In a world of growing connectivity, it is critical for fleet managers to stay one step ahead. Fatal vehicle collisions are one of the leading causes of death in the U.S., according to the Centres for Disaster Control. While this may not come as much of a surprise considering these types of fatalities frequently makes headlines, what should be more alarming is the fact that many of these collisions are preventable.
With the nature of fleet business often revolving around its employees being on the road, fleets are significantly impacted by these statistics. Automotive fleet reports most company drives average 20,000 miles per year, with more fleets experiencing an increase in preventable accidents. The primary cause of this uptick in preventable fleet accidents points to driver distraction and violations that now contributes to the 25-30% of all fleet-related accidents, reports Automotive Fleet.
It is with this prime concern that our client, a company whose mission is the development of value added services in terms of security, monitoring and telemetry, approached us. Predictive analytics is ideal for risky driving analysis, for example. The challenge with identifying high-risk drivers is the most transparent records such as driving records, behaviour violations, and accident reports, and may not always indicate a driver with the highest potential risk. Armed with information on driving behaviours through telematics data, companies can put drivers in safety training programs tailored to their risky driving behaviours before a collision occurs. Information about a driver’s behaviour can also be used to determine how likely an individual is to be involved in an accident as well as the costs associated with that risk. The project also aims to do behaviour analysis, accident analysis, predictions based on the past data which will help the customer to take preventive actions.
At Pearl, we turn vehicles in use into a stream of data that can be used to enhance product performance, improve workforce productivity, understand traffic flows and routing, enable better risk management and more. Our data scientists had a tough task at hand owing to the data lakes/ warehouse trying to combine inconsistent data from disparate sources, thus encountering errors.
The toughest of the challenges though was handling huge data. A fleet of thousands generates a huge amount of data for fleet managers to review. The larger the fleet, the more information that’s generated. Consider that GPS devices are generally set to update their position every two minutes. That’s 30 updates every hour. 720 every day. If you have a fleet of 50,000 vehicles (the size of some of the fleets using fleet management software), then that’s around 36 million packets of information being sent in to fleet managers for their review every day. Our data engineers provided tools to cope by filtering all the information that was not important to fleet managers, and only showing them the data that matters. One example of this is the real-time alert functionality. Of course monitoring thousands of vehicles could result in a lot of alerts, so we provided a KPI dashboard that allowed them to easily monitor how the fleet is performing overall, with specific areas of concern color-coded for easy identification.
The data was collected from external GPS systems like Mzone and Vialon Fleet management system. We poll these systems in a frequent interval and collect data from GPS device which return the data in JASON format. Alert generation on real time ,Collecting data from multiple vehicles at same time and processing the same was critical .These systems may go out of network range and can send late data also ..
We had more challenges as the client had a lot of disparate information systems, all of which needed to function harmoniously and efficiently. Inconsistent data, duplicates, logic conflicts, and missing data all lead to data quality challenges. Large live data handling was a big challenge.
Using historically mined data at an individual level, a team of 4 developers and analysts developed a model to predict likely destinations, trajectories, stop locations, speeding and braking events at different days of the week and times of day for individual drivers. Adept in Python and Spark, developers needed about 3 months to develop and host it in AWS and chart it out in D3js.
At Pearl we realised that a fleet’s bottom line is impacted by the way its vehicles are maintained. A proactive preventive maintenance program can help fleet managers keep repair costs to a minimum while lowering vehicle downtimes. Preventive maintenance is typically limited to scheduled maintenance specified by OEM’s. Using fleet specific data with usage patterns, we created an innovative Vehicle Health Report that measures and reports the status of key parameters of each vehicle so that maintenance issues can be identified and corrected before they become failures.
Integrated with advanced tools for machine learning and AI, we are empowered to do time series analysis and other approaches across incredibly large and varied data sets.
The client was successful in safekeeping the driver, rerouting the fleet to avoid traffic, achieve fuel efficiency, optimise resources, and plan maintenance on the basis of usage patterns thereby adding value to his business while making profits. The software identified driver behaviours and give real time alerts and warning to avoid violations. The solution was released to production in Feb 2018.
Comments are closed.