Research On Stability Evaluation Method Of Adaptive Optics System

Adaptive optics(AO)technology is widely used in various optical systems to improve the optical performance of the system.For a long time,the research of AO system mainly focuses on its optical performance index.In order to obtain higher imaging ability,researchers try to increase the number of corrector units,improve the detection accuracy of detection units,and improve the signal processing speed.These improvements also increase the complexity of AO system and challenge the stability of the system.This challenge mainly comes from two aspects: on the one hand,in the face of complex systems,it is impossible to establish an accurate evaluation model to measure the impact of system anomalies on the whole system;on the other hand,when AO system is developing towards super large and miniaturization,the application scenarios become complex,and the system state detection relying on manual becomes unsustainable and the risk of accidents increases.Intelligent detection of AO system instability caused by abnormalities in each link is the focus of this paper.The research of this paper is based on the AO system composed of classical shack Hartmann detector and continuous face discrete deformable mirror.Through the simulation and analysis of the closed-loop process of the AO system,the signal characteristics of the unstable system and the influence of each abnormal link on the system are studied.At the same time,this paper explores the recognition methods of these anomalies in machine learning,and applies them to AO system instability detection to achieve the purpose of intelligent evaluation of system stability.The specific work is as follows:1.A 127 units AO system simulation platform is established to simulate all aspects of the AO system.The intermediate process can be observed and controlled,which avoids the inconvenience of data acquisition in the actual system and the high cost of destructive test.The platform can simulate the normal closed-loop process,and can introduce anomalies into the detector and deformable mirror respectively to simulate the abnormal process of instability.2.By analyzing the structure and principle of AO system,the unstable factors affecting the system operation in practical application are sorted out,including the abnormal wavefront detector and the abnormal deformable mirror.Among them,there are four kinds of representative anomalies in the control voltage of deformable mirror: random saturation,oscillation,critical and excessive local voltage difference.There are four kinds of representative anomalies of detector slope: centroid edge sticking,jitter,bad spot and local spot gathering.The instability anomaly is characterized by numerical simulation,and the instability anomaly data is generated.3.This paper simulates system instability by inserting abnormal data in the closedloop process.By comparing the slope anomaly data with the voltage anomaly data,it is found that the root mean square(rms)value of the control voltage can accurately reflect the unstable state of the system with less noise.Select three representative algorithms from machine learning methods: clustering(K-means),classification(K-NN)and prediction(ARIMA)to identify and detect RMS values of normal and abnormal voltage.By comparing the recognition effects of the three methods,it is found that K-means has false alarm(normal recognition as anomaly),ARIMA and K-NN have underreport(anomaly recognition as normal)phenomena,but the three methods can be accurately identified at the first time of anomaly occurrence and meet the basic detection requirements.4.Three detection methods are used to evaluate the operation data of the experimental system.Four kinds of interferences are introduced into 127 units AO system and 265 units AO system: light path occlusion,stray light interference,optical axis offset and partial driver failure.Through the comparison of expert evaluation and the detection results in this paper,it is verified that K-means can effectively identify the stable state of AO system,K-NN has underreport,ARIMA has false alarm,which is basically consistent with the simulation results.In this paper,based on three machine learning methods,through the detection of RMS value of AO system control voltage to identify the stable state of the system,the feasibility of intelligent stability detection in unattended AO system is verified,which provides a reference for the next step of AO system stability improvement.