Research On Method Of Temperature Compensation For Piezoresistive Pressure Sensor

The piezoresistive pressure sensor is applied most extensively at present, the sensor hasadvantages such as high-sensitivity, high-precision, high stability, small size and is liable tobe Integrated etc. It has been applied in general state-of-the-art technology and industrial fieldsuch as biology, medical treatment, aerospace, ocean engineering, atomic power, etc.Since the semiconductor material is highly sensitive to temperature, it is inevitable thatthe variations of temperature influence the output of sensor, which is called as temperaturedrift. Temperature drift has affected stability and precision of piezoresistive pressure sensorgreatly, for the sake of higher performance, artificial neural network is applied in temperaturecompensation widely, BP neural network is most typical, RBF neural network and waveletneural network are universal.BP neural network, RBF neural network and wavelet neural network have its advantagesand disadvantages. BP neural network has higher approximation of functions, but itsconstringency speed is low. RBF neural network has better fitting function, fasterconvergence speed, and has no local minimization problem, but its dependence on trainingsamples lead to appearance of ill-conditioned data easily. Wavelet neural network has fastconvergence speed, but gets in the local minimum easily.The improvement is conducted to eliminate the defects of neural network in temperaturecompensation for piezoresistive pressure sensor.1. On the basis of study on the principles of BP neural network and principal componentanalysis, according to the low speed of constringency of BP neural network, so thecompensation model based on BP neural network with principal component analysis isproposed, then the model used to compensate temperature drift of piezoresistive pressuresensor is simulated.2. On the basis of study on the principles of RBF neural network and factor analysis,according to the ill-conditioned data problem, so the compensation model based on RBFneural network with factor analysis is proposed, then the model used to compensatetemperature drift of piezoresistive pressure sensor is simulated.3. On the basis of study on the principles of wavelet neural network and geneticalgorithm, according to the wavelet neural network immerging in partial minimum frequently,so the compensation model based on the wavelet neural network with genetic algorithm is proposed, then the model used to compensate temperature drift of piezoresistive pressuresensor is simulated.The results show that the model can depress the effect of temperature on the output ofpiezoresistive pressure sensor, and the stability and accuracy of the sensor are improvedgreatly. The compensation model based on BP neural network with principal componentanalysis has extracted prime information in temperature compensation and eliminated noiseerror in data, so cause neural network to fit smoothly, and accelerates the training speed, andhas overcome the disadvantages of BP neural network with low constringency speed. For thecompensation model based on RBF neural network with factor analysis, the factor analysishas realized dimension reduction and selection of primitive information, it has reduced theinput of network, and simplified the network structure. The learning rate and generalizationability of RBF neural network has been enhanced, and it has overcome the disadvantages ofRBF neural network with the appearance of ill-conditioned data. For the compensation modelbased on the wavelet neural network with genetic algorithm, it has given full play to globalsearch ability of genetic algorithm and function approximation of wavelet neural network, andit has overcome the disadvantages of wavelet neural network with the local minimizationproblem.