Railroad Industry A New Era of Predictive Maintenance
Currently, much maintenance is either time-based or usage-based. By contrast, predictive maintenance can result in significant cost savings that can justify transformation of entire maintenance systems.
Typical rail fleets are operated for 30 to 50 years. During this time, roughly half of the overall fleet cost is maintenance. The goal of most rail operators is to increase profits; this can be done by boosting both fleet availability and reliability. Currently, much maintenance is either time-based or usage-based. By contrast, “predictive maintenance” uses multivariable data inputs and analysis in conjunction with machine learning systems, which raise alarms before critical components can fail. Predictive maintenance can result in significant cost savings that can justify transformation of entire maintenance systems.
In Europe, it’s estimated that between 1.6 and 6.8 billion Euros could be saved via implementation of predictive maintenance, an overall gain of at least 10 percent of maintenance expenditures for the rail industry. This can be split out to a savings of between .3 and 1.3 billion Euros for OEMs, between .8 and 3 billion Euros for operators, and between .5 and 2.5 billion Euros for third parties.
But predictive maintenance is not usually easy to set up. Not only does it require surveying conditions of equipment, but it also mandates monitoring factors such as weather and power flows that affect those conditions. All of these data sources need to be tapped into, managed, and analyzed so advanced predictive computer models can be built.
In many of today’s rail maintenance systems, the entity in charge of maintenance (known in most of the EMEA region as ECM 3) that schedules maintenance jobs and manages train fleets is going to become automated. This means maintenance tasks will automatically be defined and sent to an ECM 4 group, which executes and delivers all jobs. (ECM 1 is typically responsible for a fleet’s safe operation, while ECM 2 sets maintenance standards.)
In some cases—particularly with legacy equipment lacking sensor technology—regulatory requirements necessitate manual maintenance commissioning. There are also cases with newer fleets where predictive maintenance can’t cover all types of equipment; therefore, manual maintenance commissioning will remain an important practice for ECM 3 for most organizations going forward.
As time goes on, it’s possible that ECM 3 may merge with either ECM 2 or ECM 4 groups, which could potentially lead to large productivity gains and employee count decreases. But these cases will largely depend on an entire maintenance ecosystem’s ability to automate.
As analytics tools evolve, processing and decision rules will need to be adapted and integrated into maintenance operations. Low-frequency, irregular review of regulatory compliance will need to switch to a systematic predictive-maintenance approach. Internal processes within groups ECM 2 or 3 will also need to be adapted continuously based on fleet maturity.
Within the rail industry, rolling stock OEMs own a significant amount of valuable data—track data, operations data, and sensor data—from train fleets. This means they have a leg up on other players at the table, who may only have analytics or IT knowledge, or a very narrow view of the full maintenance value chain. Therefore, analytics startups, IT platform providers, and system suppliers will typically play only a supporting role for rolling stock OEMs or rail operators (or both).
The two categories of rail operators that need to adopt predictive maintenance relatively quickly are urban/regional passenger rail operators and cargo rail operators. For long-distance passenger rail operators, competitive pressures are lower, so there’s less urgency to do so.
How Can Predictive Maintenance be Implemented?
In order to begin preparing to implement predictive maintenance processes, it’s recommended that operators:
- 1. Determine their market segment and competitive context—Is the market segment urban/regional passenger rail, cargo rail, or long-distance passenger rail?
- 2. Quantify their fleet characteristics—Is the fleet in question saturated with legacy equipment? Is there significant heterogeneity?
- 3. Quantify their number of assets—More assets usually mean an operator will tend to want to perform maintenance in-house, versus outsourcing it.
- 4. Quantify their number of environmental (urban vs. rural, for example) operating contexts—A more heterogeneous set of operating contexts makes implementing predictive maintenance easier.
- 5. Measure their market share of rail maintenance—Those operators that are already significantly involved in rail maintenance will have strong incentives to stick to it.
- 6. Consider the infrastructure conditions of the country being operated in—In countries such as the UK, Germany, or France, there are typically many legacy assets; thus, it’s easy for incumbent maintenance players to continue to dominate the market, whereas, in regions like Southeast Asia or the Middle East, rail infrastructure is still being built up.
Operators should then define a target maintenance state and strategize a roadmap to get there. Partnerships may be necessary to achieve targets.
The matter of which party owns which type of data is important for the success of a new maintenance ecosystem. Both component/sensor data and operations/maintenance data are needed for adapting to predictive maintenance. Where data ownership is distributed amongst several entities, all of the players need to be in conversation and formulate agreements that take into account the target maintenance state. Careful consideration should be given to answering:
- 1. What kind of data needs to be generated to implement predictive maintenance? How can this data be acquired in real-time or near-real-time?
- 2. Who owns the different parts of the data, and what data rights should the different parties have?
- 3. Which data is shared between the parties, and how is it shared?
- 4. Which IT solution should be used as a platform? Should it be exclusive to the parties, or could it be seen as potentially beneficial to the sector as a whole?
- 5. Should any of the data be shared with third parties (for example, system suppliers or analytics startups) that are involved, and if so, how should it be shared?
- 6. How can asset manipulation and/or data breaches be avoided, and who’s responsible for cybersecurity?
In some cases, neither rail operators nor rolling stock OEMs may be willing to hand over or sell the data concerned, but at the very least, sharing access to it may provide a fuller picture of predictive maintenance’s true potential.
At some point, very large amounts of data will be generated and extracted from fleets in real-time, and at that point, there could be a danger of data being manipulated by hackers or used as a gateway to a train’s operating system. Thus, integrating cybersecurity will become crucial for predictive maintenance efforts. Regulatory bodies may need to grant permission in some of these cases.
A recent survey in the U.S., Germany, China, and Japan showed that roughly 30 % of logistics industry firms had engaged in predictive maintenance pilots. It was found that pure analytics did not deliver desired results, because of:
1. Poor quality of data—Much data was not rich enough to predict the failure of specific subcomponents of complex systems.
2. Unreliable correlations—Prediction models revealed apparent correlations between failure codes and sensor data that were later determined to be incorrect. Interpretation of results and adaptation of models were needed, which likely could only be made successful with the close cooperation of analytics and engineering experts.
3. Not enough lead time—Findings of prediction models could often not be incorporated into maintenance processes because the times between failure alerts and component failures were insufficient.
Therefore, the results of initial proofs of concepts were that analytics and prediction models were not sufficient, precise, or comprehensive enough to support predictive maintenance schemes.
Given the data available today, the descriptive analyses of failure data in conjunction with rail engineering expertise show more promise than purely analytical approaches.
Organizational Collaboration is Key
Rolling stock OEMs or rail operators have to find ways to couple rail engineering expert knowledge and analytic power because only by working together can analytics scientists and rail experts develop powerful models. To get to such a point, these groups can either build up in-house analytics functions—consisting of analytics and rail experts working together—or buy their analytics as a service where providers work on-site with rail experts.
One particular challenge is that rail engineering know-how is generally fragmented amongst the separate divisions of a rail operator. In order to effect a real change, all rail knowledge needs to be collocated in a specific physical location with an analytics team. Relevant rail experts from any siloed functions of fleet management, procurement, and maintenance planning need to be pulled together in this location.
An example of this is Siemens’ Mobility Data Services Centers in Munich-Allach, Atlanta, and Moscow. In these locations, Siemens tries to achieve optimized train operation by intelligently utilizing rail system data for predictive maintenance. But Siemens doesn’t just rely on virtual train data. In Allach, where Siemens assembles and maintains locomotives, the firm combines real and virtual worlds to make sure its train experts don’t lose sight of what really matters—physical trains.
Beyond the technical, companies need to consider organizational issues related to cross-company or cross-departmental collaborations, culture clashes between rail engineering experts and data analysts, and/or change management and transformation of maintenance processes throughout the organization. In some cases, even defining a business case in the first place can be arduous, but that doesn’t mean it’s not worthwhile to do so.