Turbine fault diagnosis method and system and computer readable

Abstract

The invention provides a turbine fault diagnosis method, an applicable system thereof and a computer readable medium, and relates to the technical field of turbines. The turbine fault diagnosis method comprises the following steps that a training data set and a proxy model corresponding to an engine are constructed, and the proxy model comprises a mapping function of the simulation calculation process of a coupling simulation model corresponding to the engine; constructing a neural network model, and training the neural network model according to the training data set to obtain a trained neural network model as a turbine fault diagnosis model; acquiring real-time monitoring data of the turbine, and inputting the real-time monitoring data into the agent model to obtain corresponding real-time turbine internal state data; and constructing an extended feature vector according to the real-time monitoring data and the corresponding real-time turbine internal state data, inputting the extended feature vector into the turbine fault diagnosis model to obtain a fault distribution vector, and obtaining a turbine fault diagnosis result of the turbine according to the fault distribution vector.

Description

TECHNICAL FIELD

[0001] The present application mainly relates to the technical field of turbine, in particular to a turbine fault diagnosis method, a system suitable for the method and a computer readable medium. BACKGROUND

[0002] The turbocharger, i.e. the turbine, is an important component of a marine diesel engine, which can significantly improve the power of the diesel engine, reduce the fuel consumption rate, and ensure the high efficiency of the diesel engine in the full working condition range through the technology of sequential supercharging switching. The health management of the operation state of the turbine helps to effectively ensure the stability and reliability of the marine diesel engine. At present, the health monitoring means of the turbine can be mainly divided into two categories: one is to select effective characteristic parameters and establish a fault state model by adding multiple sensor measuring points; and the other is to monitor the efficiency change of the compressor or the turbine to indirectly evaluate the overall health state of the turbine.

[0003] At present, the application of the turbine fault diagnosis based on the turbine prediction model is relatively limited. The existing method often excessively relies on the sensor arrangement, which not only increases the complexity and cost of the system, but also limits its applicability in actual engineering; or is limited to the monitoring of a single parameter such as efficiency, which cannot realize real-time and accurate fault positioning, especially under the complex working conditions of the ship, it is difficult to obtain effective data for fault diagnosis, and thus accurate and comprehensive fault type diagnosis results cannot be obtained.

[0004] Therefore, there is an urgent need for a real-time and accurate turbine fault diagnosis method. SUMMARY

[0005] The technical problem to be solved by the present application is to provide a turbine fault diagnosis method, a system suitable for the method and a computer readable medium, which can obtain accurate turbine fault diagnosis results in real time.

[0006] To solve the above technical problems, the present application provides a turbine fault diagnosis method suitable for a turbine located in an engine, which comprises the following steps: constructing a training data set and a corresponding proxy model of the engine, wherein the proxy model comprises a mapping function corresponding to the simulation calculation process of the coupled simulation model of the engine; constructing a neural network model, and training the neural network model according to the training data set to obtain a trained neural network model as a turbine fault diagnosis model; obtaining real-time monitoring data of the turbine, and inputting the real-time monitoring data into the proxy model to obtain corresponding real-time internal state data of the turbine; constructing an extended feature vector according to the real-time monitoring data and the corresponding real-time internal state data of the turbine, and inputting the extended feature vector into the turbine fault diagnosis model to obtain a fault distribution vector, and obtaining a turbine fault diagnosis result of the turbine according to the fault distribution vector.

[0007] Optionally, the steps of constructing the training dataset and the surrogate model corresponding to the engine further include: constructing a coupled simulation model corresponding to the engine; obtaining historical monitoring data of the turbine, constructing a training dataset based on the coupled simulation model and the historical monitoring data; and constructing a surrogate model corresponding to the coupled simulation model based on the training dataset.

[0008] Optionally, the step of constructing the coupled simulation model corresponding to the engine further includes: constructing a one-dimensional simulation model of the turbine corresponding to the turbine; constructing a whole-machine simulation model corresponding to the engine; and coupling the one-dimensional simulation model of the turbine and the whole-machine simulation model to obtain the coupled simulation model. During the operation of the coupled simulation model, the one-dimensional simulation model of the turbine generates corresponding turbine data based on the whole-machine data generated by the whole-machine simulation model at the current time step, and uses the turbine data as the input of the whole-machine simulation model at the next time step.

[0009] Optionally, the one-dimensional turbine simulation model includes multiple control surfaces, including control surfaces corresponding to the volute, nozzle ring, impeller, and diffuser section. The real-time turbine internal state data includes control surface pressure data and control surface temperature data. The control surface pressure data includes the total outlet pressure corresponding to each control surface, and the control surface temperature data includes the total outlet temperature corresponding to each control surface.

[0010] Optionally, the formula for calculating the total outlet pressure is: In the formula For the first Total outlet temperature of each control surface. For the first Total inlet temperature of each control surface For the first The inlet circumferential velocity of each control surface For the first The exit circumferential velocity of each control surface This is the specific heat capacity at constant pressure.

[0011] Optionally, the formula for calculating the total outlet pressure is: In the formula For the first Total outlet pressure of each control surface For the first Total inlet pressure of each control surface The adiabatic index of the gas. The loss coefficient, For the first The exit Mach number of each control surface For the first The inlet circumferential velocity of each control surface For the first The exit circumferential velocity of each control surface This is the specific heat capacity at constant pressure.

[0012] Optionally, the historical monitoring data includes multiple sets of monitoring sub-data corresponding to the turbine under different operating conditions. The step of acquiring the historical monitoring data of the turbine and constructing a training dataset based on the coupled simulation model and the historical detection data further includes: acquiring the historical detection data; adjusting the parameters corresponding to the coupled simulation model according to different turbine fault types, and inputting each set of monitoring sub-data into the adjusted coupled simulation model corresponding to each turbine fault type to obtain the corresponding turbine internal state prediction data; and using each set of detection sub-data, the corresponding turbine internal state prediction data, and the corresponding turbine fault type as the training dataset.

[0013] Optionally, the step of constructing a surrogate model corresponding to the coupled simulation model based on the training dataset further includes: constructing a Kriging model, wherein the correlation function of the Kriging model is a Gaussian correlation function; and obtaining an optimized Kriging model as a surrogate model by maximizing the log-likelihood function based on the training dataset and a numerical optimization algorithm.

[0014] Optionally, the fault distribution vector includes the probabilities corresponding to multiple diagnostic results, the fault distribution vector includes multiple vector elements, the multiple vector elements correspond to the predicted probabilities of multiple diagnostic results, the multiple diagnostic results include normal state and multiple different turbine fault types, and the turbine fault diagnosis result is the diagnosis result corresponding to the largest predicted probability in the fault distribution vector.

[0015] Optionally, the monitoring data includes intake pressure, outlet pressure, intake temperature, outlet temperature, and turbocharger speed, while the turbine internal status data includes turbine efficiency, output torque, control surface pressure data, control surface temperature data, and Mach number.

[0016] Optionally, the extended feature vector includes multiple input features corresponding to each data point in the real-time monitoring data and each data point in the real-time turbine internal state data. The turbine fault diagnosis model includes: an input layer for receiving the extended feature vector; an attention module for generating a corresponding attention weight vector based on the extended feature vector, and generating a weighted feature vector based on the attention weight vector and the extended feature vector, wherein the attention weight vector includes multiple importance weights corresponding to each input feature; a hidden layer for generating a hidden layer output based on the weighted feature vector; and an output layer for generating a fault distribution vector based on the hidden layer output.

[0017] Optionally, the final diagnostic results for the turbine include turbine fault diagnosis results and attention weight vectors.

[0018] To address the aforementioned technical problems, this application provides a turbine fault diagnosis system, comprising: a memory for storing instructions executable by a processor; and a processor for executing the instructions to implement the aforementioned turbine fault diagnosis method.

[0019] To address the aforementioned technical problems, this application provides a computer-readable medium storing computer program code, which, when executed by a processor, implements the aforementioned turbine fault diagnosis method.

[0020] Compared with existing technologies, this application has the following advantages: By inputting real-time monitoring data into a surrogate model, real-time turbine internal state data reflecting the turbine's internal condition is generated. Since this type of internal state data cannot be collected by actual sensors, using real-time monitoring data and the corresponding real-time turbine internal state data as input to the turbine fault diagnosis model allows the model to perform fault analysis based on more turbine-related data, resulting in more accurate turbine fault diagnosis results. Furthermore, because the surrogate model includes a mapping function for the simulation calculation process of the corresponding coupled simulation model, and the turbine fault diagnosis model is constructed from a neural network model, both the surrogate model and the turbine fault diagnosis model have fast response speeds. This enables the rapid generation of corresponding turbine fault diagnosis results based on real-time monitoring data feedback, achieving real-time monitoring of the turbine's condition. Attached Figure Description

[0021] The accompanying drawings are included to provide a further understanding of this application; they are incorporated into and constitute a part of this application. The drawings illustrate embodiments of this application and, together with this specification, serve to explain the principles of this application. In the drawings: Figure 1 This is a schematic flowchart of a turbine fault diagnosis method according to an embodiment of this application; Figure 2 yes Figure 1 A flowchart illustrating the sub-steps of step S1; Figure 3 yes Figure 2 A flowchart illustrating the sub-steps of step S11. Figure 4 This is a schematic diagram of the flow components and control surface of a volute according to an embodiment of this application; Figure 5 yes Figure 2 A flowchart illustrating the sub-steps of step S12; Figure 6 yes Figure 2 A flowchart illustrating the sub-steps of step S13; Figure 7 This is a schematic diagram illustrating an embodiment of a turbine fault diagnosis method according to this application; and Figure 8 This is a block diagram of a turbine fault diagnosis system according to an embodiment of this application. Detailed Implementation

[0022] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are merely some examples or embodiments of this application. For those skilled in the art, these drawings can be applied to other similar scenarios without creative effort. Unless obvious from the context or otherwise specified, the same reference numerals in the drawings represent the same structures or operations.

[0023] As indicated in this application and claims, unless the context clearly indicates otherwise, the words "a," "an," "an," and / or "the" are not specifically singular and may include plural forms. Generally speaking, the terms "comprising" and "including" only indicate the inclusion of explicitly identified steps and elements, which do not constitute an exclusive list, and the method or apparatus may also include other steps or elements.

[0024] Unless otherwise specifically stated, the relative arrangement, numerical expressions, and values ​​of the components and steps described in these embodiments do not limit the scope of this application. It should also be understood that, for ease of description, the dimensions of the various parts shown in the drawings are not drawn to actual scale. Techniques, methods, and devices known to those skilled in the art may not be discussed in detail, but where appropriate, such techniques, methods, and devices should be considered part of the specification. In all examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not as limitations. Therefore, other examples of exemplary embodiments may have different values. It should be noted that similar reference numerals and letters in the following drawings denote similar items; therefore, once an item is defined in one drawing, it need not be further discussed in subsequent drawings.

[0025] In the description of this application, it should be understood that the orientation or positional relationship indicated by directional terms such as "front, back, up, down, left, right", "horizontal, vertical, horizontal" and "top, bottom" is usually based on the orientation or positional relationship shown in the accompanying drawings, and is only for the convenience of describing this application and simplifying the description. Unless otherwise stated, these directional terms do not indicate or imply that the device or element referred to must have a specific orientation or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation on the scope of protection of this application; the directional terms "inner" and "outer" refer to the inner and outer contours relative to the outline of each component itself.

[0026] For ease of description, spatial relative terms such as "above," "on top of," "on the upper surface of," "above," etc., are used herein to describe the spatial positional relationship of a device or feature as shown in the figures to other devices or features. It should be understood that spatial relative terms are intended to encompass different orientations in use or operation beyond the orientation of the device as described in the figures. For example, if the device in the figures were inverted, a device described as "above" or "on top of" other devices or structures would subsequently be positioned as "below" or "under" other devices or structures. Thus, the exemplary term "above" can include both "above" and "below." The device may also be positioned in other different ways (rotated 90 degrees or in other orientations), and the spatial relative descriptions used herein will be interpreted accordingly.

[0027] Furthermore, it should be noted that the use of terms such as "first" and "second" to define components is merely for the purpose of distinguishing the corresponding components. Unless otherwise stated, these terms have no special meaning and therefore should not be construed as limiting the scope of protection of this application. In addition, although the terminology used in this application is selected from commonly known and used terms, some terms mentioned in this application's specification may have been chosen by the applicant according to his or her judgment, and their detailed meanings are explained in the relevant sections of this description. Moreover, this application should be understood not only through the actual terms used, but also through the meaning implied by each term.

[0028] It should be understood that when a component is referred to as "on another component," "connected to another component," "coupled to another component," or "in contact with another component," it can be directly on, connected to, coupled to, or in contact with that other component, or there may be an intervening component. In contrast, when a component is referred to as "directly on another component," "directly connected to," "directly coupled to," or "directly in contact with" another component, there is no intervening component. Similarly, when a first component is referred to as "electrically contacting" or "electrically coupled to" a second component, there is an electrical path between the first and second components that allows current to flow. This electrical path may include capacitors, coupled inductors, and / or other components that allow current to flow, even if there is no direct contact between the conductive components.

[0029] Flowcharts are used in this application to illustrate the operations performed by the system according to embodiments of this application. It should be understood that the preceding or following operations are not necessarily performed in exact order. Instead, various steps can be processed in reverse order or simultaneously. Furthermore, other operations may be added to these processes, or one or more steps may be removed from these processes.

[0030] Reference Figure 1 One embodiment of this application provides a turbine fault diagnosis method 100, which is applicable to turbines located in engines. Exemplarily, in some embodiments the engine is a marine diesel engine and the turbine is a turbocharger.

[0031] Continue to refer to Figure 1 The turbine fault diagnosis method 100 includes the following steps: Step S1 is to construct a training dataset and a proxy model corresponding to the engine. The proxy model includes a mapping function for the simulation calculation process of the coupled simulation model corresponding to the engine. Step S2 is to construct a neural network model and train it based on the training dataset to obtain a trained neural network model as the turbine fault diagnosis model. Step S3 is to acquire real-time monitoring data of the turbine and input the real-time monitoring data into the proxy model to obtain corresponding real-time turbine internal state data. Step S4 is to construct an extended feature vector based on the real-time monitoring data and the corresponding real-time turbine internal state data, input the extended feature vector into the turbine fault diagnosis model to obtain a fault distribution vector, and obtain the turbine fault diagnosis result based on the fault distribution vector.

[0032] By employing a surrogate model and real-time monitoring data, the turbine fault diagnosis method 100 can generate real-time turbine internal state data that cannot be detected by actual sensors. Furthermore, using both real-time monitoring data and real-time turbine internal state data as input to the turbine fault diagnosis model enriches its analytical data range, enabling it to obtain more accurate turbine fault diagnosis results based on inputs containing more diverse information. Moreover, since the surrogate model, through a mapping function, can improve the response speed of outputting real-time turbine internal state data while maintaining near-coupled simulation model accuracy, and the turbine fault diagnosis model is constructed from a fast-responding neural network model, the turbine fault diagnosis method 100 can generate real-time turbine fault diagnosis results based on real-time monitoring data. This avoids the problem of turbine monitoring lagging behind actual turbine operating conditions, giving monitoring personnel more time to act and improving the accuracy of their decisions.

[0033] The following provides further details on steps S1 to S4. (Refer to...) Figure 2 In some embodiments, step S1 includes the following sub-steps. Step S11 is to construct the coupled simulation model corresponding to the engine. Further refer to... Figure 3In some embodiments, step S11 includes the following sub-steps. Step S111 is to construct a one-dimensional simulation model of the turbine. For example, in some embodiments, the average streamline method is used to simulate the internal flow process of the turbine, assuming that the flow inside the turbine is a one-dimensional quasi-steady-state flow and that the specific heat ratio of the fluid remains constant in the turbine. Specifically, the turbine is divided into several interconnected flow components according to its internal structure, and a one-dimensional simulation model of the turbine is constructed based on the key geometric parameters of the flow components. In the one-dimensional simulation model of the turbine, the flow characteristics within each flow component are consistent, and the inlet and outlet interfaces of each flow component are corresponding control surfaces.

[0034] For example, further refer to Figure 4 In some embodiments, the flow components include a volute, a nozzle ring, an impeller, and a diffuser section. The corresponding key turbine geometric parameters preferably include the volute inlet area, volute inlet radius, nozzle ring inlet blade height, nozzle ring inlet radius, nozzle ring outlet airflow angle, impeller inlet blade height, impeller inlet radius, outlet blade tip radius, outlet blade root radius, inlet blade angle, outlet blade angle, and number of blades. Furthermore, the corresponding control surfaces preferably include a volute inlet control surface 1, a volute outlet control surface 2 (i.e., a nozzle ring inlet leading edge control surface), a nozzle ring inlet trailing edge control surface 3, a nozzle ring outlet control surface 4, an impeller inlet control surface 5, an impeller outlet control surface 6, and a diffuser section outlet control surface 7.

[0035] Continue to refer to Figure 4 In some embodiments, the parameters of the control surfaces in the one-dimensional turbine simulation model are averaged values ​​within the flow channel cross-section. Combining mass conservation, momentum conservation, energy conservation, and turbine flow losses, the flow characteristics of each control surface are calculated. The impeller rotor is simulated in a relative reference frame, taking into account the loss model and velocity triangle. Correspondingly, the flow in the absolute reference frame within the nozzle ring is similar to the flow in the impeller rotor relative to the reference frame, therefore a similar incident loss model is used. Furthermore, the flow in the remaining stators differs mainly in the different mass flow rate formulas resulting from the velocity angle. Based on the above settings, the calculation expression for the universally applicable flow segment in the one-dimensional turbine simulation model can be obtained as follows: , In the formula For quality flow, The gas constant is For the first Total inlet temperature of each control surface For the first The outlet flow area of ​​each control surface, For the first Total inlet pressure of each control surface For the first The outlet airflow angle of each control surface The adiabatic index of the gas. For the first The exit Mach number of each control surface The loss coefficient, For the first The inlet circumferential velocity of each control surface For the first The exit circumferential speed of each control surface The specific heat capacity is given at constant pressure. The loss coefficient is selected based on the specific turbine structure, and setting this loss coefficient can improve the accuracy of the one-dimensional turbine simulation model. For example, in some embodiments, for mixed-flow turbines, the losses corresponding to the loss coefficient preferably include secondary flow losses, friction losses, impeller angle-of-attack losses, impeller passage losses, and tip clearance losses.

[0036] Furthermore, based on the calculation expression for the aforementioned universal flow segment, the total outlet temperature and total outlet pressure of each control surface can be calculated. Specifically, during the calculation, an assumption is made regarding the mass flow rate at the turbine inlet. For each segment, the Mach number is first assumed, then determined iteratively, and finally the flow parameters of each control surface are calculated. In the calculation within the stator absolute reference frame, the corresponding inlet circumferential velocity and outlet circumferential velocity are both 0. Based on the above settings, the first... Total outlet temperature of each control surface The calculation expression is: .

[0037] Correspondingly, the first The formula for calculating the total outlet pressure of each control surface is: .

[0038] Continue to refer to Figure 3 Step S112 involves constructing a complete engine simulation model. Specifically, the complete engine simulation model includes models of multiple components in the engine, excluding the turbine. For example, the complete engine simulation model includes models of the intake and exhaust pipes, compressor, intercooler, cylinders, and control valves. Step S113 involves coupling the one-dimensional turbine simulation model and the complete engine simulation model to obtain a coupled simulation model. For example, in some embodiments, a one-dimensional turbine simulation model is built in Simulink software, and a complete engine simulation model is built in GT-Power software. Then, Simulink-Harness is used to connect the one-dimensional turbine simulation model and the complete engine simulation model to achieve data interaction, thus obtaining the coupled simulation model. Further, in some embodiments, a complete engine test is conducted, and multiple parameters collected from the test are used to calibrate and correct the coupled simulation model, thereby obtaining a coupled simulation model that more closely reflects the engine's operating state.

[0039] Furthermore, during the operation of the coupled simulation model, the turbine one-dimensional simulation model generates corresponding turbine data based on the overall machine data generated by the overall machine simulation model at the current time step, and uses the turbine data as the input for the next time step of the overall machine simulation model. For example, in some embodiments, the overall machine data includes turbine inlet total temperature, mass flow rate, and turbocharger speed, while the turbine data includes turbine inlet pressure and turbine output torque. Through steps S111 to S113, a refined simulation model corresponding to the turbine can be constructed. By connecting it with the overall machine simulation model corresponding to the engine, the turbine one-dimensional simulation model can obtain more realistic external turbine data, i.e., overall machine data, thereby more realistically and accurately simulating the turbine's operating conditions and obtaining more realistic and accurate turbine data.

[0040] Continue to refer to Figure 2 Step S12 involves acquiring historical monitoring data of the turbine and constructing a training dataset based on the coupled simulation model and the historical monitoring data. Further reference... Figure 5 In some embodiments, step S12 includes the following sub-steps. Step S121 is to acquire historical detection data. The historical monitoring data includes multiple sets of monitoring sub-data corresponding to the turbine under different operating conditions. For example, in some embodiments, the monitoring sub-data includes engine intake pressure, outlet pressure, intake temperature, outlet temperature, and turbocharger speed. These data signals can be obtained by deploying sensors at the turbine inlet and outlet positions and can effectively reflect the turbine's flow characteristics, thermodynamic state, and potential faults. For instance, intake pressure and outlet pressure can reflect pressure loss to determine if there is leakage; intake temperature and outlet temperature can reflect temperature anomalies to determine if there is overheating; and turbocharger speed can reflect Mach number to determine if there is blockage. Specifically, in some embodiments, by setting up an engine and corresponding sensors on a test bench, various monitoring sub-data of the engine's turbine under different health states are acquired. Simultaneously, the turbine fault type corresponding to the detection sub-data under fault states is determined through various methods such as manual diagnosis, auxiliary instruments, or historical records.

[0041] Continue to refer to Figure 5Step S122 involves adjusting the parameters of the coupled simulation model according to different turbine fault types, and inputting each set of monitoring sub-data into the adjusted coupled simulation model corresponding to each turbine fault type to obtain the corresponding turbine internal state prediction data. In other words, by setting the coupled simulation model to normal operating conditions or different fault states, and inputting each set of monitoring sub-data into the set coupled simulation model, turbine internal state prediction data for the corresponding state is obtained. For example, in some embodiments, the turbine internal state prediction data includes predicted turbine efficiency, predicted output torque, predicted control surface pressure data and predicted control surface temperature data for each control surface, and predicted Mach number. These turbine internal state prediction data are influenced by the turbine's key geometric parameters and can reflect the flow state inside the turbine. Since turbine faults are correlated with turbine key geometric parameters, these turbine internal state prediction data can reflect turbine faults. Step S123 involves using each set of detection sub-data, the corresponding turbine internal state prediction data, and the corresponding turbine fault type as a training dataset. Understandably, the training dataset contains multiple sets of data, and each set of data contains a turbine fault type, a set of detection sub-data, and the turbine internal state prediction data corresponding to that turbine fault type.

[0042] For example, in some embodiments, steps S121 to S123 are performed as follows: First, sampling points, i.e., monitoring sub-data, are designed using the Design of Experiments (DOE) method across the entire operating range of the engine. This ensures that the samples are uniformly distributed in the multi-dimensional input space, thereby obtaining the most representative data with the fewest simulations. Then, the sampling points are input into a coupled simulation model for batch simulation, and the corresponding turbine internal state prediction data is recorded after each simulation run. Finally, all sample points and the corresponding turbine internal state prediction data are collected to form a training dataset.

[0043] Continue to refer to Figure 2 Step S13 involves constructing a proxy model corresponding to the coupled simulation model based on the training dataset. Further refer to... Figure 6 In some embodiments, step S13 includes the following sub-steps. Step S131 is to construct a Kriging model, wherein the correlation function of the Kriging model is a Gaussian correlation function. It should be noted that the Kriging model is essentially a Gaussian process regression, whose advantage lies in that it not only provides a better linear unbiased estimate of the predicted value, but also quantifies the uncertainty of the prediction result, which is crucial for evaluating the reliability of the surrogate model at unknown operating conditions. For example, the Kriging model... The expression is: , In the formula For global regression models, For a given set of variables with zero mean and covariance of , A random process. Among them, These are basis function vectors, and their types include constants, linear or quadratic polynomials, etc. For the regression coefficient vector, For process variance, The correlation function is used to describe the sample points. and Spatial correlation between them For related functions The parameters.

[0044] Furthermore, in the Kriging model, for a new prediction point... Its corresponding Kriging prediction value and mean square error The calculation expressions are as follows: , , in, , , , In the formula is n The correlation matrix of n, this correlation matrix The j-th element in the i-th row The expression is , The observed response vector is composed of the true output values ​​at all known sample points. , This is the design matrix. The middle line represents the basis function value for each sample point. , These are the regression coefficients estimated by generalized least squares. It is the correlation vector between the new predicted point and all known sample points.

[0045] Furthermore, in some embodiments, the expression for calculating the Gaussian correlation function is as follows: , In the formula As an input dimension, in some embodiments the total number of data types in the monitored sub-data is equal to the input dimension. These are the hyperparameters to be optimized, which control the range of influence of each input variable on the output.

[0046] Continue to refer toFigure 6 Step S132 involves maximizing the log-likelihood function using the training dataset and a numerical optimization algorithm to obtain an optimized Kriging model as the surrogate model. Specifically, the hyperparameters... Determined through maximum likelihood estimation, for example, by maximizing the log-likelihood function using a numerical optimization algorithm: ,in, It should be noted that this application does not limit the type of numerical optimization algorithm. In some embodiments, the numerical optimization algorithm includes genetic algorithms, and in other embodiments, the numerical optimization algorithm includes quasi-Newton methods.

[0047] Furthermore, in some embodiments, the trained surrogate model is validated using a set of independent test samples not used in training. The surrogate model whose prediction error is within an engineering-acceptable threshold is then used as the final surrogate model. For example, 10%–20% of the training data is extracted from the training dataset as independent test samples, and the remaining training data is used to optimize the Kriging model. Subsequently, the optimized Kriging model is evaluated using the aforementioned independent test samples, and it is determined whether each evaluation metric is within a threshold. The evaluation metrics include the coefficient of determination R0. 2 Key parameters include average absolute error and root mean square error, as well as efficiency and output torque. For example, the coefficient of determination R for output torque... 2 Whether it meets the requirement of being greater than 0.99 and whether the average relative error meets the requirement of being less than 1%. It should also be noted that in some embodiments, the final proxy model is deployed to the corresponding fault diagnosis system in the form of a generated dynamic link library or C code, so that the subsequent fault diagnosis system can quickly run the proxy model to output the corresponding data in milliseconds, and achieve the effect of real-time data monitoring and real-time data feedback.

[0048] In some embodiments, the turbine fault diagnosis model constructed in step S2 sequentially includes an input layer, an attention module, a hidden layer, and an output layer. In this embodiment, the input layer receives an extended feature vector. The extended feature vector includes multiple input features corresponding to each data point in the real-time monitoring data and each data point in the real-time turbine internal state data. Specifically, the number of neurons in the input layer is equal to the total number of data types in the real-time monitoring data and the real-time turbine internal state data, thereby enabling the input of the extended feature vector constructed from the real-time monitoring data and the real-time turbine internal state data into the turbine fault diagnosis model. In this embodiment, the attention module generates a corresponding attention weight vector based on the extended feature vector, and generates a weighted feature vector based on the attention weight vector and the extended feature vector. The attention weight vector includes multiple importance weights corresponding to each input feature. Specifically, in this embodiment, the attention module is a lightweight sub-network that receives the extended feature vector... The attention weight vector is generated by sequentially passing the data through a fully connected layer and a sigmoid activation function. This example's attention weight vector... The calculation expression is: , In the formula and These are the trainable parameters. Use the Sigmoid activation function and ensure the attention weight vector ,in This represents the number of neurons in the input layer.

[0049] The expression for calculating the weighted eigenvector in this embodiment is: , In the formula For weighted eigenvectors, The symbol for element-wise multiplication.

[0050] In this embodiment, the hidden layer is used to generate the hidden layer output based on the weighted feature vector. The hidden layer can be one or more layers, and the number of neurons in the hidden layer is configured according to the input dimension, the number of output categories, and the model complexity requirements. In this embodiment, when there is one hidden layer, the calculation expression for the hidden layer output is: , In the formula For output of the hidden layer, For activation function, and These are all parameters of the hidden layer.

[0051] In this embodiment, the output layer is used to generate the fault distribution vector based on the output of the hidden layer. The output layer comprises N neurons, where N is the total number of healthy states and all fault types. The output layer applies a Softmax activation function to output a fault distribution vector representing the probabilities of multiple categories. This fault distribution vector contains multiple vector elements, which sequentially represent the predicted probabilities of the diagnosed turbine system belonging to a normal state and multiple specific turbine fault types; that is, each vector element represents the predicted probability of a diagnostic result. The turbine fault diagnosis result is determined by the diagnosis result corresponding to the highest predicted probability in the fault distribution vector. The calculation expression for the fault distribution vector in this embodiment is: , In the formula The fault distribution vector, and For the parameters of the output layer, This is the Softmax activation function.

[0052] In some embodiments, to improve the fault diagnosis accuracy of the turbine fault diagnosis model, one or more of the following training and optimization methods are adopted for the turbine fault diagnosis model. Regarding the optimization and adjustment of the network parameters of the turbine fault diagnosis model, the Xavier method is used to ensure consistent activation variance, weights are extracted from a uniform distribution, bias is set to zero, and the initial learning rate is 0.001; the training process follows a supervised learning paradigm. Regarding model training, data from the training dataset is input into the turbine fault diagnosis model to be trained, and the predicted output is calculated through forward propagation. Subsequently, the error is quantified using the cross-entropy loss function, and the weights are updated through backpropagation. Furthermore, the training iterations are 1000 to 5000 epochs, the accuracy on the validation set is monitored, and an early stopping mechanism is applied to prevent overfitting. Regarding optimization and regularization, the following strategies are comprehensively adopted to improve training efficiency and enhance the model's generalization ability: mini-batch gradient descent strategy, where the batch size is 32~128; L2 regularization strategy, where λ=0.001; and dropout strategy, where the dropout rate is 0.2. The above strategies can be used individually or in combination. Furthermore, the Adam optimizer is employed to dynamically adjust the learning rate, ensuring training stability. For performance evaluation, metrics such as accuracy, recall, and F1 score are used, and macro-averaging is employed to address multi-class imbalance issues. Additionally, the training dataset is further divided into training, validation, and test sets. The training and validation sets are used for model training, while the test set is used to validate the effectiveness of the trained turbine fault diagnosis model in practical turbine fault diagnosis. For example, the ratio of the training, validation, and test sets is 8:1:1.

[0053] In some embodiments, the monitoring data includes intake pressure, outlet pressure, intake temperature, outlet temperature, and turbocharger speed. Turbine internal state data includes turbine efficiency, output torque, control surface compression data, control surface temperature data, and Mach number. The turbine internal state data corresponds to the turbine internal state prediction data mentioned earlier, thus reflecting turbine fault conditions from the perspective of turbine internal state. This allows the turbine fault diagnosis model to further combine the turbine internal state data, which cannot be obtained through sensors, with the monitoring data from the sensors to diagnose turbine faults, resulting in more accurate turbine fault diagnosis results.

[0054] The details of the turbine fault diagnosis method 100 have been explained above. The following will refer to... Figure 7 A complete description of the entire implementation of the turbine fault diagnosis method 100 in some embodiments is provided. For example... Figure 7As shown, the process begins with the construction of a physical model. Specifically, this involves building a turbine prediction model (i.e., a one-dimensional turbine simulation model) based on key turbine geometric parameters, and then correcting the GT-Power / Simulink coupled simulation model based on the turbine prediction model using experimental data. This involves connecting the one-dimensional turbine simulation model built in Simulink software and the whole-engine simulation model built in GT-Power software to obtain the coupled simulation model. Multiple experimental data obtained from whole-engine tests are then used to calibrate and correct the coupled simulation model. Next, turbine surrogate models and fault diagnosis models are constructed. In the turbine surrogate model construction, a turbine prediction model (i.e., a Kriging model) is first built, followed by simulation datasets under various operating conditions (i.e., training datasets). Finally, the turbine prediction model is trained using these simulation datasets to obtain the turbine surrogate model (the trained surrogate model). In the fault diagnosis model construction, datasets under healthy and fault conditions (i.e., training datasets) are first built based on the whole-engine coupled simulation model (i.e., the coupled simulation model). Then, a BP neural network is trained using these training datasets to obtain the turbine fault diagnosis model. Finally, the fault diagnosis system inference phase is performed. Specifically, the system acquires detection signals X (monitoring data) through sensors. This monitoring data is then input into a turbine surrogate model to obtain turbine state parameters Z (turbine internal state data). Finally, an extended feature vector constructed from the monitoring data and turbine internal state data is input into a fault diagnosis model (turbine fault diagnosis model) to output the turbine fault diagnosis result. The monitoring data includes inlet pressure, outlet pressure, inlet temperature, outlet temperature, and turbocharger speed. The turbine internal state data includes turbine efficiency, output torque, control surface pressure data, control surface temperature data, and Mach number. Furthermore, the final turbine diagnosis result includes the turbine fault diagnosis result and the attention weight vector. The attention weight vector visually demonstrates the contribution of each data point in the real-time monitoring data and real-time turbine internal state data to the turbine fault diagnosis result, thus providing a basis for achieving physically interpretable fault localization.

[0055] An embodiment of this application also proposes a method such as Figure 8 The turbine fault diagnosis system 200 shown is illustrated. According to... Figure 8 The turbine fault diagnosis system 200 may include an internal communication bus 21, a processor 22, a read-only memory (ROM) 23, a random access memory (RAM) 24, and a communication port 25. When applied to a personal computer, the turbine fault diagnosis system 200 may also include a hard disk 26.

[0056] The internal communication bus 21 enables data communication between components of the turbine fault diagnosis system 200. The processor 22 can perform judgments and issue prompts. In some embodiments, the processor 22 may consist of one or more processors. The communication port 25 enables data communication between the turbine fault diagnosis system 200 and external systems. In some embodiments, the turbine fault diagnosis system 200 can send and receive information and data from a network via the communication port 25.

[0057] The turbine fault diagnosis system 200 may also include different types of program storage units and data storage units, such as a hard disk 26, a read-only memory (ROM) 23, and a random access memory (RAM) 24, capable of storing various data files used for computer processing and / or communication, as well as possible program instructions executed by the processor 22. The processor 22 executes these instructions to implement the main part of the turbine fault diagnosis method. The results processed by the processor 22 are transmitted to the user equipment via a communication port and displayed on the user interface.

[0058] In addition, this application also proposes a computer-readable medium storing computer program code, which implements the above-described turbine fault diagnosis method when executed by a processor.

[0059] The basic concepts have been described above. Obviously, for those skilled in the art, the above disclosure is merely illustrative and does not constitute a limitation of this application. Although not explicitly stated herein, those skilled in the art may make various modifications, improvements, and corrections to this application. Such modifications, improvements, and corrections are suggested in this application, and therefore remain within the spirit and scope of the exemplary embodiments of this application.

[0060] Furthermore, this application uses specific terms to describe embodiments of the application. For example, "an embodiment," "one embodiment," and / or "some embodiments" refer to a particular feature, structure, or characteristic related to at least one embodiment of the application. Therefore, it should be emphasized and noted that "an embodiment," "one embodiment," or "an alternative embodiment" mentioned twice or more in different locations in this specification do not necessarily refer to the same embodiment. In addition, certain features, structures, or characteristics in one or more embodiments of the application can be appropriately combined.

[0061] Similarly, it should be noted that, in order to simplify the description of the present application and thus aid in the understanding of one or more embodiments, the foregoing description of the embodiments of the present application sometimes combines multiple features into a single embodiment, drawing, or description thereof. However, this disclosure method does not imply that the subject matter of the present application requires more features than those mentioned in the claims. In fact, the embodiments contain fewer features than all the features of the single embodiments disclosed above.

[0062] In some embodiments, numbers describing the quantity of components and attributes are used. It should be understood that such numbers used in the description of embodiments are modified in some examples with the terms "approximately," "approximately," or "generally." Unless otherwise stated, "approximately," "approximately," or "generally" indicates that the numbers are allowed to vary by ±20%. Accordingly, in some embodiments, the numerical parameters used in the specification and claims are approximate values, which may be changed depending on the characteristics required by individual embodiments. In some embodiments, numerical parameters should take into account specified significant digits and employ a general method of digit reservation. Although the numerical ranges and parameters used to confirm their breadth of scope in some embodiments of this application are approximate values, in specific embodiments, such values ​​are set as precisely as feasible.

[0063] Some aspects of this application can be executed entirely by hardware, entirely by software (including firmware, resident software, microcode, etc.), or by a combination of hardware and software. The aforementioned hardware or software may be referred to as a "data block," "module," "engine," "unit," "component," or "system." The processor may be one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DAPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), processors, controllers, microcontrollers, microprocessors, or combinations thereof. Furthermore, aspects of this application may manifest as computer products residing in one or more computer-readable media, including computer-readable program code. For example, computer-readable media may include, but are not limited to, magnetic storage devices (e.g., hard disks, floppy disks, magnetic tapes, etc.), optical discs (e.g., compressed CDs, digital multifunction DVDs, etc.), smart cards, and flash memory devices (e.g., cards, sticks, key drives, etc.).

[0064] A computer-readable medium may contain a propagated data signal containing computer program code, for example, on baseband or as part of a carrier wave. This propagated signal may take various forms, including electromagnetic, optical, and so on, or suitable combinations thereof. A computer-readable medium can be any computer-readable medium other than a computer-readable storage medium, which can be connected to an instruction execution system, apparatus, or device to enable communication, propagation, or transmission of a program for use. The program code located on the computer-readable medium can be propagated through any suitable medium, including radio, cable, fiber optic cable, radio frequency signals, or similar media, or any combination of the above media.

[0065] Although this application has been described with reference to specific embodiments, those skilled in the art should recognize that the above embodiments are only used to illustrate this application, and various equivalent changes or substitutions can be made without departing from the spirit of this application. Therefore, any changes or modifications to the above embodiments within the essential spirit of this application will fall within the scope of the claims of this application.

Claims

1. A turbine fault diagnosis method, characterized in that, Applicable to turbines located in engines, the turbine fault diagnosis method includes the following steps: Construct a training dataset and a proxy model corresponding to the engine, wherein the proxy model includes a mapping function for the simulation calculation process of the coupled simulation model corresponding to the engine; A neural network model is constructed, and the neural network model is trained based on the training dataset to obtain a trained neural network model as a turbine fault diagnosis model; The real-time monitoring data of the turbine is obtained, and the real-time monitoring data is input into the proxy model to obtain the corresponding real-time turbine internal state data; An extended feature vector is constructed based on the real-time monitoring data and the corresponding real-time turbine internal state data. The extended feature vector is then input into the turbine fault diagnosis model to obtain a fault distribution vector. Finally, the turbine fault diagnosis result is obtained based on the fault distribution vector.

2. The turbine fault diagnosis method as described in claim 1, characterized in that, The steps of constructing the training dataset and the agent model corresponding to the engine further include: Construct the coupled simulation model corresponding to the engine; Acquire historical monitoring data of the turbine, and construct the training dataset based on the coupled simulation model and the historical monitoring data; The proxy model corresponding to the coupled simulation model is constructed based on the training dataset.

3. The turbine fault diagnosis method as described in claim 2, characterized in that, The steps of constructing the coupled simulation model corresponding to the engine further include: Construct a one-dimensional simulation model of the turbine corresponding to the turbine; Construct a full-engine simulation model corresponding to the engine; The coupled simulation model is obtained by coupling the one-dimensional turbine simulation model and the whole machine simulation model. During the operation of the coupled simulation model, the turbine one-dimensional simulation model generates corresponding turbine data based on the whole machine data generated by the whole machine simulation model at the current time step, and uses the turbine data as the input of the whole machine simulation model at the next time step.

4. The turbine fault diagnosis method as described in claim 3, characterized in that, The one-dimensional simulation model of the turbine includes multiple control surfaces, including control surfaces corresponding to the volute casing, nozzle ring, impeller, and diffuser section. The real-time turbine internal status data includes control surface pressure data and control surface temperature data. The control surface pressure data includes the total outlet pressure corresponding to each control surface, and the control surface temperature data includes the total outlet temperature corresponding to each control surface.

5. The turbine fault diagnosis method as described in claim 4, characterized in that, The formula for calculating the total outlet pressure is: ， In the formula For the first The total outlet temperature of each control surface, For the first Total inlet temperature of each control surface For the first The inlet circumferential velocity of each control surface For the first The exit circumferential velocity of each control surface This is the specific heat capacity at constant pressure.

6. The turbine fault diagnosis method as described in claim 4, characterized in that, The formula for calculating the total outlet pressure is: ， In the formula For the first The total outlet pressure of each control surface For the first Total inlet pressure of each control surface The adiabatic index of the gas. The loss coefficient, For the first The exit Mach number of each control surface For the first The inlet circumferential velocity of each control surface For the first The exit circumferential velocity of each control surface For isobaric specific heat capacity, For the first Total inlet temperature of each control surface.

7. The turbine fault diagnosis method as described in claim 2, characterized in that, The historical monitoring data includes multiple sets of monitoring sub-data corresponding to the turbine under different operating conditions. The step of acquiring the historical monitoring data of the turbine and constructing the training dataset based on the coupled simulation model and the historical monitoring data further includes: Obtain the historical detection data; The parameters of the coupled simulation model are adjusted according to different turbine fault types, and the monitoring sub-data of each group are input into the adjusted coupled simulation model corresponding to each turbine fault type to obtain the corresponding turbine internal state prediction data. The detection sub-data of each group, the corresponding turbine internal state prediction data, and the corresponding turbine fault type are used as the training dataset.

8. The turbine fault diagnosis method as described in claim 2, characterized in that, The step of constructing the proxy model corresponding to the coupled simulation model based on the training dataset further includes: Construct a Kriging model, wherein the correlation function of the Kriging model is a Gaussian correlation function; The optimized Kriging model, obtained by maximizing the log-likelihood function using the training dataset and numerical optimization algorithm, serves as the surrogate model.

9. The turbine fault diagnosis method as described in claim 1, characterized in that, The fault distribution vector includes multiple vector elements, each vector element corresponding to a predicted probability of a multiple diagnostic result. The multiple diagnostic results include normal state and multiple different turbine fault types. The turbine fault diagnostic result is the diagnostic result corresponding to the largest predicted probability in the fault distribution vector.

10. The turbine fault diagnosis method as described in claim 1, characterized in that, The monitoring data includes intake pressure, outlet pressure, intake temperature, outlet temperature, and turbocharger speed. The turbine internal status data includes turbine efficiency, output torque, control surface pressure data, control surface temperature data, and Mach number.

11. The turbine fault diagnosis method as described in claim 1, characterized in that, The extended feature vector includes multiple input features corresponding to each data point in the real-time monitoring data and each data point in the real-time turbine internal state data, respectively. The turbine fault diagnosis model includes: Input layer, which is used to receive the extended feature vector; An attention module is configured to generate a corresponding attention weight vector based on the extended feature vector, and to generate a weighted feature vector based on the attention weight vector and the extended feature vector. The attention weight vector includes multiple importance weights corresponding to each of the input features. Hidden layer, the hidden layer being used to generate a hidden layer output based on the weighted feature vector; An output layer is used to generate the fault distribution vector based on the output of the hidden layer.

12. The turbine fault diagnosis method as described in claim 11, characterized in that, The final diagnostic result of the turbine includes the turbine fault diagnosis result and the attention weight vector.

13. A turbine fault diagnosis system, comprising: Memory is used to store instructions that can be executed by the processor; And a processor for executing the instructions to implement the turbine fault diagnosis method as described in any one of claims 1-12.

14. A computer-readable medium storing computer program code that, when executed by a processor, implements the turbine fault diagnosis method as described in any one of claims 1-12.

Images