If you own a wind farm in the sea, you really want to know how well each turbine is performing. Clearly, if one turbine is producing less electricity than the norm, you can see this, but it’s not good enough. You also need to know why, so you can fix it. Moreover it’s also pretty important to know precisely why, as it’s a difficult journey just to get to the turbine let alone climbing up and down the mast – you wouldn’t want to find you’ve brought the wrong parts or tools!

What has this got to do with train maintenance, you might ask? It was just one of the examples of successful remote condition monitoring mentioned during a two day conference in December 2015.

Over 160 people gathered in London for the fourth Annual Fleet Maintenance Congress organised by London Business Conferences. The snappy subtitle for the congress was “Reducing Cost Through Integrating CBM and Data Analytics”.

Whilst the language may be challenging, the numbers attending suggested that there was a great deal of interest in this topic. Moreover, the title might sound as though it was for computer experts and statisticians but, in fact, it was for and delivered by real train maintenance managers grappling to deliver more for less in a competitive environment, whilst getting to grips with the latest tools to do so, just as has been happening over the history of railways.

Advance warning

It is worth a small aside to mention some of the principles that underpin this article. The basis of remote condition monitoring is that you gain knowledge of the performance of a piece of equipment by monitoring its performance. This might include fitting sensors to items of equipment or using measurements already made within the equipment. The sensors might be on the train or they might be trackside.

As an example, door systems often monitor door speeds so that any reduction in speed (increasing friction in the system, say) can be corrected by increasing the power of the motor. Changes over time in the current drawn by the motor will therefore indicate deterioration in the mechanical guides or some other fault. Data about the current consumed is sent back to base over a wireless communications link (perhaps 3G/4G) into a shore-based database which can be accessed by the maintainer. This is Remote Condition Monitoring (RCM).

Condition-Based Maintenance (CBM) is the process of turning RCM data into information. With CBM, analysis is often carried out by experts in both train maintenance and data processing and, sometimes, by combining data from a number of sources. In the simple doors example for basic RCM, if the motor current goes consistently above a defined threshold, the maintenance planner can call the train into the workshop for attention. Generally, this will be far less frequently than routinely checking the door system.

For the data analyst, the door currents of all the doors over days, weeks, months or years can be analysed which might deliver insights into door system, operator and customer behaviour. This is a very simplistic illustration, but it shows the basic principle.

Conference basics

The conference included a good-sized exhibition space occupied by providers of RCM and CBM tools and solutions. The conference itself was a mixture of railway managers describing the challenges they face that are leading them into RCM/CBM, some case studies, both successful and others that were more challenging, and finally some presentations from companies selling RCM/ CBM technology and expertise.

The scene was set really well by Tom Hopkins, Heathrow Express head of engineering, who reminded us of the basics:

  1. Introducing RCM and CBM is not just introducing new technology. Most technology programmes are really, at their heart, business change programmes and Tom emphasised that unless people, process and technology receive equal weight supportedby excellent leadership, there is a good chance the project will, at best, not deliver the expected benefits and, at worst, fail.
  2. Today’s new trains should come with systems that capture train and sub-system status and parameters with software to capture and analyse this information.
  3. On older fleets, monitoring systems can be retrofitted “at a price”. Some owners or operators have capitalised on projects such as accessing OTMR (On Train Monitoring Recorder) information, taking advantage of the opportunity provided by fitting customer Wi-Fi. Others have chosen to fit sensors to monitor unreliable equipment.
  4. The biggest challenge for all the systems, whether new or old, is analysing the mass of data that is produced and deciding the trigger levels. This cannot just be decided by the CBM vendor alone; it needs input from sub system designers, maintainers and operators using analysis techniques such as Reliability Centred Maintenance.
  5. Finally, the RCM and CBM systems should be at least linked to the asset management system.

Extra benefits

This theme of retrofitting sensors was picked up by Justin Southcombe of Perpetuum (UK) and by Ross Balcombe of South Eastern Trains (SET). Perpetuum makes self-powered wireless vibration sensors (the product of research by Southampton University) and SET has used these devices to monitor axle bearings on its fleet. The train operator wanted to be able confidently to extend the interval between wheelset overhauls, and the vibration sensors were able to provide early warning of any issues with the wheel bearings.

London_Underground_1992_Stock_at_Theydon_Bois_by_tompagenet [online]

Justin said that they had carried out a limited trial to validate the principle before rolling it out across 1000s of wheels. The analysis of the vibration sensors is able to identify the ‘signature’ of early wear of bearings. A great deal of ‘noise’ had to be filtered out of the signals in order to deliver the information SET required, and it was soon realised that this ‘noise’, which was actually vibration signals from wheels and rails, could be valuable. By filtering in different ways, wheel and track defects could be monitored too. As a result, wheel and rail defects are being advised both to South Eastern Trains and to Network Rail.

While track information had not been the initial objective there was no possibility of understating its significance now, and examples of how careful analysis of data can yield far more than expected were highlighted throughout the conference.

RCM meets Oyster

Gaining unexpected information was evident in the presentation by Steve Foot and Chris Welford of London Underground. They are working on LU’s “Predict and Prevent” programme which aims to produce more and more availability and reduced failure rates at a time of increased ridership – indeed the latest record of 4.8 million customers in a day was announced during the conference.

They acknowledged that LU’s newest trains have very sophisticated RCM systems, a comment echoed by many speakers about modern trains in general. However, they chose to highlight a case study of the Central line trains which were fitted with an early Train Control and Monitoring System from new in 1992/3. This system always had the ability to download data from the train to shore, but LU’s more recent work had seen installation of Wi-Fi to allow the data to be extracted more frequently, and the use of data scientists working with the engineers to bring additional insights into the data.

LU was fortunate to be able to access some of the Transport for London Oyster card data scientists who have experience in finding patterns in seemingly random Oyster use and they helped Steve’s team develop their own job specification for the Predict and Prevent team’s own data scientists. Steve’s team delivered a computer tool for maintainers to interrogate the system and had brought about a number of benefits including an increase in the number of orders for work, resulting in an improvement in reliability. However one particular success was the enthusiasm with which the new system was received by the staff.

Increased capacity through RCM

Relationships, reducing failures of the whole railway system and coping with increasing ridership were themes of Neil O’Connor’s presentation. Neil is fleet depot manager for South West Trains. The introduction to his presentation was illustrated by a seething mass of people waiting for trains under the famous clock at Waterloo Station.

Ridership on SWT has increased significantly over the years. Waterloo alone has seen passenger arrivals and departures increase by 50% in the last 15 years and it is forecast to increase by another 30% over the next 15. The solution is more trains, but Neil only had a certain sized depot to fit them into. His solution was to reduce the number of interventions on the older trains by, inter alia, fitting new traction equipment and using various RCM techniques to improve the maintainers’ knowledge of the fleets.

Understanding the data

In his opening remarks, Tom Hopkins stressed the importance of setting the right trigger levels in the analytics. This was reinforced by a comment from Steve Foote who mentioned that point machines all seem to have different current-time signatures even where the machine and the points are of the same type and configuration.

Perpetuum and SET had adopted one technique for learning about how to use data – by using their pilot installation. Tsuyoshi Ichigi from the Technology R&D Centre of Japan Railways East Group described the approach JR East had taken to understand how to turn the data on the performance of doors and the heating/air conditioning (HVAC) system of the prototype E235 train into information.

Examples of the doors and HVAC had been set up so that engineers and researchers could validate their understanding of the data coming off the train. For example, they were able to apply controlled amounts of contamination to HVAC filters in the laboratory in order to understand both the impact on performance and how this is relayed by the monitoring system. The laboratory work was carried out for a much larger range of abnormal conditions than would be practicable to perform on the prototype train. This presentation introduced the use of Principal Component Analysis, which is a technique used to emphasise variation and bring out strong patterns in a dataset. It is often used to make data easy to explore and visualise – and is far too complex to explain here.

European perspective

Later on, Bas Sprangers and Jan Luijben of the Amsterdam Public Transport operator GVB presented on the challenges for introducing CBM on a fleet that has been supplied by different suppliers, all of whom have provided different systems and some of which do not send all the appropriate data, such as mileage. Like others, they recognised the value of analytics and CBM; for them it will help them improve availability to help meet ever-increasing demand. They described a particular challenge they have related to EU procurement rules for public bodies, which makes it difficult to take the knowledge gained from a small-scale contract (perhaps a small supplier’s intellectual property), and apply that knowledge to a bigger contract which might need to be opened to competition.

Whilst talking about EU regulations, the issue of approval of changes to maintenance regimes was discussed. All operators will understand the importance of safety in all railway operations. In the EU at least, any change in the maintenance regime has to be justified and certified and it is no good saying that this or that maintenance interval can be extended just because “the computer says it is OK”.

Although the RCM and CBM computer tools bring much greater clarity and certainty to maintenance decisions, it is still maintenance engineers who have to justify what is or is not in the fleet maintenance regime and this has to go into a well-argued case with evidence for certification and safety regulators to consider.

Belgium Railways (SNCB), Italian Railways (Trenitalia), French Railways (SNCF) and Finnish Railways (VR) all presented their experience both of retrofitting existing trains and of learning to manage with the vast quantity of data that new trains deliver. Two particularly noteworthy points were made by VR and Trenitalia.

Firstly, from VR and to illustrate the diverse data sources there is the short hand, “the three Vs of data”:

» Variety – forms, structure, sources;

» Volume – How much of it in kilobytes, megabytes, gigabytes, terabytes;

» Velocity – streamed or occasional and does it all arrive in the right order.

Secondly, Trenitalia had specified a performance requirement for the RCM systems on its newest train fleet that:

» >80% of all maintenance work orders should come from the RCM system;

» <5% of these notifications should be “no fault found”;

» <5% repeat repairs.


ROSCO benefits

A slightly different perspective was presented by Olivier André of Porterbrook Leasing with the aim of showing how it can be to a leasing company’s commercial advantage to provide some or all of the technology to enable RCM and possibly to sell the service of providing the data or information from these systems to lessees (or others?). He highlighted that C4 Underframe and C6 (carbody) overhauls present the opportunity to add facilities such as an Ethernet backbone for IP connectivity, CCTV, Wi-Fi (passengers and RCM data) sensors, and track to train wireless links.

Olivier suggested the benefits would be:

To the train operating company (TOC):

  • Possible standard architecture/open system;
  • Future proofing an older train;
  • Freedom to choose his own back office (not being tied to train supplier’s solution).

To the leasing company:

  • Protecting the value of the asset;
  • Differentiation of Porterbrook as a lessor;
  • Ability to cross reference across fleets operating on different TOCs;
  • Possibly to reduce/eliminate heavy maintenance and add tasks to routine maintenance.

After two long days

At the end of a full two-day programme, delegates had a much better appreciation of the benefits, and pitfalls, of RCM and CBM. Probably the most discussions revolved around:

» New trains will generally come with RCM, but work is still necessary to determine the CBM requirements for the particular railway’s environment. There was criticism of some suppliers (no names quoted) who seek to monetise the data these systems produce and the lack of open data structures. There were no presentations from rolling stock suppliers and perhaps these issues might be addressed next time.

» CBM is in its infancy on rolling stock compared with RCM, which is seen as a mature technology. Some of the presenters reported on their struggle to make a business case for retrofitting RCM because they need the benefits of CBM to justify RCM and CBM benefits are, often, educated guesswork. Maturity of CBM will only come as operators and suppliers pool data and are able to review others’ results.

» Many asset managers see the opportunity, but don’t have the skills in-house to carry out the development work on CBM for their network. There are outside companies that have these skills, but, for the public sector at least, the challenges of contracting in a scalable form for these services are real.

Like all good conferences, despite the amount of information received, there were still questions unanswered. No doubt these will be taken up next year.

Written by Malcolm Dobell, former head of train systems engineering at London Underground.