本文在对机械结构振动响应的时间序列建立ARMA模型前,先对时序进行了去趋势处理和正态性、平稳性检验。由于采用了分频段模态参数识别技术,用自行设计的数字滤波软件将时序滤波,从而降低了模型阶次,减少了计算机时和占用内存,并且提高了对小能量模态的识别精度。建模采用以矩法和最小二乘法相结合的方式,避免了发散,加快了收敛。全套程序在实验室考证和对某雷达结构的应用中均取得了较满意的结果。 In this paper, prior to establishing the ARMA model for the time series of the vibration response of the mechanical structure, the timing is processed to the trending and the normality and the stationarity test. Due to the adoption of sub-band modal parameter identification technology, the self-designed digital filtering software filters the timing so as to reduce the model order, reduce the computer time and memory consumption, and improve the recognition accuracy of small energy modal. The modeling uses a combination of moment method and least square method to avoid the divergence and speed up the convergence. A full set of procedures in the laboratory research and the application of a radar structure have achieved more satisfactory results.