Deep Learning Assisted Pressure Measurement Laser-Induced Grating Scattering Technique

As one of the fundamental parameters in thermodynamics,pressure is the most direct reflection of combustion process.It is of great significance for technical verification,model design and parameter optimization to realize high accuracy measurement of pressure in aeroengine,gas turbine and internal combustion engine.However,how to accurately measure the local pressure in the harsh and changeable environment is still a difficult problem puzzling scientists all over the world.Gasphase pressure measurement can be divided into contact type and non-contact type.The contact pressure measurement usually requires the sensor to be built into the measurement area.Although its accuracy is high,the intrusion of the sensor will destroy the local flow field and cannot record any local pressure changes in real time.Actually,High temperature will also lead to the damage of the sensor.On the contrary,the non-contact type does not require direct contact with the measurement area,which has strong anti-interference ability,high temperature resistance,and wide range of pressure measurement,but the system generally is more complicated.Femtosecond laser-induced grating scattering technique is a non-contact optical measurement technique with high precision and fast response speed,having great potential in realtime diagnosis of gas phase pressure.In recent years,as the most popular machine learning method,deep learning has powerful self-organization and self-adaptation capabilities.It learns data characteristics by simulating the neurons of humans or other organisms to masters the highly nonlinear relationship between input and output so that it has the ability to fit complex functions.To our knowledge,the potential of deep learning in the field of femtosecond Laser-Induced Grating Scattering(fs-LIGS)has not been explored.In this paper,we use the neural network model to learn the fs-LIGS signal characteristics of signal,so that the model can grasp a complex and highly nonlinear relationship between signal and pressure,and realize accurate prediction of pressure.In the experiment,nitrogen and air were selected as the excitation gas,an 800 nm Nd:YAG laser was used as the pump light source while 532 nm continuous wave laser provided a probe beam.We input experimental data into the fully connected neural network(FCN)model,deep neural network(DNN)model and convolutional neural network(CNN)model respectively for training and prediction.The final prediction results show that the three network models are accurate in predicting pressure with high precision,which indicates that the deep learning method has a good grasp of the nonlinear mapping relationship between signal and pressure.