GA—HCM混合聚类算法及其在液体火箭发动机故障检测中的应用

GA-HCM HYBRID CLUSTERING ALGORITHM AND ITS APPLICATION IN FAULT DETECTION FOR LIQUID ROCKET ENGINE SYSTEM

关键词:  液体推进剂火箭发动机  故障检测  故障诊断  模式识别  数值仿真

Key words:  Liquid propellant rocket engine  Fault detection  Fault diagnosis  patternrecognition  Numerical simulation

摘要:

针对HCM聚类算法对初始值和学习参数具有较强依赖性的缺点，提出了GA－HCM混合聚类算法。应用改进的基因算法为HCM算法选取初始种子解，使滑动数据窗上的聚类算法以功能层次分明的“导师一学生”制智能结构出现，从而实现了“精”与“初”相结合的解空间搜索算法，使HCM聚类算法能较快收敛到问题的最优解。同时针对液体火箭发动机系统动力学的特殊性，利用我们提出的一种适用于离线或在线系统故障检测与诊断的算法框架，基于实际试车数据对GA－HCM混合聚类算法进行了准实时的数字仿真。仿真结果表明该算法基本上克服了HCM算法的缺点，能有效地用于液体火箭发动机的事后故障分析或在线故障诊断。该故障诊断框架能区分干扰噪声、永久性故障或间歇性故障所引起的异常数据现象，并能形成当前系统的故障特征模式。对缓变故障的早期检测能力使该算法框架极富应用前景。

Abstract:

Presents a GA-HCM hybrid clustering algorithm with two leveled "director-student"type of intelligent hierarchy, considering the fatal drawbacks of HCM clustering algorithm due to itssubjectivity to the initial values and learning rate. This GA-HCM hybrid clustering algorithm uses theGenetic Algorithm to select the initial seed solutions f0r the HCM algorithm, which is used to preevaluate the initial seed solutions’ fitness. Meanwhile, this hybrid algorithm has been implemented inthe particular fault detection and diagnosis framework specially designed by us before, in order toevaluate its Performance as a multivariate clustering means for the sliding data vectors window. Thenumerical simulation result shows that this hybrid algorithm is capable of finding a globally optimal ornear-optimal clustering centroids for the sliding data window fast than the only Genetic Algorithm oronly HCM, and completely overcome HCM’S drawbacks listed above. The detection ability for theincipient fauIts ensures it a perspective and practical fault detection algorithm.