基于时间卷积网络的移栽机齿轮箱轴承退化趋势研究

Research on the Degeneration Trend of Transplanting Gear Box Bearing Box Based on Time Convolution Network

摘要: 针对移栽机齿轮箱轴承振动信号传递路径长，难以进行基于振动信号的轴承退化趋势预测问题，以基于振动信号进行机器学习的轴承退化趋势预测算法研究，可实时监测齿轮箱退化趋势，最大程度减少损失。利用时间卷积网络通过不同尺寸的卷积核对轴承振动信号进行卷积，实现对历史信息的特征提取；其次通过链式结构实现网络的参数共享，有效解决长时间序列下的低预测精度等问题；通过引入因果卷积、跳层连接与残差网络等结构，使得其在CNN与RNN的基础上能够有效解决梯度消失或爆炸等问题，捕捉序列的长期退化趋势与轴承振动信号间的非线性关系，由此建立齿轮箱轴承退化趋势预测模型。进行实验探究时间卷积网络模型对退化趋势预测准确率的影响并验证其有效性。研究结果表明，基于在不同尺寸、不同故障类型、不同传感器采集到的振动信号对齿轮箱轴承退化趋势都具有显著的预测效果，满足设计要求。

Abstract: For the long way to transmit the vibration signal of the gear box of the transplanted machine, it is difficult to conduct the prediction of the bearing degradation trend based on the vibration signal. Reduce loss to the greatest extent. Using time convolutional networks to convolve the bearing vi-bration signals through convolution kernels of different sizes to achieve the characteristics of his-torical information to extract the characteristics of historical information; secondly, the parameter sharing of the network through the chain structure can effectively solve the problems of low predic-tion accuracy under long sequence; by introducing the structure of causal convolution, jumping layer connection and residual network, it enables it to effectively solve the problems such as gradi-ent disappearance or explosion on the basis of CNN and RNN. Relations, thereby establishing a pre-dictive model of gear box bearing trend. The effect of experimental exploration of the accuracy of the accuracy of degradation trend predicts and verifies its effectiveness. The results of the study show that the vibration signals collected by different sizes, different faulty types, and different sen-sors have significant predictive effects on the trend of the bearing of the gear box to meet the design requirements.