Anomalous Data Detection for Roller-Integrated Compaction Measurement

Abstract

This paper presents a theoretical and experimental analysis of an anomalous data detection treatment for roller-integrated compaction measurement (RICM) data. Anomalous data, which may be discovered during the collection of the RICM data, can significantly influence the evaluation of the compaction quality and misrepresent the real compaction situation of the layer. Two types of anomalous data are investigated, and corresponding methods are presented to identify these types. A bidimensional anomalous data identification method is proposed to distinguish anomalous data in calibration tests, and a neighboring weighted-estimation method is presented to reject anomalous data during the compaction quality assessment. The RICM data from three field construction sites are analyzed to verify the applicability and validity of the proposed methods. The results suggest that the first method renders a more accurate correlation, whereas the second method improves the precision of the compaction evaluation.