Research On Intelligent Calibration Model Of Wind Tunnel Balance

The wind tunnel balance is the main equipment for force measurement experiments in aerodynamic research,which ensuring the measurement accuracy is a prerequisite for subsequent force measurement work.The main purpose of force measurement is to measure the six component forces on the vehicle model through the wind tunnel balance,and find out whether there is any unreasonable design in the vehicle model through force analysis,so as to reduce the possibility of unexpected factors in the actual construction.After the wind tunnel balance is stressed,multiple loads are detected by Wheatstone bridges,and then the loads are accurately converted into individual component forces.Therefore,the most important thing is how to calculate the mapping relationship from the load to the component forces.The existing balance calibration method assumes that the component forces are linearly generated by the combination of multiple loads and calculates the coefficients of each term in the polynomial by the least squares method,which does not fully take into account the influence of various nonlinear factors that may occur;at the same time,the balance calibration requires sufficient calibration data,which is time-consuming and labor-intensive to obtain.Therefore,it is meaningful to carry out research on wind tunnel balance calibration through machine learning methods.In order to solve the problem of difficulty in obtaining calibration data for wind tunnel balances,an adversarial Gaussian-Bernoulli restricted Boltzmann machines algorithm is proposed in this thesis to generate realistic balance calibration data.This algorithm incorporates the idea of adversarial generative network in the training process of Gaussian-Bernoulli restricted Boltzmann machines,and further guides the model to learn the probability distribution space of calibration data through the adversarial idea,and samples from the distribution to achieve the purpose of data enhancement.The experimental results show that this algorithm can generate enough realistic data to solve the problem of difficulty in obtaining calibration data and improve the performance in the subsequent calibration work.In order to improve the existing balance calibration method,this thesis proposes a multi-granularity scanning cascade regression algorithm based on an improved deep forest algorithm.The algorithm effectively combines multiple base learners through the idea of integration,and the integrated learning model obtained through training can be used in the balance calibration work.The main purpose of balance calibration is to obtain the mapping relationship between load and each component force.In this thesis,the problem of how to calculate the mapping relationship is converted into a regression problem,and the existing balance calibration method is replaced by the integrated learning model,which solves the defects in the original calibration work.Finally,this model is proved to have high accuracy performance in the calibration work through several comparative experiments.In this thesis,we analyze,design and implement a wind tunnel balance intelligent calibration system that incorporates the two proposed algorithmic models,which calibration data for training can be obtained by data augmentation when only a small amount of data is input.Then,the integrated learning model obtained by training the calibration data with the wind tunnel balance calibration deep forest algorithm can complete the balance calibration work.The implementation of this system significantly reduces the human and material resources required in the balance calibration and improves the efficiency of the calibration work.