RAILWAY AGE, SEPTEMBER 2019 ISSUE – Using Big Data, algorithms and IoT tools to manage locomotive assets: Freight volumes on North American railroads have been showing reductions on monthly year-on-year comparisons throughout 2019. The AAR reported that in the first half the year, cumulative volume was down 3.1% from the same point last year and intermodal units were down 3.5% from last year. While this economic situation continues to play out, the railroads are in the midst of a revolution in the availability of massive amounts of data being produced and captured. Expanding sets of information flow into organizations daily from GPS systems, signaling, Positive Train Control (PTC), Internet of Things (IoT) sources and locomotive on-board event recorders.
These huge volumes of data cannot be processed effectively with most planning or operational software that exist in the rail industry today. So-called Big Data, combined with the appropriate tools, can be used to uncover new insights, which will lead to better decisions and strategic business outcomes. This collection and use of data is now supported by advanced IoT solutions, data warehousing and processing tools, which will allow the railroad to analyze and act on the data.
Data for data’s sake can be a drag on an organization. If info is collected and stored without governance, it only adds costs and confusion. But when used to manage assets, operating and capital costs can be reduced and powerful levers are made available for management to make a difference in real time.
Big Data is being managed by many analytical tools, and they don’t have to be expensive. Hadoop is a widely used open source technology that was developed to manage Big Data and discover patterns and hidden relationships in data. A newer technology, Apache Spark, supports analytics on streaming real-time data rather than the big one-off style jobs that Hadoop supports.
Other new and significant technologies have become available to manage and analyze data and the costs related to this have dropped dramatically in recent years. A good example is in the cost of building and operating software, and maintaining databases in the cloud. Today, most large businesses have made a commitment to moving IT assets to providers such as Amazon’s AWS, Microsoft’s Azure and Google Cloud. This reduces costs, expands internal availability and provides redundancy, in a secure environment.
Also important are the software tools to build apps and manage data. The availability of high-level software development languages such as Python make it easier and cheaper to deploy data interfaces, so that data can be integrated and analyzed in all parts of an organization. It has also become more cost effective for capital intensive businesses such as railroads, to have specialized algorithms developed for analyzing data and optimizing assets—from the network, to rolling stock and crew. Additionally, the newest advances in artificial intelligence (AI) and machine learning are starting to make an impact in asset management.
A great example of where the new technologies are now impacting railroads is in the management of locomotives, both from the planning and allocation, to the real-time dispatching in a large complex network.
A freight railroad’s locomotive fleet is one of its main sources of capital costs and operating expenses. Planning the allocation of locomotives to trains is a complex task. Current locomotive planning methods often rely on “top down” methods of developing locomotive allocation plans by adapting old plans over and over again. Over time, this approach can introduce significant inefficiencies leading to oversized fleets and excessive fuel consumption. New technologies allow railroads to deploy optimization tools to build allocation plans using a “bottom up” approach, perform scenario analysis, such as locomotive procurement plans, and to minimize the ongoing locomotive requirements for a train schedule.
Integrated data provides the necessary framework for developing algorithmic solutions for apportioning locomotives in an operating plan and for dispatching locomotives in real time. Based on work that my company, Biarri Rail, has done with freight railroads in Australia, North America and the U.K., we believe there are four steps to using data and systems to improve locomotive operations.
Visualize: Most railroads allocate their locomotives using a manual process, often involving Excel or “home grown” software with a basic user interface. The centralized Train Management System (frequently a mainframe-based system) acts as the “database of record” for locomotive allocations but provides limited functionality outside of that. Scenario testing and what-if analysis are hard to perform on the central system.
Locomotive planners must also refer to multiple additional systems, which are often separate, to collect all of the data required for locomotive planning (e.g. current location, fuel level, maintenance status, existing planned future allocations). Any notes or calculations required by the locomotive manager, while pulling all this information together, must be done external to these systems.
Railroads now have access to technology and tools to pull all of this data together and visualize it within a single integrated user interface. I believe that this ability, the visualization of the data, along with providing a high-quality user experience (UX), can be as important as the algorithms and analytic tools and indeed are a valuable and necessary first step in the journey to more sophisticated application of algorithmic decision support.
Allocate: Locomotive allocation is at the core of the asset planning and management processes. Once all the relevant data has been collected, the locomotive operations team can allocate locomotive units to train services, and determine how the units should transfer between those services. With the appropriate software, these allocations are updated in real time and the locomotive operations team can then respond to the changing operational environment.
Currently, most rail operations teams only have Excel models to make these decisions. Given new methods for data accumulation and analysis and the tools to build the necessary user interfaces (UI) and optimization engines, the users can make the difficult decisions that consider complex interactions between train services in the past, present and future. In the case where sophisticated decision support tools are not available, locomotive allocation decisions tend to fall back to guidelines or rules-of-thumb when faced with a variable real-time environment.
Execute: The quality of the locomotive plan is irrelevant if the plan isn’t able to be implemented. The execution of the allocation plan requires clear and concise communication to yard managers and locomotive operators. Currently, most communication is managed through a very manual and time-consuming process that takes away from the time that the locomotive operations team could be using to decide locomotive allocations more efficiently.
An improved tool for execution must:
- Allow information to be distributed quickly, to keep everyone up-to-date.
- Generate and send reports automatically, to remove manual effort.
- Allow for feedback if the selected locomotive allocation has problems or issues (e.g. if a locomotive is positioned inconveniently within a yard, the yard manager can be given a tool to suggest an alternative).
Review and Report: A review of the plan execution is needed for future planning. An important part of running efficient operations is a focus on continuous improvement, to ensure that future planning and operations benefit from experience. This requires that reports and summary data of operations are readily available.
An important component of the review phase is integration of planning and forecasting with actual events and inventory. The data collected during operations can be used to further train an optimization algorithm for making automated suggestions in the Allocation part of the process, to improve the quality of the suggestions. New AI techniques in machine learning can also be applied, to identify patterns that are non-intuitive to a human, which will support better decision-making.
This step has a dual function of both collecting a larger data set to improve suggestions in general, and ensuring that the optimization algorithm stays up to date as network conditions and cargo volumes change.
I have had first-hand opportunities to see the results of the confluence of Big Data and the explosion of decision support tools to interpret the data, and the use of algorithms to improve asset utilization. In one such case, Biarri Rail was able to help a Class I railroad by integrating data from many sources (locomotive event recorder, fuel data, train planning system, GIS and other sources) to successfully analyze the causes for differing fuel usage patterns in regions of the networks.
This success is not limited to Biarri Rail, as there are more than a few railroads and software firms exploring the use of recently available data to provide decision-making tools to railroads. Freight volumes may soon start to rise again, but we are just seeing the beginning of Big Data, AI, deep learning and other advanced techniques in supporting more efficient railroading.