Abstract: In response to the issue of the fixed standard results for process time at different stages of service life,a variable process time prediction method considering equipment degradation was proposed.For one single condition,a process time prediction method based on BiGCU-MHResAtt model was constructed,and BiGCU was used to extract the local features.The influence relationships between different features were captured with multiple head residual self-attention networks,and the Remaining Useful Life (RUL) was optimized with a fully connected layer while implementing machining time rate prediction through Weibull probability distribution function.For multiple working conditions,a large dataset and feature transfer model were designed in combination with the single working condition model.Clustering and curve fitting were employed to generate a machining time prediction spectrum.The effectiveness of the proposed method was validated through model training and prediction by using the C-MAPSS dataset.

Key words: BiGCU-MHResAtt-Weibull model, equipment deterioration, variable processing time, remaining useful life prediction

摘要： 针对加工设备在不同寿命阶段造成的机加工工时波动问题,提出了考虑设备劣化的加工工时预测方法。针对单工况,构建了基于BiGCU-MHResAtt-Weibull模型的机加工工时预测模型,采用双向门控卷积单元(BiGCU)提取局部特征,利用多头残差自注意力网络,获取不同特征间的影响关系,全连层输出优化后的残余寿命值并通过Weibull概率分布函数实现加工工时倍率映射;针对多工况,结合单工况模型,设计了大数据集和特征迁移模型,并通过聚类和曲线拟合生成加工工时预测谱系。最后,采用美国国家航天局提供的涡扇发动机数据集完成了模型训练和预测,验证了所提方法的有效性。

关键词: BiGCU-MHResAtt-Weibull模型, 设备劣化, 加工工时, 残余寿命预测