Detection of square wave impulse interference in eddy current rail defectograms

Traffic safety in rail transport requires continuous monitoring of the rail condition for timely detection and elimination of defects. One of the methods of non-destructive testing of rails is eddy current flaw detection. The data obtained from eddy current flaw detectors (defectograms) are characterized by a significant volume, which makes it necessary to develop effective methods for their automatic processing and analysis. The analysis of defectograms is complicated by various interferences and noises present in the data. One of the most dangerous interferences that significantly distort the shape of useful signals is prolonged impulse interference. They are characterized by a pronounced square wave shape. Unlike instant impulse interference, prolonged noise cannot be eliminated by classical methods. There are no proven effective methods not only for suppressing square wave interference, but even for detecting it. This article attempts to eliminate this drawback and proposes an effective method for detecting square wave impulse interference on eddy current defectograms, which has good explanatory power. Square signals are explored from the point of view of their probability distribution. SW-characteristic was introduced, which allows to estimate the likelihood of data to the distribution of bipolar impulse signals. The smaller the value of SW-characteristic, the more similar the data distribution is to the distribution of bipolar impulse signals (upon condition that the data are normal). Square wave signals are particular example of bipolar impulse signals. The properties of SW-characteristic were examined. SW-characteristic were calculated for the normal distribution and the distribution of a homoscedastic mixture of two Gaussians. It was shown that the value of SW-characteristic for the normal distribution approximately separates the bimodal mixture of two Gaussians from the unimodal case. These and other properties of SW-characteristic allow using it to detect square wave signals in data by comparing with a threshold, which should satisfy a number of conditions. The application of the criterion based on SW-characteristic was demonstrated on the examples of eddy current defectograms; a comparison was made with criteria based on the EM-algorithm and multiscale disperse entropy. The criterion proposed in the article showed the best results. The use of SW-characteristic for detecting square wave noise has proven its effectiveness in the analysis of eddy current defectograms and can be adapted in the future to work with other types of data.