Implementation of the Deep Temporal Clustering Neural Network in Stratum Identification for Continuous Flight Auger Piles
Year: 2026
Type: article
Abstract: The stability of pile foundations is a fundamental prerequisite for the safety of transportation infrastructure. Accurately determining the extent of pile penetration into bedrock is crucial for ensuring the load-bearing capacity of the foundation. This study proposed a procedure consisting of a data cleaning operation followed by a deep temporal clustering neural network (DTCNN) to identify strata based on measurement-while-drilling (MWD) time series data. The DTCNN model utilizes bidirectional long short-term memory (Bi-LSTM) to capture time-dependent characteristics. The major contribution of this study is to propose an end-to-end procedural framework for stratigraphic identification, which consists of data cleaning, temporal representation learning, latent-space clustering, and physical-domain reconstruction for the CFA piles. For the algorithm, the data analysis was performed in both the physical domain and the latent domain. In the latent domain, a clustering layer and a supervised classifier head are arranged as two parallel branches after the Bi-LSTM and attention layers, and a decoder reconstructs the standardized input sequence. The proposed procedure accurately recognizes the relationship between MWD data and the stratum layer, achieving an overall precision up to 0.63 and an individual precision up to 0.89 for the MWS layer. The evaluation metrics from the proposed approach significantly outperform those of conventional deep feed-forward neural networks and K-means clustering methods, which treat MWD data as time-independent points. The results validate the feasibility and superiority of the proposed procedure for stratum identification and provide an intelligent technique to improve quality control in field engineering.
Source: Applied Sciences
DOI: 10.3390/app162211357
Authors: Xin Yan, Hongye Yan, Yilin Li
Institution: China Academy of Railway Sciences
Language: English
Subfield: Civil and Structural Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $2,896
OpenAlex ID: W7234577043
