摘要
车辆与路面间相互作用力中的车轮六分力是车路间的唯一耦合,获取车轮六分力是开展整车可靠性与耐久性评价的关键。针对传统的车轮六分力获取方法成本高、周期长、效率低的问题,提出数据驱动的车轮载荷快速预测的方法。首先,针对实车道路非平稳随机信号,采用基于自适应噪声完备集合经验模态分解(Complete Ensemble Empirical Mode Decomposition with Adaptive Noise,CEEMDAN)、排列熵(Permutation Entropy,PE)以及小波阈值降噪(Wavelet Threshold Denoising,WTD)的联合方法进行数据去噪;其次,以轮心加速度、减振器位移、质心加速度等容易获取且获取成本低的数据为输入,设计包含非线性传递关系的不同神经网络架构进行多路面下车轮六分力预测,并建立时域、频域、损伤域多维度载荷预测评估体系;最后,为克服训练样本大且获取代价高的缺点,提出基于神经网络输入与输出相关性-相干性分析的输入通道压缩方法,提出最小载荷信号片段划分指标并确定各路面最小片段时长,进行训练集压缩。经过模型不断迭代,车轮六分力的预测值与实测值较为接近,载荷特征也得以保留,计算效率提高28.85%,证明了最小数据集模型能够以较少的输入通道数量、较短的载荷片段时长复现较高期望的预测精度。
The six-component forces at the wheel-road interaction represent the sole coupling between the vehicle and the road surface,and obtaining these forces is critical for conducting reliability and durability assessments of the entire vehicle.In response to the high cost,long cycle,and low efficiency associated with traditional methods for obtaining wheel sixcomponent forces,a data-driven approach for rapidly predicting wheel loads was proposed.Firstly,for the non-stationary random signals on real vehicle roads,a joint method of the complete ensemble empirical mode decomposition with adaptive noise(CEEMDAN),permutation entropy(PE),and wavelet threshold denoising(WTD)was applied for the data denoising.Secondly,the easily obtainable and low-cost data,such as wheel center acceleration,damper displacement,and center of mass acceleration,were used as inputs.Various neural network architectures with nonlinear transfer relationships were designed for multi-surface wheel six-component force prediction.A multi-dimensional load prediction evaluation system was established in the time domain,frequency domain,and damage domain.Finally,in order to overcome the challenges of a large and costly training dataset,an input channel compression method based on the correlation and coherence analysis of neural network inputs and outputs was proposed.Minimum load signal segment division criteria were introduced,and the minimum segment duration for each road surface was determined to compress the training dataset.Through continuous model iterations,the predicted values of the wheel six-component forces closely match the measured values,and the load characteristics are preserved.This demonstrates that the minimal dataset model can achieve a high level of prediction accuracy with fewer input channels and shorter load segment durations,resulting in a 28.85%improvement in computational efficiency.
作者
冯金芝
李增宏
张东东
刘东俭
赵礼辉
FENG Jinzhi;LI Zenghong;ZHANG Dongdong;LIU Dongjian;ZHAO Lihui(School of Mechanical Engineering,University of Shanghai for Science and Technology,Shanghai 200093,China;CMIF Key Laboratory for Strength and Reliability Evaluation of Automotive Structures,Shanghai 200093,China;Shanghai Technical Service Platform for Reliability Evaluation of New Energy Vehicles,Shanghai 200093,China;CATARC Automotive Proving Ground Co.,Ltd.,Yancheng 224100,China)
出处
《机械强度》
北大核心
2025年第10期1-15,共15页
Journal of Mechanical Strength
基金
国家自然科学基金项目(51705322)
产学研合作项目(H-2022-304-042)。
关键词
轮心六分力
载荷预测
神经网络
损伤评估
疲劳耐久分析
Six-component force of the wheel center
Load prediction
Neural network
Damage assessment
Fatigue durability analysis
