基于混合神经网络模型的磨粒电荷检测方法研究

由于传统的滑油磨粒在线监测方法无法获取电荷分布位置信息，难以准确测量荷电颗粒数目及其携带的电荷量。为此，本文提出一种基于静电层析成像（Electrostatic tomography，EST）技术和深度学习算法的荷电颗粒检测方法。对EST传感器测量数据采用BP神经网络算法重建出测量截面上电荷的分布图像，采用卷积神经网络（Convolutional neural networks，CNN）算法分析重建图像以识别荷电颗粒数目，将识别的颗粒数目和传感器测量数据组合成输入向量，通过1个多层前馈网络确定带电颗粒数目、感应电荷值与颗粒电荷量值之间的映射关系，得到准确的各颗粒的电荷量值。实验结果表明：混合神经网络模型对数据样本的测量误差为9%，可满足滑油监测对于准确性的要求。

Charge Detection Method of Debris Based on Hybrid Neural Network

It is difficult to determine the distribution of charging debris in lubricating oil and accurately measure the number and charge value of charging debris with traditional on-line monitoring methods. To address this problem, a charging debris detection method based on electrostatic tomography (EST) technology and deep learning algorithm is proposed in the study. First, BP neural network algorithm is adopted to reconstruct the image of charge distribution on the measured section. Thereafter, the reconstructed image is analyzed with convolutional neural network (CNN) algorithm to identify the number of charging debris. Finally, the number of charging debris and the measured data obtained by the EST sensor are combined into an input vector and then transferred to a multilayer feedforward network. Therefore, the mapping relationships between the amount of charging debris, the induced charge value and the charge value of each charging debris can be formed. In this way, the accurate charge value of each charging debris can be obtained. The experimental results demonstrated that the measurement error of the mixed neural network model was about 9%, which could meet the requirement of on-line monitoring of lubricating oil.