Without Falkonry

A solution provider to the rail industry produces an automated condition monitoring application for rail switches to improve maintenance efficiency and to reduce operations interruptions. Their application centrally tracks the operations of many geographically distributed rail switches and helps operators manage maintenance activities. The application collects telemetry data from remote switches that includes switch events, ambient weather conditions, and motor current traces during the switch events.

A key function of the application is to classify the state of a switch based on an engineering analysis of motor current trace data. This state would guide users as to when a maintenance intervention would be required. Their challenge is that the effort to build models is quite high. Each model requires detailed analysis by skilled engineers of trace data for many events.

There are many different switch models and manufacturers, and in some cases a particular switch may require its own model due to the uniqueness of the installation. Once created, model accuracy can drift due to environmental or installation changes.

There is a strong desire to be able to build models much more rapidly and cost effectively through the use of automated machine learning techniques.

Applying Falkonry Condition Recognition

The application vendor already has a central repository for managing the collected telemetry data. The vendor’s team uses the Falkonry C# Agent DevKit to connect their data store to a Falkonry server deployed in the cloud. Their integration includes the ability for their application to share a set of past switch events, associated current traces, associated temperature measurements, and previously assessed switch conditions.

This integration allows the vendor to use Falkonry to rapidly train a model for an individual switch, evaluate its accuracy, and put it into live operational use.

During live operation, inspection data from maintenance activities are continually fed back into the application. The vendor uses Falkonry to regularly update the condition models based on new switch and inspection data. The Falkonry engine provides easy to use model management features that allow exploratory creation of new models during live operation with a previous model, and the ability to “hot swap” new models when desired.

Falkonry’s ability to rapidly create and manage large numbers of models at low cost through automation allows the vendor’s application to scale much more effectively, and opens it up to a broader range of devices.

Falkonry AI builds a model that classifies switch condition. Though Falkonry supports the use of single shared predictive model across a very similar set of devices (like a set switches of same make and model), it also makes a “one predictive model per device, that is frequently updated” approach feasible – something that is required in the case of many industrial device settings.


Keywords: Condition Monitoring, Monitoring Application, Intermittent Operation, Transportation, Predictive Maintenance, Rail

Cool, I am interested in learning more…