Young-Keun Kim

This paper proposes a novel model-free reinforcement learning (RL) control algorithm for a semi-active tunable vibration absorber (TVA) with nonlinear properties that are operated under nonstationary and multi-frequency excitations. The research addresses two critical challenges in vibration absorber control that are often treated separately: (i) the time-varying and nonlinear stiffness-damping characteristics, and (ii) the complex and nonstationary nature of real-world excitations. To address these challenges, a modified Q-learning algorithm is proposed by integrating vibration-domain knowledge derived from Parseval’s theorem and frequency response functions. This integration not only enables the controller to effectively minimize vibration energy without requiring an explicit model of the plant but also significantly reduces the computational complexity of the learning process. The proposed controller is experimentally validated under nonstationary multi-frequency excitation using a semi-active TVA with highly time-variant stiffness and damping properties. Experimental results demonstrated accurate real-time control performance, achieving an R-squared value of 0.994 compared to an optimal control baseline, and up to 58% reduction in vibration energy. These results provide strong evidence that reinforcement learning control strategies, when guided by vibration-domain knowledge, can offer generalizable, efficient, and adaptive solutions to complex mechanical vibration control problems.

This study presents the design optimization and semi-active resonance control of a small-scale vortex-induced vibration (VIV) bladeless wind turbine (BWT) equipped with a power efficient binary resonance controller. The proposed system integrates a smart-material-based stiffness-tuning module that adaptively adjusts the structure frequency of the BWT to match varying wind speeds. A coupled mechanical–electromagnetic model for BWT was formulated to quantify the relationships among key design parameters, including mast geometry, pivot length, and rod dimensions, and the resulting induced voltage. Multi-parameter optimization was performed to maximize energy-harvesting efficiency under mass and geometric constraints. Experimental evaluation verified an 88.9 % resonance shift capability, broadening the operational lock-in wind speed range from 1.7 to 3.2 m/s. The results confirm the potential of the semi-active BWT control concept for compact, low-noise, and adaptive wind-energy harvesters.

LiDAR-vehicle external calibration involves estimating the relative pose of the LiDAR body from the ego-vehicle coordinate frame, which is required after the LiDAR is assembled on the vehicle. Current methods necessitate direct wiring to the LiDAR controller unit to access 3D point cloud data. However, situations can limit a third party’s access to the LiDAR control unit for security and other practical reasons. Can the LiDAR-vehicle pose be estimated under such circumstances? Therefore, this study addresses this issue by proposing a new approach to the LiDAR extrinsic calibration method that does not require accessing 3D point cloud data. The proposed solution uses a designed camera-target board system to capture images of LiDAR beams projected on a cylindrical target. By utilizing geometrical modeling equations, it can estimate the relative 6D pose of the LiDAR body on the reference frame. Experiments demonstrated accuracy within 0.05 degrees and 0.7 mm. This one-shot LiDAR-vehicle alignment inspection system when accessing 3D point cloud data is restricted for various practical applications.

The bladeless wind turbine (BWT) using vortex-induced vibration is a new class of wind turbine that does not have traditional rotating blades and converts wind energy into vibration energy and into electrical energy based on vortex-shedding principles. Since conventional BWTs are only efficient for a small range of wind speeds near the structural resonant frequency, this study proposes a novel bladeless wind turbine that can tune the resonant frequency for a wider range of wind speeds to improve the effective power generation region. This study designed a discrete on–off resonance-shifting module based on a smart material with variable stiffness that can easily tune the structural frequency of the BWT to two different wind speed classes to enhance power generation efficiency. Experiments were conducted to confirm that the designed BWT can shift the resonant frequency of the structure by 60% and can operate in the ranges of light breezes and gentle breezes. Furthermore, a series of experiments present the power generation effectiveness of the proposed BWT under these different wind speed conditions.

We propose an accurate and noise-robust deep learning model to diagnose bearing faults for practical implementation in industry. To achieve high classification accuracy in a noisy environment, we designed a time-frequency multi-domain fusion block, incorporated bearing-fault physics into the model parameters, and employed attention modules. The proposed model individually extracts essential features from the time-domain vibration signal and the corresponding spectrum in a parallel pipeline. Subsequently, multi-domain feature maps are fused to capture a wider representation of bearing fault signals. The performance was enhanced by incorporating physical knowledge of fault frequencies in the design of the frequency-domain feature extraction network. The employment of an attention mechanism to selectively focus on high-importance fault characteristics on the multi-domain feature maps further improved the accuracy under high noise levels. Experiments on bearing datasets with artificially added noise demonstrated the effectiveness of the proposed model compared to other benchmark models.

This paper proposes a noise-robust and accurate bearing fault diagnosis model based on time-frequency multi-domain 1D convolutional neural networks (CNNs) with attention modules. The proposed model, referred to as the TF-MDA model, is designed for an accurate bearing fault classification model based on vibration sensor signals that can be implemented at industry sites under a high-noise environment. Previous 1D CNN-based bearing diagnosis models are mostly based on either time domain vibration signals or frequency domain spectral signals. In contrast, our model has parallel 1D CNN modules that simultaneously extract features from both the time and frequency domains. These multi-domain features are then fused to capture comprehensive information on bearing fault signals. Additionally, physics-informed preprocessings are incorporated into the frequency-spectral signals to further improve the classification accuracy. Furthermore, a channel and spatial attention module is added to effectively enhance the noise-robustness by focusing more on the fault characteristic features. Experiments were conducted using public bearing datasets, and the results indicated that the proposed model outperformed similar diagnosis models on a range of noise levels ranging from −6 to 6 dB signal-to-noise ratio (SNR).

In this study, we propose an accurate automatic light detection and ranging (LiDAR) alignment inspection system designed for manufacturing environments. The proposed inspection system is a single-shot, static method that uses a target system composed of a planar target with an NIR camera. Each laser beam spot projected on the target surface is segmented and located by analyzing the NIR images. We also propose a method to extract and match the corresponding beam feature points from the 3D point cloud and NIR image. The correspondence points are used to estimate an extrinsic LiDAR pose related to the target frame, which is then converted into a pose relative to the vehicle’s body. We validated the performance of the proposed system with experiments using a low-resolution LiDAR, and it exhibited an accuracy as high as 0.02 degrees and 0.4 mm for rotation and translation alignment, respectively.

Automobile datasets for 3D object detection are typically obtained using expensive high-resolution rotating LiDAR with 64 or more channels (Chs). However, the research budget may be limited such that only a low-resolution LiDAR of 32-Ch or lower can be used. The lower the resolution of the point cloud, the lower the detection accuracy. This study proposes a simple and effective method to up-sample low-resolution point cloud input that enhances the 3D object detection output by reconstructing objects in the sparse point cloud data to produce more dense data. First, the 3D point cloud dataset is converted into a 2D range image with four channels: x, y, z, and intensity. The interpolation on the empty space is calculated based on both the pixel distance and range values of six neighbor points to conserve the shapes of the original object during the reconstruction process. This method solves the over-smoothing problem faced by the conventional interpolation methods, and improves the operational speed and object detection performance when compared to the recent deep-learning-based super-resolution methods. Furthermore, the effectiveness of the up-sampling method on the 3D detection was validated by applying it to baseline 32-Ch point cloud data, which were then selected as the input to a point-pillar detection model. The 3D object detection result on the KITTI dataset demonstrates that the proposed method could increase the mAP (mean average precision) of pedestrians, cyclists, and cars by 9.2%p, 6.3%p, and 5.9%p, respectively, when compared to the baseline of the low-resolution 32-Ch LiDAR input. In future works, various dataset environments apart from autonomous driving will be analyzed.

Magneto-rheological gel (MRG) has been the subject of recent research due to its versatile applications. Especially, the magneto-induced electrical properties of MRGs under different levels of magnetic field enables them to be used as magneto-sensors. However, conventional MRG shows a low level of electrical conductivity, complicating its use in sensor applications. In this regard, in the present study, the carbon nanotube (CNT) and graphene oxide (GO) are added to fabricate new types of MRG. Herein, four different MRG samples were fabricated with reference to an amount of CNT and GO. The microstructural images of carbonyl iron powder (CIP)-based chain structures with CNT and GO were observed using SEM images. Then, their magneto-induced electrical impedances were investigated under four levels of magnetic field (i.e., 0, 50, 100, and 150 mT) and input frequencies (1, 2, 5, and 10 Hz). Based on the experimental results, three electrical models, including first-order series and parallel, and first- and half-order complex models, were proposed, and their accuracy was examined, showing the highest accuracy when first- and half-order complex models were used. The simulated results indicated that the incorporation of both CNT and GO can improve the magneto-induced electrical sensitivity; thus, it can be concluded that MRG with CNT and GO can be a possible method to be used in magneto-sensor applications.