THE DESIGN OF A SLIDING MODE CONTROLLER WITH OPTIMIZATION USING THE SGD AND FPA METHODS FOR LIDAR-BASED NONLINEAR GANTRYÂ CRANEÂ SYSTEM

• Main Author: M P Napitupulu, Cristian
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/86911
• Institution: Institut Teknologi Bandung

The increasing loading and unloading activities at Indonesian ports continue to grow rapidly, in line with the rising demand and population. Data from the Badan Pusat Statistik (BPS) in 2024 shows that the volume of cargo handling at Indonesia's five major ports has increased more than sixfold over the past three decades, with Tanjung Priok Port handling an average of 3,184.12 tons of goods per hour. To address these challenges, more efficient and reliable cargo handling technology is required. A key component in this process is the gantry crane, which is used to move container boxes. However, camera-based control systems and linear controllers like PID, as seen in other studies, still have limitations, especially in low-light conditions and when applied to nonlinear systems. This research proposes the development of a LiDAR-based nonlinear gantry crane control system using a Sliding Mode Controller (SMC) with Stochastic Gradient Descent (SGD) and Flower Pollination Algorithm (FPA) methods for optimizing the nonlinear gantry crane model parameters and SMC. The LiDAR-based system is chosen for its independence from external lighting conditions and its ability to accurately measure key variables such as rope length and sway angle. Meanwhile, SMC is selected for its reliability in handling nonlinear and under-actuated systems. LiDAR and YOLO (You Only Look Once) algorithm were implemented on a gantry crane prototype to measure rope length and sway angle. The measurement results showed average Mean Absolute Error (MAE), Mean Squared Error (MSE), and Root Mean Squared Error (RMSE) values for rope length of 0.00135 mm, 2.55 mm², and 1.52 mm, respectively. For sway angle, the average MAE was 0.453 degrees, MSE was 0.262 degrees², and RMSE was 0.508 degrees. For estimating the nonlinear gantry crane model parameters using SGD and FPA methods, 18 datasets were collected from the gantry crane prototype. The results showed that FPA provided better parameters with an average cost function of 0.0895 compared to 0.1227 for SGD. In simulations of 10 control scenarios, the FPA-optimized SMC parameters also outperformed, with an average cost function of 1.3427 compared to 2.7002 for SGD. Furthermore, controlling the gantry crane prototype using SMC with FPA- optimized parameters proved to be reliable and effective. This was demonstrated by stable performance in both bright and dark conditions, with an average cost function of 6.2776 in bright conditions and 5.8249 in dark conditions. As a comparison, the performance of SMC was also compared with controllers designed using Physics-Informed Neural Networks (PINN) and PID strategies from the references. The test results showed that SMC had the best performance with an average cost function of 9.0140, better than PINN (10.8572) and PID (10.7205). Keywords: gantry crane, nonlinear, LiDAR, Sliding Mode Controller, Stochastic Gradient Descent, Flower Pollination Algorithm, You Only Look Once