Research On Water Seepage Detection Of Concrete Based On Percussion Method With MFCC As The Sound Feature

There are plenty of unsaturated hydraulic concrete structures,whose part or all surfaces are operated under conditions that are exposed to water.The extreme rainstorm also has made the concrete members often work in the short-term water environment.Water penetration into concrete has a significant impact on the durability and safety of these structures and members.Therefore,it is of vital significance to detect the water seepage of concrete structures.Concrete permeability is essentially associated with the moisture content in concrete,thus the moisture content is chosen as an index to evaluate the concrete water seepage.For large-volume hydraulic concrete structures,water seepage depth is used as an apparent index to evaluate the moisture content of concrete.This paper develops a novel percussion-based method to detect the moisture content and water seepage depth in concrete.The method of percussion refers to tapping and listening,however,there are few types of research and applications about this method in civil engineering at present.At the same time,the Mel-frequency cepstral coefficient(MFCC)in sound recognition has been chosen as the feature of impact-induced sound during the percussion process to explored the feasibility in water seepage detection of concrete.In the study of concrete moisture content detection based on the percussion method,four concrete cube specimens with different moisture content were used as detected objects.A microphone was employed to collect the impact-induced sound signals by tapping the cube specimens with a hammer during the percussion process.The MFCCs of the sound signals were extracted as the feature of impact-induced sound,which were divided into training sets and testing sets.A support vector machine(SVM)based-machine learning was utilized to train and test,in order to classify the different moisture content in concrete.The experimental results demonstrated that the proposed percussion method using the MFCCs and SVM can identify different moisture levels in concrete cube specimens with accuracies more than 98%.However,concrete cube specimens are generally used for laboratory research,but not suitable for practical engineering.In the research about water seepage depth detection of concrete based on the percussion method,three concrete columns with different water seepage depth were chosen as detected objects.In order to further explore the feasibility of using MFCCs as the sound feature,the linear frequency cepstral coefficients(LFCCs)and the Fusion Coefficients in addition to MFCCs were extracted as sound features to analyze.During the percussion process,a hammer was used to tap the concrete column and a microphone was utilized to collect the impacted-induced sound signals.Three sound features were extracted respectively as the feature of impact-induced sound and divided into training sets and testing sets respectively.The training sets were divided into two groups: “water” and “no water”,aiming to establish a model of water seepage depth based on the random forest.The testing sets were classified according to the conditions of water seepage depth.The water seepage depth in concrete columns was predicted by random forest.Then the sound feature with the highest prediction accuracy was chosen and the performances of the water seepage depth model based on SVM and based on random forest(RF)were compared.The experimental results indicated that the MFCCs were more suitable to be used as the feature of impact-induced sound in water seepage depth detection compared with the LFCCs and the Fusion Coefficients.In addition,the prediction performance of the water seepage depth model based on the RF is better than that based on the SVM when MFCCs were chosen as the sound feature.The accuracy of the percussion method combined with the MFCCs and the RF can reach more than 90%.