摘要：机械故障信号通常是具有非线性时频关系多分量信号，其频谱占有较宽的频带，且各分量的频谱常常相互交叠，给故障诊断带来了很大的障碍。在传统信号分解的基函数线性逼近方法和时频重排的基础上，提出了基于时频二维逼近的信号分量提取方法。该方法对所要提取的特征分量进行参数建模,并计算出分量模型的时频函数以及多分量信号的重排时频分布；然后采用模型的时频函数拟合逼近原始信号的时频分布，并采用非线性最小二乘法确定出模型的各个参数值；最后设置能量下降梯度阈值控制迭代次数，用拟合得出的参数模型重构出信号分量。仿真实例验证了，上述方法对比基于时频滤波的信号提取方法，只需要少数几次拟合就能提取出所需要的信号分量，分量的重构精度较高。这种方法在轴承故障冲击分量提取中的应用表明，其不仅可以较精确地对轴承故障进行定位，而且能为故障原因及故障程度提供准确的判断依据。

Abstract：Machine fault information generally includes multiple components with non-linear time-frequency (TF) relationship, whose wideband frequency spectrums are overlapped. This brings great obstacles for machine fault diagnosis. Based on the signal composition of elementary function one-dimension approximating and TF reassignment, an novel method of signal components extraction is developed. In this method, the component is modeled according to the characteristics of component extracted firstly. Afterwards, the TF distributions of the component model and the analyzed signal are computed separately. And then, the threshold of energy descending gradient is set to control the iterative times and the elementary functions fitted out are selected to reconstruct the useful signal components. Finally, with the nonlinear least square method, the elementary function parameters are determined by using elementary function TF curve surface to fit the analyzed signal’s reassigning TF. The simulation examples verifies using the above method that the component needed can be extracted with several fitting iterative times and the reconstructing result has high accuracy. This method is applied to extract bearing fault components. The experiment results show that the proposed method has good ability of Locating fault precisly and providing corresponding basis for judging fault cause and degree.