西北工业大学学报
西北工業大學學報
서북공업대학학보
JOURNAL OF NORTHWESTERN POLYTECHNICAL UNIVERSITY
2013年
4期
535-539
,共5页
曹善成%宋笔锋%殷之平%黄其青
曹善成%宋筆鋒%慇之平%黃其青
조선성%송필봉%은지평%황기청
飞行载荷%飞行参数%支持向量机回归%主成分分析%遗传算法%交叉验证方法
飛行載荷%飛行參數%支持嚮量機迴歸%主成分分析%遺傳算法%交扠驗證方法
비행재하%비행삼수%지지향량궤회귀%주성분분석%유전산법%교차험증방법
aircraft%efficiency%genetic algorithms%maneuverability%mathematical models%measurements%prin-cipal component analysis%regression analysis%support vector machines%cross validation method%flight load%flight parameter
飞行载荷参数识别是单机寿命监控中的重要技术,主要通过建立飞行参数与飞行载荷之间的转换关系,实现间接获取关键部位的载荷谱。针对飞行参数与飞行载荷之间非线性识别问题,结合飞机典型的机动动作,提出了一种改进的支持向量机回归( SVM-R)飞行载荷识别模型。该模型首先采用主成分分析缩减SVM-R模型输入,再利用交叉验证和遗传算法优化 SVM-R模型设置参数,最后根据优化参数训练得到飞行载荷的SVM-R识别模型。通过在半滚机动动作下,飞行参数识别某一部位弯矩的实例分析,验证了优化改进的SVM-R模型对飞行载荷识别的最大残差可控制在实测载荷的20%以内,平均残差控制在实测载荷的3%以内,且优于未经优化的SVM-R模型。
飛行載荷參數識彆是單機壽命鑑控中的重要技術,主要通過建立飛行參數與飛行載荷之間的轉換關繫,實現間接穫取關鍵部位的載荷譜。針對飛行參數與飛行載荷之間非線性識彆問題,結閤飛機典型的機動動作,提齣瞭一種改進的支持嚮量機迴歸( SVM-R)飛行載荷識彆模型。該模型首先採用主成分分析縮減SVM-R模型輸入,再利用交扠驗證和遺傳算法優化 SVM-R模型設置參數,最後根據優化參數訓練得到飛行載荷的SVM-R識彆模型。通過在半滾機動動作下,飛行參數識彆某一部位彎矩的實例分析,驗證瞭優化改進的SVM-R模型對飛行載荷識彆的最大殘差可控製在實測載荷的20%以內,平均殘差控製在實測載荷的3%以內,且優于未經優化的SVM-R模型。
비행재하삼수식별시단궤수명감공중적중요기술,주요통과건립비행삼수여비행재하지간적전환관계,실현간접획취관건부위적재하보。침대비행삼수여비행재하지간비선성식별문제,결합비궤전형적궤동동작,제출료일충개진적지지향량궤회귀( SVM-R)비행재하식별모형。해모형수선채용주성분분석축감SVM-R모형수입,재이용교차험증화유전산법우화 SVM-R모형설치삼수,최후근거우화삼수훈련득도비행재하적SVM-R식별모형。통과재반곤궤동동작하,비행삼수식별모일부위만구적실례분석,험증료우화개진적SVM-R모형대비행재하식별적최대잔차가공제재실측재하적20%이내,평균잔차공제재실측재하적3%이내,차우우미경우화적SVM-R모형。
Flight load parameter identification is crucial for individual aircraft fatigue monitoring and is achieved mainly through the transformation between flight parameters and flight loads , thus obtaining the load spectrum of a key structural component indirectly .To solve the problem of nonlinear identification of flight parameters and flight loads, we take the typical maneuver actions of an aircraft into consideration and establish an improved flight load parameter identification model with support vector machine regression ( SVM-R ) , which we believe is effective . The core of the mathematical model consists of:(1) we use the principal component analysis to reduce the inputs of the SVM-R;(2) we use the cross-validation method and the genetic algorithm to globally search for and optimize the SVM-R model parameters;(3) we use the optimized SVM-R model parameters to train their identification mod-el.We verify the effectiveness of our identification model by comparing the measured bending moment of a key component of an aircraft in semi-roll flight maneuver with its identified bending moment .The verification results , given in Figs.4 and 5, and their analysis show preliminarily that the maximum relative residual value of the bending moment is 12 .3858%and that the average relative residual value is 2 .3688%, satisfying the requirements that the maximum relative residual value should be controlled within 20%of the measured load and that the average relative residual value should be within 3%, thus indicating that our flight load parameter identification model is accurate and effective .