How to optimize the rail fleet composition using GPS-data?
Data science can really help to solve complex supply chain management challenges. In our Data2Move Research Stories, you can find out how our students tackle these challenges.
This time we feature Bart Pierey’s Master thesis research at SABIC in Sittard. SABIC is a Saudi manufacturing company, active in petrochemicals, chemicals, industrial polymers, fertilizers and metals.
Where it started – challenge
With his research, Pierey wanted to optimize the rail fleet composition at SABIC based on, among other things, GPS data. This, because the rail yard close to one of the production sites of SABIC has limited capacity. Several companies share part of this yard and available spaces are assigned based on a ‘first come, first served’ principle. Consequential, the parking yard could be fully occupied when a new train arrives. If this is the case, the arriving train is rejected at the gate which leads to operational problems. On the other hand, the fleet cannot be decreased too much, because you need a high availability of rail cars to prevent production scale downs.
Data validation and preparation
During his research, Pierey discovered that the quality of the GPS data, gathered from the fleet management system, was not optimal. He executed a data preparation and validation project which led to more accurate data. As Pierey states: “The initial data was not good enough but after the preparation and validation phase it was considered to be sufficient.”
In his Master thesis, Pierey explains what steps he took to tackle the rail freight car fleet problems SABIC runs into. To get an understanding of the issue at hand and the characteristics of the system, he interviewed several stakeholders. “Planners and business both had interests, so I tried to solve the puzzle for SABIC, based mainly on GPS-data,” Pierey explains. He presented the final results to several business managers within the company and he developed a discrete event simulation model with stochastic holding and travel times. This model has not only been used to improve the general understanding of fleet behavior, but also to find an optimal fleet size which minimizes the utilization of parking space in the shared parking area.
Near-optimal fleet size
Besides the determination of the optimal fleet size and composition, Pierey also took the impact of the parameters holding and travel time into account. Pierey: “The holding time, the time rail cars stay at the customer’s production facility, has a major influence on the optimal fleet size. Travel time only has limited influence. That is why, I advise to emphasize on decreasing the holding times at customers.” Pierey concluded that SABIC’s current fleet size is near-optimal but their fleet composition should be adjusted.
Shared parking yard planning
To the users of the rail yard, Pierey recommended sharing the multi-use parking space forecasts. “At present, the forecasts on the usage of space in the multi-use parking area are not shared between the various site users because these are confidential. If they are shared, the arrival and departure planning could be adjusted, resulting in fewer problems in this multi-use area. Therefore, I recommend open conversation and sharing these forecasts.”
The train and rail car planning could also benefit from a modal shift, using other transport modes than just trains. As Pierey explains: “Another solution to decrease the safety stock at a company’s yard is to replace the train with another transport mode to fill the peaks in transport demand.”
Lessons learned from Pierey’s study:
Simulation can be of great value in understanding a system’s behavior. In addition, a simulation model can be used for testing several scenario’s and policy adjustments.
Make sure that the data you enter in a system is correct. Data cleaning and validation actions are very important, as garbage in results in garbage out.
To make sure that data quality is guaranteed, it is key to properly maintain data collecting systems. Appointing a system owner will help, as this person will be responsible for the system and its functioning within the organization.