There is a specific moment in a supplier qualification when a Cpk number appears on a slide and everyone nods. The number is real, the calculation is correct, and it tells you almost nothing about whether the process will still be in control next month. Statistical process control is not a certificate you earn — it is a loop that has to close, repeatedly, on a specific set of measurements. Most contract manufacturers measure a great deal and control very little, and the difference is invisible unless you know which questions to ask.
This guide is written for the buyer side of the table: quality and procurement engineers qualifying or auditing a PCBA partner. It covers which parameters deserve a control chart, how to read the charts without a statistics background, why Cpk is so often misused, and eight audit questions that separate a working SPC system from one assembled for a customer visit. Our own lines run 8 SMT lines with SPI, AOI and X-ray data feeding process control rather than just pass/fail gates, and the monitoring cadences below reflect what that actually requires.
The Difference Between "We Measure" and "We Control"
Inspection and SPC are both quality activities, and buyers conflate them constantly. Inspection examines output and sorts it into conforming and non-conforming. SPC examines the process that produced the output and asks whether it is behaving consistently. The difference in commercial exposure is stark.
A pure inspection strategy delivers the defect rate the process happens to be producing — you find and scrap what is bad, and the underlying cause keeps producing bad units. An SPC strategy watches a process variable drift toward a limit and intervenes before the first defective unit exists. The first approach reports DPPM after the fact. The second reduces DPPM. If a supplier's answer to "how do you control this process" is a description of their AOI stations, they are running inspection, not control.
| Dimension | Inspection | SPC |
|---|---|---|
| What it examines | Finished or in-process units | The process variable itself |
| When it reacts | After defects exist | Before defects exist |
| Output data | Defect counts, pass/fail | Trend, variation, control limits |
| Commercial effect | Reports your defect exposure | Reduces your defect exposure |
| Evidence it is working | Reject rate | Out-of-control signals and their follow-up actions |
The tell: a real SPC system has a history of out-of-control signals that were investigated and closed. A decorative one has charts that never trip. A process that never goes out of control is not a perfect process — it is an unmonitored one, or a chart with limits set so wide that nothing can ever violate them.
Which PCBA Parameters Deserve a Control Chart
Not every measurement belongs on a control chart. A chart is worth maintaining when the parameter has a genuine specification limit, varies cycle to cycle, and causes a defect when it drifts. In PCBA assembly, a defined set of parameters meets that test. The selection below reflects parameters we actually chart, with the sampling frequency that makes the chart meaningful.
| Parameter | Chart type | Sampling frequency | Why it matters |
|---|---|---|---|
| Solder paste volume (SPI) | X-bar & R | Every print / every panel | Leading indicator of bridging and insufficient solder |
| Reflow peak temperature | Individual & moving range | Per profile verification | Controls intermetallic growth and voiding |
| AOI defect rate | p-chart | Per lot | Tracks process drift across a production run |
| X-ray void percentage | X-bar & R | Per shift | Detects thermal profile and paste issues early |
| Placement offset | X-bar & R | Per shift | Prevents tombstoning and lead misalignment |
| Screw torque (box build) | X-bar & R | Per shift | Prevents over-stress and loosening in the field |
| Ionic contamination | Individual & moving range | Per cleaning run | Controls long-term electrochemical failure risk |
Notice that the leading parameters are process inputs, not outputs. Paste volume and peak temperature are charted because a shift in either predicts failures before any unit is defective. This is what separates a system designed to prevent defects from one designed to document them. If you want the metrics these parameters ultimately move, see our companion guide to quality metrics and DPPM benchmarks.
Reading an X-bar / R Chart Without a Statistics Degree
An X-bar and R chart is a pair of charts sharing a time axis. The upper chart plots the average of each subgroup — a subgroup being a small group of consecutive measurements taken together, typically three to five units. The lower chart plots the range within each subgroup, the difference between its highest and lowest value. Both carry a centre line and upper and lower control limits.
The control limits are not specification limits, and this is the most common misuse. Specification limits come from the customer or the design: the acceptable paste volume range, the permitted peak temperature window. Control limits come from the process itself: they are calculated from the observed variation of the data you have collected, typically as the mean plus or minus three times the standard deviation of subgroup averages. A process can be perfectly in control — stable, predictable, no signals — and still be producing units outside specification, if the process is centred in the wrong place. Control limits tell you the process is consistent. They do not tell you it is correct. Both questions must be asked separately.
Read the range chart first, always
The R chart measures within-subgroup variation. If the range is unstable, the X-bar chart's limits are computed from an unstable baseline and cannot be trusted. An out-of-control range means a special cause is acting on the process right now — investigate before interpreting anything on the averages chart.
Look for patterns, not just points outside the limits
A single point beyond a control limit is the obvious signal. The subtler and more useful signals are runs: seven consecutive points all above the centre line, or a steady upward trend across six points, or two of three consecutive points near the same limit. These indicate a shift beginning before any single measurement looks alarming. The set of these pattern rules is commonly called the Western Electric rules.
Distinguish common cause from special cause
Common-cause variation is the process's natural noise — it is present all the time and is changed only by redesigning the process. Special-cause variation is an assignable event: a new paste lot, a stencil change, a printer parameter altered overnight. Control charts exist to detect special causes so they can be corrected, and to prevent the far more common error of reacting to ordinary noise as if it were a signal.
Confirm the limits are recalculated when the process changes
Control limits must be recomputed after a deliberate process improvement, or they will keep the chart centred on the old performance. A supplier running a genuinely improving process will show you recalculated limits with dates. One running static limits for years is not controlling anything — they are trending. The paste-volume variant of this discipline is covered in our guide to solder paste volume and SPI process control.
Cpk, Ppk and the Within-Lot vs Between-Lot Trap
Cpk measures how well a stable, centred process fits inside its specification limits. It is the ratio of the distance from the process mean to the nearer specification limit, divided by three standard deviations of within-subgroup variation. A Cpk of 1.00 means the process barely fits; 1.33 is the customary minimum for a capable process; 1.67 is where critical-parameter capability usually sits.
The trap is that Cpk and Ppk use different estimates of variation, and the number a supplier quotes is often the flattering one. Cpk uses within-subgroup variation — how much the process varies in the short term, between consecutive units. Ppk uses overall variation — the total spread including the shifts that happen between lots, between shifts, between operators, across the whole study period. When a process drifts between lots but is consistent within a lot, Cpk looks good and Ppk looks much worse. The gap between the two numbers is the information: a large Cpk-to-Ppk spread means the process is not stable across time, and the true capability your product experiences is the Ppk, not the Cpk.
| Cpk value | Interpretation | Typical use |
|---|---|---|
| Below 1.00 | Not capable — specification is not reliably met | Requires corrective action before volume |
| 1.00 to 1.33 | Marginally capable | Acceptable for non-critical parameters with monitoring |
| 1.33 to 1.67 | Capable — customary production target | Standard for most PCBA characteristics |
| 1.67 and above | Highly capable | Expected for critical-to-function and safety parameters |
Always ask for both Cpk and Ppk
A supplier who reports only Cpk has given you the optimistic number. The pair together tells you whether capability is stable across the study window or only within short bursts.
Ask how many subgroups the study contains
Capability statistics computed from a handful of subgroups are noise. A defensible study runs enough subgroups to capture the variation you actually experience, typically twenty-five or more collected over real production.
Ask which parameter the number refers to
"Our process has a Cpk of 1.67" is meaningless without naming the characteristic. A high Cpk on an easy measurement and a low one on the parameter that actually drives your yield are entirely compatible statements.
Check the measurement system first
A capability study built on an unreliable gauge is measuring the gauge, not the process. Before trusting any Cpk, confirm a gauge R&R or measurement system analysis has validated the instrument — our guide to gauge R&R and MSA for PCBA covers the acceptance thresholds.
What Good SPC Data Actually Looks Like in a Supplier Report
When you request SPC evidence, you should be able to evaluate the packet on its own merits without trusting the summary slide. A substantive report contains the following, and the absence of any one item is a fair reason to ask again.
Twenty-five or more subgroups with individual values
Not a summary statistic. You need the plotted points to see pattern, trend and any points that were flagged and resolved.
Control limits shown on the chart with their calculation basis
The limits should be visible on the plot, with the period they were derived from. Limits dated to a process change are good evidence the system is maintained.
Annotations where out-of-control signals occurred
This is the single most convincing item in a packet. A chart with marked excursions and notes on the investigation and outcome demonstrates the loop actually closes.
Corrective action linkage
Each substantive excursion should connect to a corrective action record with a root cause, an action and a verification of effectiveness. The mechanics of that discipline are covered in our guide to quality dispute resolution.
Specification limits shown distinctly from control limits
A chart that conflates the two is not being read correctly by whoever drew it. The two lines mean different things and must be separately identified.
8 Questions That Reveal Whether SPC Is Real
Use these in an audit or a supplier review. The answers matter less than whether they come quickly and specifically — a working system produces detail without preparation.
Which parameters are on control charts today, and who reviews them?
A real answer names specific parameters and a person or role responsible for daily review.
When did a chart last go out of control, and what happened?
If nothing has tripped in a year, either the limits are wrong or nobody is looking. Ask to see the last corrective action record.
How are control limits calculated and when were they last recalculated?
The answer reveals whether the system is maintained or inherited from an initial study years ago.
What is the sampling frequency and who defines it?
Frequency should be tied to how fast the parameter can drift. "Once a month" for paste volume is not control; it is a snapshot.
Can I see raw data for the last thirty days, not a summary?
Summary statistics can be produced without a system. Raw plotted data cannot.
How do you distinguish common-cause from special-cause variation?
This question separates people who understand SPC from people who own SPC software. The answer should describe reacting to signals and not to noise.
Is the measurement system itself validated for each charted parameter?
Gauge R&R should exist for every parameter on a chart. Without it, the chart may be tracking measurement error.
How does an SPC signal reach the customer?
Buyers should know under what circumstances a control-chart excursion on their product triggers notification. This belongs in the supplier quality scorecard and the quality agreement.
Linking SPC to Your Supplier Agreement
The final step is contractual. SPC delivers value to a buyer only when the parameters that matter to the product are the ones being charted, and when excursion reporting is an obligation rather than a courtesy. A practical approach is to name a short list of critical-to-function parameters in the quality agreement, specify the chart type and minimum capability for each, require monthly data submission, and define a notification window for any out-of-control event on those parameters. Keep the list short. A dozen charted parameters that are actually reviewed beat a hundred that are generated and filed. For how these obligations fit alongside sampling and inspection requirements, see our AQL sampling and acceptance guide.
Summary and Next Steps
The question to bring to a supplier review is not "do you have SPC" but "show me a chart that reacted to something, and tell me what you did." Real process control leaves a trail of investigated signals, recalculated limits and closed corrective actions. A certificate leaves a number. The trail is what protects your DPPM and your schedule, and it is entirely checkable if you know what to ask for.
At Huaxing PCBA we chart SPI paste volume, reflow profiles and X-ray void data as process-control inputs, not pass/fail gates, and we will walk you through the raw data behind any project. Our facilities hold IATF 16949 and ISO 9001 certification across 8 SMT lines. Read our quality metrics and DPPM benchmarks guide or ask for the SPC packet on your current product — response within 24 hours.
