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 learn-ing 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 hyperparam-eter combinations,resulting in more stable prediction values and superior overall performance.
NETL
NSTL
万方数据
