基于深度学习和迁移学习的液压泵健康评估方法

Health Assessment for Hydraulic Pump Based on Deep Learning and Transfer Learning

对液压泵建立健康评估模型需要大量训练数据,然而由于其工作条件随时间和地点的变化,使得获取特定条件下的数据比较困难。为了在目标数据不足的条件下对液压泵建立健康评估模型,提出了一种深度学习和迁移学习的液压泵健康评估方法。首先,通过卷积神经网络的方法对已有大量历史条件下液压泵振动的频域信号建立预测模型,再用迁移学习的思想在少量目标液压泵数据上对深度学习模型进行微调。实验结果表明,该方法可以有效地提高预测准确率。

Building health assessment model for plunger pump needs a large amount of training data, whereas in the real world, working conditions changes all the time, which causes difficulties to obtain data in certain conditions. In order to build an effective prediction model for health assessment for hydraulic pump with limited target data, this paper proposes a deep learning and transfer learning method for health assessment for hydraulic pump. First, convolutional neural network （CNN） was applied to train a deep learning prediction model for the large number of existing historical frequency domain data of vibration of hydraulic pump. Next, transfer learning was used to fine tune the CNN model with limited target data. The experiment shows that this method can effectively improve the accuracy of prediction.