Analysis of pressure pulsation mechanism and dynamic characteristics of axial piston pump

The pressure pulsation of axial piston pump is not only an important cause of rotation speed fluctuation,vibration noise and output stability of the hydraulic system,but also the main information source for obtaining fault information.Hydraulic system is characterized by strong noise interference,which leads to low signal-to-noise ratio(SNR)of detection signals.Therefore,it is necessary to dig deep into the system operating state information carried by pressure signals.Firstly,based on flow loss mechanism of the plunger pump,the mapping relationship between flow pulsation and pressure pulsation is analyzed.After that,the pressure signal is filtered and reconstructed based on standard Gabor transform.Finally,according to the time-domain waveform morphology of pressure signal,four characteristic indicators are proposed to analyze the characteristics of pressure fluctuations under different working conditions.The experimental results show that the standard Gabor transform can accurately extract high-order harmonics and phase frequencies of the signal.The reconstructed time-domain waveform of pressure pulsation of the axial piston pump contains a wealth of operating status information,and the characteristics of pulsation changes under various working conditions can provide a new theoretical basis and a method support for fault diagnosis and health assessment of hydraulic pumps,motors and key components.

基于知识迁移的数据驱动迭代学习模型预测控制

迭代学习模型预测控制(iterative learning model predictive control,ILMPC)作为一种广泛应用于批次生产过程的数据驱动智能控制策略,能够在沿批次逐步提高跟踪性能的同时沿时间不断抑制实时干扰.现有ILMPC算法的点对点学习机制依赖于批次运行工况的强一致性,以此保证当前批次与历史批次间的有效信息传递.然而,生产需求和生产环境的变化通常会导致各批次的操作轨迹和操作周期存在差异,从而使得历史批次提供的先验知识对于后续批次呈现出不精确性和不完整性.为了提高ILMPC在变运行工况条件下的适应性和灵活性,本文提出了一种具有知识迁移机制的数据驱动ILMPC策略.建立自适应深度神经网络(deep neural network,DNN)沿批次学习ILMPC控制行为,实现历史控制经验在当前批次工况下的全面转换.为抑制DNN前期估计误差的影响,在知识迁移机制下进一步构建Tube控制结构下的ILMPC算法,保证ILMPC系统的时域稳定性和迭代域收敛性.针对非线性注塑过程的仿真实验验证了在操作轨迹和操作周期同时变化时,所提方法在跟踪精度和收敛速度方面具有明显优势.