A fault diagnosis and fault-tolerant regression method for embedded atmospheric data sensing system

Abstract

The application discloses a fault diagnosis and fault-tolerant method of an embedded atmospheric data sensing system, and comprises the following steps: estimating dimensionless atmospheric data and a pressure coefficient based on different pressure hole layouts based on a trained dimensionless input-output convolutional neural network; calculating the estimated pressure of each pressure hole in the layout based on the estimated pressure coefficient; calculating the error between the estimated pressure and the actual pressure to obtain the pressure distribution error of each layout, and determining the faulty sensor according to the pressure distribution error of multiple pressure hole layouts; and selecting the pressure hole layout with the minimum pressure distribution error based on the minimum error principle, and outputting the atmospheric data corresponding to the layout. The scheme can improve the accuracy and robustness of atmospheric data estimation.

Description

Technical Field

[0001] This invention relates to the field of atmospheric data processing technology, specifically to a fault diagnosis and fault-tolerant regression method, device, computing equipment, and storage medium for an embedded atmospheric data sensing system. Background Technology

[0002] An embedded air data sensing system (FADS) is a system that uses redundant measurements of pressure distribution on an aircraft surface to calculate atmospheric data. FADS systems typically have a series of pressure measurement ports at the front of the aircraft. Airflow is introduced through these ports into a pressure sensor array chamber to measure the pressure distribution data on the aircraft surface. This data is then used to calculate atmospheric parameters such as Mach number, angle of attack, sideslip angle, dynamic pressure, and static pressure. If sensors fail or data becomes abnormal, the accuracy of the atmospheric data is affected, potentially leading to serious consequences such as aircraft crashes. Therefore, researching and developing highly reliable and fault-tolerant air data sensing systems is crucial for ensuring flight safety.

[0003] Currently, FADS fault-tolerant algorithms mainly involve developing multiple FADS solution models, each handling different pressure orifice layouts, and outputting multiple atmospheric data based on model fusion strategies such as voting mechanisms. For example, for an FADS system with 11 pressure measurements, 40 different neural networks are used to obtain several results by employing different combinations of pressure signals, and finally, a voting strategy is used to determine the final atmospheric data. This fault-tolerant algorithm requires the development of multiple models, resulting in high complexity. Each solution model has poor generalization ability and struggles to handle complex and diverse sensor measurement biases. Furthermore, traditional voting strategies are sensitive to outliers, easily leading to misjudgments and affecting the accuracy of atmospheric data. Summary of the Invention

[0004] To address the technical problems existing in the prior art, this invention proposes a fault diagnosis and fault-tolerant regression method, device, computing equipment, and storage medium for an embedded atmospheric data sensing system. It uses a dataset containing various pressure gauge layouts to obtain a trained dimensionless input-output convolutional neural network model. This model can estimate atmospheric data based on various layouts, identify faulty sensors by comparing the errors between measured and estimated pressures for different layouts, and select the optimal pressure gauge layout based on the principle of minimum error, outputting the corresponding atmospheric data. This approach can effectively address various sensor faults and deviations, achieve more reliable fault-tolerant regression, and improve the system's robustness.

[0005] According to a first aspect of the present invention, a fault diagnosis and fault-tolerant regression method for an embedded atmospheric data sensing system is provided, comprising: estimating dimensionless atmospheric data and pressure coefficients based on different pressure gauge layouts using a trained dimensionless input-output convolutional neural network; calculating the estimated pressure of each pressure gauge within the layout based on the estimated pressure coefficients; calculating the error between the estimated pressure and the actual pressure to obtain the pressure distribution error of each layout; determining the faulty sensor based on the pressure distribution errors of multiple pressure gauge layouts; selecting the pressure gauge layout with the smallest pressure distribution error based on the principle of minimum error, and outputting the atmospheric data corresponding to that layout.

[0006] The above-described technical solution utilizes a pre-trained dimensionless input-output convolutional neural network to directly process atmospheric data with various layouts, eliminating the need for additional preprocessing or post-processing and simplifying the design and implementation of the entire system. This solution possesses real-time fault diagnosis and fault-tolerant regression capabilities. Even in the event of partial sensor failure or data anomalies, it can still perform data estimation and processing through fault-tolerant mechanisms, ensuring the stability and reliability of the system.

[0007] Optionally, in the fault diagnosis and fault-tolerant regression method for the embedded atmospheric data sensing system provided by the present invention, the dimensionless input-output convolutional neural network can be trained through the following steps: A training set is constructed, comprising input data and output data. The input data includes pressure ratio matrices for different pressure gauge layouts, and the output data includes dimensionless atmospheric data and pressure coefficients for each pressure gauge. The dimensionless input-output convolutional neural network is trained using the training set, and its parameters are optimized based on the mean squared error loss function to obtain a trained dimensionless input-output convolutional neural network. The dimensionless input-output convolutional neural network comprises multiple sequentially connected convolutional layers and multiple fully connected layers.

[0008] The above technical solution, by constructing a dataset containing multiple different pressure measurement hole layouts, can more comprehensively cover various possible situations and layout changes, thereby improving the model's generalization ability and adaptability.

[0009] Optionally, in the fault diagnosis and fault-tolerant regression method for the embedded atmospheric data sensing system provided by the present invention, the pressure reading of each pressure measuring hole in the embedded atmospheric data sensing system is obtained; the ratio between the pressure reading of each pressure measuring hole and its own and the other pressure readings is calculated to obtain a pressure ratio matrix, which is expressed as: In the formula, Ω is the pressure ratio matrix, p i (i = 1, 2, ..., n) represents the pressure reading of the i-th pressure measuring hole; Construct a health matrix and apply it to the pressure ratio matrix to obtain a dataset containing different pressure gauge layouts: Ω′=QΩQ, where Q=diag(q1,q2,...,q n ), q i The fault indication parameter is the pressure reading of the i-th pressure measuring hole. When the fault indication parameter is 0, the corresponding pressure reading is not included in the atmospheric data estimation; when the fault indication parameter is 1, the corresponding pressure reading is included in the atmospheric data estimation.

[0010] The above technical solution uses a health matrix to identify the rows and columns corresponding to fault readings, expanding the initial database into a larger fault database, enabling the obtained model to accurately predict atmospheric data based on different pressure gauge layouts.

[0011] Optionally, in the fault diagnosis and fault-tolerant regression method for the embedded atmospheric data sensing system provided by the present invention, a convolutional neural network trained based on a pressure ratio matrix with layout k is input, and the outputs the Mach number, angle of attack, and sideslip angle estimated based on layout k: and the pressure coefficient based on layout k: Where k is the number of the pressure testing hole layout, This is the estimated pressure coefficient for the i-th pressure measuring orifice based on layout k.

[0012] Optionally, in the fault diagnosis and fault-tolerant regression method for the embedded atmospheric data sensing system provided by the present invention, the free flow pressure and static pressure are calculated based on the measured pressure distribution and the corresponding health matrix based on layout k. In the formula, This is the estimated pressure coefficient for the i-th pressure measuring orifice based on layout k. For free flow pressure, For free-flow static pressure, Q(k) is the health matrix based on layout k, and pi (i = 1, 2, ..., n) is the pressure reading of the i-th pressure measuring hole; Based on the free-flowing pressure and static pressure, and the estimated pressure coefficient, calculate the estimated pressure at each pressure gauge based on layout k: In the formula, It is the estimated pressure at the i-th pressure measuring hole based on layout k.

[0013] The above technical solution introduces physical knowledge to calculate a criterion, which is used to select the most suitable layout from the pressure estimation results of each layout, and then outputs the most accurate atmospheric data estimate.

[0014] Optionally, in the fault diagnosis and fault-tolerant regression method for the embedded atmospheric data sensing system provided by the present invention, the pressure distribution error of layout k is calculated based on the difference between the actual pressure reading and the estimated pressure of the pressure measuring hole in layout k: In the formula, e(k) represents the pressure distribution error of layout k. The fault indication parameter is the i-th pressure reading in layout k; Compare the error magnitudes of each pressure test port layout; determine the faulty sensor based on the error magnitudes and fault indication parameters of the pressure readings.

[0015] Optionally, in the fault diagnosis and fault-tolerant regression method of the embedded atmospheric data sensing system provided by the present invention, the pressure orifice layout with the lowest error level is selected as the optimal layout, and the optimal layout and its corresponding atmospheric data are output.

[0016] The above technical solutions help to promptly detect and resolve faulty sensors in the system, ensuring that the output atmospheric data has high accuracy.

[0017] According to a second aspect of the present invention, a fault diagnosis and fault-tolerant regression device for an embedded atmospheric data sensing system is provided, comprising: a prediction module, a calculation module, a fault diagnosis module, and an output module. The prediction module is used to estimate dimensionless atmospheric data and pressure coefficients based on different pressure orifice layouts using a trained dimensionless input-output convolutional neural network. The calculation module is used to calculate the estimated pressure of each pressure gauge within the layout based on the estimated pressure coefficient; The fault diagnosis module is used to calculate the error between the estimated pressure and the actual pressure, obtain the pressure distribution error for each layout, and determine the faulty sensor based on the pressure distribution error. The output module is used to select the pressure measurement hole layout with the smallest pressure distribution error based on the principle of minimum error, and output the atmospheric data corresponding to the layout.

[0018] According to a third aspect of the present invention, a computing device is provided, comprising: at least one processor; and a memory storing program instructions, wherein the program instructions are configured to be executed by the at least one processor, the program instructions including instructions for performing the fault diagnosis and fault-tolerant regression method of the embedded atmospheric data sensing system described above.

[0019] According to a fourth aspect of the present invention, a readable storage medium storing program instructions is provided, which, when read and executed by a computing device, causes the computing device to perform the above-described fault diagnosis and fault-tolerant regression method for an embedded atmospheric data sensing system.

[0020] The fault diagnosis and fault-tolerant regression method and apparatus for embedded atmospheric data sensing systems provided by this invention can not only improve the accuracy and robustness of atmospheric data, but also effectively handle various sensor faults and biases. Compared with traditional voting methods, by adopting the principle of minimum error to select the optimal sensor layout, misjudgments are avoided and the processing efficiency of atmospheric data is improved. The implementation of this algorithm will bring more reliable fault diagnosis and fault-tolerant regression capabilities to aircraft systems, which is of great significance for ensuring the safety and stability of aircraft.

[0021] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and in order to make the above and other objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention are described below. Attached Figure Description

[0022] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings: Figure 1 A structural diagram of a computing device 100 according to an embodiment of the present invention is shown; Figure 2 A flowchart illustrating a fault diagnosis and fault-tolerant regression method 200 for an embedded atmospheric data sensing system according to an embodiment of the present invention is shown. Figure 3 A schematic diagram of a dimensionless input-output convolutional neural network according to an embodiment of the present invention is shown. Figure 4 A schematic diagram of an error criterion calculation process incorporating physical knowledge according to an embodiment of the present invention is shown. Figure 5 A schematic diagram illustrating the selection process of the optimal pressure measurement orifice layout based on the principle of minimum error according to an embodiment of the present invention is shown. Figure 6 A schematic diagram of the structure of a fault diagnosis and fault-tolerant regression device 600 for an embedded atmospheric data sensing system according to an embodiment of the present invention is shown. Figure 7 A simplified three-dimensional view and corresponding three-view drawing of the supersonic vehicle precursor model are shown. Figure 8 This diagram illustrates a comparison of the atmospheric data estimation errors of the DIONNFADS and FT-DIONNFAS methods on the test set for normal samples. Figure 9This diagram illustrates a comparison of the atmospheric data estimation errors of the DIONNFADS and FT-DIONNFAS methods for fault samples on the test set. Figure 10 A schematic diagram of the pressure distribution error corresponding to 26 combinations of pressure measuring holes is shown. Detailed Implementation

[0023] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0024] Figure 1 A structural diagram of a computing device 100 according to an embodiment of the present invention is shown. Figure 1 As shown, computing device 100 may include memory 106 and processor 104. Memory bus 108 can be used for communication between processor 104 and system memory 106.

[0025] Memory 106 may include operating system 120, applications 122, and program data 124. Application 122 may be configured to execute instructions on the operating system using program data 124. Application 122 includes program instructions for implementing various user-desired functions.

[0026] When computing device 100 starts up, processor 104 reads and executes program instructions from memory 106 of operating system 120. Application 122 runs on operating system 120, utilizing interfaces provided by operating system 120 and underlying hardware to implement various user-expected functions. When user starts application 122, application 122 is loaded into memory 106, and processor 104 reads and executes program instructions from memory 106 of application 122.

[0027] The computing device 100 also includes a storage device 132 and an output device 142, the storage device 132 being connected to a storage interface bus 134. An interface bus 140 facilitates communication from various interface devices (e.g., output device 142, peripheral interface 144, and communication device 146) via the bus / interface controller 130.

[0028] Peripheral interface 144 may include serial interface controller 154 and parallel interface controller 156, which may be configured to facilitate communication via one or more I / O ports 158 with external devices such as input devices (e.g., keyboard, mouse, pen, voice input device, touch input device) or other peripherals (e.g., printer, scanner, etc.). Communication device 146 may include network controller 160, which may be arranged to facilitate communication with one or more other computing devices 162 via a network communication link through one or more communication ports 164. In the computing device 100 according to the invention, application 122 includes instructions for performing the fault diagnosis and fault-tolerant regression method 200 of the embedded atmospheric data sensing system of the invention.

[0029] Figure 2 A flowchart illustrating a fault diagnosis and fault-tolerant regression method 200 for an embedded atmospheric data sensing system according to an embodiment of the present invention is shown. Figure 2 As shown, the method 200 begins at step S210, which estimates dimensionless atmospheric data and pressure coefficients based on different pressure orifice layouts using a trained dimensionless input-output convolutional neural network.

[0030] Data sets with different layouts can simulate performance under sensor failure or deviation conditions. By covering a variety of pressure gauge layouts, the performance of the fault-tolerant model under various conditions can be comprehensively evaluated, improving the model's generalization ability.

[0031] In one embodiment of the present invention, a dimensionless input-output convolutional neural network can be trained through the following steps: Construct a dataset containing multiple pressure gauge layouts. Specifically, acquire the pressure readings of each pressure gauge in the embedded atmospheric data sensing system. Calculate the ratio of each pressure gauge reading to its own and the remaining pressure readings to obtain the pressure ratio matrix, which can be represented as: In the formula, Ω is the pressure ratio matrix, p i (i = 1, 2, ..., n) represents the pressure reading of the i-th pressure measuring hole.

[0032] Then, a health matrix is ​​constructed, and applied to the pressure ratio matrix to obtain a dataset containing different pressure gauge layouts: Ω′=QΩQ, where Q=diag(q1,q2,...,q n ), q i This is the fault indication parameter for the pressure reading of the i-th pressure measuring hole.

[0033] Among them, the fault indication parameter q iIt can indicate whether the i-th pressure reading is included in atmospheric data estimation, q i The value can be 1 or 0: When the fault indication parameter q i If the value is 0, the corresponding pressure reading is not included in the atmospheric data estimation; when the fault indication parameter q i If the value is 1, the corresponding pressure reading is used in atmospheric data estimation. Therefore, the health matrix Q = diag(q1, q2, ..., q) can be used. n This allows for the creation of various pressure testing hole layouts.

[0034] Table 1 illustrates different combinations of pressure sensor ports according to an embodiment of the present invention. As shown in Table 1, for a pressure sensor array with 9 pressure sensor ports, 26 specific pressure sensor port layouts, i.e., 26 Q values, can be defined. A health matrix Q is used to identify the rows and columns corresponding to fault readings, and these values ​​are set to 0.

[0035] Table 1 Different combinations of pressure testing holes Based on the above operations, the health matrix Q = diag(q1,q2,...,q) n Applying this to the pressure ratio matrix Ω, we obtain a dataset of different pressure gauge layouts: Ω′=QΩQ. It should be noted that the 26 layouts involving 9 pressure gauges in Table 1 are merely examples; this scheme is also applicable to other numbers of pressure gauges and layouts.

[0036] Then, a training set is constructed, which includes input data and output data. The input data includes pressure ratio matrices for different pressure gauge layouts, and the output data includes dimensionless atmospheric data and pressure coefficients for each pressure gauge.

[0037] The dimensionless input-output convolutional neural network is trained using a training set. The parameters of the convolutional neural network are optimized based on the mean squared error loss function to obtain a trained dimensionless input-output convolutional neural network, which includes multiple convolutional layers and multiple fully connected layers connected in sequence.

[0038] Network performance can also be evaluated using a test set that covers all sensor failure types occurring under the current orifice layout, for example, at the corresponding q i Add a bias to sensors with a bias of 0. Adjust the network structure and hyperparameters based on the evaluation results.

[0039] Figure 3 A schematic diagram of a dimensionless input-output convolutional neural network according to an embodiment of the present invention is shown. Figure 3As shown, a dimensionless input-output convolutional neural network can include multiple sequentially connected convolutional layers and multiple fully connected layers. The network takes the pressure ratio matrix Ω′ of different pressure gauge layouts as input and outputs dimensionless atmospheric data and the pressure coefficient of each pressure gauge. In the formula, Mach number, For the angle of attack, Sideslip angle, denoted as , where n is the pressure coefficient at each pressure gauge and n is the number of pressure gauges.

[0040] It should be noted that the neural network model can be any type of supervised network model such as support vector machine or random forest; this solution does not limit this.

[0041] According to one embodiment of the present invention, a pre-trained convolutional neural network can be input with a pressure ratio matrix based on layout k, and output the Mach number, angle of attack, and sideslip angle estimated based on layout k: and the pressure coefficient based on layout k: Where k is the number of the pressure testing hole layout, This is the estimated pressure coefficient for the i-th pressure measuring orifice based on layout k.

[0042] For example, by expanding the 9 pressure gauge holes to 26 different pressure gauge hole layouts, 26 atmospheric data estimation results can be obtained. In the subsequent fault-tolerant regression process, an appropriate combination can be selected from the 26 atmospheric data estimation results.

[0043] Next, step S220 is executed to calculate the estimated pressure of each pressure measuring hole in the layout based on the estimated pressure coefficient.

[0044] In one embodiment of the invention, the measured pressure distribution can be combined with a corresponding health matrix based on layout k, and the free flow pressure and static pressure can be calculated using the following formula: In the formula, This is the estimated pressure coefficient for the i-th pressure measuring orifice based on layout k. For free flow pressure, Let Q(k) be the free-flow static pressure, Q(k) be the health matrix based on layout k, and pi (i = 1, 2, ..., n) be the pressure reading of the i-th pressure measuring hole.

[0045] Then, based on the free flow pressure and static pressure, as well as the estimated pressure coefficient, the estimated pressure at each pressure gauge hole based on layout k is calculated: In the formula, It is the estimated pressure at the i-th pressure measuring hole based on layout k.

[0046] Next, step S230 is executed to calculate the error between the estimated pressure and the actual pressure, obtain the pressure distribution error for each layout, and determine the faulty sensor based on the pressure distribution error.

[0047] Specifically, the pressure distribution error of layout k is calculated based on the difference between the actual pressure reading and the estimated pressure of the pressure measuring orifice in layout k: In the formula, e(k) represents the pressure distribution error of layout k. The fault indication parameter is the i-th pressure reading in layout k; Compare the error magnitudes of each pressure gauge layout, and determine the faulty sensor based on the error magnitude and the fault indication parameters of the pressure readings. For example, layouts 4, 15, 16, 17, and 18 have lower error levels and are at the same level (10). 1 (on the order of magnitude), the remaining layouts showed a higher pressure distribution error, approximately 10. 3 The magnitude of the signal indicates that layouts 4, 15, 16, 17, and 18 do not contain faulty sensors. If the health vector of sensor 3 is 0 in any of these five layouts, a fault in sensor 3 can be diagnosed.

[0048] Figure 4 A schematic diagram illustrating the error criterion calculation process incorporating physical knowledge according to an embodiment of the present invention is shown. Figure 4 As shown, the actual pressure values ​​measured at multiple pressure gauge holes [p1, p2, ..., p] are obtained. n ] Calculate the pressure ratio matrix, and then use the health matrix Q based on layout k. (k) =diag(q1) (k) ,q2 (k) , ..., q n (k) The pressure ratio matrix is ​​applied to a trained dimensionless input-output convolutional neural network, DIONNFADS, which outputs an estimate of atmospheric data based on layout k. and pressure coefficient

[0049] By combining the measured pressure distribution with the corresponding health matrix based on layout k, the free-flow pressure is estimated. and static pressure Based on the free flow pressure and static pressure, as well as the estimated pressure coefficient, the estimated pressure at each pressure gauge based on layout k is calculated [p1] (k) p2 (k) ..., pn (k) ].

[0050] Finally, based on the difference between the actual pressure reading and the estimated pressure, the pressure distribution error of layout k is calculated: Finally, step S240 is executed, which selects the pressure measurement hole layout with the smallest pressure distribution error based on the principle of minimum error, and outputs the atmospheric data corresponding to the layout.

[0051] Figure 5 A schematic diagram illustrating the selection process of the optimal pressure gauge orifice layout based on the principle of minimum error according to an embodiment of the present invention is shown. Figure 5 As shown, the 26 layouts will calculate 26 atmospheric data points and 26 e(k) values. By analyzing the magnitude of these 26 e(k) values, and following the principle of minimum error, fault-tolerant regression of the atmospheric data can be achieved. The layout with the lowest e(k) magnitude is selected, and the atmospheric data state corresponding to that layout is output.

[0052] Assuming the pressure reading of the i-th pressure gauge in layout k fails, if q i (k) =0, then layout k has eliminated the fault. In this scenario, the sensor, even with a dimensionless input-output convolutional neural network, can still output high-precision atmospheric data and pressure coefficients. Dynamic pressure, static pressure, and estimated pressure will all be relatively accurate, with e(k) on the order of the order of magnitude. If... This indicates that layout k includes fault readings. In this case, the dimensionless input-output convolutional neural network will output atmospheric data and pressure coefficients with significant errors. Dynamic pressure, static pressure, and the estimated pressure distribution will also be inaccurate, resulting in a high order of magnitude of e(k).

[0053] Therefore, the pressure orifice layout with the lowest error level can be selected as the optimal layout, and the optimal layout and its corresponding atmospheric data can be output.

[0054] Figure 6 A schematic diagram of the structure of a fault diagnosis and fault-tolerant regression device 600 for an embedded atmospheric data sensing system according to an embodiment of the present invention is shown. Figure 6 As shown, the device 600 may include a prediction module 610, a calculation module 620, a fault diagnosis module 630, and an output module 640.

[0055] The prediction module 610 can estimate dimensionless atmospheric data and pressure coefficients based on different pressure orifice layouts using a trained dimensionless input-output convolutional neural network.

[0056] The calculation module 620 can calculate the estimated pressure of each pressure measuring hole within the layout based on the pressure coefficient estimated by the prediction module 610. The fault diagnosis module 630 can calculate the error between the estimated pressure and the actual pressure, obtain the pressure distribution error of each layout, and determine the faulty sensor based on the pressure distribution error. The output module 640 can select the pressure measuring hole layout with the smallest pressure distribution error based on the principle of minimum error, and output the atmospheric data corresponding to that layout.

[0057] Through the collaboration among the various modules mentioned above, the fault diagnosis and fault-tolerant regression functions of the embedded atmospheric data sensing system are realized.

[0058] To verify the effectiveness of this scheme, a simplified standard model of a supersonic vehicle forebody with a relatively sharp leading edge was designed. Figure 7 A simplified three-dimensional view and corresponding three-view diagram of a supersonic vehicle precursor model are shown. (For example...) Figure 7 As shown, the model adopts a blended wing-body aerodynamic layout, measuring 50mm in length and 390mm in width. The nose of the aircraft is hemispherical with a radius of 10mm. A conical fuselage follows the hemispherical nose, with a bottom radius of 110mm. To accommodate sufficient load capacity, the wings are designed as delta wings with a certain thickness, and the leading edge blunting radius is 5mm to meet heat protection requirements. The model is symmetrical about both the horizontal and vertical planes.

[0059] Assume its flight envelope is as follows: Mach number (velocity of the object relative to the fluid) ranges from 1.5 to 5, flight altitude ranges from 10 to 25 km, corresponding static pressure ranges from 2.5 to 26.5 kPa, flight angle of attack ranges from -6 to 20°, and maximum sideslip angle is 6°. Based on the given flight envelope, a surface pressure distribution dataset, including training and test sets, is established using computational fluid dynamics numerical simulation methods.

[0060] The training set uniformly covers a single flight profile at an altitude of 15 km (corresponding to a free-flow static pressure of 12.1 kPa), with Mach numbers ranging from 1.5 to 5, taking values ​​of 1.5, 2, 3, 4, and 5 respectively; angles of attack ranging from -6 to 20°, taking values ​​that are multiples of 2; and sideslip angles ranging from -6 to 6°, also taking values ​​that are multiples of 2.

[0061] Calculate all combinations of the Mach number, angle of attack, and sideslip angle mentioned above. The purpose of the training set is to establish a relatively dense training dataset that covers a single flight profile, so that the trained model has a fine-grained descriptive ability for any region of that profile.

[0062] The test set is used to verify the estimation accuracy of the established FADS algorithm across the entire flight envelope. 300 sample points are randomly generated using the Monte Carlo method across the entire flight envelope.

[0063] First, the fault diagnosis and fault-tolerant regression results of DIONNFADS (an embedded atmospheric data sensing algorithm based on dimensionless input and output convolutional neural networks) and FT-DIONNFAS (a fault-tolerant flushing air data sensing algorithm based on dimensionless input and output convolutional neural networks) under normal operating conditions are analyzed. Table 2 shows the statistical results of atmospheric data errors of the DIONNFADS and FT-DIONNFAS methods on the test set for normal samples.

[0064] Table 2 shows the error statistics of DIONNFADS and FT-DIONNFADS on the test set for normal samples. Figure 8 This diagram illustrates a comparison of the atmospheric data estimation errors of the DIONNFADS and FT-DIONNFAS methods on the test set for normal samples. (Combined with...) Figure 8 As shown in Table 2, the average relative error of Mach number for FT-DIONNFADS is 0.4%, and the maximum relative error is 2.7%. The average absolute errors of angle of attack and sideslip angle are 0.075° and 0.052°, respectively, and the maximum absolute errors are 0.332° and 0.277°, respectively. These results are better than those of DIONNFADS. This is because FT-DIONNFADS selects a combination of pressure orifices that is more in line with physical principles based on the principle of minimum error, thereby outputting more accurate atmospheric data.

[0065] Next, the fault diagnosis and fault-tolerant regression results of the DIONNFADS and FT-DIONNFAS methods under fault conditions are analyzed. A 30% bias can be added to the measurement results of sensor 3 in the sensor array. Table 3 shows the atmospheric data estimation errors of the DIONNFADS and FT-DIONNFAS methods for fault samples on the test set.

[0066] Table 3 shows the error statistics of the DIONNFADS and FT-DIONNFAS methods on the test set for fault samples (with a 30% bias added to p3). As shown in Table 3, for FT-DIONNFADS, even if sensor 3 fails, the maximum relative error of Mach number is 2.2%, and the maximum absolute error of angle of attack and sideslip angle does not exceed 0.4°, demonstrating high-precision estimation results and proving the fault tolerance capability of FT-DIONNFADS.

[0067] Figure 9 This diagram illustrates a comparison of the atmospheric data estimation errors of the DIONNFADS and FT-DIONNFAS methods for fault samples on the test set. (Combined with...) Figure 9 As shown in Table 3, the prediction results for the sideslip angle by DIONNFADS show an overall shift, with a maximum absolute error of 4.337°. This is mainly because sensor 3 is located on the transverse meridional plane, which is crucial for sideslip angle estimation. Furthermore, the angle-of-attack error for a large number of samples exceeds 0.5°, indicating that DIONNFADS can no longer accurately predict atmospheric data.

[0068] When a 30% bias is added to the measurement of sensor 3, the pressure distribution error under each of the 26 pressure measurement orifice combinations given in Table 1 can be solved separately. Figure 10 A schematic diagram of the pressure distribution error corresponding to 26 combinations of pressure measuring holes is shown. For example... Figure 10 As shown, layouts 4, 15, 16, 17, and 18 have low error levels and are at the same level (10). 1 (on the order of magnitude), the remaining layouts showed a higher pressure distribution error, approximately 10. 3 Based on the magnitude of the signal, it can be determined that layouts 4, 15, 16, 17, and 18 do not contain faulty sensors. Since the health vector of sensor 3 is 0 in these five layouts, a fault in sensor 3 can be diagnosed.

[0069] The fault diagnosis and fault-tolerant regression method and apparatus for embedded atmospheric data sensing systems provided by this invention can not only improve the accuracy and robustness of atmospheric data, but also effectively handle various sensor faults and biases. Compared with traditional voting methods, by adopting the principle of minimum error to select the optimal sensor layout, misjudgments are avoided and the processing efficiency of atmospheric data is improved. The implementation of this algorithm will bring more reliable fault diagnosis and fault-tolerant regression capabilities to aircraft systems, which is of great significance for ensuring the safety and stability of aircraft.

[0070] Numerous specific details are set forth in the specification provided herein. However, it will be understood that embodiments of the invention may be practiced without these specific details. In some instances, well-known methods, structures, and techniques have not been shown in detail so as not to obscure the understanding of this specification.

[0071] Those skilled in the art will understand that modules, units, or components of the devices in the examples disclosed herein can be arranged in the devices described in this embodiment, or alternatively, can be located in one or more devices different from the devices in this example. The modules in the foregoing examples can be combined into a single module or further divided into multiple sub-modules.

[0072] Although some embodiments described herein include certain features that are included in other embodiments but not others, combinations of features from different embodiments are meant to be within the scope of the invention and form different embodiments.

[0073] Furthermore, some of the embodiments are described herein as methods or combinations of method elements that can be implemented by a processor of a computer system or by other means of performing the functions. Therefore, a processor having the necessary instructions for implementing the methods or method elements forms means for implementing the methods or method elements. Furthermore, the elements described herein in the apparatus embodiments are examples of means for implementing the functions performed by elements for the purposes of carrying out the invention.

[0074] Although the invention has been described with reference to a limited number of embodiments, those skilled in the art will understand from the foregoing description that other embodiments are conceivable within the scope of the invention described herein. It should be noted that the language used in this specification has been chosen primarily for readability and instructional purposes, and not for interpreting or limiting the subject matter of the invention. Therefore, many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the appended claims. The disclosure of the invention is illustrative rather than restrictive, and the scope of the invention is defined by the appended claims.

Claims

1. A fault diagnosis and fault-tolerant recovery method for an embedded air data sensing system, characterized in that, include: The training of a dimensionless input-output convolutional neural network is used to estimate dimensionless atmospheric data and pressure coefficients based on different pressure gauge layouts. The estimated pressure of each pressure gauge within the layout is calculated based on the estimated pressure coefficient. Calculate the error between the estimated pressure and the actual pressure to obtain the pressure distribution error for each layout, and determine the faulty sensor based on the pressure distribution error; Based on the principle of minimizing error, the pressure measurement hole layout with the smallest pressure distribution error is selected, and the atmospheric data corresponding to this layout is output. The dimensionless input-output convolutional neural network is trained through the following steps: constructing a training set, which includes input data and output data. The input data includes pressure ratio matrices for different pressure gauge layouts, and the output data includes dimensionless atmospheric data and pressure coefficients of each pressure gauge. The training set is used to train the dimensionless input-output convolutional neural network, and the parameters of the convolutional neural network are optimized based on the mean square error loss function to obtain a trained dimensionless input-output convolutional neural network. The dimensionless input-output convolutional neural network includes multiple sequentially connected convolutional layers and multiple fully connected layers. The steps for constructing a dataset containing multiple pressure gauge layouts include: acquiring the pressure readings of each pressure gauge in the embedded atmospheric data sensing system; calculating the ratio of the pressure reading of each pressure gauge to its own and the pressure readings of the others, thus obtaining a pressure ratio matrix. ， In the formula, Ω is the pressure ratio matrix. This represents the pressure reading of the i-th pressure measuring hole; Construct a health matrix and apply it to the pressure ratio matrix to obtain a dataset containing different pressure gauge layouts: ,in, For the health matrix, The fault indication parameter is the pressure reading of the i-th pressure measuring hole. When the fault indication parameter is 0, the corresponding pressure reading is not included in the atmospheric data estimation; when the fault indication parameter is 1, the corresponding pressure reading is included in the atmospheric data estimation.

2. The fault diagnosis and fault-tolerant regression method for the embedded atmospheric data sensing system according to claim 1, characterized in that, The steps for estimating dimensionless atmospheric data and pressure coefficients based on different pressure gauge layouts using a trained dimensionless input-output convolutional neural network include: The training convolutional neural network is input with the pressure ratio matrix based on layout k, and the outputs the Mach number, angle of attack, and sideslip angle estimated based on layout k. and the pressure coefficient based on layout k: Where k is the number of the pressure testing hole layout, This is the estimated pressure coefficient for the i-th pressure measuring orifice based on layout k.

3. The fault diagnosis and fault-tolerant regression method for the embedded atmospheric data sensing system according to claim 2, characterized in that, The step of calculating the estimated pressure of each pressure gauge within the layout based on the estimated pressure coefficient includes: Based on the measured pressure distribution and the corresponding health matrix based on layout k, the free flow pressure and static pressure are calculated: In the formula, , This is the estimated pressure coefficient for the i-th pressure measuring orifice based on layout k. For free flow pressure, For free-flow static pressure, For a health matrix based on layout k, This represents the pressure reading of the i-th pressure measuring hole; Based on the free-flowing pressure and static pressure, and the estimated pressure coefficient, calculate the estimated pressure at each pressure gauge based on layout k: , In the formula, It is the estimated pressure at the i-th pressure measuring hole based on layout k. Let be the estimated pressure coefficient of the i-th pressure measuring hole.

4. The fault diagnosis and fault-tolerant regression method for the embedded atmospheric data sensing system according to claim 3, characterized in that, The step of calculating the error between the estimated pressure and the actual pressure to obtain the pressure distribution error for each layout, and determining the faulty sensor based on the pressure distribution error of multiple pressure measuring hole layouts, includes: Based on the difference between the actual pressure reading and the estimated pressure of the pressure measuring orifice in layout k, the pressure distribution error of layout k is calculated: In the formula, For the pressure distribution error of layout k, The fault indication parameter is the i-th pressure reading in layout k; Compare the error magnitudes of each pressure test hole layout, and determine the faulty sensor based on the error magnitudes and fault indication parameters of the pressure readings.

5. The fault diagnosis and fault-tolerant regression method for the embedded atmospheric data sensing system according to claim 1, characterized in that, The step of selecting the pressure measuring orifice layout with the smallest pressure distribution error based on the principle of minimum error, and outputting the atmospheric data corresponding to the layout, includes: Select the pressure orifice layout with the lowest error level as the optimal layout, and output the optimal layout and its corresponding atmospheric data.

6. A fault diagnosis and fault-tolerant regression device for an embedded atmospheric data sensing system, characterized in that, include: The prediction module is used to estimate dimensionless atmospheric data and pressure coefficients based on different pressure gauge layouts, using a trained dimensionless input-output convolutional neural network. The calculation module is used to calculate the estimated pressure of each pressure gauge within the layout based on the estimated pressure coefficient; The fault diagnosis module is used to calculate the error between the estimated pressure and the actual pressure, obtain the pressure distribution error for each layout, and determine the faulty sensor based on the pressure distribution error. The output module is used to select the pressure measurement hole layout with the smallest pressure distribution error based on the principle of minimum error, and output the atmospheric data corresponding to the layout. The dimensionless input-output convolutional neural network is trained through the following steps: constructing a training set, which includes input data and output data. The input data includes pressure ratio matrices for different pressure gauge layouts, and the output data includes dimensionless atmospheric data and pressure coefficients of each pressure gauge. The training set is used to train the dimensionless input-output convolutional neural network, and the parameters of the convolutional neural network are optimized based on the mean square error loss function to obtain a trained dimensionless input-output convolutional neural network. The dimensionless input-output convolutional neural network includes multiple sequentially connected convolutional layers and multiple fully connected layers. Constructing a dataset containing multiple pressure gauge layouts includes: acquiring the pressure readings of each pressure gauge in the embedded atmospheric data sensing system; calculating the ratio of each pressure gauge reading to its own and the remaining pressure readings to obtain a pressure ratio matrix. ， In the formula, Ω is the pressure ratio matrix. This represents the pressure reading of the i-th pressure measuring hole; Construct a health matrix and apply it to the pressure ratio matrix to obtain a dataset containing different pressure gauge layouts: ,in, For the health matrix, The fault indication parameter is the pressure reading of the i-th pressure measuring hole. When the fault indication parameter is 0, the corresponding pressure reading is not included in the atmospheric data estimation; when the fault indication parameter is 1, the corresponding pressure reading is included in the atmospheric data estimation.

7. A computing device, comprising: At least one processor; and a memory storing program instructions, wherein the program instructions are configured to be executed by the at least one processor, the program instructions including instructions for performing a fault diagnosis and fault-tolerant regression method for an embedded atmospheric data sensing system as claimed in any one of claims 1-5.

8. A readable storage medium storing program instructions, which, when read and executed by a computing device, cause the computing device to perform a fault diagnosis and fault-tolerant regression method for an embedded atmospheric data sensing system as described in any one of claims 1-5.

Images