PILE DEFECT IDENTIFICATION BASED ON MULTI-HIGHER ORDER MOMENT FEATURE

Weixin Kang ； Yuchen Zhao ； Lingxi Liu

《Journal of GeoEngineering》 13卷2期 (2018/06) Pp. 69-77

https://doi.org/10.6310/jog.201806_13(2).3

摘要

This article is mainly devoted to pile defect identification and classification. The main objective of the work is to improve the pile features. Accordingly, a novel feature based on higher order statistical moments is proposed in this paper. The stress wave reflected signals are possessed by wavelet packet transform method. Three higher order statistical parameters-variance, skewness, and kurtosis are calculated in each wavelet packet band of the decomposed signals. The sliding window method is proposed to extract characteristics in every time interval. In particular, the Principal Component Analysis (PCA) is used to reduce dimension of the merged feature and eliminate the relevance among them. Then the features are fed into Support Vector Machine (SVM). Compared with three other existing features and the 42-dimensional feature, the multi-higher order moment feature achieves the highest classification accuracy which reaches 98%. The simulation results demonstrate that the proposed feature can be used as a suitable tool for pile defect detection. It is simple and effective.

關鍵字

Pile foundation ； defect identification ； higher order moment ； feature extraction ； PCA ； classification