3:30 PM - 3:45 PM
[PEM16-06] Anomaly detection system for analogue systems of next-generation solar wind observation systems using machine learning
Keywords:solar wind, next generation radio telescope, machine learning, anomaly detection
The interplanetary spatial scintillation (IPS) is a radio scattering phenomenon caused by density disturbances in the solar wind when the solar wind crosses the line of sight of a radio source. Nagoya University conducts IPS observations by using three large radio telescopes at 327 MHz. Although the existing radio telescopes observe about 70 objects per day, mote IPS observations are required.
Nagoya University has been developing a next generation solar wind observation system (ngSW) which has 1,024 analog inputs and a physical aperture area of 4000 square meters. The ngSW is a next-generation radio telescope with a large aperture and a wide field of view. It has powerful digital signal processing capability and implements a multi-beam function in a 2D flat digital phased array. The ngSW can generate approximately 10 times more solar wind speed data compared to the conventional observation systems, which will help to understand the structure of the heliosphere, elucidate the solar wind acceleration problem, and improve the accuracy of space weather forecasting. As a phase-1 project, a 64 channel digital backend (64-channel system), which is part of this system, has already been developed.
In the ngSW, systems have become more complex. The detection and monitoring of system anomalies are one of the key issues for high-quality observations. In particular, the failure of an antenna element causes amplitude attenuation, resulting in a deterioration of the quality of the synthesized beam. Since 16384 antenna elements will be used in the full-scale array, it is difficult to find failed antennas manually. Therefore, if the input level of each channel can be monitored and the amplitude attenuation of the analog system can be accurately detected, the maintainability of the ngSW will be improved, and more stable observations will be realized.
The purpose of this study is to develop and implement a flexible and highly accurate anomaly detection system for analog systems that can be operated in environments with poor signal-to-noise ratio (S/N) and variable radio frequency interference (RFI). We developed and implemented an AI that detects anomalies in analog signals from the output data of each channel of a 64-channel system. A logistic regression model was used in this AI, and five cross-validations were conducted. In laboratory experiments, a continuous wave was input as test signal, and 12,000 spectral data with and without a 3 dB attenuator on randomly selected analog input were used. The results show that a detection accuracy of the developed AI system is 99.925%. We confirmed this system in another experiment. We obtained 6,400 spectral data from a dipole antenna located outdoor in the Nagoya University campus. The result shows that detection accuracy is 99.531%. In conclusion, we were able to develop a flexible and highly accurate detection system of analog signal anomaly using machine learning.
Nagoya University has been developing a next generation solar wind observation system (ngSW) which has 1,024 analog inputs and a physical aperture area of 4000 square meters. The ngSW is a next-generation radio telescope with a large aperture and a wide field of view. It has powerful digital signal processing capability and implements a multi-beam function in a 2D flat digital phased array. The ngSW can generate approximately 10 times more solar wind speed data compared to the conventional observation systems, which will help to understand the structure of the heliosphere, elucidate the solar wind acceleration problem, and improve the accuracy of space weather forecasting. As a phase-1 project, a 64 channel digital backend (64-channel system), which is part of this system, has already been developed.
In the ngSW, systems have become more complex. The detection and monitoring of system anomalies are one of the key issues for high-quality observations. In particular, the failure of an antenna element causes amplitude attenuation, resulting in a deterioration of the quality of the synthesized beam. Since 16384 antenna elements will be used in the full-scale array, it is difficult to find failed antennas manually. Therefore, if the input level of each channel can be monitored and the amplitude attenuation of the analog system can be accurately detected, the maintainability of the ngSW will be improved, and more stable observations will be realized.
The purpose of this study is to develop and implement a flexible and highly accurate anomaly detection system for analog systems that can be operated in environments with poor signal-to-noise ratio (S/N) and variable radio frequency interference (RFI). We developed and implemented an AI that detects anomalies in analog signals from the output data of each channel of a 64-channel system. A logistic regression model was used in this AI, and five cross-validations were conducted. In laboratory experiments, a continuous wave was input as test signal, and 12,000 spectral data with and without a 3 dB attenuator on randomly selected analog input were used. The results show that a detection accuracy of the developed AI system is 99.925%. We confirmed this system in another experiment. We obtained 6,400 spectral data from a dipole antenna located outdoor in the Nagoya University campus. The result shows that detection accuracy is 99.531%. In conclusion, we were able to develop a flexible and highly accurate detection system of analog signal anomaly using machine learning.