Time-Series Forecasting Tool
Customer Challenge
The Air Force needed a tool to predict changing logistics metrics for aircraft availability, maintenance, and manpower.
Innovative Solution
Illumination Works developed a versatile time series forecasting application to predict aircraft metrics including inherent and induced failures, corrosion rates, equipment repair hours and actions, and workforce manhours. ILW normalized forecasts against fleet utilization measures such as aircraft age, flight
hours, and sorties to ensure more accurate and actionable insights.
Benefits/Outcomes
- 24,000 multi-variate models covering all aircraft systems and maintenance work unit codes
- Confidence interval bracketing of forecasts for rational decision-making accounting for uncertainty in the predictions
- Compute clusters with multi-threading enables rapid forecast generation
- Forecasts and model performance visualized on Tableau dashboard
- Application concept ideation to production deployment in six months
Toolbox
- Auto-regressive, dense neural network, long short-term memory (LSTM), deep LSTM, and bidirectional LSTM models
- TensorflowML Platform and Keras API
- Python on Databricks
- Amazon Cloud (DBFS)
- Tableau business intelligence
Business Value
- Accurate forecasts for manpower, maintenance planning, and workforce management decision making
- Improved maintenance planning to yield increased aircraft availability and readiness
- Enhanced efficiencies and cost savings
Domain Expertise
- Air Force weapon system maintenance and logistics data
