Imagine having a primary equipment break down at the peak of your work schedule. Imagine, if you had known it would breakdown and had got it fixed well before peak time. Imagine……….
Well, this is what one of our clients came to us with. It was peak season for their clients across the world. Every year, they would have to deal with breakdowns and part replacements, for their vehicles which were intensively used during the harvest season. Though they were able to capture real time data (which was captured via telemetry systems and stored in a SAP HANA big data system), they needed to figure out a way to inform the customers about predictive maintenance. The client wanted to give an advance warning of critical breakdowns. This way, our clients too would be prepared with spare parts at the right time. And if they could provide this information to their customers, they were sure to exceed vehicle sales in the next year.
Easy isn’t it? Except that our customer has over a million vehicles and a lot of data around vehicle details like the make, mileage, intensity of usage etc. Current mechanical data is sent real-time using telemetry systems . Our challenge? For one, weather data such as temperature and rainfall could impact breakdowns. And we had to find out which vehicle was going to breakdown or which one was going to need a spare part before the actual breakdown happened during harvest season. All the data was stored in a state of the art SAP HANA big data server. Our client has been in business for over 100 years and has been capturing data for quite some time.
When we sat down as a team for our first meeting, we had very little idea what we were going to do. But we, the data fanatics (as we like to call ourselves) ate, slept and dreamt data for the next few months. A couple of weeks later, and after dozens of empty coffee cups, we were ready to showcase our solution.
We built a decision tree model in R , integrated it with the predictive analytics library of SAP HANA , and developed a scoring equation , which spit out the probability of breakdown for each vehicle in real time. Voila! We had hit the bull’s eye! Our prediction was 90% accurate (it was about 35% before we started).
This was the first time ever that anyone had been able to take a humongous data for the farming sector and turn it into something that would eventually benefit the client in a big way (by that we mean increase sales $$$). The project was a Big success for our clients, but more importantly we felt really proud at having done something that no one else had attempted before.