Online Detection of Wear Particles Using Electrostatic Tomography and Convolutional Neural Network

Abstract

Detection of wear particles in lubrication oil can provide useful information for aircraft engine fault detection. Using electrostatic tomography (EST), the distribution of charged particles passing through the detected cross section can be reconstructed on line, make it possible to detect the information such as particle number and positions. However, due to the passive measurement mechanism of EST, the number of independent measurements is equal to the number of electrodes, leading to low imaging resolution and inaccurate judgment of the amount and positions of wear particles. To solve this problem, the convolutional neural network (CNN) is proposed to reconstruct EST images and detect the number and location of charged particles. 8000 samples were extracted from the model to train the network and reconstruct the charge distribution. And the numbers and locations of particles were detected by using Faster R-CNN. Simulation results show that the algorithm is feasible when the number of charged paticles is less than or equal to 3.