Study On Vibration Measurement Based On Technology Of Machine Learning In Laser Self-mixing Intereference

The laser self-mixing interference technique is an important and advanced method for ultra-precision measurements.By analyzing the continuous self-mixing signal,the amplitude and frequency(A and f)of the microvibration of an object can be obtained.and then the original vibration get recovered.The waveform and spectrum of the laser self-mixing signal can reflect the vibration pattern of the vibration signal.Based on this,this paper uses an optical platform to acquire a large number of laser self-mixing signals,and combines feature extraction and numerical regression to analyze the acquired signals.In this paper,multivariate features are extracted from the self-mixing signals and downscaled and preprocessed,then the amplitude and frequency of the original vibration are obtained by applying the architecture of numerical regression,and then the original periodic vibration is recovered.Firstly,the current state of application and the related theory for laser self-mixed vibration measurement are analyzed.Based on a thorough study of the principles of generating laser selfmixing interference,a program is built to simulate this system as well as to produce a large number of SMI simulation data sets corresponding to different vibrations.Secondly,an experimental platform is built to carry out the generation of a large number of self-mixing experimental data sets corresponding to the numerical simulation.And the feature extraction of the SMI signal is implemented on the simulation data to mine the features related to the original vibration.Then dimensionality reduction and data normalization are performed to reduce the feature dimension and unify the feature magnitude,avoiding complex analysis of self-mixing signals as well as improving the stability of vibration measurement.Then,based on the extracted features,this paper presents theoretical numerical simulation of the laser self-mixing vibration measurement technique combined with machine learning methods.In the numerical regression-based machine learning model,the effect of laser selfmixed interferometric vibration measurement with random forest integrated with decision tree and linear structure model with added regularization is compared and analyzed.It is shown that the amplitude and frequency of the original vibration measured by random forest fit almost perfectly with the simulated tiny vibration.To further demonstrate the accuracy and validity of the method,the method is applied to the analysis of the actual vibration measurements corresponding to the numerical simulation based on the built SMI optical platform.Finally,the micro-vibration measurement algorithm based on feature extraction and random forest is built and perfected for the laser self-mixed interferometric vibration measurement problem.In order to eliminate the influence brought by the external environment such as light and noise on the self-mixing interference system,wavelet transform is used to filter out the noise.Then,after extracting the multidimensional features of the self-mixing signal,the features are normalized and combined with random forest to measure the amplitude and frequency values of different vibrations.The experimental results show that the technique has good feasibility in laser self-mixing vibration measurement,and completely detaches from the steps of calculating the real-time phase according to the optical feedback intensity,avoiding the complexity of solving various nonlinear parameters according to the optical feedback intensity.The preliminary research and help are provided for the analysis of self-mixing signals and the measurement of practical microvibration.