基于自适应神经模糊推理系统的半导体生产线故障预测及维护调度

Failure prediction & maintenance scheduling for semiconductor wafer fabrication based on adaptive neuro-fuzzy inference system

为解决半导体生产线设备故障预测问题,采用自适应神经模糊推理系统构建故障预测模型。利用减法聚类确定模型初始结构,采用由最小二乘算法和梯度下降法所组成的混合学习算法优化模型参数。经实验数据检验,所建模型拟合能力强且精度高,能有效预测生产线下一阶段可能发生故障的设备名称等调度问题关键参数信息。在原有设备维护调度的基础上,嵌入故障预测模型,构建新的设备维护调度方案,并以某半导体生产线制造过程为例进行仿真验证,取得了良好的调度效果。

To solve the failure prediction problem of semiconductor wafer fabrication,Adaptive Neuro-Fuzzy Inference System（ANFIS）was applied to construct a failure prediction model.In this model,subtractive clustering algorithm was used to confirm the original structure of fuzzy inference model,and a hybrid algorithm consisted of the least-squares method and the back propagation gradient descent method was adopted to optimize the parameters.Through verification with the testing data,the model was with good fitting ability and the high recognition accuracy,which helped to forecast important information effectively such as equipment name of the contingent equipment failure.Finally,by embedding the failure prediction into the original maintenance scheduling,a new maintenance scheduling strategy was established.This model was simulated in a semiconductor wafer fabrication,and results revealed satisfactory scheduling performance.