In the production process, aiming at solving the problem of fault detection for nonlinear systems with unknown noise,an improved fault detection method based on particle filter is proposed.Firstly, the real-time estimation of the unknown noise characteristics was obtained by the Sage-Husa filter,introducing UKF to avoid the error caused by linearization of Sage-Hu-sa estimator,and taking full account of the latest measurement information to generate a new proposal distribution function.Then, the resampling process based on the weight-jittered firefly algorithm was optimized,which can reduce the problems of degeneracy and diversity in the particle filter.Finally,comparing the measured value based on improved particle filter with the actual meas-ured value,residual was generated,and the diagnosis was identified through analyzing the residual.Simulation results show the effectiveness of this method.%为解决生产过程中未知噪声背景下非线性系统的故障检测问题,提出了一种基于改进粒子滤波的故障检测方法.首先利用Sage-Husa估计器直接对未知噪声特性进行估计,引入UKF避免Sage-Husa估计器线性化带来的误差,并充分考虑最新量测信息产生新的建议分布函数;然后采用权值抖动的萤火虫算法优化重采样过程,缓解粒子退化和样本枯竭问题;最后依据改进粒子滤波估计的量测值与实际量测值比较产生残差,以残差为依据进行故障检测.仿真结果验证了该方法的有效性.
展开▼