This paper presents a study where artificial neural networks are used as a curve fitting method applying measured data from an axial compressor test rig to predict the compressor map. Emphasis is on models for predict...This paper presents a study where artificial neural networks are used as a curve fitting method applying measured data from an axial compressor test rig to predict the compressor map. Emphasis is on models for prediction of pressure ratio, compressor mass flow and mechanical efficiency. Except for evaluation of interpolation and extrapolation capabilities, this study also investigates the effect of the design parameters such as number of neurons and size of training data. To reduce the effect of noise, the auto associative neural network has been applied for noise filtering of the data from the parameters used to calculate the efficiency. In summary, the results show that artificial neural network can be used for compressor map prediction, but it should be emphasized that the selection of data normalisation scale is crucial for the model where compressor mass flow is predicted. Furthermore, it is shown that the AANN （auto associative neural network） can be used to the reduce noise in measured data and thereby enhance the quality of the data.展开更多
A comprehensive simulation was performed to better understand the impacts and effects of the additional technical noises on weak-light phase-locking for LISA. The result showed that the phase of the slave laser tracke...A comprehensive simulation was performed to better understand the impacts and effects of the additional technical noises on weak-light phase-locking for LISA. The result showed that the phase of the slave laser tracked well with the received transmitting light under different noise level, and the locking precision was limited by the phase readout noise when the laser frequency noise and clock jitter noise were removed. This result was then confirmed by a benchtop experimental test. The required LISA noise floor was recovered from the simulation which proved the validity of the simulation program. In order to convert the noise function into real time data with random characteristics, an algorism based on Fourier transform was also invented.展开更多
文摘This paper presents a study where artificial neural networks are used as a curve fitting method applying measured data from an axial compressor test rig to predict the compressor map. Emphasis is on models for prediction of pressure ratio, compressor mass flow and mechanical efficiency. Except for evaluation of interpolation and extrapolation capabilities, this study also investigates the effect of the design parameters such as number of neurons and size of training data. To reduce the effect of noise, the auto associative neural network has been applied for noise filtering of the data from the parameters used to calculate the efficiency. In summary, the results show that artificial neural network can be used for compressor map prediction, but it should be emphasized that the selection of data normalisation scale is crucial for the model where compressor mass flow is predicted. Furthermore, it is shown that the AANN （auto associative neural network） can be used to the reduce noise in measured data and thereby enhance the quality of the data.
基金supported by the Space Science Research Projects in Advance(Grant No.O930143XM1)the Scientific Equipment Development and Research Project(Grant No.Y231411YB1) of Chinese Academy of Sciences
文摘A comprehensive simulation was performed to better understand the impacts and effects of the additional technical noises on weak-light phase-locking for LISA. The result showed that the phase of the slave laser tracked well with the received transmitting light under different noise level, and the locking precision was limited by the phase readout noise when the laser frequency noise and clock jitter noise were removed. This result was then confirmed by a benchtop experimental test. The required LISA noise floor was recovered from the simulation which proved the validity of the simulation program. In order to convert the noise function into real time data with random characteristics, an algorism based on Fourier transform was also invented.
