Tier 1 automotive gearbox manufacturer

Nitriding process optimisation for gearbox components

Charge planning for nitriding heat treatment took about three hours a cycle by hand, and the furnaces still ran under capacity. We replaced the spreadsheet with an optimisation engine that plans component-fixture combinations against real physical constraints.

saved on every charge plan
3 hrssaved on every charge plan
gain in furnace utilisation
10%gain in furnace utilisation
productivity improvement
13%productivity improvement
units of power saved per month
20,000units of power saved per month

The client is not named on this page. The work, the numbers and the technical detail are described exactly as delivered.

The problem: a packing puzzle solved by hand, every cycle

Nitriding charges have to be planned around what physically fits: which components can share a fixture, how they stack, what the load weighs, how tall it stands, what the furnace can take. Doing that by hand takes hours, and the answer is never quite optimal — so the furnace runs with space in it.

Manual planning bottleneck

Charge planning took roughly three hours per cycle, and every hour of planning was an hour the throughput waited on.

Underutilised furnaces

Loads reached only 80–90% of capacity. The energy cost of a cycle is much the same whether the furnace is full or not.

Fixture incompatibility

Mismatches between components and fixtures surfaced late and caused repeated idle time while a plan was reworked.

No planning visibility

There was no centralised view of work in progress, and no way to look at a future load before committing to it.

High operational cost

Manual coordination meant resource and energy usage that was defensible but never actually optimal.

The system we built

This is an optimisation problem wearing a paperwork problem’s clothes. Once WIP is captured digitally and the fixture rules are expressed as constraints, the plan can be computed rather than negotiated.

Input

  • Web-based WIP loggingmodel + quantity, real time
  • Incoming component queuelive

Constraints

  • Fixture compatibility matrixwhich part fits which fixture
  • Physical limitsstacking, weight, height, tonnage

Engine

  • Automated charge planningmaximise load per cycle
  • Furnace-level distributionacross available fixtures
  • Scenario forecastingsimulate future batches

Output

  • Charge planin minutes, not hours
  • Batch sequenceoptimised
  1. 1

    Web-based WIP logging

    Operators record incoming components by model and quantity as they arrive, which turns the queue into data the planner can actually work from.

  2. 2

    Fixture compatibility matrix

    Stacking rules, weight, height and tonnage constraints encoded explicitly, so an infeasible pairing is rejected at planning time rather than discovered at the furnace.

  3. 3

    Automated charge planning

    The planner evaluates fixture-part combinations to maximise the load in each cycle — the same puzzle the planner was solving by hand, searched properly.

  4. 4

    Furnace-level optimisation

    Loads are distributed intelligently across available fixtures and furnaces, which is where the utilisation gain comes from.

  5. 5

    Scenario-based forecasting

    Future loads can be simulated before committing, so batch sequencing is a decision rather than a consequence.

What changed

3 hours saved per charge

Planning stopped being a shift activity. The time went back into throughput.

20,000 units of power saved per month

Fuller loads mean fewer cycles for the same output, and a cycle costs roughly the same energy either way.

10% furnace utilisation gain

Better packing efficiency from searching combinations the manual process could not evaluate in the time available.

13% productivity improvement

End-to-end cycle time reduced across the process, not just at the planning step.

8% cost reduction

Lower manpower, energy and rework-related cost, compounding across every cycle.

1.6% fixture cost saving

Better use of the fixtures already owned, avoiding procurement that had looked necessary.

Optimisation only works if it respects the floor

A planning algorithm that ignores stacking rules, fixture weight limits or what the crane can actually lift produces plans nobody follows. The constraint matrix was the real deliverable here; the optimiser was the easy half.

That is generally the shape of this work — the modelling is not hard, understanding the process well enough to model it honestly is.

Is a planner solving a packing puzzle in a spreadsheet?

Heat treatment, plating, paint, any batch process with fixture or capacity constraints — if the plan takes hours and still leaves the equipment under-loaded, there is usually room to compute it instead.