Evaluation of Features in Time Frequency Domain and Improvement of Sensitivity and Efficiency of Hammering Method using Neural Networks

While research on quantification and automation of hammering inspection is progressing, there are prob- lems in inspection time and cost to replace or assist inspectors. This study aimed for the improvement in sensitivity and efficiency of the hammering method by estimating the influence range of detection results. In the experiment, a concrete wall specimen with void defects was used. The features with higher influence were selected from the time-frequency analysis and multiple feature selection algorithms. As a result of defect detection and its influence range using neural networks, it is possible to detect void defects up to a depth of 8 cm. The inspection results can be efficiently visualized by estimating the influence range.