Adaptive SPC: Higher Cpk in Aerospace Heat Treatment

The engine continuously samples process data from every connected furnace zone — temperature per zone, quench pressure, atmosphere dew point, load position — at the configured interval. It maintains a rolling statistical baseline of the current process regime: the mean and standard deviation for each monitored parameter over a configurable time window, typically the most recent 25 to 50 data points per parameter. When a new furnace cycle starts with a different recipe, a different alloy grade, or a different material cross-section, the baseline resets to the new regime within the first few data collection cycles. The engine distinguishes between routine process noise (common cause variation that does not affect product quality) and a genuine regime change that requires a new baseline. This distinction — drawn automatically from the data — is what static control limits cannot make and is the primary source of both false alarms and missed detections in conventional SPC systems.

With the current baseline established, the engine calculates new Upper and Lower Control Limits for each monitored parameter at every sampling interval. The formula is standard SPC — UCL = mean + 3 sigma, LCL = mean - 3 sigma — but the mean and sigma values are drawn from the rolling baseline, not from a static historical study. When the process tightens (sigma decreases), the limits narrow, making the chart more sensitive to deviations. When the process shifts to a new regime (mean moves), the limits follow, preventing false alarms on legitimate process changes. Simultaneously, the engine recalculates Cpk for each quality characteristic — hardness range, case depth, tensile strength — using the adaptive limits as the control range. The Cpk value reflects the current process capability, not the capability that was measured during the last PQ study. This is the Cpk operator should see: the one that tells you whether the furnace is capable of producing conforming parts right now.

With adaptive limits in place, the engine applies a second filter to every potential alert: is this deviation real relative to the current process state, or is it the kind of variation that the process produces during normal operation? Because the limits are already calibrated to the current regime, most conventional SPC alerts that would have fired during recipe transitions or material changes are eliminated at the calculation stage rather than requiring operator judgement to dismiss. The alerts that do fire are genuine deviations from the current baseline — a zone 3 reading that has moved 3.5 sigma from the current mean while zones 1 and 2 remain stable, a quench delay that exceeds the current window, an atmosphere dew point that has shifted outside the adaptive band. The engine also projects the current drift trajectory: if zone 2 is rising at 2 degrees per 10-minute interval and the adaptive UCL is 15 degrees above the current reading, the operator sees how much time remains before a potential limit breach. This trend projection combined with the Cpk trend is what gives the operator actionable foresight rather than retrospective alarm.

Static limits: The operator switches from an AMS 2770 aluminum solution treat cycle to an AMS 2759 steel hardening cycle. The temperature range, soak duration, and quench medium are completely different. Static limits calibrated for the aluminum cycle generate false alarms on the first three steel loads because the steel operating temperature is outside the previous cycle's control range. Cpk for the steel cycle is artificially deflated because the variation measured against static limits includes the transition period — even though the steel process is running within its own specification.

Adaptive limits: The recipe change is detected within the first two data collection cycles. Limits recalibrate to the steel cycle's baseline. No false alarms. Cpk calculation starts from the correct baseline after the transition window. The operator sees a Cpk of 1.72 on the steel cycle by the second load, reflecting the actual process capability rather than the transition distortion.

Static limits: Zone 3 thermocouple reading begins drifting upward over four consecutive loads — 3 degrees, then 6, then 8, then 10 degrees above the nominal setpoint. Static limits eventually flag the deviation on load four, but by then the drift has been present for three loads without generating a signal. Cpk for the affected alloy grade has dropped from 1.65 to 1.18 across the four-load period, but the operator has no real-time visibility into the decline because the static Cpk update occurs after the test results, not during the drift accumulation.

Adaptive limits: The adaptive baseline detects the zone 3 drift pattern on load two — not as a single-point deviation but as a consistent directional shift across the zone's readings. The operator receives a predictive alert on load two: zone 3 trending up, estimated time to reach AMS upper tolerance limit is 2.5 loads at current trajectory. Cpk trend line shows the projected decline. The operator adjusts the zone 3 setpoint before load three. Cpk holds at 1.62 across the adjustment period.

Static limits: Seasonal humidity increases the operator's quench transfer time by 4 to 7 seconds as the furnace door seal and quench elevator mechanism respond to atmospheric conditions. Static limits on quench delay, calibrated during low-humidity months, generate an alert on every third load. The operator learns to dismiss quench delay alerts because they are seasonal, not process-critical. Cpk for quench-related quality characteristics shows increased variation that appears as a capability decline even though the metallurgical outcome is unchanged.

Adaptive limits: The adaptive baseline shifts the quench delay UCL upward as the seasonal pattern establishes itself — the system recognises that the new normal quench delay range is 4 to 7 seconds higher than winter months, and adjusts accordingly. Alerts only fire when a delay exceeds the new adaptive UCL by a statistically significant margin. Cpk for quench-related characteristics reflects the actual variation rather than the seasonal offset, holding steady at 1.58 through the humidity period.

Static limits: A partial load with 40% of the standard fixture density heats faster than a full load, reaching soak temperature 12 minutes earlier. Static limits on ramp rate and soak initiation timing, established from the standard load configuration, generate a ramp rate alert. The operator evaluates, dismisses, and notes the partial load condition. The same alert repeats on every partial load. Over time, the operator's alert dismissal rate climbs, and the Cpk for ramp rate shows elevated variation that is purely a function of load configuration rather than process capability.

Adaptive limits: The operator enters the load density or fixture configuration at cycle setup. The adaptive limits for ramp rate and soak initiation adjust to the partial load baseline — the UCL and LCL for ramp rate widen to accommodate the different heating profile, and the soak initiation control chart uses the partial load expected timing rather than the full load standard. The operator receives no false alarm. Cpk for ramp rate remains consistent across load configurations because the limits reflect the actual process behaviour for each load type.