Estimating vibration sources for industrial IoT using dilated CNN and deconvolution

Abstract

To minimize data traffic in industrial IoT applications, vibration-based condition monitoring should be conducted on sensors and without requiring machine-specific information. The proposed method enables blind estimation of vibration sources, eliminating the need for information about the monitored equipment or external measurements. Vibrations in rotating machinery primarily originate from two sources: dominant gear-related vibrations and low-energy signals associated with bearing faults. Both sources are distorted by the machine's transfer function before reaching the sensor. This method estimates both sources in two stages: first, the gear signal is isolated using a dilated CNN; second, the bearing fault signal is estimated using the squared log envelope of the residual. The effect of the transfer function is removed from both sources using a novel whitening-based deconvolution method (WBD). Both simulation and experimental results demonstrate the method's ability to detect bearing failures early without additional information. This study considers both local and distributed bearing faults, assuming the vibrations are recorded under stable operating conditions.