汽车工程
汽車工程
기차공정
AUTOMOTIVE ENGINEERING
2015年
8期
875-880,979
,共7页
刘建敏%李晓磊%乔新勇%张杰
劉建敏%李曉磊%喬新勇%張傑
류건민%리효뢰%교신용%장걸
柴油机%振动%气缸压力%希尔伯特包络%集成神经网络
柴油機%振動%氣缸壓力%希爾伯特包絡%集成神經網絡
시유궤%진동%기항압력%희이백특포락%집성신경망락
diesel engine%vibration%cylinder pressure%Hilbert envelope%integrated neural network
对某12150型柴油机进行了缸内燃烧激励的瞬态动力学计算,分析了其缸盖振动的位移、速度和加速度与缸内燃烧特征参数的对应关系。接着在此基础上,对实测振动加速度进行数字积分和平均滤波得到振动位移信号,并利用希尔伯特包络和滑动平均法提取了振动位移的趋势项。再以该趋势项为输入参数构建了Adaboost BP集成神经网络模型,最后利用此模型对不同工况下的缸内压力进行识别。结果表明:振动位移趋势项与缸内压力的良好对应关系和参数本身的简洁性有效降低了神经网络输入的复杂度,提高了神经网络的训练效率;集成神经网络模型能够准确识别不同工况下的缸内压力,其泛化性和精度均有大幅度提高。
對某12150型柴油機進行瞭缸內燃燒激勵的瞬態動力學計算,分析瞭其缸蓋振動的位移、速度和加速度與缸內燃燒特徵參數的對應關繫。接著在此基礎上,對實測振動加速度進行數字積分和平均濾波得到振動位移信號,併利用希爾伯特包絡和滑動平均法提取瞭振動位移的趨勢項。再以該趨勢項為輸入參數構建瞭Adaboost BP集成神經網絡模型,最後利用此模型對不同工況下的缸內壓力進行識彆。結果錶明:振動位移趨勢項與缸內壓力的良好對應關繫和參數本身的簡潔性有效降低瞭神經網絡輸入的複雜度,提高瞭神經網絡的訓練效率;集成神經網絡模型能夠準確識彆不同工況下的缸內壓力,其汎化性和精度均有大幅度提高。
대모12150형시유궤진행료항내연소격려적순태동역학계산,분석료기항개진동적위이、속도화가속도여항내연소특정삼수적대응관계。접착재차기출상,대실측진동가속도진행수자적분화평균려파득도진동위이신호,병이용희이백특포락화활동평균법제취료진동위이적추세항。재이해추세항위수입삼수구건료Adaboost BP집성신경망락모형,최후이용차모형대불동공황하적항내압력진행식별。결과표명:진동위이추세항여항내압력적량호대응관계화삼수본신적간길성유효강저료신경망락수입적복잡도,제고료신경망락적훈련효솔;집성신경망락모형능구준학식별불동공황하적항내압력,기범화성화정도균유대폭도제고。
The transient dynamics calculation of in-cylinder combustion excitation in a 12150 diesel engine is carried out and the correlations between the displacement, velocity and acceleration of cylinder head vibration and the characteristic parameters of in-cylinder combustion are analyzed. Then on these bases, with the digital integra-tion and average filtering of the vibration acceleration measured, vibration displacement signals are obtained, and by using Hilbert envelope and moving average method the trend of vibration displacement is extracted, with which as input parameter, Adaboost BP integrated neural network model is built. Finally based on the model the cylinder pressures in different working conditions are identified. The results show that the good correlation between the trend of vibration displacement and cylinder pressure as well as the brevity of parameters themselves effectively reduce the complexity of neural networks input, and hence improve the training efficiency of neural network, while the integrat-ed neural networks model can accurately identify cylinder pressures under different working conditions, with its gen-eralization and accuracy greatly increased.