Abstract: Real-time condition detection and evaluation of power devices plays an important role in industrial production to ensure product quality, economic efficiency and safety. The traditional operating status of power devices often requires a large number of sensor resources and occupies more wireless/wired multiplexed communication resources. In order to achieve intelligent state detection of power devices, reduce the complexity of using sensors, and enhance the reliability of the system, this paper proposes a power device component bypass multi-point state detection method based on power state detection. The innovation of the method is to bypass the power supply ripple as an important feature of the power device operating state without modifying the original circuit structure. Based on this, a signal processing algorithm is applied to analyze the frequency domain information of the power ripple and construct a data set to further distinguish the respective operating states of the power devices by machine learning methods. In this paper, the proposed correlation method is experimentally verified by detecting ripple of PWM modulation of high-power LED diodes. The experimental results show that the method has superior detection performance and is insensitive to the environment, achieving a more intelligent, reliable, and scalable intelligent detection of power device components.