Product Update, 30.03.2023

Predictive Maintenance (PdM), through continuous monitoring and sophisticated machine learning, enables users to monitor the health of an asset, such as railway switches and crossings (S&C), and the prediction of potential failures. It has become one of the most effective maintenance approaches for railway infrastructure companies and, according to Global Railway Review (2018), has been shown to increase asset reliability and reduce overall capital expenditure (CAPEX) and operational expenditure (OPEX).

While most solutions rely on simple alerts or static thresholds, which can be misleading and result in many false alarms, we at KONUX, are approaching this differently. This Spring ’23 Product Update shows you how we are applying smart algorithms as part of our PdM strategy to detect anomalies around S&C that prove valuable and actionable for our customers.

The problem with simple alerts

In the railway industry, simple alerts based on static thresholds or sudden changes in the data can limit the usefulness of condition monitoring and PdM solutions. This is because an overload of alerts can erode the trust of maintenance teams, and put large-scale adoption of PdM practices at risk. It also makes it hard to find valuable alerts amongst the already long list of detected events.

But why is integrating reliable alerts in a PdM product so challenging? It inherently requires a deep understanding of the monitored asset as a holistic physical system and knowledge of how different components impact each other. While digital twin projects aim to build an understanding of complex assets such as S&C, operationalising the knowledge can be extremely challenging. This is especially true when such solutions do not adequately take into consideration the ways of working of track engineers and maintainers. Additionally, simple alerts do not take environmental factors into account that can cause sudden changes in the monitored data. All of the above can contribute to misinterpretation of the data and false alarms.

How are we approaching it differently?

At KONUX, we combine machine learning and Industrial Internet of Things (IIoT) to increase operational efficiency for railway infrastructure maintenance and operations. In data alone, we have recorded 80 million train traces from all seasons and all traffic types in more than 10 countries. We have received and incorporated feedback from asset owners, maintenance engineers, regional managers, and other stakeholders. In addition, our holistic approach also allows us to provide insights based on our knowledge of how different components of an asset affect each other.

This experience and industry expertise has enabled us to deliver alerts on events that are meaningful and relevant to the context of the assets as well as the users. As sudden changes are usually caused by abrupt defects or ineffective maintenance activities that impact asset health, it is crucial for asset owners to be notified about them in a timely manner.

Introducing Smart Alerts

Smart Alerts are a set of notifications that inform users when important events are detected in the system, while factoring in relevant information such as S&C configuration, weather conditions, train speed, and traffic composition. The first Smart Alert available to our users relates to sudden changes in vertical acceleration. Vertical acceleration is an indicator of the force on the rail when a train runs over it and can be useful in detecting anomalies around switch geometry. For example, sudden changes in vertical acceleration at the frog (crossing) area of an asset could indicate a breakout, dent, or unsuccessful welding activity.

However, Smart Alerts on vertical acceleration are not triggered by simple threshold exceedances nor change-points in isolation, but are corrected for external dynamic effects, so that asset owners are notified of events that are meaningful and relevant to them. In other words, we notify users when the vertical acceleration of an asset suddenly changes, but not when, for example, it’s due to weather or train speed changes, because they are usually sources of false alarms.

Below is an example of an omitted alert due to weather. If a sudden change is detected when the ballast is frozen due to low temperatures, our algorithms determine whether a frost period is likely the reason for a sharp increase in acceleration and omit sending an alert, saving inspection teams valuable time in the field.

To provide our users with a comprehensive overview of the most important events, we have created the Alerts view in our user interface. Knowing that there is a high likelihood of finding a failure on the identified assets, our users can confidently take action and prevent severe damage.

With Smart Alerts, we are able to deliver meaningful insights to our customers by ensuring that they only receive alerts that matter and that can be acted upon with trust. This enables our customers to avoid unnecessary inspections, reduce downtime costs, and eventually increase operational capacity.

. . .

Would like to learn more about our solutions? Get in touch here.