Here is my idea for a PhD research proposal submission. Anyone is free to copy, share, modify, or improve this article in full or in part for their own purpose.
Introduction
In Australia, 70% of bridges exceed 50 years, with 72% in Western Australia located in remote areas, making routine inspections costly and prone to delays, increasing risks of structural failures and freight disruptions. Structural health monitoring (SHM) is vital for predictive maintenance, safety, and cost reduction. This research proposes a cost-effective, sustainable SHM system integrating piezoelectric sensors for acoustic detection and energy harvesting, ground-homing drones with thermal camera and environmental sensors (temperature, humidity, luminance), edge computing, and a digital twin framework. Piezoelectric sensors will detect structural anomalies and power sensor nodes, while solar cells power home bases and drones. Drones, triggered by sensors, capture imagery adjusted by environmental conditions, processed in real-time via edge computing, and visualized in a digital twin for smart predictive bridge maintenance.
Research Background
Bridge monitoring using vision-AI involves photographing structures with cameras and analyzing images with algorithms like CNN and YOLOv8 for real-time defect detection (Ali et al., 2022; Wang et al., 2024). However, conventional cameras struggle with low resolution in poor lighting and sensitivity to environmental noise (e.g., shadows, weather), leading to inaccurate crack detection and missed anomalies (Feng & Feng, 2021; Liu et al., 2021). Thermal cameras overcome these issues by detecting temperature gradients from defects (e.g., cracks, delamination) regardless of lighting, offering higher sensitivity to subsurface anomalies and robustness to environmental variations (Wang et al., 2024).
Image capture can be performed by scanning bridges or remote photography. Scanning is time-consuming, requiring close-range imaging for detailed data, often taking hours to cover a single bridge (Liu et al., 2021; Ellenberger & Byers, 2023). Remote capture demands high-resolution cameras, increasing costs significantly (Feng & Feng, 2021). A solution is to deploy drones triggered by piezoelectric sensors detecting structural damage, capturing targeted images only at affected areas, reducing time and cost while leveraging multi-modal data for accuracy (Winahyu et al., 2021). The piezo-sensor nodes must be robust, easy to deploy, and self-powered (sustainable) due to the absence of a power grid on the bridge.
Research Objectives
This research aims to develop a cost-effective, sustainable, deployable bridge SHM system using piezoelectric sensors, drone-assisted multi-sensor imaging (thermal-camera, environmental sensors) triggered by piezo-sensors alert, and vision-AI assisted digital twin for real-time, predictive monitoring of remote bridges.
Meanwhile, the research objectives are: (1) Develop a smart Bridge Health monitoring system includes a sustainable & damage-sensitive piezoelectric sensor network, a ground-homing drone system with multi-sensor imaging, a digital twin, and the best-fit communication network, (2) Research the best edge-AI methods for displacement/crack detection embedded in a piezo-sensor node as an initial alert to the drone station, (3) Discover the best vision-AI method running on the cloud server for more detailed structural problems based on thermal imaging captured from the drone, and find its correlation to surrounding temperature, humidity, and luminance for the best recognition, (4) Integrate data into a digital twin for 3D modeling and predictive simulations, and (5) Validate through lab and field tests, assessing accuracy, sustainability, and scalability.
Methodology
- Design and develop a piezoelectric sensor network (Picture 1), energy harvesting method, and deployment strategy alongside the bridge.
Picture 1. Piezoelectric sensor node design
- Build a mini bridge (contains concrete and steel) in a laboratory for simulating displacement, cracks, or other structural problems, create an acoustic dataset, train with TNN/TinyML (proposed), build a model, run on the edge-microcontroller (proposed ESP32S3/STM32WL55)
- Design and develop the whole system (Picture 2).
Picture 2. Network architecture design
- Develop computer vision-based multi-point structural displacement/cracks identification method.
Proposed using U-Net algorithm due to its superior semantic segmentation (95–98% IoU), robustness to noise, and compatibility with multi-modal data (e.g., thermal + visual), ideal for detecting cracks and displacement in low-resolution thermal images (Wang et al., 2024). Methods: For cracks, preprocess thermal images with temperature/humidity adjustments from environmental sensors, train U-Net to segment crack regions based on temperature gradients and apply edge detection and thresholding (>1°C) to quantify crack dimensions. For displacement, align thermal/visual images, use U-Net to segment thermal anomalies from stress-induced heat, and track displacement with optical flow, validated by piezoelectric data. These methods, integrated with edge computing and a digital twin, ensure precise, real-time anomaly detection in your drone-based SHM system (Feng & Feng, 2021; Peng et al., 2024).
- Design and develop Digital Twin framework including 3D physics-based models, thermal maps, energy status, and alerts.
- Validation and field testing, evaluating scalability (the sensors network), sustainability, and performance (refine algorithms, sensor placement, and energy systems) for robustness.
Timeline
This research is planned in 4 years with activities: (Year-1): Sensor development, signal processing, energy harvesting, ML model, scalability planning, (Year-2): Drone/home base design, data acquisition, edge computing, analysis development, (Year-3): Data fusion, vision-AI analysis, digital twin development, and (Year-4): Field testing, optimization, thesis writing.
a Technopreneur – writer – Enthusiastic about learning AI, IoT, Robotics, Raspberry Pi, Arduino, ESP8266, Delphi, Python, Javascript, PHP, etc. Founder of startup Indomaker.com