飞行员疲劳工作负荷状态识别关键技术研究

The fatigue workload evaluation of the pilot is important for the operating safety of civil aviation. It is necessary to develop the research of the pilots’ fatigue workload state recognition using multi physiological signals. The fatigue workload features extraction algorithm of the pilots’ multi-physiological signals is proposed by integrating Ensemble Empirical Mode Decomposition (EEMD), Treelet transform and entropy technique. The multi-scale elaborate features extraction algorithm for the pilots’ pupil is also proposed by SOHO wavelet and entropy technique. It can be proved that Treelet transform is simple and easy to extract many similar fatigue signal parts, which is suitable to the analysis of fatigue characteristic from the process of the pilot’ operation. This study proposes four kinds of Auto-Encoders such as deep auto-encoder network (DAEN), deep denoising auto-encoder network (DDAEN),deep sparse auto-encoder network (DSAEN) and deep contractive auto-encoder network (DCAEN) to learn the pilots’ multi-physiological fatigue workload features, they are also compared with the traditional feature extraction and dimension reduction methods. The formulated four type of multi-scale-sparse Gaussian Process models are proposed as the classifier of deep learning network by the integration of noise, sparse representation, deep learning, Gaussian process and triangular fuzzy technique. The pilots’ fatigue workload state recognition frame is established by the deep learning and multi-scale sparse Gaussian process model. The excellence of multi-scale sparse Gaussian process over standard Gaussian process is also proved to be effective. The fatigue workload state recognition and early warning system is validated by some groups of pilot’ data.

飞行员疲劳工作负荷的评估对民航运行安全具有显著意义，因此有必要开展基于生理参数信号的飞行员疲劳工作负荷状态识别研究。提出基于集合经验模态分解-小树变换-熵的飞行员多生理参数疲劳特征提取算法。提出基于SOHO小波的飞行员曈孔多尺度细节特征提取算法。证明超小波—小树变换更加简单易行并且能够提取更多的相似疲劳信号分量，非常适合于飞行员操控过程中的疲劳特性分析。比较深度自编码网络、深度降噪自编码网络、深度稀疏自编码网络和深度收缩自编码网络对飞行员疲劳工作负荷抽象特征提取效果,并与传统特征提取和降维方法对比。综合考虑噪声、稀疏表示、深度学习、高斯过程和三角模糊技术，建立基于多尺度高斯过程模型的四种类型深度学习网络分类器，证明多尺度稀疏高斯过程优于高斯过程，建立基于深度学习-多尺度稀疏高斯过程的飞行员疲劳工作负荷状态识别框架。通过国产民机多组飞行员数据验证飞行员疲劳工作负荷状态识别与预测系统的有效性。

围绕民用大飞机、下一代军机座舱工程等国家重大需求，开展座舱视脑交互机理及工程实现研究。座舱人机交互的瓶颈在于飞行的机动性不能超越人体的生理极限，由此产生三个痛点: 1)定向空间障碍引起旋转性飞行错觉; 2)过载飞行致飞行员意识丧失、失去知觉; 3)高难度飞行性能引起飞行员工作负荷变化显著，致飞行员操控飞行能力下降。这三个痛点涉及到飞行员视觉、脑认知和工作负荷，它们是相互耦合响应的。为解决这三个痛点，需要克服三个技术挑战：1) 视脑交互信号鉴别难，处理难; 2) 脑网络标定难、推理难; 3) 飞行员工作负荷建模难，求解难。由三个技术挑战形成代表性成果如下：1) 旋转性飞行员视觉注意力跟踪; 2) 飞行员脑认知网络构建与推理；3) 飞行员工作负荷评价飞行绩效。项目相关成果发表在在IEEE Transactions on Cybernetics、IEEE Transactions on Industrial Electronics、、IEEE Transactions on Neural Networks and Learning System、IEEE Transactions on Intelligent Transportation Systems，IEEE Transactions on Instrumentation and Measurement、IEEE Transactions on Cognitive and Developmental Systems、Reliability Engineering and System Safety、自动化学报和电子学报等国内外TOP杂志上发表SCI/EI期刊论文17篇，其中13篇SCI论文，3篇第一作者论文影响因子11.079。