Development of the Wireless Vibration Sensor Module with the Anomaly Detection Function Calculated by the Variational Auto-Encoder (VAE)

Abstract

An ultra-low power consumption wireless vibration sensor module system equipped with an abnormality determination function using a variable automatic encoder (VAE) has been developed as a new method of anomaly prediction solution for rotating equipment such as motors, pumps, and fans, which are widely used in factories. This system uses a cloud service to learn the vibration conditions of rotating equipment during normal operation, and after wirelessly transmitting the learning parameters to the sensor module, the sensor module itself calculates the abnormality score. In order to be able to transmit the learning parameters wirelessly via LPWA, a compression method using branch pruning and quantization has been established. In addition, it was confirmed that the monitoring by the anomaly score by VAE can detect the anomaly of the equipment more quickly than the conventional monitoring by the vibration characteristic evaluation index value, and this method was confirmed to be extremely useful.