Factoring in frost for reliable insights even in unreliable weather

Winter '24 Product Update


The unseen impact of winter

Today, Infrastructure Managers tackle visual winter problems, with solutions such as snowploughs, hot air blowers, steam jets, brushes and scrapers to clear snow and ice from the tracks. While taking care of the visual clean-up, frozen catenaries etc., which pose an immediate problem, they are missing the impact that cannot be seen – the impact of frost – which results in both higher false alerts within an array of monitoring systems and alerts that are not an immediate problem but are drawing on immediate resources. 

According to Infrastructure Managers such as Network Rail, “[t]housands of our people also work around the clock in all weathers, monitoring, maintaining and repairing the tracks so that we can run a safe and reliable service for passengers”. There is potentially more work to be done on track in winter and Infrastructure Managers are further stretched on resources, so it is crucial to know what is worth acting on and what is not. This is why it is important to know why something is a problem and where it’s a problem.

The limitation of thresholds in asset monitoring 

In the railway industry, it is very common to use defined, simple alerts based on specific thresholds. An overload of alerts can erode the trust of maintenance teams and put large-scale adoption of Predictive Maintenance (PdM) practices at risk as it becomes difficult to find valuable alerts amongst the already long list of detected events. While current practices for alerts are useful they have their limitations, especially in winter. 

Trackbed health is an important metric for ensuring optimal S&C uptime and overall performance, if compromised, it can inversely and critically impact all other S&C components. However, in winter when frost occurs, the trackbed behaves very differently as it becomes stiff in below-zero temperatures. This can mean that the data behaves abnormally and can violate predefined thresholds, which in turn creates false alerts. In winter, as much as 50% of alerts can be unreliable. This further erodes trust in digitalisation systems, such as PdM and can lead to wasting resources and time, and compromising on worker safety.

Our diagnostic approach to reliable insights

At KONUX, we do not simply provide data, but we take a diagnostic approach to understanding the health of an asset, such as trackbed. KONUX Switch, our S&C Predictive Maintenance System continuously captures data via IIoT devices, monitors the health of key switch components and analyses the patterns of asset conditions to anticipate failures. Sudden changes in the condition of an asset trigger a Smart Alert – 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. Smart Alerts, however, are not triggered by simple threshold exceedances or change points in isolation but are algorithmically corrected for external dynamic effects, such as frost. For instance, if a sudden change is detected when the ballast is frozen due to low temperatures, our algorithms determine if a frost period is the likely reason for the change in data. We know that these data variations do not suggest immediate or permanent issues with trackbed health, but rather are a consequence of something like the temporary stiffening of the trackbed due to freezing weather conditions. By omitting an alert, we are saving inspection teams valuable time in the field and reducing unnecessary safety risk. Understanding the context and reason behind this anomalous data behaviour is essential to us being able to send only relevant alerts that can be acted on with trust. This is how we are able to ensure over 90% accuracy of our prediction.

The example below shows how our system has identified and flagged a frost period when the trackbed was frozen and we saw sudden changes in our data readings. Additionally, it shows that due to the temporary trackbed behaviour changes, prediction is suspended until the frost period has passed to not provide misleading readings or alerts. Frozen trackbed can have a temporary impact on our readings but importantly, is not related to the long-term health of the asset.

Example 1: Frost period detection in our UI

As illustrated in the example below, impact force increased significantly but did not trigger a Smart Alert as our algorithms determined a frost period was the reason for the change in data and omitted sending an alert. Once the frost period passed, the impact force readings decreased and prediction was reinstated.

Example 2: Diagnostic approach to understanding asset health

Essentially, our solution not only identifies and comprehends frost-related effects but also presents this information in a user-friendly manner. This enables infrastructure managers to grasp the nuances of trackbed anomalies during frost periods, empowering them to make informed data-backed decisions and prioritise immediate required actions effectively.


Informed, data-backed decision-making

We understand our users and their needs, our systems do not simply display data but offer insights required for informed decision-making, offering clarity, context and importantly, true actionable insights. By eliminating the noise of false alerts and unreliable predictions, especially in winter conditions, infrastructure managers can focus on tackling pressing winter-related challenges that impact operations and safety. We are able to navigate the complexities of winter conditions and the resulting anomalies in trackbed behaviour, our approach significantly contributes to a more efficient use of time, resources and importantly reduces unnecessary safety risk. Through all of this, we build trust with greater confidence in digitalisation and PdM practices.

Share this post