Prof ANDREW STARR reports on progress with AUTONOM, a project collaboratively run between Cranfield University and Network Rail that is researching an autonomous maintenance system combining asset health with planning, scheduling and an assessment of the costs
Taking a planned maintenance approach on rail networks – keeping to the schedule rather than evidence of what’s happening to assets in reality – can lead to unnecessary work and service disruption. Approximately £850 million is spent each year on maintaining the British rail system, which needs to be optimised to increase reliability, safety and capacity. At the same time, unplanned maintenance due to failures that haven’t been anticipated also adds further disruption and cost.
The sensor technology to access real-time data from large and dispersed areas of assets has been available for a number of years. As has the big data analysis for making sense of the streams of information. What’s been missing has been the two working together and providing network operators with a reliable picture of what maintenance really needs doing, when, and how that works in terms of maximising efficiency within limits of available time and resources. AUTONOM (Integrated through-life support for high-value systems), a research project between Network Rail and Cranfield University, is providing the proof of what can be done with an autonomous maintenance system that joins up asset health with planning, scheduling and an assessment of the costs involved.
This kind of intelligent railway maintenance has significant advantages. The enhanced railway operation means reduced delays and an improved customer experience. There’s less need for maintenance staff to spend time on the tracks, so lower levels of risk and fewer opportunities for human error. More efficiency in operations means both cost-savings and more energy efficiency. The approach puts maintenance in line with the Future Railway and Railway Technical Strategy vision of a highly automated, safer railway.
The precise knowledge of the real-time location of trains carrying condition monitoring sensors is of paramount importance. The positional accuracy target of the UK future rail is within 2m. But a finer resolution is required for locating faults such as damage or missing parts. GPS systems on their own aren’t robust enough to provide reliable detail, and can provide poor data in tunnels and built-up areas.
The team has investigated integrated systems for location and maintenance systems, including new train camera based system and ‘location resolution algorithms’ that draw on data from different sources and cross-check. There’s a related problem with the mix of both mobile and statics assets, all of which need monitoring, and we’ve been developing approaches for combining output from different types of sensors.
The AUTONOM system is based around a user dashboard, with feeds from the train sensor, a driver’s view, specific network location and data on any faults. Faults are given a ranking from 1-5 (5 representing a critical fault, 4 a warning alert and 1-3 a healthy asset). Data is fed into a business process optimisation system. The cost equation
The sheer number of needed interventions for maintenance means that some form of prioritisation is essential on the rail networks. Determining the cost effectiveness of condition monitoring for any industry is a complex problem. The cost of an intervention, compared to the benefits arising, can be estimated – using what is known as a ‘parametric’ cost estimation. Many industries do not require an exact calculation but rather an indication of the break-even point and probability that the break-even point has been passed. This can provide important and timely information to support or automate decision-making in a timely and responsive fashion.
Using the combined information from data mining, schedule and cost information, asset managers can be supported in making their decisions on maintenance, able to make savings on budgets without any question of increasing risk. In general terms, the cost model used in the Network Rail case study looks at the maintenance fault type and data from the condition-monitoring to give an indication of whether the maintenance work is urgent or can be safely catalogued for later inspection.
Estimated maintenance activity costs (crew, materials, rate in terms of timing) plus costs from the denial of service (based on expected time needed for repairs and when these will be carried out, at peak time or off-peak). If the work can be safely delayed it allows for an optimisation process to look for the most cost-effective solution based on a prediction of maintenance cost.
Costs taken into account include:
- Cost of condition monitoring equipment (sensors, utilities to power the sensors, any monitoring of the sensor equipment)
- Analysis of delays in the UK rail industry and the fines structure that is used between operators and Network Rail
- Analysis of the effect of planned versus unplanned maintenance • Different fault types (ie. rail cracking fault compared to a points failure)
- Scheduling related issues (seasonal variations in cost, time-of-day variation in cost).
Government literature advises that where possible benefit and cost is assessed in purely financial terms – but there are currently no available guidelines on how to do this for a condition monitoring system. The potential and specific cost-savings from the approach are being assessed as part of the research. By the end of the project we will have built up a portfolio of techniques for examining condition monitoring projects (current and future), considering such topics as cost/benefit analysis, uncertainty analysis and whole life cycle costs. Uncertainty around the actual cost of a denial of service value – dependent on the criticality of the route and amount of redundancy – is a major challenge for assessing new systems. The detail is difficult – but the broad picture is clear: there are going to be huge economic benefits available to Network Rail in terms of reductions of the maintenance expenditure and improved safety and reliability. As the future rail strategy is to increase the number of trains and passengers on the UK rail network, faults will be more likely to impact multiple trains. Even minor faults will start to cause significant delays.
Work on the AUTONOM approach is ongoing. We need to make sure the risk estimation is in tune with railway terminology and standards; and check the assumed prioritisation of risks against expert opinion. There is also a need to develop the system for unsupervised use on service trains, meaning the potential for much greater frequency of measurements.