| For an adult lying supine,the value of intracranial pressure is typically 0.78 to 1.76kPa.However,various neurological diseases can lead to increasing intracranial pressure.The monitoring of intracranial pressure is very important in neurosurgery and neurology evaluation,it requires small volume,light weight and high integration of monitoring equipment.Fiber optic pressure sensor is widely used in the field of intracranial pressure monitoring.Using fiber optic pressure sensor to monitor intracranial pressure has a very high requirement in demodulation technology.In recent years,researchers at home and abroad have been proposing solutions to improve the accuracy of pressure.The main work of this paper is to design an embedded instrument for measuring the intracranial pressure based spectral analysis method of machine learning.Firstly,the hardware circuit and measuring device are designed and implemented,and the circuit is made and tested.Then,with multiple linear regression and BP(Back Propagation)neural network two kinds of machine learning spectral analysis methods,the reflection spectral information is used to demodulate the pressure,and the two methods are verified by simulation and experimental analysis.Through continuous adjustment and optimization of the designed circ uit and algorithm,the embedded instrument finally meets the standard of medical requirements.The whole system combines the fiber optic pressure sensing with the embedded system,which can monitor the temperature of the light source in real time and keep it constant,thus greatly reducing the influence of light source disturbance on the reflection spectrum.In addition,the machine learning algorithm makes the demodulation process simple and convenient,and solves the problem of large error and low precision of the results of the modulation system.The accuracy and stability of the whole intracranial pressure measurement system are greatly improved by optimizing the system circuit design and demodulation algorithm.At the same time,the embedded intracranial pressure measuring instrument has high integration degree and small size.The instrument size of the integrated spectrometer is only 20cmx 14cm,and it does not need to be used with PC,which is simple to operate and reduces the space occupation.Therefore,the embedded instrument for measuring intracranial pressure designed in this paper has high practicability and great application potential in the field of clinical medicine.Firstly,this paper describes the principle and research status of intracranial pressure monitoring technology,and then introduces the traditional spectral demodulation analysis method,and its advantages and disadvantages and adaptive conditions are introduced in detail.Then on the basis of machine learning algorithm is proposed for the analysis of the spectrum,two kinds of machine learning algorithm are a multiple linear regression fitting and BP neural network model,through two kinds of methods of simulation verifies the feasibility of its theory,and carries on the comparison to model all aspects of the performance index,after optimizing adjustment of machine learning model in order to achieve its goal.Next,the hardware part of the ICP measurement system is designed,including the light source driver module,the light source temperature measurement and control module,optical switch,optical fiber coupler,spectrometer and so on,experiments are carried out to verify the above analysis algorithm.Finally,machine learning algorithm is deployed on the main control embedded system.Detailed work of this paper is as follows:(1)We have investigated the development status and specific application fields of fiber optic Fabry-Perot pressure sensor,understood the principle and classification of the sensor.And then we make an analysis of the traditional spectral analysis method of theoretical research and make a comparison between their advantages and disadvantages.On the basis of full investigation,we proposed the optimized machine learning spectral analysis algorithm in this paper to make up for the shortcomings of traditional spectral analysis methods.It can reduce the complexity of the algorithm and improve the accuracy of the pressure.(2)According to the proposed spectral analysis method of machine learning,an intracranial pressure measurement system based on high-performance embedded processor is designed,including the constant current driving circuit of the light source,the temperature measurement and control circuit of the light source based on single-chip microcomputer,and the control circuit of spectral acquisition by optical switch,etc.The temperature control module of the light source outputs the PWM signal through the single chip microcomputer.The PWM signal changes the duty cycle to adjust the heating power of the heating resistance.Temperature experiments show that the heating circuit can be heated from the initial temperature to 40?,and it can maintain the stability of this state.By controlling the temperature of the light source to keep constant,the interference of the light source to the reflection spectrum is reduced so as to improve the accuracy of the demodulation pressure.The embedded measuring instrument has the advantages of high integration,small size,real-time monitoring,good temperature compensation,simple operation and so on.(3)The high-performance embedded processor realizes the communication with the temperature control circuit of the single-chip microcomputer and the spectrometer collection part,so as to achieve the purpose of real-time reading of the temperature of the light source and reflection of spectral information.At the same time,the machine learning model is built,the principle of the algorithm is verified and the simulation results are analyzed.During this period,parameters were optimized and algorithms were adjusted continuously.After the preset accuracy was reached,the model was established,and then the accuracy and generalization ability of the model were verified by experimental measurement.In the multiple linear regression model,the error between the true maximum wavelength and the predicted wavelength is 0.0305nm,and the fitting effect is about 0.9992,while in the BP neural network model,the average error of the predicted central wavelength is about 0.0264nm,and the linear fitting degree is about 0.99994.(4)We built pressure control system for Fabry-Perot pressure sensor,by using optical switch to collect the incidence and reflection spectrum.And then we smoothed and filtered the collected spectral data,the machine learning algorithm model built was used to realize the demodulation of pressure,and the experimental data were fitted and analyzed.The sensitivity of Fabry-Perot sensor is about 0.03097nm/kPa.The fitting degree of predicted pressure value and true pressure value is about 0.998.In the range of 0-45kPa,the demodulation pressure can be analyzed according to the sensor reflection spectrum data,which further verifies the feasibility of the theoretical analysis results. |
PDF Full Text Request