Dan Pikelny and Terry Kline
VP and Chief Analytics Officer/ SVP and CIO, Navistar
Data to Decisions
Navistar International Corporation (NYSE: NAV) is a holding company whose subsidiaries and affiliates produce International® brand commercial and military trucks, proprietary diesel engines and IC Bus™ brand school and commercial buses. The company's subsidiaries and affiliates also produce truck and diesel engine service parts. Another affiliate offers financing services. Additional information is available at www.Navistar.com/newsroom.
Navistar, as many other manufacturers, needa to be able to support their products when they are out in the field. In this instance, the product is trucks, and the key measure of success for trucks is uptime – the proportion of time that a truck is operable by a driver rather than being in the shop for repairs. The repairs present a second challenge for Navistar, specifically warranty exposure. From an analytics perspective the challenge is one, measuring the warranty exposure, and two, detecting problems in the vehicle before customers experience it in the field so that they may be brought in for preventive maintenance and lower warranty costs.
Navistar uses a remote diagnostics system: OnCommand Connection. The system gives the company access to vehicle and engine data, on all makes and models, by interfacing with telematics systems that fleet customers use such as PeopleNet and Omnitracs. From an analytic perspective, the IoT data is available to train machine learning models to estimate warranty exposure and predict which vehicles are at risk and which components are potentially problematic. The Analytics team developed algorithms to leverage vehicle usage data for warranty exposure and also detect emerging problems in as few as five vehicles in a population of tens of thousands. The issue detection algorithm predicts the lifetime failure rate and sets alerts on future risk instead of focusing on historical experience. In addition, the team uses AI techniques to predict which individual vehicles based on IoT sensor data are highest risk for the resulting failures.
Through these new capabilities Navistar was able to view the data in an entirely new way that has allowed for insight discovery: visualization, estimation, and predictive analysis of warranty exposure and mitigation through intervention. First Navistar is able to leverage vehicle actual usage almost real-time estimation of warranty exposure. Prior, warranty risk exposure was based on only age and general assumptions of usage. Second, issues were typically not discovered until warranty claims reached critical levels at which point problems became systemic. Those systemic issues not only impacted Navistar’s bottom line, but also deteriorated customer experience and reduced vehicle uptime. Now Navistar is able to identify issues before becoming widespread and proactively make repairs and design changes while minimizing customer downtime.
Telematics data is the backbone of this project and is handled in multiple phases through both streaming and batch processing; data is constantly ingested at a rate of over 150 million unique records per day from multiple telematics providers that cover over 300,000 vehicles. Additional data sources are also ingested daily to blend and enhance the telematics data along with multiple analytic batch processes executed daily and weekly for different levels of analysis.
The results / metrics from using the data described above include the ability to:
- Understand warranty exposure and appropriately reflect exposure in financial warranty accruals based on actual vehicle usage
- Detect major issues for customers occurring 4-6 months earlier using a predictive model that identifies potential maintenance issues for customers, predicting failures for more than 40,000 combinations of diagnostic trouble codes by make, model and year of vehicle.
- Identify the top fleets that were starting to experience issues and contacting those customers to set up a proactive repair program to change a part before it failed, reducing maintenance costs up to 40 percent for connected vehicles. One company’s fleet decreased the cost-per-mile maintenance from between 12 to 15 cents to less than three cents.
- Save an estimated $21 million in warranty costs over four years using the predictive models.
Using Hadoop and Cloudera we ingest, store, process, and blend millons of telematics data points and existing internal sources that include warranty claims and sales data. Navistar uses systems such as Flume to handle millions of records streaming in daily from over 300,000 vehicles from multiple sources, Sqoop for scraping data from existing internal data sources, Hive & Impala for data preparation, blending, and aggregation, and Hadoop Streaming for analytics and processing.
No other transportation manufacturer has been able to leverage telematics data to this extent. We used an open architecture system, OnCommand Connection that allowed us to partner with and use the data from multiple Telematics Service Providers (TSPs) to serve our customers better and improve our product performance. From an analytic perspective, this allows us to have access to data on different types of vehicles and helps us serve customers with mixed fleets. This results in Navistar being able to bring the knowledge and information management that is typically available to only large fleets, to much smaller operators.
While Analytics developed the analytic methods, successful implementation of the project required participants from multiple areas of the company. Reliability and Quality reviews the output of the models on a weekly basis and works with Engineering to review potential failures and design fixes. Customer service works directly with the customers to ensure pro-active actions are taking place, and Warranty Finance reviews trend data and makes appropriate warranty accrual adjustments.