The UK’s London Underground transportation system is managed by Transport for London (TfL) and has been around since 1863, facilitates 1.4 billion trips annually, and is now integrating data science into its maintenance structure in the form of predictive maintenance.
The Role of Predictive Maintenance
Predictive maintenance is a solution to the ageing system’s ongoing issues with disruption to the daily train schedule, the upkeep of which accounts for almost 60 percent of TfL’s budget. A lot of things can break down unexpectedly: anything from engines, to stations, to tracks, to elevators, and one glitch, even minor, can disrupt the entire transportation system for hours, or even days.
The goal of predictive maintenance is to collect and analyse data in order to mitigate risk, but breakdowns are not solely attributable to a 150 year old infrastructure. Weather patterns, which can be extreme both in temperature and humidity in London, also factor in. Additionally, utilisation, departure location, and existing maintenance rate have all been identified as important sources of data that factor into failure rates.
As they study failure risks and patterns, TfL’s team of data scientists and reliability analysts have not only identified these four risk factors, as they study the current state and they are developing a system that examines data-sets from each subset, TfL assets, failures, maintenance, service operation, and external issues, including weather.
Factors studied in this predictive maintenance model include the impact of each subset on failure rate, the severity of impact as related to its frequency of occurrence, and the cost of failure each subset represents. In this way, they hope to develop an overview of all the interacting factors and compare each piece in order to determine what and how to mitigate.
The predictive maintenance model is not without its challenges. Missing data, information silos, limited data ranges due to infrequency of failure, and an infrastructure that is constantly being updated all serve as potential barriers to gathering and compiling accurate data in a timely manner.
Also, because of the complex nature of the system, data scientists have had to carefully coordinate with multiple departments in order to ensure they are receiving a complete picture of the system in its entirety.
The Results of Predictive Maintenance
The goals of predictive maintenance for TfL are to allow staff to make decisions on maintenance and upkeep priorities based on data analytics in order to mitigate failures as well as upkeep costs.
Too rigorous of a maintenance schedule will result in money wasted due to over-maintaining as well as unneeded disruptions to train schedules.
Instead, they hope to use historical failure and maintenance data as a means to predict the probability of issues and identify causes of these issues.
This system could eventually lead to the establishment of a fixed maintenance schedule, but it does not completely eliminate preventable failures. Scientist’s true goal is to get to a point where they are able to predict a failure right before it takes place.
This is the goal of TfL’s existing data science projects. A number of algorithms are being developed and modified toward this end within the predictive maintenance project, and as the team integrates this algorithm into daily operations, currently with 75 percent accuracy, TfL is saving money through predictive maintenance, both for themselves as well as for visitors and commuters.
Read more of our latest articles here.