基于深度迁移学习的刀具剩余寿命预测

Prediction of Tool Remaining Useful Life Based on Deep Transfer Learning

针对生产条件变化导致刀具衰退规律发生较大改变,历史衰退规律下的刀具剩余寿命预测模型难以对新规律下的刀具进行准确预测,而新衰退规律下没有足够的带标签样本训练新模型的问题,提出一种基于数据分布自适应的深度迁移学习方法。利用历史性能衰退规律下的刀具过程监控数据样本,训练刀具剩余寿命预测模型;在模型中引入自适应层,对历史样本和新衰退规律下的样本进行领域自适应,更新刀具剩余寿命预测模型参数对新衰退规律下的刀具进行剩余寿命预测。以轮槽铣刀加工过程为例进行实验验证,与迁移前的模型相比,迁移后的模型提高了新衰退规律下刀具剩余寿命预测的准确性。

The change of production conditions leads to great changes of tool performance degradation mode,tool remaining useful life prediction model with historical performance degradation mode could not accurately predict the tool remaining useful life with new performance degradation mode,and there were not sufficient labeled data to train the new model.Thus,a deep transfer learning method based on data distribution adaptive was proposed.The prediction model of historical tool remaining useful life is trained using the tool process monitoring data samples with historical performance degradation mode,and an adaptive layer was introduced to perform domain adaptation for the historical samples and the samples with new performance degradation mode to update the parameters of the prediction model.The example of wheel groove milling cutter proved that the model after transfer can improve the accuracy of tool remaining useful life prediction with new degradation mode compared with the model before transfer.