基于机器视觉的振动筛运动特性快速诊断方法

Machine vision-based vibrating screen motion characteristics rapid diagnostic method

在分析振动筛运动特性主流诊断方法的基础上,引入Lucas-Kanade光流估计算法,提出了一种基于特征点对应关系的快速定性诊断策略。该方法用特征点的光流矢量描述动态目标在相邻2帧中的投影关系,通过检测连续图像序列中目标区域内的特征点数目和光流矢量、设定光流阈值,并应用正反向误差校正理论,完成了对振动筛运动特性的规律捕捉与分析。现场试验结果表明,该方法可有效消除由系统噪声和粉煤粒脱落造成的错误追踪,光流追踪算法耗时约为2.9 ms,在预定义评估时间的15 s内,应用阈值下限评定法,即可针对由各种振动筛局部隐性机械故障引起的筛体运动特性紊乱现象,作出微观、快速、精确的定性诊断。

Based on analysis of the mainstream vibrating screen motion characteristics diagnostic methods currently available,a rapid qualitative diagnostic method with the introduction of the Lucas-Kanade optical flow estimation algorithm and working based on characteristic point correspondence is proposed.The method can perform regular capture and analysis of motion characteristics of vibrating screen by using the light flow vector of feature points to describe the projection relationship of dynamic targets in 2 adjacent frames,detecting the number of feature points and light flow vectors in the target area in the continuous image sequence,setting the light flow threshold,and applying the positive and reverse error correction theory.Result of field test shows the method can effectively eliminate error tracking caused by system noise and shedding of fine coal particles;the algorithmic time for light flow tracking takes about 2.9 ms which is well within the predefined evaluation time of 15 s;and by using the lower-limit threshold evaluation method,the disorder in motion of vibrating screen caused by any local recessive mechanical fault can get rapidly diagnosed in a micro,accurate and qualitative manner.