公开(公告)号：KR100793244B1

公开(公告)日：2008-01-10

申请号：KR1020060077372

申请日：2006-08-16

Applicant: 재단법인서울대학교산학협력재단

Inventor： 성동원 ,  이수갑 ,  김재환 ,  정원태

IPC: F04C25/02 ,  F04C25/00

Abstract: A vacuum pump monitoring system is provided to compress a plurality of sensor signal data to a single statistic(T2) for users to diagnose system states easily and precisely. A vacuum pump monitoring system includes an acceleration speed sensor(220), a suction pressure sensor(230), a current sensor(240), and a microphone sensor(270) mounted to a booster pump, and a current sensor(250), a discharge pressure sensor(260), and 3-axis acceleration speed sensors(280) mounted to a dry pump. Signals of the above sensors are available as input values for the monitoring system to obtain an output value(T2). The system carries out internal analysis through the steps of collecting data serving as references for system state diagnosis from data of a normal operation section, and carrying out real state diagnosis with reference to the reference data. In the first step, normal operation signals(HDS;Historical Data Set) are input for a predetermined initial time period from the sensors, batches are extracted therefrom to make length of the batches uniform, the data are compressed via MPCA(Multiway Principal Component Analysis), and a value(T2) is calculated to obtain an UCL(Upper Control Limit). In the second step, batches are generated by collecting sensor signals after the predetermined initial time period, and length of the batches becomes uniform to calculate principal components for obtaining the value(T2). The value(T2) is compared with the UCL and the abnormality is determined as the result of the comparison.

Abstract translation: 提供了一种真空泵监测系统，用于将多个传感器信号数据压缩为单个统计量（T2），以便用户容易且精确地诊断系统状态。 真空泵监测系统包括加速度传感器（220），吸入压力传感器（230），电流传感器（240）和安装在增压泵上的麦克风传感器（270），以及电流传感器（250） 排出压力传感器（260）和安装在干式泵上的3轴加速度传感器（280）。 以上传感器的信号可作为监控系统的输入值获得输出值（T2）。 该系统通过从正常操作部分的数据收集用作系统状态诊断的参考的数据并参照参考数据进行实际状态诊断的步骤进行内部分析。 在第一步中，从传感器输入正常操作信号（HDS;历史数据集）预定的初始时间段，从中提取批次以使批次的长度均匀，数据通过MPCA（多路主成分分析 ），并且计算值（T2）以获得UCL（上限控制）。 在第二步骤中，通过在预定的初始时间段之后收集传感器信号来生成批次，并且批次的长度变得均匀以计算用于获得值（T2）的主成分。 将值（T2）与UCL进行比较，并且作为比较的结果确定异常。