Title: Single-Laser 3D Vibrometry with Depth Camera Integration

Keywords: 3D Vibrometry, Depth Camera, Point Cloud Registration, CSLDV

This PhD research explores a new approach to 3D vibrometry using a single scanning laser Doppler vibrometer in combination with a depth camera. Traditional 3D modal analysis relies on multiple sensors, but this work proposes repositioning a single sensor and reconstructing its spatial orientation through point cloud registration.

By capturing the geometry of the object at each measurement viewpoint with an RGB-D camera, the system estimates the transformation between views, enabling accurate 3D vibration mapping with minimal hardware. The research also investigates optimal sensor placement, using algorithms to minimize measurement redundancy while maintaining modal resolution.

Applications include structural health monitoring, particularly in power transformers, where this technique can simplify equipment setup and reduce costs. Validation is performed through comparisons with conventional multi-sensor setups and piezoelectric accelerometers. This project contributes to the fields of experimental modal analysis, robotic perception, and vibration measurement innovation.

• Rui Moreira
• Vítor Santos
• Sérgio Tavares

3D Vibrometry plays a vital role in the experimental analysis of structural dynamics, enabling full-field vibration measurements in all spatial directions. Several systems have been developed for 3D vibrometry, including the Polytec1 PSV-500-3D, which employs three synchronized Laser Doppler Vibrometers (LDVs) arranged orthogonally to capture out-of-plane and in-plane motion. Other approaches include setups using high-speed 3D digital image correlation [1] and scanning LDVs mounted on robotic arms [2].

Some researchers have explored Continuously Scanning Laser Doppler Vibrometry (CSLDV) to improve spatial resolution and efficiency [3]. These systems typically use two orthogonally mounted mirrors to steer a single laser beam along a predefined path. This contrasts with conventional point-by-point scanning systems and enables faster data acquisition [4]. CSLDV has been applied to curved or rotating structures [5], often incorporating reflective mirrors [6] and image-based tracking [7]. However, such setups are complex, requiring precise mirror control and advanced signal processing.

A notable gap in the literature is the absence of approaches using depth cameras in 3D vibrometry. Depth cameras can capture dense point clouds in real time and are widely used in robotics and computer vision. They could provide a low-cost, non-contact method to track a repositioned single LDV across viewpoints. This offers a promising direction for affordable, flexible 3D vibrometry systems. However, integrating these point clouds requires robust registration techniques that can handle noise, partial overlaps [8], and changing perspectives.

Point cloud registration is an active area of research. Traditional methods, such as ICP [9], SHOT [10], and FPFH [11], align point clouds based on geometric features or correspondences. Although widely used, these methods often struggle with large misalignments, non-distinctive surfaces, or occlusions.

Recent work has explored learning-based methods, including 3DMatch [12], FCGF [13], DCP [14], and GeoTransformer [15]. These learn local or global features and predict correspondences or transformations directly. Additional approaches such as DIP [16] and GeDi [17] extend this line of work by incorporating iterative refinement and geometry-guided diffusion, respectively. While effective, they require large labeled datasets and often generalize poorly to new structures.

To address these challenges, recent research has also explored featureless registration strategies. One such approach is the Exhaustive Grid Search (EGS) [18], which performs a brute-force search over rotation and translation parameters to maximize correlation between voxelized point clouds. EGS offers strong performance in challenging conditions, such as low overlap or sensor noise, and does not rely on learned features or prior alignment.

The main objective of this Ph.D. project is to develop a flexible and cost-effective methodology for 3D vibrometry using a single scanning Laser Doppler Vibrometer (LDV). This method is designed to overcome the expense and complexity of conventional multi-LDV systems by simplifying setup and reducing calibration requirements.

1. Depth Camera-Based Tracking and Registration

The system will use a depth camera to track the LDV’s position and orientation during repositioning. At each viewpoint, point clouds will be captured and aligned using robust registration algorithms such as Exhaustive Grid Search (EGS), which are effective even in noisy, partially overlapping, or occluded scenarios.

2. Viewpoint Planning and Optimization

This component addresses how to minimize the number of scanning positions required while maintaining full structural coverage and accurate modal representation. The process involves solving a combined geometric and experimental optimization problem.

3. Continuous Scanning with Galvanometer-Mounted Mirrors

The project will implement a continuous scanning mechanism where galvanometer-mounted mirrors steer the laser across predefined trajectories. This system will be evaluated against traditional point-by-point scanning methods, comparing speed, spatial resolution, and overall complexity.

4. Robotic Automation with a UR10e Arm

To automate LDV repositioning, a UR10e robotic arm will be integrated into the system. This automation improves measurement precision and repeatability, especially for complex or hard-to-access geometries.

5. Practical Applications and Deployment

The methodology will be validated for applications such as structural health monitoring and non-contact vibration testing in field and lab environments. Its flexibility makes it especially valuable for inspecting large or remote structures where setup time and manual access are limited.

• Can 3D vibrometry be accurately performed using a single scanning LDV, repositioned across multiple viewpoints and tracked using depth camera data?

  This question explores the feasibility of replacing traditional multi-LDV setups with a lower-cost, single-sensor approach by estimating rigid transformations between measurements using point cloud registration.

• How can the number and placement of scanning viewpoints be optimized to achieve complete and accurate 3D modal analysis?

  This question investigates strategies to minimize the number of required laser positions while ensuring full coverage of the structure and the reliability of the extracted modal parameters.

• How can continuous scanning using galvanometer-mounted mirrors be implemented to ensure that the laser accurately targets consistent points on the structure?

  This question focuses on the engineering challenges of synchronization, calibration, and scan trajectory design, aiming to ensure spatial consistency across repeated measurements and across different scanning methods.