Machine learning-assisted sparse observation assimilation for real-time aerodynamic field perception

Accurate aerodynamic distribution perception and real-time flight state evaluation are crucial for flight safety,e.g.,stall detection.However,the observations are usually sparse due to limitations in sensor mounting space and cost,and a reconstruction technology is urgently required.Herein,a machine learning-assisted assimilation method based on sparse observations has been proposed.Different from the traditional reconstruction methods focusing on boundary condition correction,the proposed method formulates the flow field pressure distribution as a linear superposition of flow field modes,thereby forming a real-time reconstruction pattern that combines offline modal extraction using computational fluid dynamics(CFD)with real-time determination of modal weights using a neural network.In this study,CFD simulations were conducted under 800different operating conditions for common modal extraction and model training.The weights of these modes were determined online based on merely five observations for reconstructing the full pressure field.A pressure reconstruction with a relative error of 6.1%and a mean square error of 0.003 was achieved within the prescribed condition range.The computational cost was just2 ms for each reconstruction run,significantly faster than the 20 min required by the classical reconstruction ensemble transform Kalman filter.It also showed that the method maintains almost the same accuracy amidst 1.5%measurement noise.As practical examples,shock waves and the change of lift coefficient were analyzed using the proposed method,providing remarkable evidence for the capability of the method in supporting stall detection.These validate the method’s effectiveness and explore its potential in real-time and accurate monitoring of an aircraft.

Generating high-resolution climate maps from sparse and irregular observations using a novel hybrid RBF network

Sparse and irregular climate observations in many developing countries are not enough to satisfy the need of assessing climate change risks and planning suitable mitigation strategies.The wideused statistical downscaling model(SDSM)software tools use multi-linear regression to extract linear relations between largescale and local climate variables and then produce high-resolution climate maps from sparse climate observations.The latest machine learning techniques(e.g.SRCNN,SRGAN)can extract nonlinear links,but they are only suitable for downscaling low-resolution grid data and cannot utilize the link to other climate variables to improve the downscaling performance.In this study,we proposed a novel hybrid RBF(Radial Basis Function)network by embedding several RBF networks into new RBF networks.Our model can well incorporate climate and topographical variables with different resolutions and extract their nonlinear relations for spatial downscaling.To test the performance of our model,we generated high-resolution precipitation,air temperature and humidity maps from 34 meteorological stations in Bangladesh.In terms of three statistical indicators,the accuracy of high-resolution climate maps generated by our hybrid RBF network clearly outperformed those using a multi-linear regression(MLR),Kriging interpolation or a pure RBF network.