基于改进BiGRU的刀具磨损预测

Intelligent Tool Wear Prediction Method Based on Improved BiGRU

针对双向门控循环神经网络(bidirectional gated recurrent unit,BiGRU)超参数难以确定以及对重要特征捕捉能力较弱的问题,提出了一种改进模型用于刀具磨损预测。模型采用经过下采样的多通道传感器数据作为输入,使用随机搜索算法自适应的确定深度学习模型的最优超参数组合,并引入注意力机制与指数搜索算法增强对全局特征与局部趋势的捕捉能力。模型在PHM2010数据集上进行了实验验证,结果表明,该方法可快速确定超参数组合,并获得更稳定的预测值,具有更好的综合性能。

Aiming at the problem that the hyperparameters of the bidirectional gated recurrent unit(BiGRU)are difficult to determine and the ability to capture important features is weak,a improved model is proposed for tool wear prediction.The model utilized down-sampled multichannel sensor data as input and employed a random search algorithm to adaptively determine the optimal hyperparameter combination for the deep learning model.Additionally,attention mechanism and exponential search algorithm were introduced to enhance the capturing capability of global features and local trends.Experimental validation was performed on the PHM2010 dataset,and the results demonstrated that this approach enabled rapid determination of hyperparameter combinations,resulting in more stable prediction values and superior overall performance.