Abstract: Aiming at the problem that engine operation data are multi-modal and difficult to achieve effective life prediction, a branching life prediction method is proposed that integrates multi-modal features of the potential relationship between images and engine time operation data. First, a sliding window was used to segment the engine operation data to construct sequence samples of engine operation data. The Gramian Angular Field (GAF) was used to convert the constructed sequence samples into images. Next, the sequence samples and images were used as a Bi-Directional Long Short-Term Memory (BiLSTM) and a Convolutional Neural Network (CNN) to obtain potential relationship features between sensors such as trends and cycles. Finally, the cross-attention mechanism (CAM)was introduced to achieve fusion of the two modal features and realize life prediction of the engine. Experimental results based on the public C-MAPSS dataset show that the R2 (R-squared) of the prediction method is higher than 0.99 and the RMSE (Root Mean Square Error) is less than 1. It can improve computational efficiency while ensuring the prediction accuracy.

Key words: life prediction, multi-modal fusion, Gramian Angular Field &#40, GAF&#41

摘要： 针对发动机运行数据面临多模态、难以实现有效的寿命预测问题，提出一种融合图像和发动机时间运行数据潜在关系的多模态融合特征并分支寿命预测方法。首先，滑动窗口被用于对发动机运行数据的分割，以构造发动机运行数据的序列样本。格拉姆角场(GAF)被用于将构造的序列样本转化为图像；接着，序列样本和图像分别被用于双向长短期记忆网络(BiLSTM)和卷积神经网络(CNN)获取趋势、周期等传感器之间的潜在关系特征；最后，引入交叉注意力机制(CAM)实现两种模态特征的融合并实现发动机寿命预测。在公开C?MAPSS数据集上的实验结果证明，该预测方法R2高于0.99, RMSE在1以内，能在保证预测精度的同时，改善计算效率。

关键词: 寿命预测, 多模态融合, 格拉姆角场, 卷积神经网络, 交叉注意力机制