Estimating Time to Failures

Progressive Deterioration of Components

All components have a limited lifespan. Factory tested components are often supplied with data on expected time to failure and mean time between failures. Such certification is based around testing components under repeated operational stress conditions in a test environment. In some cases, the durability of the components may be based on their material properties (e.g. rail, sleepers etc.) and in others on the electrical and mechanical properties (e.g. sensors, pcbs, processors, storage etc.) In the real-world, components are subjected to outdoor conditions that are often not easy to simulate in factory test environments. For example, changes in outdoor temperatures, effect of rain, humidity, vibrations, unexpected loads and several other factors can cause damage to the component and lead to its premature failure. Predictive maintenance is built around a central premise, i.e. “early prediction of future failures can be used for timely maintenance that increases overall time to failure of that component and is lower in cost than repairs when the component fails completely”. For this, regular monitoring of component conditions is crucial to generate enough statistical data on their deterioration that can be used to predict their future maintenance needs.

Estimation – what does it involve?

Estimating time to failure is a mathematical modelling process that is optimised to the needs of a specific rail network. Whether we are estimating the time to failure for a rolling stock component or a track component, the process is the same which includes:

  • Quantifying material properties of the component as an input into the process that uses data on how these materials degrade over time.
  • Quantifying railway usage data, e.g. load, vibrations, etc. subjected to the component
  • Quantifying environmental decay factors and their effect on component life
  • Choosing a decay model, linear, quadratic or polynomial

Once this data is available, we can use two different approaches to estimating time to failure include:

  1. A probabilistic modelling approach which generates a mathematical model using the above factors
  2. A neural network based predictor using the above as inputs and generating a time to failure estimate

In both cases, substantial data on past failures is needed from the railway to use the experience in such a design and verify whether the system is capable of predicting even historic events to demonstrate that it can also predict future events.