‘Significant reduction of point failures possible’

Author: David Vermeij

Placed on: 06 September 2016

Tags: point failures, application, digitisation, railways, availability

DSC_4987[1]

David Vermeij

Strukton Rail

Door gebruik te maken van big data technieken en kunstmatige intelligentie wordt nu de stap gezet van diagnose naar prognose, het voorspellen van toekomstige storingen.

Digitisation is rapidly changing the world. The world of railway maintenance is keeping pace with this development, with the aim of reducing the number of malfunctions and increasing the availability of railways. The basis is set by continuous monitoring of the status of the railways, or in other words, diagnosis. Big data techniques and artificial intelligence are applied to take the step from diagnosis to prognosis, the prediction of future malfunctions. Together, the maintenance engineers of Strukton Rail and the big data analysts of Anchormen have faced the challenge of developing a prognosis tool, the first application of which should be operational by the end of this year. Results show that mechanical failures of points are predictable and that, as a consequence, more specific maintenance is possible. Strukton Rail believes that intelligent maintenance allows the railway sector to reduce the number of technical malfunctions by 50%. The new prognosis tool is in line with Strukton’s development towards becoming an international service and technology provider.

Increasingly high demands are made for the availability of railways. Failures tend to spread across the timetables due to the growing density of rail traffic. Time available to do maintenance work is decreasing. The Dutch railway manager ProRail has introduced performance contracts in the deregulated market in the Netherlands, stimulating maintenance service providers to keep a sharp eye on the price and quality of their services. A service provider that outperforms the contracted norm for malfunctions receives a bonus. A penalty is imposed for every extra minute of failure delay in case the service provider performs below standard.

“It is therefore important to detect and solve potential causes of failures at the earliest possible stage, before they actually lead to a malfunction’”

David Vermeij, Manager R&D at Strukton Rail, the company that maintains 35% of the Dutch railways and is also active on an international basis, with branches in Sweden, Denmark, Belgium, Italy, Australia and the US.

Reduction of point failures

Railway maintenance has been substantially improved in recent years. The number of failures is decreasing through the introduction of new parts, measurement and monitoring systems and integrated maintenance. Nevertheless, points are still one of the main causes of train delays. On average, a point fails once a year. However, it is often critically located points that cause malfunctions more frequently and lead to delays. Strukton Rail and Anchormen show that the performance of these points can be improved.

But how? Simply expanding maintenance activities is certainly not the solution for the problem of faltering points. After all, time for maintenance is scarce and expensive. Premature replacement of expensive parts is simply a waste. As such, a high price is paid for reduction of failures. Strukton Rail believes that deploying big data and artificial intelligence for predictive maintenance is the solution and a solution that lead to greater availability of tracks at lower costs.

More efficient maintenance

Strukton Rail uses the Preventive Maintenance and Breakdown Diagnosis System (POSS®) for its maintenance activities. The POSS® system was developed in-house and continuously monitors the status of points. About half of the 3,000 points under the management of Strukton Rail in the Netherlands, and globally as many as 10,000 points, are equipped with POSS® sensors that record data on the energy consumption of point motors during the movement of the switch blades. Irregularities in energy consumption can be a sign of an upcoming failure. POSS® has already led to a substantial reduction in the number of point failures world-wide.

The combination of POSS® with big data and artificial intelligence will further improve the results. The system already automatically issues a warning if a point is consuming too much energy. However, this often happens just before the failure occurs, with the result that there is insufficient time to plan maintenance work in a regular manner.

Anchormen and Strukton Rail have joined forces to develop an application that monitors and analyses the energy consumption data in real time, on the basis of which point failures can be predicted. Based on patterns in historical data, a computer model can detect aberrant behaviour before it is actually observable. By learning the behaviour that accompanies a particular type of malfunction, it can identify this behaviour at an early stage and issue a warning. Specific maintenance actions can then be taken without obstructing rail traffic, in order to prevent the actual occurrence of the malfunction. Test results show that a substantial number of failures can already be predicted two weeks in advance with a high degree of reliability.

“ProRail’s ambition is to minimise avoidable failures. Our new application can make a significant contribution to that”

Vermeij

Prognosis tool

Strukton Rail will expand POSS® with a prognosis tool to provide a warning on upcoming failures. The application is not only suitable for the Dutch railways, but can also be used in other countries.

“A first application will be operational before the end of the year”

Vermeij

Powerful combination

In 2015, Strukton Rail and Anchormen successfully completed a pilot project in order to research a model that can predict point failures. Both parties have now built a test environment. An Artificial Intelligence (A.I.) Team of Anchormen develops the specific features and algorithms. Strukton Rail supplies the data and the know-how on the operation of the points. 

“The combination of the two companies is powerful because they each contribute their own expertise. The Anchormen data scientists have a background in A.I. and programming, while the Strukton Rail maintenance engineers are experts in the operational domain”

Jeffrey van der Eijk, partner at Anchormen

 

Anchormen has always had a preference for combining the two domains and aims for partnerships for this. 

“It is possible to train computers in recognizing when maintenance is needed simply with self-learning algorithms. But such systems will always perform better when domain experts are involved. In this case, we involve people who know how and why points can fail and know which data you need to predict specific malfunctions”

Van der Eijk

 

“A self-learning system needs many consistent failure data before it can make reliable predictions. Until then, it will regularly give incorrect warnings, which is far too costly. Furthermore, track workers and train operation managers will get reluctant to act when they continue to be called out for no reason. Above all, a prediction has to be reliable and give the right information on the cause, so that a maintenance team can be sent out with proper instructions. You need both data and domain knowledge for that”

Vermeij adds