基于多维球体信息融合的航空发动机故障分类研究

Research of Fault Classification in Aero-Engine Based on Multi-Dimension Sphere Information Fusion

摘要: 针对航空发动机各类故障表征参数存在耦合且故障类别不易区分的问题，引入多维几何球体距离信息融合的方法，旨在解决航空发动机多故障模式分类的问题。基于发动机自适应模型生成对应的性能参数和测量参数组成故障样本，利用样本的先验状态作为类别标签。同时去掉异常故障样本，并将每一类故障数据看成是一个多维几何球体，利用带动态邻域的自适应粒子群算法确定最优的几何球体的中心以及半径，得到的球体能更好地区分不同的故障类别。同时融合多个几何球体距离的信息，克服了以往仅仅依据样本点与故障类别中心的距离进行分类的弊端。将该方法应用到某航空发动机故障分类中。结果表明，故障分类精度达到96.67%，且计算时间比其他故障分类方法短。

Abstract: Aiming at the problem that fault parameters are coupled and classification is difficult to distinguish, distance information fusion on multi-geometrical sphere was introduced to solve the classification problem for multi-faults of aeroengine. This method used self-adaptive model to produce corresponding performance parameter; these parameters were composed of fault sample; the preview state of the sample was taken as classification label; exceptional sample was eliminated, every kind of fault was looked upon as a multi-dimension geometrical sphere, the core and radius of sphere were determined by self-adaptive particle swarm optimization with dynamic neighborhood, thus the sphere can classify fault well, and distance information of spheres was fused, overcoming the defects that classified simply by distance between sample point and core. This method was applied to fault classification in the aeroengine. The result indicates the accuracy of classification reaches 96.67%, and computation time is shorter than other methods.