Abstract
Inspection of water pipes has traditionally relied on the superior auditory skills of experienced inspection personnel to detect the sounds of water leaks. More recently, technology is being applied to make the inspections more efficient. An existing technical approach is to detect leaks based on threshold values of sound pressure data indicating sound intensity. Since the decision is based on a single variable, however, it is difficult to distinguish actual leak sounds from other simulated water leak sounds (false positives) occurring in the field, such as mechanical sounds or the sound of sewage water flow. Considering the precision needed for determining the location of leaks, the existing discrimination model may not be adequate for the task. This study focuses on the shape characteristics obtained when a histogram is made from sound pressure data of leaks. We devised a multi-variable discrimination model based on studies of the factors contributing to the three groups Leakage/ Simulated water leak sounds (false positives)/ No leakage, and compared the performance of this model with the existing method. The proposed method yielded a three-variable discrimination model. While the discrimination rate was not markedly better than the existing method for distinguishing “No leakage” from “Other (Leakage or Simulated water leak sounds),” it was found to be effective for discriminating between “Leakage” and “Simulated water leak sounds.”
