Last year, in my first ever article for Rail Engineer (issue 136, February 2016), I reported on the fourth annual Rolling Stock Fleet Maintenance Summit. It was therefore with some trepidation I undertook to do the same for the fifth event, held in December 2016. My worry was that I might have to send in a single word article – “ditto”.
I should not have worried! Some of last year’s speakers provided useful updates about their progress and there were several new speakers.
The title, “Rolling Stock Fleet Maintenance Summit focussed on improving fleet reliability, availability, maintainability and safety at reduced cost though leveraging data analytics and condition based maintenance” was not snappy or concise, but provided plenty of scope to allow a wide variety of issues to be presented. The venue was bigger and more people attended, so the organisers – London Business Conferences Group – clearly know their market.
There were nearly 200 delegates – suppliers of equipment and/or sensors, suppliers of analytical tools, system integrators, consultants and operators. This led to a variety of different styles of presentation, which kept the event lively.
By the way, the title of this piece refers to two uses of the acronym RCM. The most used form was ‘remote condition monitoring’, but many speakers emphasised that, to get the best from RCM, they would need the output of a ‘reliability centred maintenance’ exercise which I have termed RCM2. If you’re not familiar with either term, all will become clear.
One year on
Last year, my impression was that operators were struggling to convince management that providing RCM tools (on-train sensors, data collection, wireless transmission, back office servers and analysis) was a sensible thing to do in order to meet objectives of improved reliability and availability with reduced costs. The challenge was to convince the ‘powers that be’ that RCM is a wise investment in the absence of a guarantee of the benefits it could bring as the methodology was in its infancy. Last year’s presentations went some way to provide the evidence.
This year, RCM provision was almost taken as read and there was more emphasis on the workplace structures, processes and competence necessary to use it. Put another way, the industry is slower to exploit data gathering and analysis technology than it has been in deploying it. Speakers also highlighted the importance of cooperation between practical engineers, who understand the systems being monitored, and data scientists who understand how to seek out and manipulate vast data sets to identify relationships and conclusions.
The continental view
Héloïse Nonne, from SNCF, introduced the work carried out by her data science department using examples of pantograph monitoring, state of toilet fluid tanks, and lineside vegetation management.
SNCF has set up a central unit to develop the principles of big data and embed the practice of monitoring and machine learning, although Héloïse emphasised the importance of comparing what the data scientists suggest with the expert view before any decisions are made. She also recommended that only simple decision-tree algorithms be used, at least in the early days, so that asset experts can read them. Until confidence is built in machine learning, ‘black box’ algorithms should be avoided.
Highly accurate results are necessary if maintainers are to trust the machine learning output. It is different and more complicated than predicting what other products customers might like to buy whilst reviewing their on-line shopping baskets!
Philippe de Laharpe, also from SNCF, provided more information about the toilet trials. This issue was to ensure that tanks were filled or emptied on TGVs in order not to have toilets out of use.
The first trials on four trains had led the conclusion that the overall mean level of water in the tanks is 80 per cent. This knowledge had allowed SNCF to reduce the frequency of expensive replenishment in stations, and cut by 50 per cent the number of replenishments in the workshop. The system should allow management to identify low-level tanks (20 per cent) which can be flagged and filled before assigning the TGV on a commercial journey.
Philippe also highlighted work being done on remote monitoring of diesel fuel tanks, on battery condition, and on the French KVB speed monitoring system.
Wan-Jui Lee of Dutch Railways (NS) discussed experiments analysing compressor on and off times. Analysis of this data provided very clear evidence of leaks long before they became evident to maintenance staff. In the example, the train had been called to the depot after the anomaly was detected (coloured yellow in the illustration) but no fault was found. A month later the leak was identified and fixed.
Whilst this showed the success of the system in identifying the defect, it is a problem if the defect is not readily identifiable to a human (for example, a large number of possible leakage sources on a multiple unit). Efforts need to be made to help the human determine where to look or what to change.
The UK view
Justin Southcombe of Perpetuum, standing in for Mark Johnson of Southeastern Trains, provided an update on the deployment of vibration sensors on Southeastern’s fleet. Last year, it was reported that the company had equipped all of its Electrostar trains with axlebox vibration sensors with the aim of extending the life between wheel bearing overhauls.
This was only the start and with further analysis, engineers can now identify wheel defects and, using GPS and time/date data, have identified trends with track condition. Justin provided updates and illustrations of wheel and track defects identified by the system. He also reported that Southeastern has achieved the original objective of extending wheel bearing life by one year, and are now exploring a further extension.
For wheels, wheel-turning activities can now be planned more accurately. There has been a significant reduction in the backlog of work for the wheel lathe and average wheel condition has improved.
Stephen Foote and Chris Welford of London Underground updated the conference on their progress with RCM on the 1992-stock Central line trains. Having demonstrated that data collection via Wi-Fi on this middle-aged train works and can be used to provide useful information, LU has built a dashboard that integrates all data from the various sub-systems on all 85 trains to allow the users to identify issues and trends. This is useable by anyone used to drop down filters in systems such as ExCel. It can show, for example whether Automatic Train Operation faults are occurring randomly over the line (indicating train-borne faults) or clustered in one location (indicating an infrastructure fault).
They also illustrated a one-page Wheel Slide Protection dashboard integrating all the alerts from 85 trains’ data covering some 680 electronic boxes and over 2,700 tachometers. All this provides the maintenance planners with a clear picture of which trains need attention and the reports can be easily customised.
Like many others, Stephen and Chris emphasised the importance of involving and training the front-line staff (maintainers and their managers), and where necessary, reviewing the organisation structure to embed the right skills with the right authority into the maintenance process.
Abhinay Ramani, from First TransPennine, and Babakalli Alkali of Glasgow Caledonian University presented a case study on door maintenance of ScotRail (when operated by First Group) Class 158 trains. They carried out an RCM2 exercise, covering 100 components and 345 failure modes, to establish a base line against which they could make decisions.
The process identifies the function of the component, a description of the functional failure, the failure mode and its effect. Then, for each failure, it considers the safety, environmental, operational and even hidden consequences.
Information about frequency of use is also required. Were they all operating at more or less the same frequency? To gain this information, the team monitored the door interlock switches and the guard’s key switch operations by way of the signals input to the On-Train Monitor Recorder (OTMR) fitted to each train (wireless downloads available). Much to everyone’s surprise, it was discovered that the most-used doors were operating at up to 10 times the frequency of the least used. This was not expected and shows the value of obtaining real data.
The monitoring also picked up other defects, for example slow doors. The result was less frequent maintenance of the door system and better reliability, backed by a safety argument enabled by the RCM2 analysis.
Neil O’Connor from South West Trains updated his presentation from 2015. The performance and capacity issues affecting the Wessex route continued unabated with significant growth in demand year-on-year with little extra capacity to absorb it. Many of the initiatives to address this issue are coming to fruition; the Class 458 fleet expansion from 30, four-car units to 36 five-car units has been completed, the first of 30 five-car Class 707 units has been built and the conversion of the class 455 units from DC camshaft to AC propulsion is proceeding, despite delays along the way. As well as these developments with the trains, SWT has worked on a number of softer issues to improve performance such as:
- Shared Fleet/Operations functional goals;
- On-train equipment defect reporting matrix involving drivers, guards and platform staff;
- Disruption management plans;
- Incident management plans;
- NEXALA spectrum fleet control manager;
- “Cut and Run”.
By way of a pre-amble to his progress report on RCM, Neil revealed that SWT is also working with Network Rail on the plans for major remodelling at Waterloo in 2017 including a major blockade during which their franchise changes hands!
Neil described the provision of passenger facing Wi-Fi on Class 159 DMUs which had provided the opportunity to connect the OTMR to enable wireless downloads of the existing channels and some additional channels provided by SWT. He added that wireless downloads are also available from the Train Management Systems on Classes 444, 450 and 458 and, using WSP (wheel-slide protection) activity and GPS, these can identify locations prone to wheel slide. They can also access the trains’ CCTV.
The download of the OTMR and other RCM data provides more than fault information. It can be combined with data from train running systems to identify why performance might deviate from timetable (for example variable dwell times or restrictive signals). This can help improve the timetable by using the results from hundreds or thousands of runs to provide real, rather than modelled, performance information.
Several service trains have been fitted with equipment to provide unattended track geometry measurement. This allows early identification of track defects and the ability to check that rectification work has indeed fixed the issue. They also have several units that monitor the geometry of the conductor rail and monitor ride comfort.
Keeping wheels in good condition generates a significant volume of work. SWT has taken a different approach from Southeastern in the use of a small number of track geometry trains together with Wheelview and Discview at Wimbledon and Salisbury depots, plus wheel management software that receives and analyses the output of wheelset monitoring systems such as Wheelchex, RailBAM, and Gotcha.
Neil also described his project to upgrade the central IT system to integrate data and management processes currently spread across various IT systems and databases.
The suppliers’ view
Justin Southcombe of Perpetuum returned to provide an update on further applications of his company’s product – the robust self-powered vibration sensor that transmits data wirelessly. He illustrated one of the challenges experienced by many rail industry suppliers of new and novel equipment, namely that a customer wants proof that the equipment does what is claimed and refuses to accept the evidence from another railway.
As an example, Justin described an exercise carried out collaboratively between five companies to demonstrate the system. Based on the damage to bearings removed from service on another railway, two new bearing were ‘damaged’ using spark erosion techniques. These were fitted to test wheelsets, which were then run in service for 4000km and compared with a known good wheelset. The value of the system was clearly demonstrated when it identified a third defective bearing during the test. A wheel that was outside normal limits was also detected. With the co-operation of all parties, this work, including the safety case, was completed in three months.
Rob Spence of Danburykline drew comparisons between the aviation industry and the railway. He was critical of the railway practice of each operator inventing its own process for developing its maintenance regime, and he was also critical of some of what had been presented at the conference. He was clear that both wireless downloading and clever people were necessary, but they had to do their clever work before that next time the train comes to depot at which point the maintainer can go directly to the train with the right tools, equipment and parts to get the job done with minimum downtime.
Given Rob’s experience as a senior manager in aircraft maintenance, he was a great fan of the MSG-3 process (simplistically, a standardised form of the RCM2 philosophy as described in the ScotRail example) that is not only applied but is required to be applied to the development of the maintenance regime for most commercial aircraft in service today. This is a standardised process so that anyone in the aircraft industry would be familiar with it. As an example, it is also normal to have standardised fault codes across the industry.
Rob emphasised the importance of deciding what maintenance needs to be done before trying to determine how it should be done efficiently. He outlined the benefits to performance and cost that had been achieved in the rail industry by the application of MSG-3. This presentation was particularly relevant, given that some of the operators were complaining about the lack of standardisation and had felt they were almost reinventing the wheel from scratch.
And after all that?
After twenty-seven varied presentations, some of which are detailed above, what did I learn?
Data Scientists and Rolling Stock Engineers working together generally deliver better results than working individually,
Machine learning is, for the railway, in its infancy,
It is sensible to start by learning to do simple things with the data to deliver some results. It’s easy to be too clever, too early,
Data is an adjunct to learning about failure modes effects and criticality, not a substitute,
Possibly the most important point was that most speakers identified how much more they learned from their data sets than they had originally expected.