基于实验设计和神经网络的冲压信号特征提取与诊断  

Feature Extraction from Stamping Tonnage Signals and Fault Diagnosis by Neural Networks and Design of Experiment

作　　者：吴洪飞[1] 潘尔顺[1] 

机构地区：[1]上海交通大学机械与动力工程学院,上海200240

出　　处：《机械科学与技术》2007年第11期1385-1389,共5页Mechanical Science and Technology for Aerospace Engineering

基　　金：国家自然科学基金项目(50405016)资助

摘　　要：制造过程中的各变量因素交互作用复杂,对其进行在线质量控制的前提是考虑交互影响,并获得足够的实时噪声因素信息来支持相应的控制策略。随着传感技术的发展,在计算能力和信号处理技术的强力支持下,可以实现对噪声因素的实时估计,进而对加工参数进行相应调整。基于此,本文提出了一种方法,在实验设计基础上,结合主成分分析(PCA)用于信号生成和预处理,并使用神经网络分类器对冲压信号进行特征提取与诊断,以获取更多的实时信息。论文还以一个实例证明方法有效且实用。The interactions between variables in manufacture process are complex. It is important to acquire enough real-time noise information with consideration of these interactions for using the online control strategy. With the development of sensing technology, high power computing and signal processing, it is feasible to estimate noise factors in real-time so that the process parameters can be adjusted. This paper proposes a new methodology information, which focuses on feature extraction from stamping tonnage signals and fault diagnos to get more real is based on NN （ Neural Networks） classifier. Design of experiment （DOE） and principal component analysis （PCA） are employed for signals generation and pretreatment. A real case study was presented to verify that the above-mentioned methodology is efficient and practical.

关 键 词：神经网络 主成分分析 实验设计 实时诊断 

分 类 号：TG386[金属学及工艺—金属压力加工]