Long Nguyen Khac

In this study, we propose applying the Backstepping control method to the Delta robot, incorporating Fuzzy Logic to handle fuzzy input values and provide solutions based on fuzzy sets. This approach aims to enable the robot to approximate the desired motion trajectory within a short timeframe and maintain system stability in the face of disturbances that hinder movement between joint torques. The system’s stability is ensured and successfully demonstrated based on Lyapunov stability theory. The performance of the controller is evaluated through numerical simulation results using Matlab & Simulink tools on the figure-eight trajectory. Comparing the Backstepping controller combined with Fuzzy Logic to traditional Backstepping, the results demonstrate the superior performance of the Backstepping controller combined with Fuzzy Logic. It can effectively control the motion of the Delta robot along the desired trajectory with small errors, and short response times, providing a more flexible and adaptive control approach that can handle uncertain inputs, especially when the system is affected by unknown external disturbances.

The lack of awareness can damage workers’ safety and even their lives. Indeed, an escalating number of occupational injuries and fatalities on real construction sites underline the pressing need for better comprehension of personal protective equipment (PPE) usage. To address this concern, we employ a Yolov5s deep learning model to detect PPE items. This research endeavor involves the creation of a novel dataset, sourced from security cameras on real construction sites in Vietnam. The dataset encompasses five classes: person, vest, gloves, boots, and helmet. Our data collection criteria are stringent, encompassing various aspects like object angles, overlapping degrees, and distance ranges. To enhance model accuracy and identify the optimal architecture, we compare the Yolov5s model across six different activation functions: FReLu, LeakyReLu, Mish, HardSwish, SiLU, and AconC. Among these, SiLU activation coupled with Yolov5s yields remarkable results, demonstrating an outstanding mAP@0.5 of 83.9%. This achievement significantly outperforms other models, thereby showcasing the efficacy of SiLU activation in the PPE detection task using Yolov5. This study underscores the potential of leveraging various activation functions to elevate the performance of deep learning models in real-world scenarios.