贫数据条件下民用航空发动机制造成本估算建模

Manufacturing Cost Estimation Modeling of Civil Aero-Engine with Few Data

民用航空发动机的设计方案一旦冻结,其制造成本就已基本确定。若要提高发动机的经济性,就要在设计阶段就估算出发动机的制造成本并改进设计方案。然而,我国的民用航空发动机行业尚处于起步阶段,在成本估算工作中可参考的历史成本数据很少。为了克服国内民用航空发动机行业积累的制造成本数据量少对成本估算工作精度的制约问题,本文提出了一种虚拟数据生成方法,用于在贫数据条件下对航空发动机制造成本数据集进行扩充。然后,将提出的虚拟数据生成方法与BP神经网络结合运用,建立了一个制造成本估算模型,并且采用贝叶斯优化算法对模型的超参数进行了优化。算例分析结果表明,提出的虚拟数据生成方法可以显著提高制造成本估算模型的精度,建立的成本估算模型的性能也显著优于常见的参数法估算模型。研究成果可为民用航空发动机设计方案优化中的制造成本控制工作提供方法支撑,还能够为航空发动机的产品定价和市场营销等决策提供依据。

Once the design scheme of a civil aviation engine is frozen, its manufacturing cost has been basically determined. To improve the economy of the engine, it is necessary to estimate the manufacturing cost of the engine and improve the design at the design stage. However, the civil aviation engine industry of China is still in its infancy, and there are few historical cost data that can be referenced in cost estimation. To overcome the restriction on the accuracy of cost estimation caused by the small amount of manufacturing cost data accumulated in domestic civil aeroengine industry, a virtual data generation method is proposed to expand the aeroengine manufacturing cost data set under the condition of few data. Then, a manufacturing cost estimation model is established by combining the proposed virtual data generation method with BP neural network, and the hyper parameters of the model are optimized by Bayesian optimization algorithm. The results of example analysis show that the proposed virtual data generation method can significantly improve the accuracy of manufacturing cost estimation model, and the performance of the established cost estimation model is also significantly better than the common parameter estimation model. The research results can not only provide method support for manufacturing cost control in civil aeroengine design scheme optimization, but also provide the basis for aeroengine product pricing and marketing decision-making.