<![CDATA[PT. PAL Indonesia is a company that operates in the field of national-scale ship production, which has the potential for fire hazards during the ship fabrication process. Therefore, a fire protection system must implemented. This research began with observation to determine the suitability of active and passive fire protection systems in the workplace based on several standards such as SNI 03-3985-2000, NFPA 13, Permenaker no. 04/1980, Permen PU no. 26/PRT/M/2008, and SNI 03-1745-2000. Observational data collection used a checklist with a cross-sectional research design. This production site has the potential for danger, causing fires in large areas, including Irian Dock and Sumatera Dock. Active fire protection has several types such as alarms, detectors, sprinklers, fire extinguishers, and hydrants. In passive fire protection, the assessment is based on the building structure. Field observations showed that active protection systems such as alarms were in the good category, detectors in the good category, sprinklers in the good category, light fire extinguishers (APAR) in the good category, hydrants in the good category, and passive fire protection systems in the good category. To support active and passive fire protection systems, this research proposes a fire spot recognition system based on YOLOv5s by utilizing CCTV facilities that are installed in the PT PAL work area. Dataset collection was carried out using image samples for each class, four classes were used in this research, including RMO class (type of work that causes a combination of flash points and sparks such as grinding and welding work), spark class, fire spot class and finally the fire class. The research used 1971 training data, 515 validation data, and 262 testing data. The best results were obtained with an IoU threshold value of 0.5 which had an mAP value during testing for all classes of 0.919. The accuracy produced through the confusion matrix is 0.755 or 75.5% with object detection testing on running videos showing a fairly high and stable accuracy value]]>
Copyrights © 2025
