Page 1
A Schedule Optimization Tool for Destructive and Non-Destructive Vehicle TestsJeremy Ludwig, Annaka Kalton, and Robert Richards
Stottler Henke Associates, Inc. Brian Bautsch, Craig Markusic, and Cyndi Jones
Honda R&D Americas, Inc.
ICAPS-SPARK, June 2016
Page 2
Overview• Introduction• Scheduling Framework• Scheduling UI• Domain Customization• Methods• Results• Deployment• Conclusion
Page 3
Introduction• Create a schedule for testing new and
refreshed vehicle models• Only some tasks are destructive
• Most tasks are non-destructive but may have other constraints
• Test vehicles hand-built• Build order
• Not all available at once• Variety of models
• Frame, Market, Drivetrain, and Trim
• Project end date defined externally• Limited personnel and facility resources
Page 4
Constraints• Temporal• Calendar• Ordering• Resource • Build Pitch• Exclusive• Destructive
Page 5
Aurora Scheduling Framework• Create a high-quality schedule
• Based on a model of temporal, calendar, ordering, and resource constraints
• Uses graph analysis techniques and heuristic-based scheduling
• Customized for domain• Handle special kinds of tasks
• Exclusive, Destructive• Minimize the number of vehicles required
• Select the types of vehicles built• Select a build order for the vehicles
Page 9
Initial Schedule by Resource
Page 10
Scheduling Framework Components• Schedule Initialization
• Preprocessor• Prioritizer
• Scheduling Loop• Priortizier• Scheduler• Quality Criterion• Conflict Manager
• Schedule Finalization• Postprocessor
Page 11
Differences From Prototype• Testing on more complex models that
require over 100 vehicles• Utilizing facility and personnel constraints
when creating a schedule • Supporting the transition of the software
into the hands of the actual planners
Page 12
Domain Specific Customization• User Interface
• Wizard• Model Verification during Import
• Build Pitch• Manage Vehicles• Long Tasks• Optimization Dashboard
• Scheduling Components
Page 17
Optimization Dashboard
Page 18
Scheduling Component Customization• Preprocessor
• Override project end date• Prioritizer
• Exclusive Task• Long Task• Destructive Task• Tight Window• End Based• Load Based• Subsequent Duration
Page 19
Methods• Actual Model
• ~340 tasks• ~4000 days of work • ~30-50 vehicle types
• Manual Solution• Not attempted
Page 20
Results• Aurora Solution
• 60-200 Vehicles• Adhere to all constraints• Represents 6% reduction from best estimate• Prototype found 12% reduction in direct comparison
• Schedule created in 2 minutes from model vs. days of labor• Spend time using ‘What-if’ capability further
improve the schedule• Time• Build Pitch• Negotiation
Page 21
Optimized Schedule
Page 22
Deployment• Deployed and in use by novice planners
• Previous solution no longer used• Integrated with enterprise system
• Input data extracted from external data• Results exported to corporate format
• Providing huge savings and other benefits with every new test suite
Page 23
Conclusion• Complex, real-world, scheduling problem • Added domain-specific heuristics to a
general intelligent scheduling framework• Added help for novice planners• Generated schedule for vehicle testing
• Significant reduction in the number of prototype vehicles required
• Still completed in the given timeframe• Extend to multiple projects in future
work