Achieving lightweight real-time object detection necessitates balancing model compression with detection accuracy, a difficulty exacerbated by low redundancy and uneven contributions from convolutional layers. As an alternative to traditional methods, we propose Rigorous Gradation Pruning (RGP), which uses a desensitized first-order Taylor approximation to assess filter importance, enabling precise pruning of redundant kernels. This approach includes the iterative reassessment of layer significance to protect essential layers, ensuring effective detection performance. We applied RGP to YOLOv8 detectors and tested it on GTSDB, Seaships, and COCO datasets. On GTSDB, RGP achieved 80% compression of YOLOv8n with only a 0.11% drop in mAP0.5, while increasing frames per second (FPS) by 43.84%. For YOLOv8x, RGP achieved 90% compression, a 1.26% mAP0.5:0.95 increase, and a 112.66% FPS boost. Significant compression was also achieved on Seaships and COCO datasets, demonstrating RGP’s robustness across diverse object detection tasks and its potential for advancing efficient, high-speed detection models.
This article presents an adaptive logarithm-based sliding mode control (ALSMC) for an active suspension system to enhance ride comfort while maintaining the road holding and suspension space limits. The main objective of the control design is to devise a controller that does not require precise and accurate model parameters while ensuring resilience to model uncertainty and external disturbance. To achieve this objective, a natural logarithm sliding mode control (lnSMC) approach is adopted. Unlike the conventional sliding mode control that utilizes a linear sliding surface, the lnSMC is designed based on a natural logarithmic function, which creates the boundary layer of the controlled states. Thus, the selection of the controller parameter will be more practical as its value is dependent on the desired amplitude of the controlled state. This prevents trial and error when determining the controller parameter. In this study, the system’s dynamics are represented by a nonlinear quarter-car suspension model (e.g., nonlinear springs and piecewise dampers) with unknown model parameters. An adaptive law is incorporated to estimate all unknown model parameters, which is subsequently used for updating the control law. A simulation study is carried out to illustrate the effectiveness of the proposed control law under various types of disturbances, including sinusoidal road profile, bump road profile, and random road profile disturbance.
25 April 2024 in IEEE Transactions on Intelligent Vehicles
A neuroadaptive controller based on natural logarithm sliding mode control (lnSMC) is proposed for an active suspension system with unknown nonlinear dynamics and uncertain model parameters. The proposed control scheme ensures that the controlled states are constrained within the desired bound of heave and pitch motions, thereby eliminating the need for trial and error in determining the lnSMC controller parameters. Moreover, the unknown nonlinear system dynamics are approximated by a radial basis function neural network (RBFNN), which updates its weights continuously in real-time. Considering the high degree of parameter uncertainties in suspension systems, an adaptive law based on a gradient algorithm with a projection operator is incorporated to estimate the unknown parameters (e.g., vehicle mass and mass moment of inertia). Simulation studies on a half-car active suspension model are carried out to evaluate the performance and robustness of the proposed controller under various road disturbances, including bumps and random road profiles. For comparative purposes, neuroadaptive controllers based on classical sliding mode and terminal sliding mode are designed as benchmark controllers. The simulation results indicated that the proposed controller achieves a better suspension performance indicators compared to the benchmark controllers.
28 November 2023 in Journal of Vibration and Control
This paper presents the control synthesis of a nonlinear sliding mode control, known as the “ natural logarithm sliding mode control” (lnSMC), for a general n-th order system. The lnSMC offers a simple and straightforward design approach because it only requires tuning two design parameters. The first parameter has a physical meaning corresponding to the bounds of the state variables, which can be easily determined without a trial-and-error process. The second parameter is related to the controller’s feedback gains, which ensure that the system dynamics converged to the sliding surface. Lyapunov stability theory is used to analyze the closed-loop system and finite-time convergence stability. As a case study, the natural logarithm sliding mode controller is designed to suppress the vibration in a simple spring-mass system and an active suspension system. The simulation study shows that the proposed controllers have significant vibration suppression performance. Furthermore, the simulation study indicates that the sliding mode controller designed by using the natural logarithm sliding manifold outperforms the standard linear sliding manifold counterpart. In addition, due to the nature of the nonlinear switching function that creates boundary states, lnSMC has a huge potential to be applied in a wider area of vibration control.
Traffic flow prediction is critical for Intelligent Transportation Systems to alleviate congestion and optimize traffic management. The existing basic Encoder-Decoder Transformer model for multi-step prediction requires high computational complexity, making them less efficient for real-time applications. Therefore, this paper presents an Encoder-Only Transformer that eliminates the decoder component while evaluating the model’s ability to capture long-term spatial-temporal dependencies. This simplification was found to reduce computational overhead while maintaining a reasonable predictive accuracy comparable to the basic Encoder-Decoder Transformer and Long Short-Term Memory (LSTM) models. Experimental evaluations on two real-world datasets, Minnesota and California, for hourly prediction of 6, 12, and 24-hour horizon tasks, demonstrate the proposed model’s effectiveness. The proposed Encoder-Only Transformer outperforms LSTM networks across all horizons task for the Minnesota dataset, achieving up to 17.33% improvement in Mean Absolute Error. The proposed model excels in 24-hour horizon task for the California dataset but underperforms for shorter horizons compared to the LSTM and basic Encoder-Decoder Transformer models. These findings highlight the Encoder-Only Transformer’s application for multi-step traffic flow prediction while emphasizing dataset-specific variations in model performance.
Robustness of the controller becomes major important factor in active engine mounting system to ensure the vibration of the engine can be attenuated in entire range of operating speed. As robustness is the main advantage of sliding mode control, it has been proposed for vibration control applications. However, most of the proposed sliding mode control are using linear sliding surface. In addition, there is no schematic method to determine the optimum sliding function. In this paper, natural logarithm-based sliding function is introduced. By its nature, natural logarithm function is nonlinear function. Since it consists of one parameter only, which is related to maximum allowable vibration level, it is more easily to be determined by the designer. Moreover, the advantage of fast on-line training using Extended Minimum Resource Allocating Network (EMRAN) algorithm, made it possible to design feedback control law without having to obtain the precise system model which is normally becomes major difficulties of sliding mode approach.