Intelligence series: Predictive Analytics for Optimised Rail Operations
Latest update: September 2019
Since 2018, there have been developments in the uses of predictive analytics in the healthcare, aerospace, and oil and gas industries. For example, the Cambridge University Hospitals NHS Foundation Trust (CUH) has started to use predictive models to help make quick decisions, for example using predictive analytics to identify if a patient has sepsis. The model incorporates a support system that assists clinicians when making decisions about appropriate investigations and antibiotics to prescribe. In 2017, 55% of sepsis patients arriving in the emergency department were prescribed the correct medication within 90 minutes and in 2018, this number increased to 100%, demonstrating the value of this using this model.
Research into different components of predictive analytics such as deep learning has resulted in a wider range of possible implementations for the rail industry. For example, Internet of Things (IoT)-based predictive analytics are used to ensure rail and rolling stock are in the correct condition for efficient use.
What is predictive analytics?
Predictive analytics is a form of advanced analytics that uses big data, statistics, modelling and machine learning techniques to identify the likelihood of future outcomes based on historical data. There are two types of predictive models that are used – classification and regression. Classification models make a prediction from data inputs and present a binary result or a possible set of finite categorical outcomes. Whereas regression models are not restricted to a finite set of outcomes and are presented as a continuous quantitative value.
The above diagram illustrates how rail utilizes Internet of Things (IoT) based predictive analytics. BNSF Railway (a freight company based in North America) deployed a number of detectors and vision cameras to capture data to identify defects in both freight cars and railway track. This data is communicated to the cloud for analysis and run through algorithms using machine learning to reveal unhealthy data which could be implicative of future breakages. This solution helps to reduce train delays caused by faulty components.
What industries use predictive analytics?
Within the insurance industry, predictive analytics is used to account for risk exposure and the cost needed to cover the risk. For example, health insurance providers analyse data from past medical claims and records from labs and pharmacies to predict the chances of member illness, default and bankruptcy.
Boeing Analytx provides aircraft health management, maintenance and crew planning optimisation, and engine fleet planning, to companies within the aviation industry. Aircraft health management services can use predictive analytics to allow proactive maintenance of aircraft systems which reduces costs and malfunctions. A Scheme Boeing provided AirBridgeCargo (a Russian cargo airline) was to deploy a Fuel Dashboard service across its fleet. These services has helped to reduce fuel consumption with total savings averaging at over 4%.
In the healthcare industry, Healthcare Catalyst provides analytic services to hospitals. The company collaborates with hospital staff to gather patient data and uses machine learning algorithms to predict the likelihood of a patient developing a chronic disease and risk of not showing up to scheduled appointments.
In the oil and gas industry, software has been used to help maintenance engineers to gather historic data embedded in sensors in the refineries. For example, ARC Resources (a Canadian oil and gas producer), has managed to reduce operational costs by approximately $30k a year per well.
How will predictive analytics impact the rail industry?
The rail industry is already using predictive analytics for some operations. These techniques facilitate predictive maintenance and asset management. They are combined with machine learning to analyse large volumes of multi-variable data whilst establishing a comprehensive profile for each asset. This delivers detailed information of the asset’s performance at varying operating rates. The use of real-time dashboards, automated diagnostics and root-cause analysis provides alerts of potential issues before a breakdown, thus reducing downtime and accident risks, and optimising maintenance and asset life management.
Moreover, predictive analytics can be implemented for safety monitoring and risk analysis to obtain the real-time status of the overall system, as well as providing tactical and planning recommendations. This can improve safety and reduce risk throughout the rail network. Train delay times can be predicted throughout the network using predictive analytics in conjunction with machine learning, therefore improving route-planning and train operations.
What should the rail industry do?
Investing in the training and recruitment of data scientists would support creating predictive models that are suitable, thus reducing bias and preventing any unwanted outcomes.
A decrease in fuel consumption may be achieved by using predictive analytics. Analysis of idle times and when these can be reduced, will not just save fuel and cost, but also decrease its impact on the environment.
Anomaly and failure agents could be deployed to learn the patterns of normal operating behaviour. Any deviations from normal conditions could help spot miniscule pattern changes that would usually result in failures. This could help move from a ‘run to failure’ operating mode to a preventative operation mode. The subsequent change from unplanned to planned maintenance could help to save cost and increase the longevity of rolling stock.
What is the current state of R&D
Predictive analytics have been researched for use in the maintenance of assets such as manufacturing machines, vehicles and offshore wind farms. For example, proof-of-concept predictive analytics models have demonstrated capability to monitor bogie performance and being able to recommend actions prior to failure.
The Swiss Federal Railway and CSEM (Swiss research and development centre) are researching deep learning, a subset of machine learning. Possible advances for this capability include the detection and classification of faults in railway tracks, minimising the number of onsite inspections and reducing the time experts spend on false positives.
What uncertainties remain?
Historical datasets may show a correlation between two variables, but the model won’t adapt if the variables have recently changed due to external factors. Moreover, predictive modelling may require large datasets. Older equipment currently in service may not be capable of collecting and transmitting data with the required volume, speed and accuracy to make the most of advanced analytical systems. Furthermore, the implementation of this equipment could require a large investment, so a careful cost-benefit case should be made and considered before any major investment is made.