Study On Method Of Multiple Pressure Sensor Data Fusion

Multi-sensor data fusion has attracted broad attention.The cross-sensitivity problem widely exists in the pressure sensor, whose static characteristics are not only subjected to the variety of target parameters but also affected by a number of non-target parameters.Because it is determined by the corresponding parameters, performance of the sensor's outputs are very unstable and inaccuracy.In this paper,we addressed the concern on the problem of the unstable output voltage of pressure sensor caused by environmental factors such as temperature, noise,and power supply fluctuations.After reading a large number of references, we proposed in this paper a novel approach, which makes use of BP neural network tool to solve the data fusion compensation problem based on the voltage threshold and some sensor output voltage preprocessed method.This new method improves the performance of robustness,fault tolerance and real-time.The main research work of this paper are as follows:First of all,we do intensive research as well as make some comparisons among many data fusion methods.Based on the different characteristics and applicable conditions of these methods,we present a new appropriate fusion algorithm which can satisfy the desire of our system.Secondly, we study on the selecting methods corresponding to neural network input value which is used for data fushion and present three rules on the selecting method of neural network input:the data network integration capabilities, class separability and the impact on the class separability after considering the system noise.Thirdly, we construct a new data fusion approach of the pressure sensor based on BP neural network,which can make multi-layer process to the data and reduce the impact of noise.Then, we can obtain more accurately dataset after completing data fusion with the neural network method.At last,we make some comparision with the experiental resluts and also prove the feasibility of our approach.