Optimizing Scheduling at GM’s Cold Weather Testing Facility

The Problem

General Motors’ cold weather vehicle testing facility in Canada had a monumental scheduling challenge on its hands. GM autos under development are required to endure the rigors of extreme cold, as part of the “full-vehicle validation” process, before moving into mass production. Yet even in Kapuskasing, the northernmost town in Ontario accessible by road, weather cold enough to analyze vehicles’ performance only lasts several weeks a year. Besides weather, constraints on the testing facility’s capacity to perform all the testing GM required were availability of drivers, limits on the number of daily tests that can be performed on any vehicle over a particular period of time, and various others. In combination, these constraints overwhelmed the test facility’s ability to provide all the required analysis.

The Analytics Solution

In its effort to improve the “throughput” of testing at the Kapuskasing facility, GM brought in an analytics team. The team built a mathematical model for daily test scheduling. With 50 vehicles and 40 drivers to line up within all the constraints, the model necessarily contained some three million variables and thus needed too much computational time using traditional off-the-shelf solution software. So GM’s analytics experts developed a customized “heuristic” method of solution.

The heuristic solution method proceeds by generating a series of potentially viable schedules, then selecting the best one. It accomplishes this by constructing a complete schedule through a series of stages, each of which ends at a particular time, and it obtains a feasible partial schedule at each stage. Ultimately, the system reaches the end of a full scheduling time period with a complete feasible schedule, matching vehicles and drivers.

The Value

This scheduling model developed by the analytics team immediately reduced the typical time required to create a daily testing schedule from nearly four hours to a few minutes. In addition, the new scheduling system was 192% more accurate than the old one, resulting in a 96% reduction in wasted test mileage. These results, in turn, led to a 129% increase in the number of vehicles that completed testing during the first winter the model was implemented. Bottom-line financial benefits to GM — measured by reduced expenditures on under-warranty vehicle repairs resulting from design flaws missed because of insufficient testing — are estimated in the millions of dollars.