摘要
针对滚珠丝杠副多故障问题、单一传感器的不确定性及Dempster合成法则本身存在的不足。提出基于改进DST及RBF神经网络的滚珠丝杠副故障诊断方法。将RBF神经网络的输出结果作为各焦元的基本概率分配,滚珠丝杠副的4种故障类型作为系统的识别框架,引入证据可信度概念,根据新的合成公式进行决策级融合。试验结果表明,改进的新合成公式提高了滚珠丝杠副故障诊断的准确率,使冲突证据合成的结果更理想,取得较好的效果。
According to more problems of ball screw,the uncertainty of single sensor and synthesis of Dempster rule itself exists insufficiency,proposing fault diagnosis methods of ball screw based on improved D-S evidence theory and RBF neural network.The RBF neural network output as the BPA of each focal element,four kinds of fault types of the ball screw as a recognition framework system,introducing a concept of credibility of evidence,and according to a new synthesis formula for decision level fusion.The results show that the improved new synthetic formula increases the accuracy of fault diagnosis ball,the conflict of evidence synthesis results better,to achieve better results.
出处
《制造技术与机床》
北大核心
2014年第9期163-166,共4页
Manufacturing Technology & Machine Tool
基金
国家自然科学基金项目(51075220)
青岛市科技计划基础研究项目(12-1-4-4-(3)-JCH)
山东省高等学校科技计划项目(J13LB11)
