織物欠点データ収集システムの性能

抄録

A new data collection system was developed by the suitable algorithms for purposes of the detection and evaluation of fabric defects according to the similar procedure to the pattern recognition like the common run of men.
Our development system has the performance that is able to retrieve the data of scannings including the abnormal signals only, according to the light scanning system in the fabric width direction by using the light spot having high resolving power, in the first place.
In the next place, the defect patterns are made by data setting of these scannings with abnormal signals. The shapes, sizes, existing positions and distribution of the electro-optical output levels of the abnormal points of defect patterns are obtained by the pattern recognition procedures like the Mesh dividing method or Sequential dividing method.
Then, the algorithms of this system are organized by mean of the following three parts.
1. The variations of mean values of electro-optical output in each scanning in the fabric feeding direction and the histograms of electro-optical output in the scanning direction for the proper length of the normal fabric part are calculated.
The DC bias superposed on the electro-optical output from the former and the standard deviation σ needed for detecting defects from the latter are obtained respectively.
2. The abnormal levels corresponding to the defects in each scanning electro-optical output are detected. For the purpose of these processings, the DC bias obtained according to algorithm 1 are eliminated from the electro-optical output and the histograms with 32 classes are made by the integrating results in each enlarged ripple parts. Next, the upper and lower limits of histograms in each scanning are set up by K times as much as standard deviation σ of histogram for the normal part of fabric, and the existing position, size and level of abnormal signals that are beyond over these limits of the histograms are regarded as the informations of defects and discriminated.
3. The shape, size, existing position and distribution of the electro-optical output levels of abnormal points of the defect patterns that are made by data setting of the scannings with the abnormal signals are recognized.
It is said that our development system realizes these algorithms in good agreement with the experimental results for some specimens.