密封电子元器件微粒碰撞噪声检测的组件信号识别方法研究

Sealed electronic components are widely used in high reliability requirements of the occasion, the Remainders are a major cause of failure of the relay. Particle impact noise detection (PIND) test is a widely used pre-test for sealed electronic components. For now, there are many miscarriages and difficult identifications for the components signals, and the evaluation method of confidence level is the lacked.This study proposed a method to identify the components signals based on k- mean clustering. High degree of precision to identify the components signals and the remainders was excited with a periodic vibration. Feature Recognition of components signals based on the HHT and machine learning methods have been proposed. High degree of precision to identify the components signals and the remainders was excited with aperiodic vibrations. Confidence level evaluation method of identification based on the consistency of the component forecast of transduction inference was proposed. On this basis, confidence level evaluation method of identification with refuse recognition was studied. Finally, an improved method of identifying components signals for sealed electronic components have been proposed, revised proposal was prepared. The detection accuracy of PIND will be improved.

密封电子元器件广泛应用于对可靠性要求高的场合，其内部多余物是引起继电器失效的主要原因之一。微粒碰撞噪声检测（PIND）试验是检测密封电子元器件内部多余物的出厂前必做试验。针对目前组件信号的误判较多且识别困难、缺乏置信度评价的现状，本研究通过提出基于k-平均聚类的组件信号特征识别方法，实现周期性振动激励下组件信号和多余物信号的高精度区分。提出基于希尔伯特黄变换、机器学习理论和数据融合理论的组件信号特征识别方法，实现非周期振动激励下组件信号的识别。提出带有置信度评价的密封电子元器件组件信号识别方法，建立基于转导推理一致性预测的组件识别置信度评价方法，在此基础上，研究具有拒绝识别能力的组件信号置信度评价方法。最后提出改进的密封电子元器件组件信号识别方法，并编制密封电子元器件相关PIND国家标准的修订建议，提高我国PIND试验的检测精度。

微粒碰撞噪声检测(PIND)试验作为密封电子元器件出厂前的必做筛选试验，是预防密封电子元器件内部多余物的最后一道屏障。对组件信号的识别是密封电子元器件多余物检测的重点和难点，本项目的研究正是在此背景下展开的。首先分析组件信号的产生机理；研究周期性振动激励下组件信号和多余物信号的特征参数，提出基于聚类分析的整体识别法和基于k-近邻算法的单脉冲识别法，实现周期性振动激励下组件信号和多余物信号的高精度区分；研究非周期振动激励下组件信号和多余物信号的特点，提出基于改进的随机森林算法的，非周期振动激励下组件信号识别方法，实现非周期振动激励下组件信号的高精度识别；分析常见置信度评价方法对密封电子元器件组件信号的置信度评价能力，分别设计软分类器和硬分类器的单次检测可信度评价指标和量化分析方法，实现单次检测的可信度评价。最后，将研究成果集成到现有多余物自动检测系统中。实机测试结果表明，本研究提出的改进的密封电子元器件组件信号识别方法，可以有效检测周期和非周期振动激励下的密封电子元器件组件信号。