基于Transformer-Bi-LSTM模型的武器装备剩余寿命预测方法

Weaponry residual life prediction method based on Transformer-Bi-LSTM model

摘要:

武器装备担负保卫国土安全的重要使命，其保持稳定运行状态具有重大国防、政治意义；因其装备运行状态不便中断、故障定位过程复杂，使得传统维修方式效率较低；装备使用数据具有连续性、长期性、不平稳性，甚至一些深度学习模型无法处理其中的退化状态历史依赖与关联问题；通过构建元器件层级的剩余寿命预测架构，对特征工程、退化指标构建以及Transformer-Bi-LSTM模型开展研究，采用距离编码技术，实现针对深度学习模型的技术创新，优化模型预测效果；基于某型武器装备主要器件正常试样数据，进行本方法分析验证，在器件已运行时间达到90%设计试验寿命长度时能够进行有效且准确的剩余寿命预测，所提方法满足武器装备器件寿命预警及更换提醒、保障装备战备完好性的应用需求。

Abstract:

Weaponry is responsible for the important mission of safeguarding national security, and its stable operation is of great nation defense and political significance. Due to the inconvenient interruption of the operation status of the equipment and the complex fault location process, the traditional maintenance method is inefficient. The equipment usage data is continuous, long-term, and instability, and some deep learning models cannot deal with the historical dependence and association of degraded states. By constructing the remaining life prediction architecture at the component level, the feature engineering, degradation index construction and Transformer-Bi-LSTM model are studied, and distance coding are used to realize the technological innovation of the deep learning model and optimize the prediction effect of the model. Based on the normal sample data of the primary components of a weapon equipment, this method has been analyzed and validated. It can effectively and accurately predict the remaining life when the device has been in operation for 90% of its designated test life span. The proposed method meets the requirements for early warning and replacement reminders for weapon equipment devices, ensuring equipment combat readiness integrity.