CN-121981002-A - CFD-based S-shaped pitot tube calibration coefficient prediction method

CN121981002ACN 121981002 ACN121981002 ACN 121981002ACN-121981002-A

Abstract

The invention discloses a CFD-based calibration coefficient prediction method for an S-shaped pitot tube, which is realized by the following steps of S1, carrying out CFD numerical simulation based on parameterized modeling to obtain calibration coefficient sets of the S-shaped pitot tube under different installation angle deviations, flow velocity range conditions and environmental parameters, S2, integrating simulation data to construct a feature vector data set of key influence parameters, S3, dividing the data set into a training set and a testing set, S3, constructing a high-precision nonlinear prediction model by utilizing the training set based on an RBF algorithm, and S4, verifying model performance by utilizing the testing set and judging whether prediction accuracy accords with expectations or not. The CFD simulation technology and the machine learning algorithm are integrated, the problems that the traditional real-flow calibration device is limited in capability and cannot reproduce on-site complex conditions are solved, and a solution is provided for high-precision prediction and coefficient correction of calibration coefficients under on-site complex installation conditions and non-uniform flow fields.

Inventors

GENG ZILONG
LI HAIYANG
XU JIANGUANG
LI JIALONG
Gu Zeyi

Assignees

上海第二工业大学

Dates

Publication Date: 20260505
Application Date: 20260123

Claims (9)

1. The CFD-based S-shaped pitot tube calibration coefficient prediction method is characterized by comprising the following steps of: s1, constructing a calibration coefficient database based on parameterized CFD simulation, namely, establishing a three-dimensional parameterized geometric model of an S-shaped pitot tube and a flow field domain, setting a plurality of working conditions comprising installation angle deviation, a flow velocity range and environmental parameters by a system, and acquiring a calibration coefficient set under different working conditions through CFD numerical simulation; S2, integrating the simulation data obtained in the step S1, extracting key influence parameters to construct a feature vector set, wherein the feature vector at least comprises an installation angle deviation parameter, a flow field parameter and an environment parameter, and taking a corresponding calibration coefficient as an output quantity; S3, dividing the CFD simulation processed data obtained in the step S1 and the step S2 into a training set for model training, a verification set for super parameter tuning and a test set for final performance evaluation; S4, dividing the data obtained in the step S3 into a training set and a testing set, constructing a Radial Basis Function (RBF) neural network model based on the training set data, determining a network structure, a radial basis function center point and an implicit layer output function of the neural network, and performing system optimization on model super parameters by using a verification set; S5, constructing an RBF neural network prediction model, evaluating the prediction performance of the model constructed in the step S4 by using the test set, and taking Root Mean Square Error (RMSE), average relative error (MRE) and decision coefficient (R) as core evaluation indexes; And S6, when the performance of the model meets the preset requirement, the model is applied to the calibration coefficient prediction of the actual working condition, if the performance does not meet the requirement, the step S4 is returned to adjust the model parameters until the model meets the requirement, and a model updating mechanism is established to continuously optimize the model.
2. The method according to claim 1, wherein in step S1, the CFD numerical simulation specifically includes establishing a 1:1 three-dimensional parameterized model, meshing with unstructured mesh technology, encrypting a pitot tube local area and a near wall surface, and performing mesh independence verification. A Realizable k-epsilon turbulence model is selected, boundary conditions such as a mass flow inlet and a pressure outlet are set, a pressure-based coupling solver and a second-order discrete format are adopted, and convergence criteria are that residuals of each control equation are reduced to below 10 and the change of the monitoring physical quantity tends to be stable.
3. The method according to claim 1, wherein for the steps S1 and S2, the S-shaped pitot tube is used as a standard flow rate value in a calibration process using the above-mentioned non-opposite direction, using a standard L-shaped pitot tube already calibrated as a standard, and the calibration is based on Bernoulli 'S principle, and the calibration coefficients are calculated according to the Bernoulli' S equation, as shown in the formula (1) (1) (1) Where K is expressed as a calibration coefficient S-type pitot tube calibration coefficient dependent on yaw angle alpha and pitch angle beta, Representing the differential pressure between the total pressure hole and the static pressure hole of the pitot tube to be calibrated, Is the density of the fluid in the pipeline, Is the current flow rate at the position of the pitot tube to be tested.
4. The method of claim 1, wherein the different conditions in step S1 include a change in the amount of angular offset of the installation (yaw angle, pitch angle), a calibration factor Ksimulation of numerical modeling, a real flow comparison with a real flow condition calibration factor Kreal flow, a range of media flow rates, a change in media temperature, and an accuracy of differential pressure gauge measurements.
5. The method according to claim 1, wherein the parameters of the eigenvector in step S2 are set to six dimensions, medium flow rate range, pitch angle, yaw angle, medium temperature, differential pressure gauge measurement accuracy, calibration coefficient K simulation of numerical simulation (whether real flow condition calibration coefficient K real flow is met or not).
6. The method according to claim 1, wherein in steps S3 and S4, the model training and evaluation is assisted by a cross-validation method to enhance the robustness and generalization ability of the model.
7. The method of claim 1, wherein the process of constructing the RBF neural network model in step S4 includes defining a complete raw data set comprising 140 samples, each sample having 6 input features: (2) Where Xi is the 6-dimensional input eigenvector of the ith sample and Yi is the target output value (pitot tube calibration coefficient) of the ith sample. Randomly arranging 140 samples, selecting the first 110 samples as training sets and the last 30 samples as test sets; random permutation index Irand = pi ({ 1,2,., 140 }) (3) Wherein formula (3), pi ({ 1,2,., 140 }) is a random permutation function; respectively organizing the training set and the testing set into a matrix form, wherein each column of the input feature matrix corresponds to one sample, and the output vector corresponds to the target value of each sample; Training set: test set: And constructing a radial basis function neural network, wherein the number of neurons of an implicit layer of the network is equal to the number of training samples by using a Gaussian radial basis function, and each training sample corresponds to the center of one radial basis function. For the j-th radial basis function (j=1, 2,., M, where m=110 is the number of training samples); (4) Wherein (4), Is the normalized feature vector of the jth training sample and serves as the center of the radial basis function; the expansion speed parameter is 1000 and is used for controlling the smoothness of the function; representing the euclidean norm.
The output of the rbf neural network is a linear combination of hidden layer outputs: (5) Wherein, the formula (5), w0 is a bias term, wj is a weight coefficient corresponding to the j-th radial basis function, M=110 is the number of neurons of the hidden layer (equal to the number of training samples), and the weight coefficient is solved by a least square method, so that the error between the output of the network on the training set and the target value is minimized. Defining an implicit layer output matrix, wherein the first column is 1, and the first column corresponds to a bias term: The method according to claim 1, wherein in step S5, the model performance is preset such that the average relative error (MRE) of the test set is less than 1% and the decision coefficient (R) is greater than 0.98. Root Mean Square Error (RMSE) of the model on the training set and the test set is calculated, the training set root mean square error: (6) wherein (6) Is the original training set target value. Root mean square error of test set: (7) Wherein (7) Is the original test set target value.
9. The method according to claim 1, wherein step S2 further includes a data preprocessing step, normalizing (normalizing) the feature vectors, the normalization strategy needs to consider the physical meaning and the value range of each parameter, linearly map the original data to the [0,1] interval, eliminate the dimensional difference between different features, improve the training stability of the model, use the same normalization parameters as the training set in the test set, and j=1, 2 for each input feature dimension, 6 has: normalized eigenvalues: Output normalization: Wherein: Converting the predicted value of the normalized scale back to the original data scale; inverse normalization of training set: Vector form: Test set inverse normalization: The method of claim 1 further comprising a model updating and impact analysis step of updating model parameters in a manner of incremental learning or periodic full-weight retraining to enable the model to have continuous evolution capability when new CFD simulation data or field actual measurement verification data are obtained, and performing parameter sensitivity analysis and interaction effect research based on the updated model and CFD database to quantitatively evaluate the influence rule of factors such as installation angle deviation on the calibration coefficient and provide data support for optimizing the field installation specification.

Description

CFD-based S-shaped pitot tube calibration coefficient prediction method Technical Field The invention relates to the technical fields of flow and velocity measurement, computational Fluid Dynamics (CFD) simulation and machine learning intersection, in particular to an S-shaped pitot tube calibration coefficient prediction method based on CFD numerical simulation and machine learning algorithm, which is particularly suitable for high-precision prediction and correction of calibration coefficients under complex field installation conditions and non-uniform flow fields. Background The S-shaped pitot tube is used as a flow velocity measuring device which has a simple structure, is firm and durable and is easy to manufacture, and is widely applied to flow velocity/flow velocity measurement in the fields of wind tunnel experiments, industrial flue monitoring, heating ventilation air conditioning and the like. The working principle is that the flow velocity of a point in fluid is calculated according to the Bernoulli equation by measuring the difference between the total pressure and the static pressure of the point. However, due to the S-shaped structure of the pitot tube, the fluid can be separated and complicated to bypass when flowing through the pitot tube, so that the measured pressure difference is different from the pressure difference in an ideal state, and therefore, a calibration coefficient K must be introduced for correction: ; wherein K is expressed as a calibration coefficient S-type pitot tube calibration coefficient depending on the yaw angle alpha and the pitch angle beta, Representing the differential pressure between the total pressure hole and the static pressure hole of the pitot tube to be calibrated,Is the density of the fluid in the pipeline,Is the current flow rate at the position of the pitot tube to be tested. The accuracy of the calibration coefficients directly determines the accuracy of the flow rate measurement. At present, the main method for obtaining the calibration coefficient of the S-shaped pitot tube is a wind tunnel real-flow calibration method, which is a traditional and accepted direct calibration mode, and basically operates by installing the S-shaped pitot tube to be calibrated in an experimental section of a standard wind tunnel, the wind tunnel flow rate is regulated through the system, the measured flow rate of the pitot tube is compared with the wind tunnel reference flow rate point by point (usually calibrated by a laser Doppler velocimeter or a standard pitot tube calibrated by a first-level calibration), and then the calibration coefficient of the wind tunnel reference flow rate is determined. Although this approach has long been regarded as a benchmark, there are several significant inherent drawbacks. Firstly, the method has extremely strict requirements on experimental conditions, and needs to rely on high-precision and high-cost standard wind tunnel facilities and matched precise instruments, so that the initial investment and subsequent maintenance cost of equipment are huge, the whole calibration process is time-consuming and tedious, fine installation and positioning, repeated debugging and point-by-point testing are required each time, and the overall efficiency is low. In practical engineering application, pitot tubes often generate yaw angles and pitch angles with different degrees due to installation deviation or directional change of a flow field, while traditional wind tunnel experiments are limited by mechanical structures and experiment cost, and are difficult to perform systematic calibration test on all possible angle combinations (especially large-angle and compound-angle working conditions), and generally only provide coefficient results facing the incoming flow direction or under a limited number of angles, and other angles which are not actually measured are often dependent on empirical estimation or rough interpolation, so that remarkable measurement uncertainty is introduced. In addition, when a pitot tube is measured in a wind tunnel, the surrounding flow field is constrained and disturbed by the wall of the tunnel, i.e. the so-called "tube wall effect" or "blocking effect". Although the actual flow calibration actually contains the combined effect of the effect in the results, it is difficult to effectively isolate and quantitatively evaluate the effect itself. Further, the method lacks good expansibility and flexibility, and once the size of the wind tunnel pipeline or experimental conditions are changed, the whole set of calibration experiment needs to be re-implemented, so that the workload is increased, and the application value of the method in different application scenes is limited. Disclosure of Invention The invention aims to overcome the defects of the conventional wind tunnel real flow calibration technology and provides a calibration coefficient prediction method based on fusion of a computational fluid dynamics numerical metho