Data and mechanism fused aero-engine gas path system fault diagnosis method

Abstract

The invention belongs to the technical field of aero-engine gas path system fault diagnosis, and particularly relates to a data and mechanism fused aero-engine gas path system fault diagnosis method, which comprises a fault detection method, and the fault detection method comprises the steps: 1, constructing a BP neural network, and initializing the weight and bias of the BP neural network; 2, based on the input and measurement output data of the gas path system, optimizing the weight and bias of the BP neural network by using a particle swarm optimization (PSO); 3, training the BP neural network based on the input and measurement output data of the gas path system to obtain a BP neural network output estimation model; and 4, calculating an output estimation sequence of the gas path system in real time by using the BP neural network output estimation model, calculating a residual sequence and a residual evaluation function thereof, and carrying out fault detection on the gas path system.

Description

Technical Field

[0001] This application belongs to the technical field of aero-engine gas path system fault diagnosis, and particularly relates to a method for diagnosing aero-engine gas path system faults by integrating data and mechanism. Background Technique

[0002] As one of the most important components of an aircraft, an aero-engine is a complex aerodynamic and thermal system integrating technologies such as aircraft, electricity, gas, and fluid. It usually works for a long time in a harsh environment, with extremely high temperature, pressure, rotational speed, vibration, and its load. Therefore, with the increase of operation time, the reliability of the aero-engine will gradually decrease, and the occurrence of faults is inevitable.

[0003] In the aero-engine system, faults are usually divided into gas path system faults, bearing system faults, and control system faults. Among them, gas path system faults are usually caused by factors such as component aging and increased mechanical friction, and are the most difficult to accurately detect and locate. Accurately and effectively diagnosing gas path system faults and taking corresponding remedial and maintenance measures are of great significance for improving the stability and safety of aero-engines and extending the service life of high aero-engines.

[0004] Currently, for the diagnosis of aero-engine gas path system faults, there are mainly model-based diagnosis methods and data-driven diagnosis methods.

[0005] Most of the model-based diagnosis methods are designed using Kalman filters, such as linear Kalman filters, extended Kalman filters, unscented Kalman filter designs, etc. However, due to the complexity of the structure of the aero-engine gas path system, it is difficult to measure its model-related parameters. Therefore, the model-based diagnosis methods cannot be well applied, and even false alarms or missed alarms may occur.

[0006] The data-driven diagnosis method uses the historical data of aero-engines for fault diagnosis, and mostly adopts the method based on subspace identification. This method requires multi-step identification represented by kernel functions, including strict equation operations and prior knowledge such as the order of the system, and has the following significant defects: 1) Directly introducing neural networks for training and verification, lacking theoretical analysis and proof, and having poor interpretability; 2) There are certain limitations for the fault diagnosis of linear systems with unknown parameters, including requiring the known system order and the matrix to be full row rank, etc.; 3) Using the neural network method, the network optimization is insufficient, the algorithm execution cycle is long, the operation efficiency is low, and it has high requirements for computer performance, and is not suitable for the rapid processing and analysis of massive aero-engine operation data and fault detection.

[0007] In view of the existence of the above technical defects, this application is proposed. Summary of the Invention

[0008] The objective of this application is to provide a fault diagnosis method for the gas path system of an aero-engine that integrates data and mechanism. It is a fault diagnosis method with no strict requirements for the gas path system. Without relying on the parameters of the airflow system, it can achieve the purpose of fault detection and isolation only through the input and output data of the operating gas path system of the aero-engine, so as to overcome or mitigate at least one aspect of the known technical deficiencies.

[0009] The technical solution of this application is as follows: A fault diagnosis method for the gas path system of an aero-engine that integrates data and mechanism, including a fault detection method, and this fault detection method includes: Step 1: Construct a BP neural network and initialize the weights and biases of the BP neural network; Step 2: Based on the input and measured output data of the gas path system, optimize the weights and biases of the BP neural network using the particle swarm optimization (PSO) algorithm; Step 3: Based on the input and measured output data of the gas path system, train the BP neural network to obtain the BP neural network output estimation model; Step 4: Use the BP neural network output estimation model to calculate the output estimation sequence of the gas path system in real time, calculate the residual sequence and its residual evaluation function, and perform fault detection on the gas path system.

[0010] Optionally, in the above-mentioned fault diagnosis method for the gas path system of an aero-engine that integrates data and mechanism, step 2 of the fault detection method is specifically as follows: S21: For the weights and biases of the BP neural network, initialize the particle swarm in the particle swarm optimization (PSO) algorithm; S22: Based on the input and measured output data of the gas path system, calculate the fitness function value of the particle swarm optimization (PSO) algorithm for the BP neural network; S23: Calculate the best position of a single particle and the best position of the particle swarm ; S24: Determine whether the maximum number of iterations has been reached, or whether the fitness function value is less than the set error threshold: If so, obtain the best position of a single particle and the best position of the particle swarm , and obtain the optimized values of the weights and biases of the BP neural network; If not, based on the best position of a single particle and the best position of the particle swarm , update the particle positions and velocities, and return to S22.

[0011] Optionally, in the above method for diagnosing faults in the gas path system of an aero-engine by integrating data and mechanism, in step four of the fault detection method, the output estimation sequence of the gas path system is calculated in real time by using the BP neural network output estimation model, specifically as follows: ; Among them, is the output estimation sequence of the gas path system, which is a stacked column vector with a length of starting from time ; is the input sequence of the gas path system, which is a stacked column vector with a length of starting from time ; is the input sequence of the gas path system, which is a stacked column vector with a length of starting from time ; is the health parameter sequence of the gas path system, which is a stacked column vector with a length of starting from time ; is the health parameter sequence of the gas path system, which is a stacked column vector with a length of starting from time ; is the measured output sequence of the gas path system, which is a stacked column vector with a length of starting from time ; are the hyperparameters of the BP neural network; is the structure function of the BP neural network.

[0012] Optionally, in the above method for diagnosing faults in the gas path system of an aero-engine by integrating data and mechanism, in step four of the fault detection method, the residual sequence is calculated, specifically as follows: ; Among them, is the residual sequence; is the measured output sequence of the gas path system, which is a stacked column vector with a length of starting from time ;

[0013] Optionally, in the above method for diagnosing faults in the gas path system of an aero-engine by integrating data and mechanism, in step four of the fault detection method, the residual evaluation function is calculated, specifically as follows: ; Among them, is the residual evaluation function; is expectation.

[0014] Optionally, in the above method for diagnosing faults in the gas path system of an aero-engine by integrating data and mechanism, in step four of the fault detection method, the fault detection of the gas path system is specifically as follows: Compare the residual evaluation function with the residual threshold to determine whether a fault has occurred in the gas path system. If , it is determined that no fault has occurred in the gas path system. If , it is determined that a fault has occurred in the gas path system.

[0015] Optionally, in the above method for diagnosing faults in the gas path system of an aero-engine by integrating data and mechanism, in step four of the fault detection method, the residual threshold is specifically calculated as: ; Among them, is the acceptable significance level of the gas path system fault; is the degree of freedom of the chi-square distribution.

[0016] Optionally, in the above method for diagnosing faults in the gas path system of an aero-engine by integrating data and mechanism, it further includes a fault isolation method, and this fault isolation method includes: Step 1: Construct a fault database of the gas path system, where each fault data includes a process input and a measurement output; Step 2: Determine the fault types corresponding to each fault data and set corresponding labels for each fault data; Step 3: Select the fault data as the input and the corresponding label as the output to train the fault isolation neural network to obtain a fault isolation neural network model; Step 4: Use the fault isolation neural network model to perform real-time calculations on the fault data of the gas path system, output the corresponding label, and then judge the corresponding fault type.

[0017] Optionally, in the above method for diagnosing faults in the gas path system of an aero-engine by integrating data and mechanism, in step three of the fault isolation method, the fault isolation neural network uses a BP neural network.

[0018] This application has at least the following beneficial technical effects: A fault diagnosis method for the gas path system of an aero-engine integrating data and mechanism is provided. A model for estimating the output is designed using a BP neural network for fault detection, which is feasible and interpretable. By calculating the residual sequence and its residual evaluation function and comparing it with the residual threshold for fault judgment, fault detection can be achieved only through input and output data, without relying on the parameters of the gas path system, and it has high engineering application value. On the basis of fault diagnosis, a fault isolation model is constructed using the neural network model to identify the fault type with high accuracy. Brief Description of the Drawings

[0019] Figure 1 is a schematic diagram of the BP neural network provided by an embodiment of the present application; Figure 2 is a schematic diagram of the fault detection provided by an embodiment of the present application; Figure 3 is a schematic diagram of the fault diagnosis method for the gas path system of an aero-engine integrating data and mechanism provided by an embodiment of the present application.

[0020] To better illustrate this embodiment, some contents in the drawings will be omitted, enlarged or reduced, which are only for exemplary illustration and should not be construed as a limitation to the present application. Detailed Embodiment

[0021] To make the technical solutions and their advantages of the present application clearer, the technical solutions of the present application will be further described clearly and completely in conjunction with the drawings. It can be understood that the specific embodiments described herein are only part of the embodiments of the present application, which are only used to explain the present application and not to limit the present application. It should be noted that for the convenience of description, only the parts related to the present application are shown in the drawings, and other related parts can refer to the general design.

[0022] In addition, unless otherwise defined, the technical terms or scientific terms used in the description of the present application should be the ordinary meanings understood by those of ordinary skill in the art to which the present application belongs. The "including" used in the description of the present application means that the concept appearing before this word covers the concepts listed after this word and their equivalents, without excluding other related concepts.

[0023] Under the data-driven framework, an embodiment of the present application designs a fault diagnosis method for the gas path system of an aero-engine integrating data and mechanism. Aiming at the limitations of the subspace identification method, by introducing a BP neural network, a residual generator is designed, a residual evaluation function is established, and without relying on system parameters, only through the input and output data of the aero-engine operation, fault detection and fault isolation are realized. At the same time, aiming at the problems existing in the traditional BP neural network, the particle swarm optimization algorithm PSO is introduced to optimize the relevant parameters of the network, improving the performance and fitting accuracy of the BP neural network.

[0024] The discrete time-invariant equations of the aero-engine gas path system are established as follows: …………(1.1) Where, is the state of the gas path system at the th moment, is the process input of the gas path system at the th moment, is the health parameter of the gas path system at the th moment, is the measured output of the gas path system at the th moment, is the process noise of the gas path system at the th moment, is the measurement noise of the gas path system at the th moment, and and are both uncorrelated zero-mean white noise signals.

[0025] For a given operating point, equation (1.1) is linearized and can be expressed in the following form: …………(1.2) Where, 、 、 、 、 、 are gas path system parameters with appropriate dimensions.

[0026] In the aero-engine system, the component efficiency and flow rate of the gas path system are key health parameters, which are unmeasurable parameters, and through the regression linearization process, the gas path system matrix is unknown. In particular, if the gas path system is affected by health parameter faults, equation (1.2) can be changed into the following form: …………(1.3) Where, is the gas path system fault vector to be diagnosed.

[0027] In the aero-engine system, for abnormal and unexpected events, faults will cause sudden changes in health parameters. For example, if there is a fault in the fan in the aero-engine system, health parameters such as fan efficiency and fan air flow will also change accordingly.

[0028] Due to the complex structure of the aero-engine gas path system, it is difficult to accurately measure the model parameters of the gas path system. Without relying on the gas path system parameters, fault detection and fault isolation can be achieved through input and output data. Equation (1.3) can be written in the form of the following transfer function: …… (1.4) where is the extended observability matrix, and its specific form is as follows: ………… (1.5) is the block Toeplitz matrix, and its specific form is as follows: ………… (1.6) is the healthy parameter-output transfer matrix, is the process noise-output transfer matrix. Both have the same structure, and In addition, is the measurement noise-output transfer matrix, which is the identity matrix.

[0029] is the measurement output sequence of the gas path system, which is a stacked column vector with a length of starting from time and is defined as: ………… (1.7) The gas path system input sequence , the healthy parameter sequence , the fault sequence , and the process noise sequence , the measurement noise and have the same structure.

[0030] Both the performance degradation and component abnormal faults of the aero-engine will cause changes in the healthy parameters. The changes in the healthy parameters caused by performance degradation are relatively slow and are generally treated as no faults. The occurrence of component abnormal faults is often sudden, which will cause changes in the healthy parameters related to the faults and cause the actual output to deviate from the normal output. For related fault diagnosis, the following assumptions can be made: Assumption 1: Under normal conditions, the healthy parameters of the aero-engine remain unchanged throughout the entire operating cycle; Assumption 2: In each fault case, only the healthy parameters related to the fault will change, and the remaining healthy parameters remain unchanged, with a default value of 1.

[0031] Currently, for a linear system with unknown parameters, the fault diagnosis method based on subspace identification can write Equation (1.3) in the form of equation state estimation as follows: …… (2.1) where is the estimated state of the gas path system at the th moment, is the covariance matrix of the zero-mean signal, is the Kalman gain.

[0032] Based on Equation (2.1), Equation (1.4) can be rewritten in the following form: …… (2.2) where is a block Toeplitz matrix, and has the same structure, is a stacked column vector, similar to the output sequence .

[0033] The corresponding observer of Equation (1.3) can be established in the following form: …… (2.3) where , , .

[0034] To eliminate , which is usually unknown in practice, the following form of equation can be obtained from Equation (2.3): …… (2.4) Since has eigenvalues inside the unit circle, for a large , ≈ 0, then the state can be further expressed in the following form: …… (2.5) In the case of no fault, substituting Equation (2.5) into Equation (2.2), we get: …… (2.6) In Equation (2.6), all the gas path system parameters are unknown. The fault diagnosis method based on subspace identification is developed by identifying the left null space of , denoted by , and a data matrix 。 composed of past input and output data is introducedHowever, when this method is applied, there are limitations such as the system order must be known, and the matrix must be row full rank. Once the order is unknown or the matrix is not row full rank, it cannot be used.

[0035] Aiming at the limitations of the subspace identification diagnosis method, the powerful fitting regression ability and learning ability of neural networks can be utilized, and it can be designed by combining the periodic operation data of aeroengines.

[0036] Generally speaking, the learning process of the BP neural network includes two stages: signal forward propagation and error backpropagation. It includes an input layer, a hidden layer, and an output layer. As Figure 1 shown, the mathematical relationship between the three layers can be described as follows: …………(3.1) Among them, 、 、 are the input of the input layer, the value of the hidden layer, and the output of the output layer respectively, 、 are the activation functions of the hidden layer and the output layer respectively; 、 、 、 are the weights and biases of the hidden layer and the output layer respectively.

[0037] The hyperparameters of the BP neural network are defined as 、 , and the overall model of the BP neural network is established as follows: …………(3.2) Among them, 、 are the input and output of the BP neural network respectively, is the structure function of the network.

[0038] Through equation (2.6), the BP neural network of equation (1.2) can be expressed as follows: …(3.3) Among them, is the output estimation sequence of the gas path system, and is the self The stacked column vectors of length starting from the moment ; is the input sequence of the gas path system, and is from The stacked column vectors of length starting from the moment ; is the input sequence of the gas path system, and is from The stacked column vectors of length starting from the moment ; is the sequence of health parameters of the gas path system, and is from The stacked column vectors of length starting from the moment ; is the sequence of health parameters of the gas path system, and is from The stacked column vectors of length starting from the moment ; is the measured output sequence of the gas path system, and is from The stacked column vectors of length starting from the moment .

[0039] Based on Equation (3.3), the loss function of the BP neural network is specifically as follows: ………… (3.4) The hyperparameter can be calculated as follows: ………… (3.5) The residual signal of the BP neural network , is calculated as follows: ………… (3.6) The above constructs the representation equation of the BP neural network of the gas path system, and establishes the connection between the BP neural network model and the state space model. The feasibility of the related fault diagnosis method can be theoretically proven. In addition, by using the available data of the aero-engine health management cycle, the BP neural network can be trained and learned to reduce the residual signal .

[0040] Although the BP neural network can be used to learn the input-output relationship of the gas path system, it has some limitations, such as slow convergence speed and long training time. On the other hand, the selection of the initial weights and biases of the BP neural network has a great impact on its network performance. For example, improper selection of the initial weights and biases of the network will affect the training time and convergence, and even lead to local optimization. To address this problem, the particle swarm optimization (PSO) algorithm can be introduced to optimize the initial weights and biases, improving the performance and fitting accuracy of the BP neural network.

[0041] The particle swarm optimization (PSO) algorithm is a global stochastic algorithm for searching the optimal solution, which obtains optimization by tracking the best position of each individual particle and the best position of the particle swarm and dynamically updates the position of the particle in each iteration of the search process.

[0042] Suppose particles constitute a particle swarm in the -dimensional search space, and there is a population . Then the position of the -th particle in the search space is , the velocity is , the individual extreme value of the -th particle is , and the global extreme value of the population is

[0043] In each iteration of the search process, the velocity and position of the particle are updated through the best position of each individual particle and the best position of the particle swarm . The update is as follows: ………… (4.1) where the superscripts , +1 represent the iteration number, is the inertia weight, 、 are non-negative acceleration factors, and 、 are random numbers between [0, 1].

[0044] The PSO algorithm does not directly optimize the objective function, but optimizes it through the fitness function corresponding to the objective function. Therefore, to optimize the initial weights and biases of the BP neural network, the fitness function of the PSO algorithm can be constructed as follows: ; where is the a measurement output, which is the th output estimation of the BP neural network.

[0045] Based on the above, for the fault diagnosis method of the aero-engine gas path system integrating data and mechanism, a fault detection method is designed, as Figure 2 shown, and the specific reference is as follows.

[0046] Step 1: Construct a BP neural network.

[0047] Set the number of neurons, the number of network layers, the learning rate, and the activation function of the BP neural network, and initialize the weights and biases of the BP neural network.

[0048] Step 2: Based on the input and measurement output data of the gas path system, optimize the weights and biases of the BP neural network using the particle swarm optimization algorithm PSO.

[0049] S21. For the weights and biases of the BP neural network, initialize the particle swarm in the particle swarm optimization algorithm PSO, including particle positions, velocities, and including the number of particles , the search space dimension , the inertia weight , the non-negative acceleration factor 、 and random numbers 、 etc.

[0050] S22. Based on the input and measurement output data of the gas path system, calculate the fitness function value of the particle swarm optimization algorithm PSO for the BP neural network.

[0051] S23. Calculate the best position of a single particle and the best position of the particle swarm .

[0052] Specifically, the fitness function value of each particle can be calculated and compared with the fitness of the historical position to obtain the best individual position , and compare the fitness of all particle positions in the current iteration to obtain the optimal global position .

[0053] S24. Determine whether the maximum number of iterations has been reached, or whether the fitness function value is less than the set error threshold, that is, determine whether the constraint condition is satisfied: If so, obtain the best position of a single particle and the best position of the particle swarm , and obtain the optimized values of the weights and biases of the BP neural network; Otherwise, based on the best position of a single particle and the best position of the particle swarm , update the particle position and velocity, specifically referring to Equation (4.1), and return S22.

[0054] Step 3: Based on the input and measurement output data of the gas path system, train the BP neural network to obtain the BP neural network output estimation model, specifically expressed referring to Equation (3.3), where the hyperparameters , the optimal hyperparameters can be taken * .

[0055] The above uses the PSO algorithm to optimize the initial values of the weights and biases of the BP neural network by the particle swarm algorithm PSO for the BP neural network, which can make the BP neural network have optimized initial values of weights and biases, can improve the performance of the BP neural network, and enable the BP neural network to have a better ability to learn the input-output relationship of Equation (3.3), and efficiently obtain the BP neural network output estimation model.

[0056] The BP neural network output estimation model is trained using the normal input-output data of the gas path system. If the BP neural network training is accurate, then once a fault occurs, there will be a deviation between the output of the BP neural network output estimation model and the actual output. Therefore, a fault can be detected by designing the residual.

[0057] Step 4: Use the BP neural network output estimation model to calculate the output estimation sequence of the gas path system in real time, calculate the residual sequence and its residual evaluation function, and perform fault detection on the gas path system.

[0058] Calculate the residual sequence as follows: …………(5.1) where is the residual sequence.

[0059] Calculate the residual evaluation function as follows: …………(5.2) where is the residual evaluation function; is the expectation of.

[0060] Assume that the process noise and measurement noise follow a Gaussian distribution. Then, the test can be used for residual evaluation, , where is the degree of freedom of the chi-square distribution, which determines the shape of the distribution and usually takes natural numbers.

[0061] Compare the residual evaluation function with the residual threshold to determine whether the gas path system has a fault. If , it is determined that the gas path system has no fault. If , it is determined that the gas path system has a fault.

[0062] For the residual threshold , it can be specifically designed and calculated as follows: ; where is the acceptable significance level for gas path system faults. The specific value can be selected and determined by those skilled in the art according to the specific actual situation when applying the technical solution disclosed in this application.

[0063] In the above-mentioned method for diagnosing faults in the gas path system of an aero-engine by fusing data and mechanism, a fault detection method based on the PSO-BP neural network is designed within the data-driven framework. The BP neural network is constructed and trained using the normal data designed. In order to overcome the disadvantages of the BP neural network such as slow convergence speed and long training time, the particle swarm optimization algorithm PSO is used to optimize the initial weights and biases of the BP neural network, and the BP neural network output estimation model is efficiently trained. Then, the output estimation sequence is calculated based on the BP neural network output estimation model, and the residual sequence and its residual evaluation function are calculated to realize the detection of faults in the gas path system.

[0064] After detecting a fault in the gas path system, it is necessary to further identify the fault type to achieve fault isolation. It has strong classification ability and can be used to isolate faults. Based on this, a fault isolation method is designed as follows: Step 1: Construct a fault database for the gas path system, where each fault data includes a process input and a measurement output.

[0065] As much historical fault data of the gas path system as possible can be collected to establish an over-full fault database, so as to have a greater degree of freedom in selecting candidate fault data and make the number of fault types included in the candidate fault data much larger than the required number.

[0066] Step 2: Determine the fault type corresponding to each fault data and set the corresponding label for each fault data.

[0067] Step 3: Select the fault data as the input and the corresponding label as the output to train the fault isolation neural network to obtain the fault isolation neural network model.

[0068] Specifically, a BP neural network can be adopted for the fault isolation neural network.

[0069] Step 4: Use the fault isolation neural network model to perform real-time calculation on the fault data of the gas path system, output corresponding labels, and then judge the corresponding fault types to achieve fault isolation.

[0070] After fault detection, input the data corresponding to the occurrence of a fault in the gas path system into the fault isolation neural network model for calculation. If the type of fault that occurs coincides with one of the fault types corresponding to the gas path system faults in the fault database, the label calculated by the fault isolation neural network model will be significantly close to 1 for this fault type, while close to 0 for other labels. The following fault table can be constructed:

[0071] In the method for diagnosing faults in the gas path system of an aero-engine by fusing data and mechanism disclosed in the above embodiments, two neural network operations are performed in the designed fault detection and isolation methods. First, in the fault detection stage, a neural network is trained using the normal input and output data of the gas path system to obtain a neural network model for output estimation, and then residuals are used for fault detection. This can be considered as utilizing the fitting and regression ability of the neural network. In the fault isolation stage, another neural network is trained using the fault data and labels of the gas path system to obtain a neural network model for fault isolation for fault identification. This can be considered as utilizing the classification ability of the neural network, as Figure 3 shown.

[0072] In the method for diagnosing faults in the gas path system of an aero-engine by fusing data and mechanism disclosed in the above embodiments, a model for estimating output is designed using a BP neural network for fault detection, which is feasible and interpretable. Moreover, by calculating the residual sequence and its residual evaluation function and comparing it with the residual threshold for fault judgment, fault detection can be achieved only through input and output data without relying on the parameters of the gas path system, having high engineering application value. On the basis of fault diagnosis, a model for fault isolation is constructed using the neural network model to identify fault types with high accuracy.

[0073] In addition, in the method for diagnosing faults in the gas path system of an aero-engine by fusing data and mechanism disclosed in the above embodiments, aiming at problems such as slow convergence speed, long training time, and high computational cost of the BP neural network, the particle swarm optimization algorithm PSO is introduced to optimize the relevant parameters of the network, improving the performance and fitting accuracy of the BP neural network and enabling the rapid construction of the estimation output model of the BP neural network.

[0074] So far, the technical solutions of this application have been described in conjunction with the preferred embodiments shown in the accompanying drawings. Those skilled in the art should understand that the protection scope of this application is obviously not limited to these specific embodiments. Without departing from the principle of this application, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the protection scope of this application.

Claims

1. A method for diagnosing faults in an aircraft engine gas path system by integrating data and mechanism, characterized in that: A fault detection method is included, the fault detection method comprising: Step 1: Construct a BP neural network and initialize the weights and biases of the BP neural network; Step 2: Based on the input and measured output data of the gas path system, the weight and bias of the BP neural network are optimized using the particle swarm algorithm PSO; Step 3: Based on the input and measured output data of the gas path system, the BP neural network is trained to obtain a BP neural network output estimation model; Step 4: Use the BP neural network output estimation model to calculate the output estimation sequence of the gas path system in real time, calculate the residual sequence and its residual evaluation function, and perform fault detection on the gas path system; Step 2 of the fault detection method is specifically as follows: S21, for the weight and bias of the BP neural network, initialize the particle swarm in the particle swarm algorithm PSO; S22, based on the input and measured output data of the gas path system, calculating the fitness function value of the particle swarm algorithm PSO for the BP neural network; S23. Calculate the optimal position of a single particle in the particle swarm algorithm PSO and the optimal position of the particle swarm ; S24, determine whether the maximum number of iterations has been reached, or whether the fitness function value is less than the set error threshold: If so, the optimal position of a single particle is obtained and the optimal position of the particle swarm , get the optimized values ​​of weight and bias of BP neural network; If not, then based on the best position of a single particle and the optimal position of the particle swarm , update the particle position and velocity, and return to S22.

2. The method for diagnosing faults in an aircraft engine gas path system by integrating data and mechanism according to claim 1, characterized in that: In step 4 of the fault detection method, the output estimation sequence of the gas path system is calculated in real time using the BP neural network output estimation model, specifically: ； in, is the output estimation sequence of the gas system, The length from time A stacked column vector of ; is the input sequence of the gas system, The length from time A stacked column vector of ; is the input sequence of the gas system, The length from time A stacked column vector of ; is the health parameter sequence of the gas system, The length from time A stacked column vector of ; is the health parameter sequence of the gas system, The length from time A stacked column vector of ; is the measurement output sequence of the gas system, The length from time A stacked column vector of ; is the hyperparameter of BP neural network; is the structural function of the BP neural network.

3. The method for diagnosing faults in an aircraft engine gas path system by integrating data and mechanism according to claim 2, characterized in that: In step 4 of the fault detection method, the residual sequence is calculated, specifically: ； in, is the residual sequence; It is the measurement output sequence of the gas system. The length from time A stacked column vector of .

4. The method for diagnosing faults in an aircraft engine gas path system by integrating data and mechanism according to claim 3 is characterized in that: In step 4 of the fault detection method, the residual evaluation function is calculated, specifically: ； in, is the residual evaluation function; for expectations.

5. The method for diagnosing faults in an aircraft engine gas path system by integrating data and mechanism according to claim 4, characterized in that: In step 4 of the fault detection method, the gas circuit system is subjected to fault detection, specifically: The residual evaluation function With residual threshold Compare and judge whether the gas system fails. , it is judged that there is no fault in the gas system. , it is judged that the gas system is faulty.

6. The method for diagnosing faults in an aircraft engine gas path system by integrating data and mechanism according to claim 5, characterized in that: In step 4 of the fault detection method, the residual threshold The specific calculation is: ； in, The acceptable significance level of gas path system failure; is the degrees of freedom of the chi-square distribution.

7. The method for diagnosing faults in an aircraft engine gas path system by integrating data and mechanism according to claim 6, characterized in that: Also included is a fault isolation method, the fault isolation method comprising: Step 1: construct a fault database of the gas circuit system, where each fault data includes a process input and a measurement output; Step 2: determine the fault type corresponding to each fault data, and set a corresponding label for each fault data; Step 3: Select fault data as input, take corresponding labels as output, train the fault isolation neural network, and obtain a fault isolation neural network model; Step 4: Use the fault isolation neural network model to perform real-time calculations on the fault data of the gas circuit system, output corresponding labels, and then determine the corresponding fault type.

8. The method for diagnosing faults in an aircraft engine gas path system by integrating data and mechanism according to claim 7, characterized in that: In step three of the fault isolation method, the fault isolation neural network adopts a BP neural network.

Images