The challenge
Melting furnaces run around the clock, and energy totals mean little without production context. To reduce consumption, the plant needed to compare each batch with batches of the same alloy, a similar scrap mix and a similar input weight.
Our approach
- Phase 1: collect meter readings, temperatures, door states, batch events and material records, then clean and align them into one record per casting batch.
- Benchmark each batch against the best quarter of comparable historical batches, only where enough history exists.
- Phase 2A: machine-learning models that predict consumption, adapt the benchmark to current conditions and explain deviations.
- Phase 2B: a data-driven digital twin to explore door timing, loading schedules and air-to-gas ratio.
Results & features
- Comparable, per-batch energy indicators instead of plant-wide totals.
- Deviations from best practice become visible while the batch is still running.
- Operators stay in the loop: recommendations are reviewed before any change is made.
- Designed to extend to further furnaces by configuration, without redesign.