Skip to content

Porosity Statistical Process Control

Process-control monitor — instantiates Functional Porosity Design

Keeps a production run's void architecture inside spec by sampling a few critical metrics, watching for drift, and correcting the process before defects accumulate.

Porosity Statistical Process Control governs the void architecture across a production run, not within a single part. It is a control loop: sample a few critical porosity metrics, plot them over time, distinguish ordinary scatter from a genuine shift, and correct the process only when the data warrant it — so the whole population of parts stays inside the defect band. The defining idea is holding a distribution stable, not designing or characterizing one body. It presupposes that someone else has set the target and someone else has proven the process capable; its job is to keep every unit landing on that target as the line runs.

Example

A metal additive-manufacturing shop prints load-bearing brackets by laser powder-bed fusion, where stray porosity is the main defect. Rather than X-ray every part, the shop samples coupons on a plan and measures their void fraction and maximum pore size, plotting each on a control chart whose limits come from the process's own early variation. Random part-to-part scatter within the limits is left untouched. But a slow upward creep in porosity — traced later to a degrading laser optic and damp powder — trips a control limit while parts are still in spec, prompting a parameter and powder correction before any bad brackets ship. The line is held to its capability instead of being inspected after the fact.

How it works

The distinguishing loop is sample → chart → separate signal from noise → act only on signal. A few critical-to-function metrics are chosen (void fraction, maximum pore size, defect count), sampled on a defined plan, and plotted on control charts whose limits are set from the process's own natural variation. The chart's whole purpose is to separate common-cause scatter, which is left alone, from special-cause shifts, which trigger a correction fed back to the void-creation parameters. Unlike a one-part characterization, it is watching a time series of a distribution and defending its stationarity.

Tuning parameters

  • Controlled metrics — which few porosity metrics are charted; more coverage catches more failure modes but multiplies destructive sampling.
  • Sampling rate & size — how often and how many units are pulled; tighter sampling catches drift sooner at higher test cost.
  • Control-limit width — how far from target trips an alarm; tight limits detect fast but raise false alarms and invite over-adjustment.
  • Reaction rule — what a signal triggers — adjust, hold, quarantine; over-reacting to noise injects variation instead of removing it.
  • Capability target — the process-capability level a run must clear before it is trusted to produce to spec.

When it helps, and when it misleads

Its strength is turning "the parts are usually fine" into a monitored, capable process that catches drift before scrap piles up and that tells real shifts from noise. It misleads in three ways. It can hold a process to spec but cannot rescue one that was never capable; it controls only what is charted, so an uncharted defect mode runs free; and its worst misuse is tampering — adjusting a stable process in response to every individual sample, which provably increases variation rather than reducing it, the lesson of Deming's funnel demonstration.[1] Control also assumes a stationary process, so a slow wear trend can walk the mean while each point still reads in-limit. The discipline is to act only on genuine special-cause signals, verify capability before trusting the chart, and re-center deliberately for known trends rather than chasing scatter.

How it implements the components

SPC fills the production-control components — the ones about holding the population in spec over time:

  • process_variability_and_defect_band — its core object: the statistical band of void variation it holds inside spec, separating common-cause scatter from special-cause shifts.
  • void_fraction_budget — total void fraction is charted as a critical metric and held on target across the run.
  • void_size_and_shape_distribution — the pore-size distribution is sampled and controlled so the whole population, not merely its mean, stays in spec.

It does not design or physically produce the voids — that is Topology Optimization for Void Placement and the fabrication methods — nor manage in-service clogging and degradation, which is the operational Clogging and Regeneration Protocol rather than the production line.

References

[1] Deming's funnel experiment — adjusting a stable, common-cause-only process in reaction to each individual measurement makes its output more variable, not less; acting only on statistically significant special-cause signals is what keeps a control chart from degenerating into tampering. ↩