Civil Engineering & Deep Learning

Recurrent neural network long short term memory model to detect the pile toe using raw data of pile integrity test

This research introduces a novel deep learning model, specifically a Recurrent Neural Network with Long Short-Term Memory (RNN-LSTM) architecture, to automate the detection of pile toe locations using raw data from Pile Integrity Testing (PIT). Traditional Low-Strain Integrity Testing (LSIT) relies heavily on expert interpretation, introducing subjectivity and efficiency limitations. The proposed model aims to overcome these challenges by learning wave propagation behavior from acceleration inputs and accurately generating reflectograms. The study involved collecting LSIT data from various Egyptian driven pile projects, preprocessing raw acceleration signals into digitized velocity-time series, and training several RNN-LSTM networks. An optimized six-layer, 32-neuron LSTM model achieved high accuracy (R² of 0.9126 for training, 0.8778 for validation) and demonstrated satisfactory generalization, with visual inspection confirming 'Good' toe location predictions for 84% of validation cases. This deep learning approach offers greater reliability and reduced reliance on human experience compared to conventional methods.