Research on Human-ware for Safety Pilot Operation

为了提高飞行安全性，对驾驶员操作进行了分析。为了建立人工操作的“人-机”，本文从三个方面进行了研究：（1）驾驶员信息流与控制的神经网络分析;（2）驾驶员的风险分析;（3）低风险最优控制。在神经网络分析中，试图用神经网络建立一个数学模型，以模拟视觉信号作为输入数据，飞行员的操作作为输出数据之间的关系。该模型不仅利用了飞行模拟器的飞行数据，而且利用了真实的飞行数据。采用遗传算法提高模型的泛化能力。分析数据可以清楚地显示出飞行员的不同之处，以及每个视觉线索所起的作用。在飞行员的管理技能分析中，利用记录的飞行数据来分析决策技能。分析的数据清楚地表明了老飞行员和新飞行员在管理技能上的差异。此外，还开发了基于PC的管理技能评估分析工具。在最优控制分析中，神经网络被优化为飞行控制器，以最大限度地提高在几种条件下的着陆成功率。通过对优化控制器的分析，研究了飞机的最优着陆问题。总之，本研究开发了可用于提高飞机飞行安全性的分析工具或方法。

Analysis of human pilot's operation is investigated to improve the flight safety. In order to establish "human-ware" of manual operation, researches are carried out in the following three ways ; (1)Neural Network Analysis of Pilot's Information Flow and Control, (2)Pilot's Workload Analysis, and (3)Optimal Control with Low Risk. In the Neural Network Analysis, the mathematical model which can emulate the relationship between the visual cues as input data and the pilot's operation as output data is tried to obtained by using Neural Network. Flight data obtained from not only Flight Simulator but also Real Flight are utilized to make the model. Genetic Algorithms are employed to improve the generalization capacity of the model. The analysis data could make clear the difference of a veteran and a freshman pilot and the role of each visual cue. In Pilot's Workload Analysis, decision making skills were analyzed by using the recorded flight data Analyzed data made clear the difference in management skill between a veteran and a freshman pilot. Additionally, analysis tool based on PC was developed to evaluate the management skill. In Optimal Control Analysis, the neural network is optimized as a flight controller to maximize the success ratio of landing under several conditions. By analyzing the optimized controller, optimal landing operations are investigated. In summary, this research developed the analysis tools or methods which can be utilized to improve the aircraft flight safety.