Motor bearing life prediction method and system based on edge heterogeneous computing

Abstract

This invention provides a method and system for predicting the lifespan of motor bearings based on edge heterogeneous computing, belonging to the field of motor component lifespan prediction. The method includes: acquiring real-time vibration signals of the motor bearing; constructing a multi-level heterogeneous wake-up scheduling mechanism driven by both events and time sequences; waking up a neural network processor under preset conditions; extracting the absolute time span difference between the last sleep state and the current wake-up time of the neural network processor; constructing a liquid neural network and inferring the hidden state of the neural network processor at the current wake-up time; adaptively adjusting the liquid time constant in the ordinary differential equation based on the energy characteristics of the real-time vibration signal; performing lightweight inference based on the approximate algebraic closed-form solution of the adjusted ordinary differential equation to obtain the remaining lifespan of the motor bearing; and controlling the neural network processor to enter sleep mode. This invention can dynamically allocate computing power according to the device status, achieving accurate capture of sudden anomalies with extremely low power consumption.

Images