PROTOTYPE DESIGN OF GANTRY CRANE WITH COMPUTER VISION BASED CONTROL SYSTEM

• Main Author: Raihan Miransyahputra, Muhammad
• Format: Final Project
• Language: Indonesia
• Subjects: Teknik (Rekayasa, enjinering dan kegiatan berkaitan)
• Online Access: https://digilib.itb.ac.id/gdl/view/61210
• Institution: Institut Teknologi Bandung

Along with the increasing rate of worldwide shipping, the automation trend at the port is being intensively developed to balance the growth of loading and unloading process at the port, one of them is development of gantry crane system. Until today, the operation of gantry crane at the port still depends on the human operators with different abilities. It affects the consistency of loading and unloading activity, but nowadays the development of gantry crane system which slowly minimizes human intervention has started. This final project developed a computer vison-based gantry crane control system and applied the result to a lab-scaled gantry crane prototype that has hoisting, moving, placing feature, and twist lock mechanism to mimicking the actual container gantry crane system, with the data communication method is carried out with Robot Operating System (ROS) software and microcontroller, so the proposed method can provide an alternative to existing development of gantry crane system. In this final project, the first step taken was gantry crane control system modelling using two dimensions approach. The gantry crane control system has two output variables, the first one is trolley position corresponding to the track and the second one is load’s sway angle corresponding to the trolley. The system is equipped with two controllers for load moving process, PID for trolley position and PD for the load’s sway angle. Both of the controller was tuned by using three optimization algorithm which are Particle Swarm Optimization (PSO), Stochastic Fractal Search (SFS), and Flower Pollination Algorithm (FPA). From the simulation results obtained that in terms of convergence speed, FPA and SFS have the best performance, but PSO has the superiority in terms of computation speed per iteration. Position and sway angle feedback become an important part in this system, therefore a SSDMobileNetV2 based container object detection is proposed which can provide bounding box and object midpoint that can be used as an input for the prototype system. The test results with the designed prototype have resulted in an object detection system as a setpoint input with 95% confidence interval of the error 0.21 ± 1.71 mm based on rotary encoder measurement and -3.30 ± 4.10 mm based on direct measurement on the object. The sway angle measurement obtained from digital image processing of the mark point at the hoisting arm to track the movement from the camera and converted to sway angle using geometrical projection formula. Based on final testing, FPA tuning gives the best simulation cost function error with a value of 0.00043 and PSO gives the best prototype cost function error with a value of 0.00357. From the whole test, obtained that the prototype was functioning properly using computer vision-based control system that had been designed, as shown by the prototype responses mimicking the simulation responses.