Bayesian inference-based automatic analysis method for aerodynamic heating test data, electronic device and storage medium

Abstract

The application discloses a method for automatically analyzing aerodynamic heating test data based on Bayesian inference, an electronic device and a storage medium, and belongs to the technical field of aerodynamic heating test of aircrafts. The method solves the problems of large data processing result error and low abnormality screening automation degree of the traditional aerodynamic heating test data analysis method in the prior art. The method first pre-processes the aerodynamic heating test data, which includes abnormal value elimination and completion, time sequence alignment and voltage signal smoothing. Secondly, oscillation identification is respectively performed on the voltage signal and the heat flow signal, the maximum value of the maximum oscillation end time of the voltage signal and the heat flow signal is taken as a global cutoff time, the oscillation segment data before the time point is eliminated, and the smooth segment data is obtained. Finally, the CFD simulation data is taken as a physical constraint, the statistical characteristics and the physical law are fused, and the abnormal sensor in the smooth segment is identified based on Bayesian inference. The application improves the abnormality detection precision and can be applied to analyze the aerodynamic heating test data.

Description

Technical Field

[0001] This invention relates to the field of aerodynamic thermal testing technology for aircraft, specifically to an automatic analysis method, electronic equipment, and storage medium for aerodynamic thermal test data based on Bayesian inference. Background Technology

[0002] Aerothermal environment is a core challenge for hypersonic flight of aircraft. The intense compression and friction between the fuselage and the air can cause the surface temperature to exceed 1000℃. Accurate acquisition of aerothermal data is the key to ensuring the structural safety of the aircraft and promoting the development of the model. Wind tunnel testing is the core means of simulating aerothermal environment. Among them, high-enthalpy pulse wind tunnels have the advantages of long operating time and high flow field stability. However, during the test, aerothermal data is prone to fluctuation and anomalies due to factors such as incoming flow disturbance, heat flow sensor drift, installation error, and hardware failure.

[0003] Therefore, aerodynamic thermal data processing needs to solve two core problems: First, accurate identification of stable segments, which requires screening out local stable segments with "small heat flow fluctuations and high synchronization of heat flow sensors" to provide a basis for heat flow mean calculation; Second, detection of abnormal data in stable segments, which requires identifying abnormal values ​​caused by heat flow sensor failures to avoid misleading thermal protection design; However, existing data processing methods have significant defects: (1) Stable segment identification relies on manual experience to judge stability, and different people have different judgment results, resulting in low data utilization and long processing time for a single set of data, which cannot meet the needs of rapid iteration; (2) Anomaly detection is only screened by the naked eye, which is easy to misjudge data that "conforms to statistical laws but violates physical principles" or miss hidden faults that "are close to the threshold but have abnormal spatial correlations", resulting in low detection efficiency and accuracy; (3) Stable segment identification and anomaly detection are carried out independently and do not form a technical closed loop. If there is a heat flow sensor failure in the manually screened stable segment, it is necessary to spend extra time to investigate, which further reduces processing efficiency; (4) Data processing is disconnected from aerodynamic thermal physical laws, and anomaly detection lacks physical constraint support, resulting in insufficient reliability of results.

[0004] In summary, there is an urgent need for a high-precision, physically consistent, Bayesian inference-based automatic analysis method, electronic equipment, and storage medium for aerodynamic thermal test data. Summary of the Invention

[0005] A brief overview of the invention is given below to provide a basic understanding of certain aspects of it. It should be understood that this overview is not an exhaustive summary of the invention. It is not intended to identify key or essential parts of the invention, nor is it intended to limit the scope of the invention. Its purpose is merely to present certain concepts in a simplified form as a prelude to the more detailed description that follows.

[0006] In view of this, in order to solve the problems of large errors in data processing results and low automation of anomaly screening in the traditional aerodynamic thermal test data analysis methods in the prior art, the present invention provides an automatic analysis method for aerodynamic thermal test data based on Bayesian inference, electronic equipment and storage medium.

[0007] Technical solution one is as follows: an automatic analysis method for aerodynamic thermal test data based on Bayesian inference, including the following steps:

[0008] S1. Preprocess the aerodynamic thermal test data collected from the wind tunnel test. The aerodynamic thermal test data includes raw voltage data and raw heat flow data, and the preprocessed heat flow signal and voltage signal are obtained.

[0009] S2. Perform oscillation interval detection on the preprocessed voltage signal and heat flow signal respectively, locate the interval of violent data fluctuation, extract the oscillation interval, and obtain the maximum end time of the corresponding oscillation interval of the voltage signal and heat flow signal respectively;

[0010] S3. Take the maximum value among the maximum end times of the oscillation interval as the global cutoff time, remove the oscillation segment data before the current time point, and obtain the stable segment data. If no effective oscillation interval is detected, directly retain the preprocessed data of the entire time period as the stable segment.

[0011] S4. Using CFD simulation data as physical constraints, combining statistical characteristics and physical laws, anomaly detection is performed on the stationary data based on Bayesian inference to identify abnormal heat flow sensors within the stationary segment and obtain anomaly detection results.

[0012] S5. Save the stable segment data, features and anomaly detection results, form standardized output, and complete the automatic analysis of aerodynamic thermal test data.

[0013] Furthermore, step S1 includes the following steps:

[0014] S11. The extreme values ​​of the original heat flow data are removed using the 3σ criterion. At the same time, physically meaningless negative values ​​are also removed. The missing parts of the original heat flow data are filled by linear interpolation to ensure the continuity of the time series.

[0015] In S11, the 3σ criterion is expressed as:

[0016] |x-μ|>3σ

[0017] Where μ is the data mean, σ is the standard deviation, and x is the raw heat flux data;

[0018] S12. Using the wind tunnel test trigger time as a reference, all raw heat flow data are aligned to a uniform time axis by linear interpolation resampling to complete the time series alignment and obtain the preprocessed heat flow signal.

[0019] S13. Gaussian filtering is used to smooth the raw voltage data. A sampling frequency is set, and the standard deviation of the Gaussian filter kernel of the filtering window is dynamically calculated based on the sampling frequency. This leads to the preprocessed voltage signal;

[0020] In step S13, the dynamic calculation process of the sampling frequency is as follows:

[0021]

[0022]

[0023] in, The sampling frequency is expressed in Hz. Filter window length.

[0024] Furthermore, step S2 includes the following steps:

[0025] S21. By calculating the fluctuation intensity and setting the oscillation threshold, the voltage oscillation interval is extracted, merged, and optimized to obtain the strong oscillation interval. The oscillation cutoff point of the voltage signal is then determined, thereby realizing voltage signal oscillation identification.

[0026] In step S21, the first-order differential absolute mean is used to characterize the fluctuation intensity of the voltage signal. That is, to calculate the mean of the absolute values ​​of the first-order differences of each heat flux sensor. ;

[0027] Fluctuation intensity Represented as:

[0028]

[0029]

[0030] in, For the first A heat flow sensor in Voltage signal at time, For the first A heat flow sensor in The voltage signal at a given moment;

[0031] Data collected 1 second after the wind tunnel test is taken as the baseline for the stable period. Based on the mean and standard deviation of the stable period fluctuation intensity, the voltage oscillation threshold is set as the mean of the stable period + 2.5 times the standard deviation. If the amount of effective data in the stable period is insufficient, the mean of the fluctuation intensity over the entire period + 2.0 times the standard deviation is used as the voltage oscillation threshold.

[0032] If there is sufficient valid data during the stable period, the threshold calculation formula is as follows: ;

[0033] If the amount of valid data during the stable period is insufficient, the threshold is calculated using the full-time volatility intensity series. ;

[0034] in, The threshold for judging voltage oscillation of the heat flow sensor. The arithmetic mean of the volatility intensity sequence during the stable period. The statistical standard deviation of the volatility intensity sequence during the stable period. The arithmetic mean of the full-time volatility intensity sequence. Statistical standard deviation of the volatility intensity series over all time periods;

[0035] By traversing the data throughout the entire time period through a sliding window, the intervals where the fluctuation intensity exceeds the threshold are marked, and the extremely short noise intervals with a duration of <0.005s are filtered out to obtain the voltage oscillation intervals;

[0036] Voltage oscillation intervals with adjacent voltage oscillation intervals or gap fluctuations exceeding the oscillation threshold by a set percentage (70%) are merged, while strong oscillation intervals with fluctuations ≥ 1.2 times the oscillation threshold are retained;

[0037] Extract the end time of all strong oscillation intervals, and take the maximum value of the oscillation end time as the oscillation cutoff point of the voltage signal;

[0038] S22. By quantifying global fluctuations and setting a heat flow oscillation threshold, the original heat flow oscillation intervals are screened and merged to obtain the main oscillation intervals, determine the oscillation cutoff point of the heat flow signal, and complete the heat flow signal oscillation identification:

[0039] In step S22, the standard deviation of all heat flux sensor data at each time point is calculated. As an indicator of global volatility intensity;

[0040] Standard deviation of all heat flux sensor data at each time point Represented as:

[0041]

[0042] in, Indicates the first A heat flow sensor in Heat flux density at any given time , For all heat flow sensors in The average heat flux density at any given time;

[0043] Based on the mean of global fluctuations over the entire period with standard deviation The heat flow oscillation threshold was set to the mean plus 1.5 times the standard deviation.

[0044] Extract the original heat flow oscillation intervals with a duration of ≥0.05s, merge adjacent heat flow oscillation intervals with a gap of ≤0.08s, and select the main oscillation intervals by the condition that the cumulative proportion of fluctuation intensity is ≥80%.

[0045] The maximum end time of the main oscillation interval is taken as the oscillation cutoff point of the heat flow signal.

[0046] Furthermore, step S4 includes the following steps:

[0047] S41. Based on the CFD simulation data and the coordinate position of the heat flux sensor, extract the CFD numerical deviation characteristics and concave point abrupt change characteristics:

[0048] In step S41, the CFD simulation data is interpolated to the coordinate positions of each heat flux sensor to calculate the average heat flux during the steady-state segment and the theoretical CFD heat flux value, thereby obtaining the CFD numerical deviation. ;

[0049] CFD numerical deviation Represented as:

[0050]

[0051] in, For the first The average heat flux during the steady-state segment of each heat flux sensor. The theoretical heat flux value is interpolated to the current heat flux sensor location using CFD.

[0052] The relative deviation of the heat flux mean between the heat flux sensor and the adjacent non-faulty heat flux sensor is calculated. If the relative deviation of the heat flux mean is >0.1, it is determined to be a concave point abrupt change feature.

[0053] S42. Mark the extreme CFD deviation heat flux sensor and its adjacent heat flux sensor. If the true mutation rate of the adjacent heat flux sensor, that is, the relative deviation between the adjacent heat flux sensor and the preceding normal heat flux sensor, is <0.1, it is determined to be a false concave point. The mutation value of the false concave point is corrected to 0 to obtain the corrected concave point mutation feature.

[0054] S43. Weight and standardize the CFD numerical bias features and the corrected concave point mutation features, divide the collected initial samples, fit the divided samples respectively, use the Bayesian inference model to calculate the posterior probability of the anomaly and set its threshold, and output the anomaly label and probability.

[0055] In S43, Min-Max standardization is used to map the two types of features to the [0,1] interval;

[0056] Based on the 65th percentile of the total feature score as the dividing threshold, the data collected by the heat flow sensor is divided into initial abnormal samples and initial normal samples.

[0057] Gaussian kernel density is fitted to estimate the likelihood function for the initial abnormal samples and the initial normal samples respectively. When the sample size is insufficient to meet the set value, a normal distribution is used for fitting.

[0058] The anomalous posterior probability of each heat flux sensor is calculated based on Bayes' theorem. ;

[0059] Possible posterior probability for each heat flux sensor Represented as:

[0060]

[0061] in, = =0.5, The likelihood value of the initial outlier sample. This represents the likelihood value of the initial normal samples;

[0062] Using the 65th percentile of the abnormal posterior probability as the threshold for the abnormal posterior probability, heat flow sensors with an abnormal posterior probability higher than the threshold are marked as abnormal, and the abnormal label and abnormal posterior probability are output.

[0063] Technical solution two is as follows: An electronic device includes a processor and a memory, wherein the memory stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, the steps in the method described in technical solution one are performed.

[0064] Technical solution three is as follows: a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, it performs the steps of the method as described in any one of the technical solutions.

[0065] The beneficial effects of the present invention are as follows: (1) The present invention has high accuracy and automation in identifying stable segments: the oscillation identification strategy based on the unique characteristics of voltage / heat flow signals does not require a preset stabilization time, has a high degree of overlap with the real stable segments jointly labeled by humans, and solves the pain points of missing short stable segments and misjudging pseudo-stable segments in traditional manual stability judgment, thus effectively improving the data utilization rate; (2) The anomaly detection accuracy and physical interpretability are high: the Bayesian inference model that integrates CFD physical constraints and statistical features achieves anomaly detection accuracy, recall rate and physical rationality rate of 100%, successfully detecting overt faulty heat flow sensors and latent anomalies. (3) The processing efficiency is greatly improved: the processing time of a single set of data is 20s, which is more efficient than the traditional manual processing process. After FPGA hardware acceleration, it can be further compressed to meet the real-time early warning requirements of wind tunnel test. (4) Strong engineering implementation: It can be directly integrated into the wind tunnel data acquisition system to form a standardized and automated integrated post-processing solution. The output data has a high degree of fitting with the CFD theoretical curve, and the physical laws are consistent with the aerodynamic and thermal theories, providing accurate data guarantee for the design of thermal protection system. It has wide adaptability. Attached Figure Description

[0066] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this invention, illustrate exemplary embodiments of the invention and are used to explain the invention, but do not constitute an undue limitation of the invention. In the drawings:

[0067] Figure 1 This is a flowchart illustrating an automated analysis method for aerodynamic thermal test data based on Bayesian inference.

[0068] Figure 2 is a schematic diagram of the heat flux sensor changing over time, where (a) is a schematic diagram of the voltage signal of the heat flux sensor changing over time, and (b) is a schematic diagram of the heat flux density of the heat flux sensor changing over time.

[0069] Figure 3 is a comparison diagram of the average heat flux density of the steady segment identified by human identification and automatic identification. Among them, (a) is a comparison diagram of the average heat flux density of all heat flux sensors identified by human identification and automatic identification, and (b) is a comparison diagram of the average heat flux density of the heat flux sensor with the largest deviation removed by human identification and automatic identification.

[0070] Figure 4 is a comparative diagram of heat flux density data from the abnormal heat flux sensor. Detailed Implementation

[0071] To make the technical solutions and advantages of the embodiments of the present invention clearer, the exemplary embodiments of the present invention will be further described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not an exhaustive list of all embodiments. It should be noted that, unless otherwise specified, the embodiments and features in the embodiments of the present invention can be combined with each other.

[0072] Example 1: Reference Figure 1 Figure 4 illustrates in detail the automatic analysis method for aerodynamic thermal test data based on Bayesian inference in this embodiment, specifically including the following steps:

[0073] S1. Preprocess the aerodynamic thermal test data collected from the wind tunnel test. The aerodynamic thermal test data includes raw voltage data and raw heat flow data, and the preprocessed heat flow signal and voltage signal are obtained.

[0074] S2. Perform oscillation interval detection on the preprocessed voltage signal and heat flow signal respectively, locate the interval of violent data fluctuation, extract the oscillation interval, and obtain the maximum end time of the corresponding oscillation interval of the voltage signal and heat flow signal respectively;

[0075] S3. Take the maximum value among the maximum end times of the oscillation interval as the global cutoff time, remove the oscillation segment data before the current time point, and obtain the stable segment data. If no effective oscillation interval is detected, directly retain the preprocessed data of the entire time period as the stable segment.

[0076] S4. Using CFD simulation data as physical constraints, combining statistical characteristics and physical laws, anomaly detection is performed on the stationary data based on Bayesian inference to identify abnormal heat flow sensors within the stationary segment and obtain anomaly detection results.

[0077] S5. Save the stable segment data, features and anomaly detection results, form standardized output, and complete the automatic analysis of aerodynamic thermal test data.

[0078] Specifically, in this embodiment, referring to the aerodynamic thermal test data of a certain FL64 test model, the test model is a designed and manufactured model. Heat flow measurement points are arranged along a straight line on the lower surface of the model. The test model adopts a tail support form, and the model is connected to the upper curved blade of the quick-insertion mechanism through a support rod and an adapter. The test adopts the tail support method, and the total pressure control is used to determine the Mach number. The flow field Mach number accuracy control error is ≤1%. The test process collects and records test data through the acquisition system. During the test, after waiting for the test flow field to be established, the quick-insertion mechanism moves, which drives the model to quickly insert into the stable flow field in 0.2s. After the set test time interval, the quick-insertion mechanism quickly leaves the flow field, and the test ends, completing the aerodynamic thermal test data acquisition.

[0079] Referring to Figure 3(b), the deviation of the maximum deviation heat flux sensor is 21.59%.

[0080] Furthermore, step S1 includes the following steps:

[0081] S11. The extreme values ​​of the original heat flow data are removed using the 3σ criterion. At the same time, physically meaningless negative values ​​are also removed. The missing parts of the original heat flow data are filled by linear interpolation to ensure the continuity of the time series.

[0082] In S11, the 3σ criterion is expressed as:

[0083] |x-μ|>3σ

[0084] Where μ is the data mean, σ is the standard deviation, and x is the raw heat flux data;

[0085] S12. Based on the wind tunnel test trigger time, all raw heat flow data are aligned to a uniform time axis through linear interpolation resampling to eliminate the asynchronous problem caused by the acquisition channel delay, complete the time series alignment, and obtain the preprocessed heat flow signal.

[0086] S13. Gaussian filtering is used to smooth the raw voltage data. A sampling frequency is set, and the standard deviation of the Gaussian filter kernel of the filtering window is dynamically calculated based on the sampling frequency. (The filter window size is 0.001 times the sampling frequency, and half of the filter window size is used as...) value) While preserving the true oscillation characteristics, high-frequency noise is suppressed, thus obtaining the preprocessed voltage signal;

[0087] In step S13, the dynamic calculation process of the sampling frequency is as follows:

[0088]

[0089]

[0090] in, The sampling frequency is expressed in Hz. Filter window length.

[0091] Specifically, in this embodiment, the mean and standard deviation of the raw data from 23 heat flow sensors are calculated, extreme values ​​are removed, and the wind tunnel test is triggered at 0s.

[0092] Furthermore, step S2 includes the following steps:

[0093] S21. By calculating the fluctuation intensity and setting the oscillation threshold, the voltage oscillation interval is extracted, merged, and optimized to obtain the strong oscillation interval. The oscillation cutoff point of the voltage signal is then determined, thereby realizing voltage signal oscillation identification.

[0094] In step S21, the first-order differential absolute mean is used to characterize the fluctuation intensity of the voltage signal. That is, to calculate the mean of the absolute values ​​of the first-order differences of each heat flux sensor. Sensitively captures rapid changes;

[0095] Fluctuation intensity Represented as:

[0096]

[0097]

[0098] in, For the first A heat flow sensor in Voltage signal at time, For the first A heat flow sensor in The voltage signal at a given moment;

[0099] Data collected 1 second after the wind tunnel test is taken as the baseline for the stable period. Based on the mean and standard deviation of the stable period fluctuation intensity, the voltage oscillation threshold is set as the mean of the stable period + 2.5 times the standard deviation. If the amount of effective data in the stable period is insufficient, the mean of the fluctuation intensity over the entire period + 2.0 times the standard deviation is used as the voltage oscillation threshold.

[0100] If there is sufficient valid data during the stable period, the threshold calculation formula is as follows: ;

[0101] If the amount of valid data during the stable period is insufficient, the threshold is calculated using the full-time volatility intensity series. ;

[0102] in, The threshold for judging voltage oscillation of the heat flow sensor. The arithmetic mean of the volatility intensity sequence during the stable period. The statistical standard deviation of the volatility intensity sequence during the stable period. The arithmetic mean of the full-time volatility intensity sequence. Statistical standard deviation of the volatility intensity series over all time periods;

[0103] By traversing the data throughout the entire time period through a filter window, the intervals where the fluctuation intensity exceeds the threshold are marked, and the extremely short noise intervals with a duration of <0.005s are filtered out to obtain the voltage oscillation intervals.

[0104] Voltage oscillation intervals with adjacent voltage oscillation intervals (gap ≤ 0.05s) or gap fluctuation intensity exceeding the oscillation threshold by a set percentage (70%) are merged, and strong oscillation intervals with fluctuation intensity ≥ 1.2 times the oscillation threshold are retained;

[0105] Extract the end time of all strong oscillation intervals, and take the maximum value of the oscillation end time as the oscillation cutoff point of the voltage signal;

[0106] S22. By quantifying global fluctuations and setting a heat flow oscillation threshold, the original heat flow oscillation intervals are screened and merged to obtain the main oscillation intervals, determine the oscillation cutoff point of the heat flow signal, and complete the heat flow signal oscillation identification:

[0107] In step S22, the standard deviation of all heat flux sensor data at each time point is calculated. As an indicator of global volatility intensity;

[0108] Standard deviation of all heat flux sensor data at each time point Represented as:

[0109]

[0110] in, Indicates the first A heat flow sensor in Heat flux density at any given time , For all heat flow sensors in The average heat flux density at any given time;

[0111] Based on the mean of global fluctuations over the entire period with standard deviation The heat flow oscillation threshold is set to the mean + 1.5 times the standard deviation to adapt to the slow change characteristics of the heat flow signal;

[0112] Extract the original heat flow oscillation intervals with a duration of ≥0.05s, merge adjacent heat flow oscillation intervals with a gap of ≤0.08s, and select the main oscillation intervals by the condition that the cumulative proportion of fluctuation intensity is ≥80%.

[0113] The maximum end time of the main oscillation interval is taken as the oscillation cutoff point of the heat flow signal.

[0114] Furthermore, step S4 includes the following steps:

[0115] S41. Based on the CFD simulation data and the coordinate position of the heat flux sensor, extract the CFD numerical deviation characteristics and concave point abrupt change characteristics:

[0116] In step S41, the CFD simulation data is interpolated to the coordinate positions of each heat flux sensor to calculate the average heat flux during the steady-state segment and the theoretical CFD heat flux value, thereby obtaining the CFD numerical deviation. ;

[0117] CFD numerical deviation Represented as:

[0118]

[0119] in, For the first The average heat flux during the steady-state segment of each heat flux sensor. The theoretical heat flux value is interpolated to the current heat flux sensor location using CFD.

[0120] The relative deviation of the heat flux mean between the heat flux sensor and the neighboring non-faulty heat flux sensor is calculated as (neighbor mean - heat flux mean of the i-th heat flux sensor) / neighbor mean. If the relative deviation of the heat flux mean is >0.1, it is determined to be a concave point abrupt change feature, and spatial correlation anomalies are captured.

[0121] S42. Marking extreme CFD deviation heat flow sensor ( ≥100%, This represents the relative deviation between the measured value and the CFD simulation value of the i-th heat flux sensor.

[0122] If the true mutation rate of the adjacent heat flow sensor, i.e. the relative deviation between the adjacent heat flow sensor and the preceding normal heat flow sensor, is <0.1, it is determined to be a false concave point. The mutation value of the false concave point is corrected to 0 to obtain the corrected concave point mutation feature.

[0123] S43. Weight (emphasizing physical constraint weight) and standardize the CFD numerical bias features and the corrected concave point mutation features, divide the initial samples collected, fit the divided samples respectively, use the Bayesian inference model to calculate the posterior probability of the anomaly and set its threshold, and output the anomaly label and probability.

[0124] In S43, Min-Max standardization is used to map the two types of features to the [0,1] interval;

[0125] Based on the 65th percentile of the total feature score as the dividing threshold, the data collected by the heat flow sensor is divided into initial abnormal samples and initial normal samples.

[0126] Gaussian kernel density estimation (KDE) likelihood functions are fitted to the initial abnormal samples and the initial normal samples respectively. When the sample size is insufficient to meet the set value, a normal distribution is used for fitting.

[0127] The anomalous posterior probability of each heat flux sensor is calculated based on Bayes' theorem. ;

[0128] Anomalous posterior probability for each heat flux sensor Represented as:

[0129]

[0130] in, = =0.5 (no pre-set priors), The likelihood value of the initial outlier sample. This represents the likelihood value of the initial normal samples;

[0131] Using the 65th percentile of the abnormal posterior probability as the threshold for the abnormal posterior probability, heat flow sensors with an abnormal posterior probability higher than the threshold are marked as abnormal, and the abnormal label and abnormal posterior probability are output.

[0132] Specifically, the calculation results show that the posterior probability of heat flux sensors 1, 2, 3, 7, 8, 13, 19, and 21 being abnormal is greater than 0.99, indicating that they are abnormal heat flux sensors, which is consistent with the results of the investigation after the test.

[0133] It saves stable segment data, CFD deviation and concave point mutation characteristic values ​​of each heat flux sensor, anomaly detection tags and probabilities, and can be integrated into the FL-64 wind tunnel data acquisition system. After FPGA acceleration, it can meet the real-time early warning requirements.

[0134] In step S5, FPGA hardware acceleration technology is introduced to optimize the sliding window calculation and Bayesian model likelihood fitting process, compress the processing time of a single set of data, and meet the real-time early warning requirements; an interactive visualization interface is developed to support experimenters in fine-tuning the stable segment interval and anomaly judgment results, forming a semi-automated process with automatic as the main method and manual as the auxiliary method.

[0135] Example 2: Reference Figure 1 Figure 4 illustrates this embodiment in detail. An electronic device includes a processor and a memory. The memory stores computer-readable instructions. When the computer-readable instructions are executed by the processor, the steps in the method described in Embodiment 1 are performed.

[0136] Specifically, the computer device of the present invention may include a processor and a memory, such as a microcontroller containing a central processing unit. Furthermore, the processor executes the computer program stored in the memory to implement the steps of the aforementioned recommendation method for modifyable relationship-driven recommendation data based on CREO software. The processor may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created based on the use of the mobile phone (such as audio data, phonebook, etc.). In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0137] Example 3: Reference Figure 1 Figure 4 illustrates this embodiment in detail. A computer-readable storage medium stores a computer program thereon, which, when executed by a processor, performs the steps of the method as described in Embodiment 1.

[0138] Specifically, the computer-readable storage medium of the present invention can be any form of storage medium that can be read by the processor of a computer device, including but not limited to non-volatile memory, volatile memory, ferroelectric memory, etc. The computer-readable storage medium stores a computer program. When the processor of the computer device reads and executes the computer program stored in the memory, the steps of the above-described modeling method for modifyable relation-driven modeling data based on CREO software can be implemented. The computer program includes computer program code, which can be in the form of source code, object code, executable file, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drive, portable hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content contained in the computer-readable medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, the computer-readable medium does not include electrical carrier signals and telecommunication signals.

[0139] 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. Furthermore, it should be noted that the language used in this specification has been chosen primarily for readability and instructional purposes, and not for the purpose of 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 and not restrictive, and the scope of the invention is defined by the appended claims.

Claims

1. A method for automatic analysis of aerodynamic heating test data based on Bayesian inference, characterized in that, Includes the following steps: S1. Preprocess the aerodynamic thermal test data collected from the wind tunnel test. The aerodynamic thermal test data includes raw voltage data and raw heat flow data, and the preprocessed heat flow signal and voltage signal are obtained. S2. Perform oscillation interval detection on the preprocessed voltage signal and heat flow signal respectively, locate the interval of violent data fluctuation, extract the oscillation interval, and obtain the maximum end time of the corresponding oscillation interval of the voltage signal and heat flow signal respectively; S3. Take the maximum value among the maximum end times of the oscillation interval as the global cutoff time, remove the oscillation segment data before the current time point, and obtain the stable segment data. If no effective oscillation interval is detected, directly retain the preprocessed data of the entire time period as the stable segment. S4. Using CFD simulation data as physical constraints, combining statistical characteristics and physical laws, anomaly detection is performed on the stationary data based on Bayesian inference to identify abnormal heat flow sensors within the stationary segment and obtain anomaly detection results. S5. Save the stable segment data, features and anomaly detection results, form standardized output, and complete the automatic analysis of aerodynamic thermal test data.

2. The method of claim 1, wherein, S1 includes the following steps: S11. The extreme values ​​of the original heat flow data are removed using the 3σ criterion. At the same time, physically meaningless negative values ​​are also removed. The missing parts of the original heat flow data are filled by linear interpolation to ensure the continuity of the time series. In S11, the 3σ criterion is expressed as: |x-μ|>3σ Where μ is the data mean, σ is the standard deviation, and x is the raw heat flux data; S12. Using the wind tunnel test trigger time as a reference, all raw heat flow data are aligned to a uniform time axis by linear interpolation resampling to complete the time series alignment and obtain the preprocessed heat flow signal. S13. Smoothing the voltage raw data by using Gaussian filtering, setting a sampling frequency, and dynamically calculating the Gaussian filtering kernel standard deviation of the filtering window based on the sampling frequency Further, the preprocessed voltage signal is obtained. In step S13, the dynamic calculation process of the sampling frequency is as follows: ； ； in, The sampling frequency is expressed in Hz. Filter window length.

3. The automatic analysis method for aerodynamic thermal test data based on Bayesian inference according to claim 2, characterized in that, S2 includes the following steps: S21. By calculating the fluctuation intensity and setting the oscillation threshold, the voltage oscillation interval is extracted, merged, and optimized to obtain the strong oscillation interval. The oscillation cutoff point of the voltage signal is then determined, thereby realizing voltage signal oscillation identification. In step S21, the first-order differential absolute mean is used to characterize the fluctuation intensity of the voltage signal. That is, to calculate the mean of the absolute values ​​of the first-order differences of each heat flux sensor. ; Fluctuation intensity Represented as: ； ； in, For the first A heat flow sensor in Voltage signal at time, For the first A heat flow sensor in The voltage signal at a given moment; Data collected 1 second after the wind tunnel test is taken as the baseline for the stable period. Based on the mean and standard deviation of the stable period fluctuation intensity, the voltage oscillation threshold is set as the mean of the stable period + 2.5 times the standard deviation. If the amount of effective data in the stable period is insufficient, the mean of the fluctuation intensity over the entire period + 2.0 times the standard deviation is used as the voltage oscillation threshold. If there is sufficient valid data during the stable period, the threshold calculation formula is as follows: ; If the amount of valid data during the stable period is insufficient, the threshold is calculated using the full-time volatility intensity series. ; in, The threshold for judging voltage oscillation of the heat flow sensor. The arithmetic mean of the volatility intensity sequence during the stable period. The statistical standard deviation of the volatility intensity sequence during the stable period. The arithmetic mean of the all-time volatility intensity series. Statistical standard deviation of the volatility intensity series over all time periods; By traversing the data throughout the entire time period through a filter window, the intervals where the fluctuation intensity exceeds the threshold are marked, and the extremely short noise intervals with a duration of <0.005s are filtered out to obtain the voltage oscillation intervals. Voltage oscillation intervals with adjacent voltage oscillation intervals or gap fluctuations exceeding the oscillation threshold by a set percentage (70%) are merged, while strong oscillation intervals with fluctuations ≥ 1.2 times the oscillation threshold are retained; Extract the end time of all strong oscillation intervals, and take the maximum end time of the strong oscillation interval as the oscillation cutoff point of the voltage signal; S22. By quantifying global fluctuations and setting a heat flow oscillation threshold, the original heat flow oscillation intervals are screened and merged to obtain the main oscillation intervals, determine the oscillation cutoff point of the heat flow signal, and complete the heat flow signal oscillation identification: In step S22, the standard deviation of all heat flux sensor data at each time point is calculated. As an indicator of global volatility intensity; Standard deviation of all heat flux sensor data at each time point Represented as: ； in, Indicates the first A heat flow sensor in Heat flux density at any given time , For all heat flow sensors in The average heat flux density at any given time; Based on the mean of global fluctuations over the entire period with standard deviation The heat flow oscillation threshold was set to the mean plus 1.5 times the standard deviation. Extract the original heat flow oscillation intervals with a duration of ≥0.05s, merge adjacent heat flow oscillation intervals with a gap of ≤0.08s, and select the main oscillation intervals by the condition that the cumulative proportion of fluctuation intensity is ≥80%. The maximum end time of the main oscillation interval is taken as the oscillation cutoff point of the heat flow signal.

4. The automatic analysis method for aerodynamic thermal test data based on Bayesian inference according to claim 3, characterized in that, S4 includes the following steps: S41. Based on the CFD simulation data and the coordinate position of the heat flux sensor, extract the CFD numerical deviation characteristics and concave point abrupt change characteristics: In step S41, the CFD simulation data is interpolated to the coordinate positions of each heat flux sensor to calculate the average heat flux during the steady-state segment and the theoretical CFD heat flux value, thereby obtaining the CFD numerical deviation. ; CFD numerical deviation Represented as: ； in, For the first The average heat flux during the steady-state segment of each heat flux sensor. The theoretical heat flux value is interpolated to the current heat flux sensor location using CFD. The relative deviation of the heat flux mean between the heat flux sensor and the adjacent non-faulty heat flux sensor is calculated. If the relative deviation of the heat flux mean is >0.1, it is determined to be a concave point abrupt change feature. S42. Mark the extreme CFD deviation heat flux sensor and its adjacent heat flux sensor. If the true mutation rate of the adjacent heat flux sensor, that is, the relative deviation between the adjacent heat flux sensor and the preceding normal heat flux sensor, is <0.1, it is determined to be a false concave point. The mutation value of the false concave point is corrected to 0 to obtain the corrected concave point mutation feature. S43. Weight and standardize the CFD numerical bias features and the corrected concave point mutation features, divide the collected initial samples, fit the divided samples respectively, use the Bayesian inference model to calculate the posterior probability of the anomaly and set its threshold, and output the anomaly label and probability. In S43, Min-Max standardization is used to map the two types of features to the [0,1] interval; Based on the 65th percentile of the total feature score as the dividing threshold, the data collected by the heat flow sensor is divided into initial abnormal samples and initial normal samples. Gaussian kernel density is fitted to estimate the likelihood function for the initial abnormal samples and the initial normal samples respectively. When the sample size is insufficient to meet the set value, a normal distribution is used for fitting. The anomalous posterior probability of each heat flux sensor is calculated based on Bayes' theorem. ; Possible posterior probability for each heat flux sensor Represented as: ； in, = =0.5, The likelihood value of the initial outlier sample. This represents the likelihood value of the initial normal samples; Using the 65th percentile of the abnormal posterior probability as the threshold for the abnormal posterior probability, heat flow sensors with an abnormal posterior probability higher than the threshold are marked as abnormal, and the abnormal label and abnormal posterior probability are output.

5. An electronic device, characterized in that, It includes a processor and a memory, the memory storing computer-readable instructions that, when executed by the processor, perform the steps of the method as described in any one of claims 1 to 4.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it performs the steps of the method as described in any one of claims 1 to 4.

Images