农业工程学报
農業工程學報
농업공정학보
2015年
3期
129-136
,共8页
神经网络%遗传算法%曲轴%模态贡献因子%结构优化
神經網絡%遺傳算法%麯軸%模態貢獻因子%結構優化
신경망락%유전산법%곡축%모태공헌인자%결구우화
neural networks%genetic algorithms%crankshafts%mode contribute factors%structure optimization
为了计算得到曲轴动态特性参数，并使用模态贡献因子理论对动态特性进行研究，基于研究结果能够更加有效的实现曲轴结构参数的优化，首先基于多体动力学理论建立了某V12型曲轴的刚柔耦合多体动力学模型。基于模态贡献因子理论对曲轴动态特性进行分析，将前15阶模态缩聚为前5阶模态以减少计算规模，最大应力值的误差为0.9%，扭振角位移的误差为0.16%。利用BP（back propagation）神经网络建立了曲轴动静特性与结构参数之间的数学模型，进行了最大应力值和扭振角位移的线性回归，其输出响应的复相关系数都在0.95以上，表明此网络的泛化能力和预测性能都很好。对神经网络建立的数学模型使用遗传算法进行优化，优化后此曲轴的扭振角位移减小了2.63%，最大应力值减小了3.98%。结果表明神经网络结合遗传算法的优化方法对曲轴结构参数的动态特性和静态特性的联合优化能够满足设计预期，并且具有高效性和可行性。
為瞭計算得到麯軸動態特性參數，併使用模態貢獻因子理論對動態特性進行研究，基于研究結果能夠更加有效的實現麯軸結構參數的優化，首先基于多體動力學理論建立瞭某V12型麯軸的剛柔耦閤多體動力學模型。基于模態貢獻因子理論對麯軸動態特性進行分析，將前15階模態縮聚為前5階模態以減少計算規模，最大應力值的誤差為0.9%，扭振角位移的誤差為0.16%。利用BP（back propagation）神經網絡建立瞭麯軸動靜特性與結構參數之間的數學模型，進行瞭最大應力值和扭振角位移的線性迴歸，其輸齣響應的複相關繫數都在0.95以上，錶明此網絡的汎化能力和預測性能都很好。對神經網絡建立的數學模型使用遺傳算法進行優化，優化後此麯軸的扭振角位移減小瞭2.63%，最大應力值減小瞭3.98%。結果錶明神經網絡結閤遺傳算法的優化方法對麯軸結構參數的動態特性和靜態特性的聯閤優化能夠滿足設計預期，併且具有高效性和可行性。
위료계산득도곡축동태특성삼수，병사용모태공헌인자이론대동태특성진행연구，기우연구결과능구경가유효적실현곡축결구삼수적우화，수선기우다체동역학이론건립료모V12형곡축적강유우합다체동역학모형。기우모태공헌인자이론대곡축동태특성진행분석，장전15계모태축취위전5계모태이감소계산규모，최대응력치적오차위0.9%，뉴진각위이적오차위0.16%。이용BP（back propagation）신경망락건립료곡축동정특성여결구삼수지간적수학모형，진행료최대응력치화뉴진각위이적선성회귀，기수출향응적복상관계수도재0.95이상，표명차망락적범화능력화예측성능도흔호。대신경망락건립적수학모형사용유전산법진행우화，우화후차곡축적뉴진각위이감소료2.63%，최대응력치감소료3.98%。결과표명신경망락결합유전산법적우화방법대곡축결구삼수적동태특성화정태특성적연합우화능구만족설계예기，병차구유고효성화가행성。
As the key component of most power machines, crankshaft is under heavy cycle impact load and torque in working process, and its working performance and reliability directly affect the work efficiency and work safety. The dynamic characteristics are the important indicators of working stability and reliability of crankshaft. In order to get better performance of the crankshaft, when doing the structure optimization design of crankshaft, it is necessary to take into account the dynamic characteristics together with the static characteristics as the optimization objectives. In order to achieve this goal, in this pater, dynamic characteristics of crankshaft were analyzed at first. Based on multi-body dynamics theory, a rigid-flexible coupling multi-body dynamic model of a V12 crankshaft was built on ADAMS software platform. Using the given data by manufacturer, the boundary conditions in the working process of the crankshaft were calculated, which were correctly applied on the rigid-flexible coupling multi-body dynamic model. The angular vibration of crankshaft was obtained by calculation, which can measure the dynamic characteristics of crankshaft. The static characteristics of crankshaft were measured by maximum stress value. To calculate the stress, the method of applying load boundary conditions on corresponding journal of crankshaft was in this way: along the axial line, the loads were uniformly distributed; along the circumferential direction, the loads were distributed on 120 degrees range, and could be expressed in cosine way. Based on the modal contribution factor theory to analyze the dynamic characteristics of crankshaft, it could get a conclusion that the first modal contribution to the crankshaft dynamic response was much greater than the other order. Compared with the calculation results contained the top fifteen modes, the results of just calculating the top five modes appeared acceptable, the error of the maximum stress value and angular vibration was 0.9 percent and 0.16 percent, respectively. Therefore, when using the orthogonal experimental method to collect the samples which needed for the artificial neural network modeling, only the top five modes were calculated, and in this way, the calculation scale was reduced while the calculation accuracy was guaranteed. Through training the sample, using gradient descent learning algorithm, a BP neural network model of two-input and two-output was established, which is with a single hidden layer contained six nodes. The linear regression between the control parameters and maximum stress value, and angular vibration was processed. The multiple correlation coefficient of output response was both larger than 0.95. The results showed that the network had good generalization ability and forecast performance. Using the neural network mathematical model as the constraints and objective function of performance optimization,the calibration was optimized by genetic algorithm. The optimized angular vibration and maximum stress value of crankshaft was reduced 2.63 percent and 3.98 percent, respectively, while the mass was also reduced 0.56 percent, which catered to the modern design requirements of crankshaft lightweight design. It turned out that the method based on BP neural network and genetic algorithm can satisfy the structure parameters optimization, which combined the dynamic characteristics and static characteristics, and is feasible and efficient.
