INSTITUTING a predictive and preventative maintenance regime has long been an ambition of infrastructure owners and managers. It promises to save time and reduce costs, and the technology to introduce many of the processes and techniques required have seemingly been available for a long time.
Yet the reality on the ground is often very different. Data aggregation and collection, which is driving this new condition-based approach to maintenance, is not currently at a standard or level of penetration required to satisfy safety regulations. This leaves infrastructure managers generally relying on tried-and-tested but less efficient reactive and interval maintenance methods to keep their railways in good working order.
In Britain work is underway to change this situation. Network Rail (NR) is actively developing its “predict and prevent” maintenance strategy for its 30,000km network and related assets. And it is making progress.
According to NR’s professional head of maintenance Mr Tim Flower, who was speaking at London Business Conferences’ 3rd Track Maintenance and Asset Management Conference in Berlin on January 26, NR is now able to remotely monitor the condition of more than 42,000 fixed assets.
This includes 1000, or 95%, of power supplies, 12,000 (61%) points, 18,000 (26%) track circuits and 3000 (80%) point heating systems. It is also expecting to conclude the rollout of plain-line pattern recognition over approximately 24,000km of track and introduce automated inspection of surface cracks on 29,000 track-km within the next year. In addition, through the Offering Rail Better Information Systems (Orbis) programme, NR staff are able to use their mobile devices to access Fault Code Lookup (FCL) to improve fault diagnosis at the trackside or to capture data from specific assets and send this information to NR’s centralised Ellipse maintenance system for processing.
NR is spending approximately £100m on condition monitoring during its current five-year funding period, and this enhanced capability is already translating into improved maintenance performance. In 2016-17, NR reported 98 broken rails, a slight reduction over the 109 reported in 2015-16, but a 90% decline from the 952 reported in 1998-99. All of this was achieved in a period when traffic levels increased by 50%. In addition, since the start of the condition monitoring strategy in 2008-09, total incidents from non-track assets (excluding telecoms) have fallen steadily from 30,596 to reach 20,595 in 2015-16.
Nevertheless, Flower says that NR’s data processing capacity is still short of the standard required if it is to deploy predictive and preventative maintenance. “We understand our asset base, but we don’t necessarily understand which components are inside the box,” Flower says. “That’s the kind of information we want to drive.
“For a lot of our maintenance regimes, we utilise record cards in location cases, so we have a huge amount of information that could be driving a condition-based approach to our maintenance. Unfortunately, we don’t have the electronic systems yet to take that information in and provide the engineers with information that allows them to make better decisions. That’s a huge push over the next year.”
NR’s condition monitoring strategy is underpinned by a Failure Modes Effects and Critical Analysis (FMECA) approach to asset management. This process is widely used in aviation and charts the probability of failure modes against the severity of consequences allowing the direction of remedial efforts to where they are needed most.
However, with 70% of NR’s assets having some form of FMECA, and only 24% having FMECA loaded into NR’s fault management system (FMS), Flower says NR is pushing ahead with reviewing, updating and creating FMECA for its assets, with a goal of reaching 80% deployment.
A major step towards achieving this in the long-term was taken last month through the adoption of NR’s product acceptance procedure, which requires all products to provide evidence of meeting the new design for reliability standard. Evidence includes FCL data tables derived from FMECA and the provision for a suggested maintenance schedule that uses reliability-centred maintenance based on FMECA.
Additional standardisation for the predict and prevent strategy is being achieved through the adoption of ISO 13374 - condition monitoring and diagnostics of machine systems - for the technological rollout, and as the basis for NR’s strategy and its benefits projections.
Yet with remote condition monitoring still a reasonably new concept, Flower says integration with other systems remains at an early stage of development. For example, data retrieved from FCL and Orbis tend to be treated in isolation rather to create a single decision-making capability. And while data on individual assets such as overhead line may show compliance with individual regulations, it does not necessarily correspond with other assets which make up the complete railway system, which may also meet their relative compliance standards.
But progress is being made. Flower reports that 22 data collection systems are now integrated into a single asset data storage facility and 15 more are expected to be added by 2018, while NR is considering whether to migrate or shutdown another 30-40 systems.
Aggregating this centralised information to develop decision-support and reporting tools to aid maintenance decision making for each type of infrastructure is a major objective for NR up to the end of 2018. Beyond 2019, and NR’s next five-year control period, Flower says the focus will shift to higher-level data analytics. This will enable the process to move away from technology-specific processing, which can only detect the condition of assets, to offer a health assessment, or diagnose specific faults and failures. This will ultimately extend to prognostic assessment, which will predict failure and faults, and advise on specific maintenance actions.
“The goal for us is to optimise the whole life cost point for renewal,” Flower says. “As an asset gets older, the average annual cost diminishes, but we get to a point that the asset has degraded so much that it is going to fail at a level we can’t afford. At the moment, we haven’t been able to calculate what that level is across all of our assets to assess the timing of the intervention. The goal is for the frequency of intervention to be low. But in reality, we are very risk averse because we don’t have the data to optimise that decision. So, it’s all about closing that loop and bringing the data that we know we’ve got and turning it into information that is user friendly.”
Among the current projects which Flower says is facilitating the shift towards data analytics, and where NR is happy with the proof of concept, is a system for predicting the time to failure for points.
He says this initially used a multi-variant statistical processing analysis of the data retrieved on point condition. But with this proving too complicated for the end user, and NR not getting the benefits it expected, with the help of Birmingham University, it went back to basics to develop an expert system. This looks at what went wrong in the past, what the failure mode is, and what the trace looks like for signal processing to developing fault identification processes.
The system develops the existing points condition monitoring (PCM) system which triggers an alarm if swing time, average current and peak current exceeds a predetermined threshold, transmitting this signal to the central server via GPRS. The new system utilises algorithmic assessments of the trace against known signatures to offer engineers an extensive but more detailed view of the health of point machines. A main line pilot of the system is planned in mid-2017.
“Every time we get a new failure mode we just feed it in, and we have a level of confidence of 80% that it is failure mode A, or 20% that it is failure mode B,” Flower says. “So, it’s not saying it’s definitely right, but it gives technicians a lot of support in terms of understanding what they’re going to fix.”
The next stage of this project is to understand how quickly the asset is degrading and what is the time to failure in order to inform a controlled, rather than an emergency, maintenance procedure. Similar work is underway to reach this level of understanding for track circuit condition, with two pilots expected to start soon on two routes.
“This will make sure the algorithms we are developing are telling us what is happening, how accurate they are, and if they are completely fit for purpose, or if they come with a health warning,” Flower says. “We then need to go through the safety assessments to consider what it is they can replace or what it is they can supplement.”
Flower adds that utilising data analytics in this way will ultimately prevent NR from wasting millions of pounds on unnecessary maintenance work. But the infrastructure manager still faces a number of challenges before it can reach this stage.
For example, Flower says that its safety management processes “could be more slick” in order to facilitate this change. However, its IT capability, or lack of, is NR’s largest current bugbear. In particular, he says the process of transferring data from its sensors into the IT network is proving phenomenally difficult.
“For our Wonderwear system, we have three different suppliers involved; one to manage the wires, one to manage the access to the network, and one to manage the system,” Flower says. “It is very difficult to get anything done. We need to move to cloud-based solutions, and move to processing information outside of our estate rather than trying to process it all in house. We need to defer to the experts for it. And they are not necessarily sitting in NR.”
Placing its data in the cloud and making it readily available will potentially redefine the infrastructure manager’s interaction with suppliers. While there has been some reluctance to do this so far, it is essential to facilitate NR’s objective to offer “coopetition” in its dealing with the supply chain.
Mr Paul Barnes, senior project manager for asset management at NR’s western route, says this procurement approach promotes competition among competing parties which are using a common data set and is common in the aviation industry. He argues that keeping the data set open is critical to prevent a situation where NR is “held to ransom by the supply chain.” Specifically, he says NR must prevent intellectual property residing at the point where the algorithm is interpreted, which could limit entrants into the market place. Instead he says the emphasis should be on encouraging the supply chain to use this raw data intelligently.
Barnes added that it is also important for the infrastructure manager to avoid a scenario where algorithms are providing detailed intelligence of the condition of the system, but its staff no longer understand how the system actually works.
“My fear is that the wisdom and experience of the people on the ground who understand how our system works will be lost as everyone turns to this way of working,” he says. “Where here is the Turing test?”
Flower reiterated this desire to work with the supply chain to identify solutions currently unknown to NR to help it achieve its objectives.
For instance, he says there are significant opportunities to use train operating companies’ fleets to retrieve asset data. This is heightened by NR’s fixed data points already reaching their maximum level of deployment and the expense and restrictions of using measurement trains.
“We are moving towards it slowly with the track geometry capability on the IEP fleets for Great Western and Virgin East Coast,” Flower says. “But we need to think about what we want to use that data for. We will soon have 11 trains travelling on the Great Western Railway day in day out providing this information. Is it for compliance, or is it to tell us that we did a good job on the shift last night?”
Whatever the final decision, the rewards for the successful implementation of the strategy are significant. Independent estimates project that a complete rollout of predictive maintenance could reduce service failures by more than 70%, cut down-time following failures by 35-40%, increase workforce productivity by 20%, and, critically, reduce the time engineers spend on the trackside. It also promises to cut maintenance costs by 25-30%.
Flower says that NR is striving to achieve a 25-30% total expenditure saving from its predict and prevent strategy as well as a 35-40% reduction in downtime following failure and a 35-45% reduction in failure rate. And he believes that NR has the building blocks in place to achieve this. Among these is a “single guiding mind” throughout the organisation, with the chief executive and directors onboard with what the predict and prevent team are striving to achieve. The team’s organisational structure is also designed to give it the best possible chance of success.
Inevitably there is a great deal of consideration and work ahead of NR to achieve its objectives, which will probably take it well into the next decade. But as Flower is keen to point out, it is a journey worth taking.
“The opportunities are massive if we get this right,” he says.