden layer, and then transmits it to the next hidden layer, propagates the obtained data signal to the output layer, and then compares the data output by the output layer with the true label to calculate the loss function, and then adjusts the parameters in the model according to the error calculated by the loss function to reduce the size of the loss function, so that the output of the prediction model is close to the true data, and optimizes the prediction effect of the prediction model.

[0110] It is understandable that the loss function of MLP is selected according to actual needs and is not limited here. For example, in one embodiment, mean square error MSE or mean absolute error MAE can be used. Among them, the curve of the loss function using MSE is smooth, continuous and differentiable everywhere, and the gradient descent algorithm is used, and the gradient decreases as the error decreases. This method can converge at a faster speed, but is more sensitive to outliers. Although MAE is not sensitive to outliers, there are non-differentiable points, and its gradient will not decrease as the error decreases, which is not conducive to convergence.

[0111] After processing and analysis, the sample data of this embodiment has few outliers and is required to be differentiable everywhere, so MSE is selected as the training loss function.

[0112] Optionally, when using sample data to train the prediction model, first pre-process the M groups of data obtained from the whole machine working time to T1 hour test, divide 75% of the M groups of data into training sets for training the prediction model, and divide the remaining 25% into test sets for testing the prediction model. Combined with the above example, when 60 groups of data are obtained, 45 groups are used as training sets and 15 groups are used as test sets.

[0113] Furthermore, in order to avoid the results after training falling into local distribution and overfitting, the data set composed of M groups of data is randomly shuffled; in addition, in order to avoid some parameters from becoming larger and having a greater impact on the training results, the sample data needs to be normalized. It should be noted that random shuffling means randomly shuffling all sample data without considering the test time.

[0114] Furthermore, when training the prediction model, a small batch of sample sets can be used to train the initial model to determine the convergence of the prediction model; while ensuring that the convergence trend of the prediction model is good, a large batch of training sets can be used to train it. Afterwards, in order to speed up the convergence of the loss and reduce the prediction error rate, the prediction model is tested and the parameters are adjusted. The specific adjustment objects are the number of hidden layers and the number of nodes. Finally, the number of hidden layers and the number of nodes of the network are determined to obtain a trained prediction model.

[0115] Combined with the previous description, the following example illustrates the training process of the initial model. For example, in one embodiment, the number of input layer nodes of the MLP network is set to 8, the number of output layer nodes is set to 8, the number of hidden layers is set to 3, 2^7=128 nodes are selected as the initial number of nodes, and the mean square error MSE is selected as the training loss function. First, a small batch (epoch=30) is used for training to determine the convergence of the network. While ensuring that the algorithm converges well, a large batch is used for training (epoch=600), and the output error rate is 3.6%. From 915 hours to 1515 hours, the system is split every 15 hours to obtain component-level test data, and 40 sets of valid data are obtained. The training is continued according to the above method, and the final output error rate is 2.1%, and a trained prediction model is obtained.

[0116] The trained prediction model is as follows:

[0117] ;

[0118] in, is the compressor outlet pressure, is the compressor outlet temperature, is the combustion chamber outlet pressure, is the combustion chamber outlet temperature, is the fuel flow rate, is the power turbine exhaust temperature, for cooling the turbine outlet flow, and To cool the turbine outlet temperature, is the compressor flow, The compressor efficiency health parameter, Combustion efficiency of the combustion chamber, is the total pressure recovery coefficient health parameter, is the power turbine flow, The power turbine efficiency health parameter, To cool the turbine flow, Health parameters for cooling turbine efficiency, is the shaft speed, For environmental pressure, is the ambient temperature, is the overall output shaft power of the power unit, Heat exchange for environmental control.

[0119] S104. Based on the Kalman filter, the health parameters at the previous moment, the measurement parameters at the current moment, the predicted measurement parameters at the previous moment and the Kalman gain matrix at the current moment are calculated to perform gas path fault detection on the combined power unit according to the health parameters at the current moment.

[0120] Specifically, the health parameters at the current moment can be calculated according to the following formula:

[0121] ;

[0122] in, is the health parameter at the current moment;

[0123] is the Kalman gain matrix at the current moment;

[0124] is the measurement parameter at the current moment;

[0125] is the health parameter of the previous moment;

[0126] is the predicted measurement parameter at the previous moment.

[0127] S105, updating the state covariance matrix at the current moment according to the Kalman gain matrix at the current moment, the combined Jacobian matrix, and the state covariance matrix at the previous moment.

[0128] Specifically, the state covariance matrix at the current moment can be updated according to the following formula:

[0129] ;

[0130] Among them, the is the state covariance matrix at the current moment;

[0131] Said is the Kalman gain matrix at the current moment;

[0132] Said is the combined Jacobian matrix;

[0133] Said is the state covariance matrix of the previous moment.

[0134] Furthermore, after completing the update of the state covariance matrix at the current moment, the nonlinear Kalman filter at the next moment can be started, and the moment is updated according to the following formula:

[0135] ;

[0136] By repeating the above steps, the health parameters of the combined power plant can be solved in real time.

[0137] The method provided in this embodiment, on the one hand, combines the first initial Jacobian matrix of the simulation model and the second initial Jacobian matrix of the pre-trained prediction model to determine the combined Jacobian matrix of the simulation model, and then determines the Kalman gain matrix of the Kalman filter at the current moment based on the combined Jacobian matrix. In this way, the advantages of the simulation model and the prediction model are maximized, the description ability of each model on the system state is optimized, and the dynamic changes in the complex system are captured more accurately, thereby improving the prediction accuracy and more accurately estimating the health parameters of the combined power unit. On the second hand, by jointly using the simulation model and the prediction model, accurate detection of gas path faults can be maintained in a complex and changeable operating environment, and the robustness to external disturbances can be increased.

[0138] Figure 3 This is a flow chart of the second embodiment of the method for detecting gas path faults in a combined power device provided by the present application. Figure 3 The method provided in this embodiment, based on the above embodiment, the process of determining the combined Jacobian matrix may include:

[0139] S301. Obtain a first initial Jacobian matrix of the simulation model using a small perturbation method.

[0140] It should be noted that the basic principle of the small disturbance method is: for a stably operating system, when it receives a small disturbance, the response of the system can be approximately described by a linearized equation. The specific implementation process of the small disturbance can refer to the description in the relevant technology and will not be repeated here.

[0141] Specifically, the first initial Jacobian matrix is ​​an 8*8 matrix, as shown below:

[0142] ;

[0143] Furthermore, each row of the first Jacobian matrix can be calculated according to the following formula:

[0144] ;

[0145] ;

[0146] ;

[0147] ;

[0148] ;

[0149] ;

[0150] ;

[0151] ;

[0152] in, for or , is an infinitesimal quantity, To specify parameters.

[0153] S302: using the preset number of groups of sample data as learning data, using the prediction model to learn the learning data, and using a small perturbation method to determine a second initial Jacobian matrix of the prediction model based on the learning result.

[0154] Furthermore, a preset number of groups of sample data are used as learning data, and the learning data is learned using the prediction model. Then, the small perturbation method is used again to determine the second initial Jacobian matrix of the prediction model based on the obtained learning results. The acquisition of the second initial Jacobian matrix is ​​as described in step S401 and will not be repeated here.

[0155] S303: Determine a mean square error of the prediction model when learning the learning data, and determine a first weight coefficient of the first initial Jacobian matrix and a second weight coefficient of the second initial Jacobian matrix based on the mean square error.

[0156] Specifically, the mean square error can be calculated according to the following formula:

[0157] ;

[0158] in, is the predicted health parameter value, is the actual health parameter value.

[0159] Further, after the mean square error is obtained, the first weight coefficient and the second weight coefficient may be determined according to the following formula:

[0160] ;

[0161] ;

[0162] in, is the first weight coefficient, ranging from 0 to 1; is the mean square error; is the weight function coefficient; is the second weight coefficient, ranging from 0 to 1.

[0163] S304: Perform weighted processing on the first initial Jacobian matrix and the second initial Jacobian matrix according to the first weight coefficient and the second weight coefficient, and determine the weighted processing result as the combined Jacobian matrix.

[0164] Specifically, the combined Jacobian matrix corresponding to the simulation model can be calculated by the following formula:

[0165] ;

[0166] ;

[0167] in, is the combined Jacobian matrix corresponding to the simulation model, is the first initial Jacobian matrix of the simulation model, is the second initial Jacobian matrix of the prediction model, is the mean square error.

[0168] The method provided in the present application for detecting gas path faults in a combined power device, on the one hand, by combining the Jacobian matrices of the simulation model and the prediction model and calculating the weight coefficient based on the mean square error, it is possible to optimize the description capability of each model on the system state, so as to more accurately capture the dynamic changes in the complex system, thereby improving the prediction accuracy; on the second hand, by using the mean square error to calculate the weight coefficient, the contribution ratio of the simulation model and the prediction model to the combined Jacobian matrix is ​​dynamically adjusted, so that the determination process of the combined Jacobian matrix can be adaptively optimized according to different working conditions, ensuring that the optimal prediction effect can be obtained in different scenarios, and the robustness of fault detection can be enhanced; on the third hand, since the simulation model and the prediction model each have different advantages and limitations, the balance point can be determined in the model uncertainty through weighting, thereby reducing the error accumulation caused by the deviation of a single model and enhancing the stability of the prediction.

[0169] Figure 4 This is a schematic diagram of a third embodiment of a method for detecting gas path faults in a combined power device according to an exemplary embodiment of the present application. Figure 4 Based on the above embodiment, in the method provided by this embodiment, the combined Jacobian matrix is ​​dynamically updated, and the updating process of the combined Jacobian matrix includes:

[0170] S401. Acquire actual measurement parameters and actual health parameters of the combined power device in real time, and determine the actual measurement parameters and actual health parameters at the same time as a group of sample data.

[0171] The process of obtaining sample data is described in the previous embodiment and will not be repeated here.

[0172] S402. When the number of sample data reaches a specified number, the combined Jacobian matrix at the current moment is used as the first initial Jacobian matrix of the simulation model, and the specified number of groups of sample data are used as learning data, and the step of learning the learning data using the prediction model is executed again.

[0173] The specified number is set according to actual needs and is not limited in this embodiment.

[0174] Specifically, referring to the previous description, M groups of test data were collected during the initial T1 time. Based on the initial M groups of test data, the Jacobian matrix of the simulation model was , and the Jacobian matrix obtained based on the prediction model Fusion is performed to obtain the combined Jacobian matrix Further, after obtaining the subsequent N groups of data, the current combined Jacobian matrix is ​​used as the first initial Jacobian matrix of the initial model, that is, Grant , and then learn the N groups of sample data based on the prediction model to obtain the Jacobian matrix , further, we get the updated combined Jacobian matrix .

[0175] For example, combining the above example, initially, the small perturbation method is used to construct the Jacobian matrix of the simulation model Based on the prediction model, the 60 sets of sample data obtained from 0-915 hours are learned to obtain the Jacobian matrix , at this time the first weight is equal to 0.72, and the combined Jacobian matrix is ​​obtained , then, within the range of 915-1515 hours, predictions can be made based on the combined Jacobian matrix.

[0176] Furthermore, within the range of 915-1515 hours, 40 sets of sample data are obtained. At this time, the current combined Jacobian matrix can be assigned to , again based on the prediction model to learn these 40 sets of sample data, and get the new Jacobian matrix , at this time the first weight is equal to 0.85, so the combined Jacobian matrix can be obtained , so that after 1515 hours, predictions can be made based on the newly obtained Jacobian matrix.

[0177] The method provided in the present application for detecting gas path faults in a combined power device combines the Jacobian matrix based on a simulation model and the Jacobian matrix based on a prediction model, and continuously updates the combined Jacobian during actual use. It can more effectively respond to system disturbances and parameter changes, maintain system stability and accuracy, determine the optimal nonlinear Kalman filter model, achieve accurate diagnosis of complex faults, and ensure the effectiveness of complex fault identification.

[0178] Furthermore, in order to verify the effect of the method provided in this application, this article also provides corresponding verification experiments, which are introduced below:

[0179] Specifically, in the verification experiment, the following settings are made: , , , , At the initial moment, the health parameter value of each component is set to 100%, and at t=2s, the combustion efficiency of the combustion chamber is injected with a 3% sudden fault, and the total pressure recovery coefficient of the combustion chamber is injected with a 4% fault. The health parameters (healthiness) are predicted in real time based on the method provided in this application. The prediction results are as follows: Figure 5 As shown ( Figure 5 , which is an experimental result diagram of a verification experiment shown in an exemplary embodiment of the present application). Please refer to Figure 5 ,from Figure 5 It can be seen that the predicted values ​​of the health parameters of each component are consistent with the injected fault. The method proposed in the present invention can accurately detect gas path faults.

[0180] Corresponding to the aforementioned embodiment of a method for detecting gas path faults in a combined power device, the present application also provides an embodiment of a device for detecting gas path faults in a combined power device.

[0181] Figure 6 This is a schematic diagram of the structure of the first embodiment of the device for detecting gas path faults in a combined power device provided by the present application. Figure 6 The device provided in this embodiment includes an acquisition module 610, a processing module 620, a determination module 630, a calculation module 640 and an update module 650, wherein:

[0182] The acquisition module 610 is used to acquire the measurement parameters at the current moment and the state parameters at the previous moment; wherein the state parameters include the specified parameters and the health parameters related to the measurement parameters; the measurement parameters are the parameters related to the gas path performance among the various parameters of the compressor, combustion chamber, power turbine, and cooling turbine in the combined power device;

[0183] The processing module 620 is used to input the state parameters of the previous moment into the simulation model corresponding to the combined power device, so that the simulation model outputs the predicted measurement parameters of the previous moment according to the state parameters of the previous moment; wherein the simulation model is used to characterize the correlation between the state parameters and the measurement parameters;

[0184] The determination module 630 is used to determine the Kalman gain matrix of the Kalman filter at the current moment according to the pre-constructed state covariance matrix of the Kalman filter at the previous moment, the measurement noise matrix of the Kalman filter at the current moment, the system noise matrix of the Kalman filter at the current moment, and the combined Jacobian matrix corresponding to the simulation model; wherein the combined Jacobian matrix corresponding to the simulation model is obtained based on the first initial Jacobian matrix of the simulation model and the second initial Jacobian matrix of the pre-trained prediction model; the pre-trained prediction model is used to predict the corresponding health parameters based on the input measurement parameters; the prediction model is trained based on a preset number of groups of sample data, and each group of sample data includes actual measurement parameters and actual health parameters at any moment;

[0185] The calculation module 640 is used to calculate the health parameter at the current moment based on the Kalman filter according to the health parameter at the previous moment, the measurement parameter at the current moment, the predicted measurement parameter at the previous moment and the Kalman gain matrix at the current moment, so as to perform gas path fault detection on the combined power device according to the health parameter at the current moment;

[0186] The updating module 650 is used to update the state covariance matrix at the current moment according to the Kalman gain matrix at the current moment, the combined Jacobian matrix, and the state covariance matrix at the previous moment.

[0187] 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.

[0188] 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.

[0189] 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.

[0190] 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 method for detecting gas path faults in a combined power unit, characterized in that: The method comprises: Acquire the measurement parameters at the current moment and the state parameters at the previous moment; wherein the state parameters include the specified parameters and the health parameters related to the measurement parameters; the measurement parameters are the parameters related to the gas path performance among the parameters of the compressor, combustion chamber, power turbine, and cooling turbine in the combined power device; Inputting the state parameters of the previous moment into the simulation model corresponding to the combined power device, so that the simulation model outputs the predicted measurement parameters of the previous moment according to the state parameters of the previous moment; wherein the simulation model is used to characterize the correlation between the state parameters and the measurement parameters; The Kalman gain matrix of the Kalman filter at the current moment is determined according to the state covariance matrix of the Kalman filter at the previous moment, the measurement noise matrix of the Kalman filter at the current moment, the system noise matrix of the Kalman filter at the current moment, and the combined Jacobian matrix corresponding to the simulation model; wherein the combined Jacobian matrix corresponding to the simulation model is obtained based on the first initial Jacobian matrix of the simulation model and the second initial Jacobian matrix of the pre-trained prediction model; the pre-trained prediction model is used to predict the corresponding health parameters based on the input measurement parameters; the prediction model is trained based on a preset number of groups of sample data, and each group of sample data includes actual measurement parameters and actual health parameters at any moment; Based on the Kalman filter, the health parameter at the current moment is calculated according to the health parameter at the previous moment, the measurement parameter at the current moment, the predicted measurement parameter at the previous moment and the Kalman gain matrix at the current moment, so as to perform gas path fault detection on the combined power device according to the health parameter at the current moment; The state covariance matrix at the current moment is updated according to the Kalman gain matrix at the current moment, the combined Jacobian matrix, and the state covariance matrix at the previous moment.

2. The method according to claim 1, characterized in that The process of determining the combined Jacobian matrix includes: Obtaining a first initial Jacobian matrix of the simulation model using a small perturbation method; Using the preset number of groups of sample data as learning data, using the prediction model to learn the learning data, and using a small perturbation method to determine a second initial Jacobian matrix of the prediction model based on the learning result; Determine a mean square error of the prediction model when learning the learning data, and determine a first weight coefficient of the first initial Jacobian matrix and a second weight coefficient of the second initial Jacobian matrix based on the mean square error; The first initial Jacobian matrix and the second initial Jacobian matrix are weighted according to the first weight coefficient and the second weight coefficient, and the weighted processing result is determined as the combined Jacobian matrix.

3. The method according to claim 2, characterized in that The combined Jacobian matrix is ​​dynamically updated, and the updating process of the combined Jacobian matrix includes: Acquire actual measurement parameters and actual health parameters of the combined power device in real time, and determine the actual measurement parameters and actual health parameters at the same time as a set of sample data; When the number of sample data reaches a specified number, the combined Jacobian matrix at the current moment is used as the first initial Jacobian matrix of the simulation model, and the specified number of groups of sample data are used as learning data, and the step of learning the learning data using the prediction model is performed again.

4. The method according to claim 2, characterized in that: The determining, based on the mean square error, a first weight coefficient of the first initial Jacobian matrix and a second weight coefficient of the second initial Jacobian matrix comprises: The first weight coefficient and the second weight coefficient are determined according to the following formula: ； ； in, is the first weight coefficient; is the mean square error; is the regulating factor; is the second weight coefficient.

5. The method according to claim 1, characterized in that Determining the Kalman gain matrix of the Kalman filter at the current moment according to the pre-constructed state covariance matrix of the Kalman filter at the previous moment, the measurement noise matrix of the Kalman filter at the current moment, the system noise matrix of the Kalman filter at the current moment, and the combined Jacobian matrix corresponding to the simulation model, including: The Kalman gain matrix of the Kalman filter at the current moment is determined according to the following formula: ； ； in, is the Kalman gain matrix of the Kalman filter at the current moment; is the combined Jacobian matrix corresponding to the simulation model; is the intermediate amount; is the state covariance matrix of the Kalman filter at the previous moment; is the measurement noise matrix of the Kalman filter at the current moment; is the system noise matrix of the Kalman filter at the current moment.

6. The method according to claim 3, characterized in that The process of obtaining the set of sample data includes: Collecting whole-machine data of the combined power device according to a first preset sampling interval to collect actual measurement parameters; performing component-level data collection on the compressor, the combustion chamber, the power turbine, and the cooling turbine respectively at a second preset sampling interval, and calculating actual health parameters based on the collected raw data; wherein the second preset sampling interval is greater than the first preset sampling interval; The actual measured parameters and actual health parameters at the same moment are taken as a set of sample data.

7. The method according to claim 1, characterized in that The process of establishing the simulation model includes: According to the characteristics of the compressor, the combustion chamber, the power turbine and the cooling turbine, a first mathematical model corresponding to the compressor, a second mathematical model corresponding to the combustion chamber, a third mathematical model corresponding to the power turbine and a fourth mathematical model corresponding to the cooling turbine are respectively established; wherein each mathematical model is used to characterize the correlation between some parameters in the measurement parameters and some parameters in the health parameters; According to the flow balance relationship, pressure balance relationship and energy balance relationship of the combined power device, the first mathematical model, the second mathematical model, the third mathematical model and the fourth mathematical model are integrated to obtain the simulation model.

8. The method according to claim 1, characterized in that The calculating of the health parameter at the current moment based on the Kalman filter according to the health parameter at the previous moment, the measurement parameter at the current moment, the predicted measurement parameter at the previous moment and the Kalman gain matrix at the current moment includes: The current health parameters are calculated according to the following formula: ； in, is the health parameter at the current moment; is the Kalman gain matrix at the current moment; is the measurement parameter at the current moment; is the health parameter at the previous moment; is the predicted measurement parameter at the previous moment.

9. The method according to claim 1, characterized in that: The updating of the state covariance matrix at the next moment according to the Kalman gain at the current moment, the combined Jacobian matrix, and the state covariance matrix at the previous moment comprises: Update the current state covariance matrix according to the following formula: ； in, is the state covariance matrix at the current moment; is the Kalman gain matrix at the current moment; is the combined Jacobian matrix; is the state covariance matrix of the previous moment.

10. A device for detecting gas path faults in a combined power unit, characterized in that: The device includes an acquisition module, a processing module, a determination module, a calculation module and an update module, wherein: The acquisition module is used to acquire the measurement parameters at the current moment and the state parameters at the previous moment; wherein the state parameters include specified parameters and health parameters related to the measurement parameters; the measurement parameters are parameters related to gas path performance among various parameters of the compressor, combustion chamber, power turbine, and cooling turbine in the combined power device; The processing module is used to input the state parameters of the previous moment into the simulation model corresponding to the combined power device, so that the simulation model outputs the predicted measurement parameters of the previous moment according to the state parameters of the previous moment; wherein the simulation model is used to characterize the correlation between the state parameters and the measurement parameters; The determination module is used to determine the Kalman gain matrix of the Kalman filter at the current moment according to the pre-constructed state covariance matrix of the Kalman filter at the previous moment, the measurement noise matrix of the Kalman filter at the current moment, the system noise matrix of the Kalman filter at the current moment, and the combined Jacobian matrix corresponding to the simulation model; wherein the combined Jacobian matrix corresponding to the simulation model is obtained based on the first initial Jacobian matrix of the simulation model and the second initial Jacobian matrix of the pre-trained prediction model; the pre-trained prediction model is used to predict the corresponding health parameters based on the input measurement parameters; the prediction model is trained based on a preset number of groups of sample data, and each group of sample data includes actual measurement parameters and actual health parameters at any moment; The calculation module is used to calculate the health parameter at the current moment based on the Kalman filter according to the health parameter at the previous moment, the measurement parameter at the current moment, the predicted measurement parameter at the previous moment and the Kalman gain matrix at the current moment, so as to perform gas path fault detection on the combined power device according to the health parameter at the current moment; The updating module is used to update the state covariance matrix at the current moment according to the Kalman gain matrix at the current moment, the combined Jacobian matrix, and the state covariance matrix at the previous moment.

Images