论文部分内容阅读
针对间歇过程多工序、变量非线性、非高斯分布等特征,提出了一种基于稀疏距离的故障检测方法(FD-SD).采用稀疏距离衡量测试样本周围训练样本的分布密度,统计测试样本近距离训练样本分布特征,应用变窗宽核密度估计方法估算样本距离的累计分布函数,根据阈值计算样本的稀疏距离.根据稀疏距离的累计分布函数设定检测控制限,建立基于稀疏距离的检测模型.该方法可以避免变量服从高斯、线性分布等假设问题,同时使故障检测的准确性与可靠性得到提高.通过在模拟实例和半导体蚀刻批次过程中的仿真应用,说明该方法可以处理过程具有非线性、多模态、多工序生产特征的故障检测问题.仿真实验验证了方法的有效性.
Aiming at the characteristics of multi-process, non-linear variables and non-Gaussian distribution in batch process, a fault detection method based on sparse distance (FD-SD) is proposed. Sparse distance is used to measure the distribution density of training samples around the test samples. The distance distribution of training samples was estimated by using the variable window width kernel density estimation method to estimate the cumulative distribution function of the sample distance and the sparse distance was calculated according to the threshold.The detection limit was set according to the cumulative distribution function of sparse distance and the sparse distance based detection model This method can avoid the hypothesis that variables follow Gaussian distribution and linear distribution, and improve the accuracy and reliability of fault detection.Through the simulation application in simulation and semiconductor etching batches, it shows that the method can process Nonlinear, multi-modal and multi-process production characteristics.The simulation results show the effectiveness of the method.