11/07/2025
By Danielle Fretwell

The Francis College of Engineering, Department of Electrical and Computer Engineering, invites you to attend a Doctoral Dissertation defense by Lidan Cao on: “AI-Enhanced Field Reconstruction through the Development of a Versatile Embedded Optical Fiber Sensing System”

Candidate Name: Lidan Cao
Degree: Doctoral
Defense Date: Wednesday, November 19, 2025
Time: 9:30 - 11 a.m.
Location: Perry 215

Committee:

  • Advisor: Xingwei Wang, Professor, Electrical and computer engineering, UMass Lowell
  • Xuejun Lu, Professor, Electrical and computer engineering, UMass Lowell
  • Hualiang Zhang, Professor, Electrical and computer engineering, UMass Lowell

Abstract:
The distribution of temperature and strain fields holds significant importance across various industrial applications, emphasizing the need for accurate monitoring and mapping of these fields. Optical fiber sensors (OFS), which have emerged over the past half-century, are essential components in numerous applications due to their inherent advantages, such as small size, immunity to electromagnetic interference, and high sensitivity. Among OFS, distributed optical sensing systems, particularly those based on Optical Frequency Domain Reflectometry (OFDR), are pivotal in harnessing these advantages. OFDR, leveraging Rayleigh backscattering, enables distributed sensing with millimeter-level spatial resolution and exceptional sensitivity, making it ideal for precise measurements in diverse industrial settings. Its distributed sensing capabilities and high spatial resolution enable intricate pattern designs for enhanced data collection at specific locations.

The accurate reconstruction of spatiotemporal physical fields, such as temperature and strain, from sparse sensor data is a critical challenge across numerous scientific and industrial domains. While distributed OFS, particularly OFDR, offer high spatial resolution and sensitivity, their practical deployment is often hindered by two key obstacles: the need for robust sensor packaging to survive demanding environments, and the mathematical difficulty of accurately reconstructing 2D/3D fields from limited one-dimensional data.

This dissertation presents the end-to-end development of a versatile, AI-enhanced intelligent sensing system designed to overcome these challenges. The work is twofold. First, a modular "physical interface layer" was developed through the systematic design, fabrication, and characterization of OFDR sensors embedded in a range of materials, including elastomer, 3D-printed composites, high-temperature polyimide films, and soft silicone. This investigation established a performance baseline and selection framework for tailoring sensor hardware to specific application requirements, achieving high linearity and characterizing dynamic response.
Second, an "intelligent data reconstruction layer" was created, centered on a novel Quadratic Optimization Gaussian Radial Basis Function (QO-GRBF) algorithm. This advanced algorithm is specifically designed to address the ill-posed, underdetermined nature of field reconstruction from sparse sensor inputs, a common limitation of traditional methods such as the standard RBF or least-squares method. The efficacy of the integrated system was validated through the high-fidelity reconstruction of a 2D temperature field, where the QO-GRBF algorithm achieved an average absolute error as low as 0.21°C and demonstrated over a 30% improvement in accuracy compared to conventional GRBF in scenarios with dense data points.

Ultimately, this research delivers a complete, validated framework—from robust hardware design to an advanced AI-driven algorithm—for high-fidelity field reconstruction. This work provides a powerful and adaptable platform solution for a wide range of monitoring tasks in complex industrial and scientific environments.