ABSTRACT
At present, the effect of global warming becomes more severe due to carbon emissions. This negative effect attracts the interest of the international community to minimize it. The railway industry is an industry in which carbon emission cannot be neglected due to the sizes of projects, long operation and maintenance stages in the service life, including sources of energy that might not be clean in some areas. This study aims to minimize the negative effect of the railway industry on the maintenance aspect by using reinforcement learning to optimize maintenance activities. The maintenance in the railway industry is a complicated task and might not be optimized in terms of cost efficiency, serviceability, and environmental impact. The use of reinforcement learning can improve the overall efficiency of railway maintenance as it has been proven in many tasks in the railway industry and other industries. Data used to develop the machine learning model are based on field data collected during 2016-019. The length of the studies section is 30 km. Sources of data are from track geometry cars, maintenance reports, defect reports, and maintenance manuals of the sampled railway operator. The methodology used in the study is Proximal Policy Optimization (PPO). The results show that the use of reinforcement learning can reduce the carbon emission from railway maintenance activities by 48% which causes a significant amount of carbon emission while the railway defects are reduced by 68% which improved maintenance efficiency.
