Production Rate Breaks: The Curve Does Not Pause
Learning is perishable. When production stops, learning degrades. The longer the break, the greater the degradation. This is not just about individual worker skill — it includes supplier responsiveness, tooling condition, support equipment calibration, and organizational process memory.
| Break Duration | Workforce Retained | Typical Hours Increase | Recovery Period |
|---|---|---|---|
| 1–3 months | > 90% | 5–12% | 3–8 units |
| 3–6 months | 70–90% | 12–25% | 8–15 units |
| 6–12 months | 50–70% | 20–35% | 15–25 units |
| 1–3 years | 20–50% | 30–60% | 25–50 units |
| > 3 years | < 20% | 50–100%+ | Near full restart |
Scenario: A fighter wing program completes unit 75 at 4,200 hours (on an 85% curve, T1 = 10,000). Production halts for 8 months due to a budget sequestration. 65% of the workforce is retained.
Pre-break trajectory: Unit 76 should be approximately 4,170 hours.
Post-break restart: Unit 76 actual = 5,200 hours (25% above trend).
Recovery: Hours return to the pre-break trend line by approximately unit 95 (20 units of recovery). The total excess hours for units 76–95 are approximately 8,500 hours above what the uninterrupted curve would have predicted.
Cost of the break: 8,500 excess hours × $85/hour burdened rate = $722,500 in learning loss. This is in addition to the fixed costs of maintaining the facility during the shutdown.
Surge and Idle Effects
Rate changes without full production stops also affect the curve. A surge (rapid rate increase) and an idle-down (rate decrease) create different problems, and neither follows the standard learning curve model without adjustment.
| Rate Change | Mechanism | Effect on Hours | Modeling Approach |
|---|---|---|---|
| Surge (rate ×2) | New hires, overtime, second/third shifts, stretched supervision | +5–15% per unit during ramp | Add rate penalty factor; model new hire learning separately |
| Surge (rate ×3+) | All of the above plus facility expansion, new suppliers, process changes | +10–25% per unit during ramp | Segment curve; treat post-surge as a partial restart |
| Idle-down (rate ×0.5) | Retain best workers, longer time per unit, less pressure | –3–8% (improvement) or neutral | Often no adjustment needed; monitor for attrition effects |
| Idle-down (rate ×0.25) | Workforce reduction, skill loss, supplier disengagement | +5–15% (degradation) | Model as partial disruption with workforce retention factor |
⚠️ The Surge Paradox
Surging production adds more cumulative units quickly, which should accelerate learning. But the rate effects (new hires, overtime fatigue, supply chain strain) often outweigh the learning benefit. The net effect is usually negative — hours per unit increase during the surge ramp even though cumulative production is increasing faster. Do not let the learning curve formula convince you that doubling the rate will halve the time to reach a target unit cost.
Workforce Turnover Impact
Learning curves assume a stable workforce. In reality, aerospace production facilities experience 10–30% annual turnover depending on labor market conditions, location, and program stability. Each departing experienced worker is replaced by a novice who starts at a higher individual hour rate, dragging the program average upward.
Framework: Model the workforce as a blend of experienced and new workers, each on their own individual learning curves.
Variables:
- Annual turnover rate: T (e.g., 0.20 for 20%)
- New worker productivity: Pnew = 60–70% of experienced worker on day 1
- New worker ramp time: 3–6 months to reach 90% of experienced productivity
- Blended productivity: Pblend = (1–T) × Pexp + T × Pnew
Example: 20% turnover, new workers at 65% productivity:
Pblend = 0.80 × 1.00 + 0.20 × 0.65 = 0.93
This means the effective learning rate flattens by approximately 7% of productivity, translating to a 2–3 percentage point increase in the observed learning rate (e.g., 85% becomes 87–88%).
Modeling Disruptions as Learning Curve Resets
When a disruption is severe enough, the simplest modeling approach is to treat the restart as a new learning curve with a modified T1. The post-disruption T1 is lower than the original T1 (because retained knowledge, tooling, and processes carry forward) but higher than the pre-disruption unit hours (because of skill decay and workforce loss).
| Reset Parameter | How to Estimate | Typical Range |
|---|---|---|
| Post-restart T1 | Last pre-break unit hours × disruption multiplier | 1.15–1.60 × pre-break hours |
| Post-restart learning rate | Usually same as pre-break (fundamentals unchanged) | Same or 1–2 points flatter |
| Recovery crossover | Unit number where post-restart curve meets original curve | 10–50 units depending on severity |
After the crossover point, the post-restart curve typically follows the original curve’s trajectory. The total cost of the disruption is the area between the two curves from restart to crossover. This area represents the learning loss — hours that were permanently consumed and cannot be recovered.
The Stanford-B Model
The Stanford-B model extends the standard learning curve to handle two common situations that the basic model cannot: (1) programs that start with prior experience, and (2) curves that plateau at high unit numbers.
Y = T1 × (X + B)b
Where B is the equivalent number of prior units of experience.
| B Value | Interpretation | Use Case |
|---|---|---|
| B = 0 | No prior experience; standard Crawford model | New product, first production run |
| B = 10 | Equivalent of 10 units of prior experience | Derivative aircraft, experienced workforce from similar program |
| B = 50 | Substantial prior experience | Block upgrade of production aircraft, line restart with retained team |
| B = 200+ | Extensive experience; curve starts nearly flat | Mature product with minor configuration change |
Example: A derivative fighter uses the same assembly process as the baseline, which produced 100 units. Setting B = 40 (40% credit for transferable experience):
Unit 1 of the derivative: Y = T1 × (1 + 40)–0.2345 = T1 × 41–0.2345 = T1 × 0.432
The derivative starts at 43% of T1 rather than 100%, reflecting the carried-over experience. This is far more realistic than starting at T1 and far more accurate than simply using the baseline’s unit 101 projection.
⚠️ B Is Not Free
The B parameter adds a degree of freedom to the model, which means it can always improve the fit. But an unmotivated B value is just curve-fitting noise. Always tie B to a physical rationale: how many equivalent units of experience does this workforce, tooling set, and process library represent? If you cannot answer that question, do not use the Stanford-B model — use a segmented standard curve instead.
🎯 The Bottom Line
Production disruptions are not exceptions — they are the norm in long-running aerospace programs. A 6-month break can add 20–35% to restart unit hours and take 15–25 units to recover. Surges often increase hours despite adding cumulative volume. Workforce turnover flattens the observed learning rate by 2–4 points. The Stanford-B model handles prior experience and production restarts by adding an equivalent-experience parameter. Every program forecast must account for planned and potential disruptions, or it will systematically underestimate future hours. Next: HPU Forecasting for Proposals & EACs — turning these models into defensible estimates for bids and program baselines.
Stop reading, start modeling
Model your process flow, run simulations, optimize staffing with TOC math, and test your knowledge with 107 interactive checks — all in one platform.