Citation

Abdul Aziz, Nurul Ain (2025) Deep Learning-based construction activities monitoring through distributed fibre sensing. Masters thesis, Multimedia University.

Full text not available from this repository.

Abstract

Inadequate supervision of construction work near roads and utilities can damage infrastructure, disrupt services, and create safety hazards. Current monitoring technologies, such as cameras, satellite imagery, and piezoelectric sensors, have significant limitations. They are susceptible to electromagnetic interference, cover only small areas, require constant power, and require manual analysis of construction activities. While fibre Bragg grating sensors offer an alternative, their high cost restricts their use to short distances. A single-mode optical fibre buried beneath the road surface is used as a distributed vibration sensor to monitor construction activities. The optical fibre operates passively without power and is immune to electromagnetic interference, making it well-suited for real-time, long-range, continuous, and uninterrupted monitoring applications. Various tools, such as a sledgehammer, shovel, hoe, digger, breaker, compactor, and backhoe excavator, are employed to simulate typical activities. Mechanical impacts from these tools induce micro-deformations in the buried single-mode optical fibre, generating backscattered light signals that propagate toward a Distributed Acoustic Sensing (DAS) interrogator. These signals are captured as vibration responses with a spatial resolution of 1-5 m and a sampling frequency of 55 kHz, enabling precise localisation and characterisation of disturbances along the sensing path. The raw vibration signals are processed with a Butterworth bandpass filter with a passband from 20 Hz to 1000 Hz to remove noise while retaining relevant frequency components. Mel-frequency cepstral coefficients (MFCC) and Gammatone frequency cepstral coefficients (GFCC) are extracted from the vibration signals and used as input features to train various deep learning architectures, including Long Short-Term Memory (LSTM) networks, one-dimensional Convolutional Neural Networks (1D-CNN), and a hybrid 1D-CNN-LSTM model. Among these, the 1D-CNN-LSTM model leveraging GFCC features achieved the highest classification accuracy of 93.74%, outperforming the other models evaluated in this study. Therefore, the 1D-CNN-LSTM model demonstrates strong potential for implementing deep learning-based construction activity monitoring using distributed fibre optic sensors.

Item Type:	Thesis (Masters)
Additional Information:	Call No.: TA1815 .N87 2025
Uncontrolled Keywords:	Optical fiber detectors
Subjects:	T Technology > TA Engineering (General). Civil engineering (General) > TA1501-1820 Applied optics. Photonics
Divisions:	Faculty of Artificial Intelligence & Engineering (FAIE)
Depositing User:	Ms Nurul Iqtiani Ahmad
Date Deposited:	16 Apr 2026 01:56
Last Modified:	16 Apr 2026 01:56
URII:	http://shdl.mmu.edu.my/id/eprint/15711

Downloads

Edit (login required)