基于BP神经网络的柴油机气缸压力重构

针对缸盖螺栓头部的振动信号和气缸压力信号分别进行了功率谱分析，发现两者频域特性相差很大。因此用线性的方法求得的两者之间的传递函数，不能反映实际缸盖结构的传递特性。基于BP神经网络，在时域内建立了该振动信号与气缸压力信号之间的非线性关系，探索了重构气缸压力的神经网络方法。对信号以等时间间隔采样时，不同的机器转速需要不同的网络结构。针对这一不足，利用样条插值方法拟合采样信号，以等曲轴转角间隔重新采样。这样只要训练样本足够，建立一个网络就能适合所有转速的情况。

Reconstruction of Diesel Engine Cylinder Pressure Based on BP Neural Network

The power spectra of cylinder head bolt vibration signals and cylinder pressure signals were analyzed, respectively. It is found to find that they are much different. The transfer function between them, which was obtained in linear method, could not reflect the transfer characteristic of the cylinder head. Based on the BP neural network, non-linear relation between the acceleration signals and cylinder pressure signals was set up in time domain. The cylinder pressure of a diesel engine was reconstructed by processing the acceleration signals. If the signals were sampled at constant time interval, more than one neural network were needed when the revolution speed of the engine often changed. The spline interpolation was utilized to match the sampled signals, and the signals were re-sampled at constant crankshaft rotation angle internal, so that only one neural network was needed when the training samples were enough.