首页 >

基于民机维修文本数据的故障诊断方法研究

贾宝惠1,姜番2,王玉鑫1,王杜1   

  1. 1. 中国民航大学
    2. 中国民航大学航空工程学院
  • 收稿日期:2021-11-02 修回日期:2022-03-02 出版日期:2022-03-04 发布日期:2022-03-04
  • 通讯作者: 王玉鑫
  • 基金资助:
    国家自然科学基金——民用飞机持续安全性分析技术研究;研究生科研创新

Research on fault diagnosis method based on civil Aircraft maintenance text data

  • Received:2021-11-02 Revised:2022-03-02 Online:2022-03-04 Published:2022-03-04

摘要: 在民航检修与维护过程中积累了大量蕴含丰富故障特征的文本维修记录,然而由于维修文本本身存在复杂性,其还未实现智能诊断,数据利用率低。提出一种不断修正迭代的基于预训练语言模型双向转换器编码表示(Bidirectional Encoder Representations from Transformers,BERT)及轻量级梯度提升机(Light Gradient Boosting Machine, LightGBM)的飞机维修记录的故障原因分析方法,求解文本形式的维修记录中的故障原因,用以辅助维修人员进行正确的维修决策。首先,在基于BERT的故障诊断模型Transformer特征提取架构中引入多头注意力机制(Multi-head Self-Attention),以充分捕捉融合上下文的双向语义、更加关注于重点词汇;其次,为了提高诊断速度减少模型的参数并融合LightGBM模型来实现维修文本的故障原因分类;最后,将改进的模型与其他常用文本分析模型进行对比实验,在基于民机维修文本的故障诊断中该模型的准确率比原BERT模型、TextCNN模型和LSTM模型分别提升了40.1%、23.93%和1.57%,且BERT-LightGBM模型比BERT模型诊断速度提升了18.33%。表明所提方法在实现飞机维修文本故障诊断方面的有效性及优越性。

关键词: 飞机维修文本, 故障诊断, BERT, LightGBM, 多头注意力机制, 参数优化

Abstract: In the process of civil aviation maintenance and maintenance, a large number of text maintenance records containing rich fault features have been accumulated. However, due to the complexity of the maintenance text itself, it has not realized intelligent diagnosis, and the data utilization rate is low.A Bidirectional Encoder Representation from Transformers based on pretrained language model is presented.BERT and Light Gradient Boosting Machine (LightGBM) for aircraft maintenance record fault cause analysis method, solving the fault cause in the text form of maintenance record, for assisting maintenance personnel to make correct maintenance decisions.First, multi-head self-attention is introduced into the Feature extraction architecture of Bert-based Transformer fault diagnosis model to fully capture the bi-directional semantics of fusion context and pay more Attention to key words.Secondly, in order to improve the diagnosis speed, the parameters of the model are reduced and the LightGBM model is integrated to achieve the fault cause classification of the maintenance text.Finally, the improved model is compared with other commonly used text analysis models. In the fault diagnosis based on civil aircraft maintenance text, the accuracy of the model is 40.1%, 23.93% and 1.57% higher than the original BERT model, TextCNN model and LSTM model, respectively.The diagnostic speed of Bert-LightgBM model is 18.33% higher than BERT model.It shows the effectiveness and superiority of the proposed method in realizing the text fault diagnosis of aircraft maintenance.

Key words: Aircraft maintenance text, Fault diagnosis, BERT, LightGBM, Multi head attention mechanism, Parameter optimization

中图分类号: