SPC Control Charts: 6 Rules for PCB Assembly Lines

SPC control charts plot a measured characteristic from a PCB assembly line against time, with a centre line and a pair of limits that come from the process itself rather than from the drawing. When the points wander outside those limits, or form a pattern the process should not produce, the chart has found a signal worth investigating. SPC control charts are a working tool rather than a report prepared for a customer.

The method matters on an assembly line because so many parameters drift slowly: paste volume, placement offset, reflow temperature, hole diameter. A chart catches a shift while it is still small, which is far cheaper than sorting boards after a customer complaint. This guide covers chart choice, limits, patterns and the decisions that follow.

Why SPC Control Charts Suit PCB Assembly

Assembly processes are stable enough to be predictable and variable enough to drift, which is exactly the condition statistical control was designed for. A printed deposit that averages the right volume but spreads over a wide range will produce defects at both ends even though the average looks healthy.

Charts turn that spread into a visible trend. The operator sees the shift on the screen before the first defect appears, and the engineering team gets a record of when the process changed, which is what makes root cause analysis possible. A chart on a screen beside the line is worth more than a chart in a monthly review.

Choosing the Chart for the Characteristic

Variable data, such as paste height or placement offset, is charted with an average and range pair. Attribute data, such as the number of solder defects per panel, is charted with a proportion or count chart. Mixing the two, or charting a variable on an attribute chart, hides the signal the method is meant to reveal.

The subgroup size follows the process. A reading taken every panel gives a large number of small subgroups, while a sample of five boards every hour gives fewer, larger subgroups that are more sensitive to a slow drift.

Setting Control Limits From Real Data

Control limits are calculated from the variation the process actually shows, usually as three standard deviations either side of the mean, and they are not the same as the specification limits on the drawing. A process can be perfectly stable and still be incapable of meeting the tolerance. Where a standard is named in the control plan, such as the statistical guidance published by IPC, that reference should travel with the limits.

Limits should be recalculated only after a documented change, such as a new stencil, a new paste lot or a machine overhaul. Recalculating them after every excursion is the fastest way to hide the problem the chart was built to find.

Reading an X-bar R Chart

An X-bar R chart carries two panels: one for the subgroup average and one for the subgroup range. The average panel shows shifts in the process centre, and the range panel shows changes in spread, so the pair separates a move from a widening. Both panels have to be read together before any conclusion is drawn.

The distinction decides the response. A sudden move in the centre panel usually means a material or setup change, while a jump in the range panel points at a machine or a measurement problem, and the same logic applies when the readings come from solder paste inspection equipment.

Special Cause Patterns and What They Mean

A single point beyond a limit is the simplest special cause, but most signals arrive as patterns: seven points on one side of the centre line, a steady run up or down, or points hugging the centre line more closely than the limits predict. Each pattern has a different cause.

Runs above and below the centre line suggest a shift in a controlled input such as paste temperature or a new operator. Trends suggest tool wear, and hugging suggests that the subgroup now contains several different streams that are averaging out each other’s variation.

Attribute Charts for Defects

Defect counts per panel are charted with a proportion chart when the sample size varies and with a count chart when it does not. The mathematics differs, but the reading discipline is the same: look for the special cause, not for the number.

Where the sampling plan defines how many boards are inspected and how a lot is accepted, the chart and the plan should agree. Linking the two is the point of the work set out in AQL sampling at final inspection, because a chart that is fed by a different sample than the acceptance plan measures a different process. A pattern in the defect chart is often the first evidence that a process upstream has shifted.

Reacting to an Out-of-Control Point

The reaction should be written before the alarm occurs. The operator confirms the reading, checks the obvious inputs, and then decides whether to continue, to hold the material or to stop the line, following a documented escalation path.

Adjusting the process to bring the point back inside the limits without finding the cause is the classic error. It adds variation, and it turns a stable process into one that is chasing its own measurements.

Control Limits Versus Specification Limits

The two sets of numbers answer different questions. Control limits describe what the process does, and specification limits describe what the product needs. The gap between the two tells the team how much room the process has left. When the process is stable but the two are too close, the fix is a capability study rather than more inspection.

Capability work, such as the study described in process capability analysis, measures how well the spread fits inside the tolerance and identifies whether the centre or the spread needs to move. Charts say what happened; capability says whether it is good enough.

Records, Audits and Improvement

A chart is only as useful as its record. The log should carry the part number, the characteristic, the sample size, the limits in force and the action taken for each signal, all of it retrievable at audit. Reviewing charts in a weekly meeting keeps the habit alive and gives improvement projects their evidence.

The first article result is the natural companion, because it establishes the setting the chart is then expected to maintain. The approval process described in first article inspection and the overall standard in judging PCB quality both depend on the same habit of measuring rather than assuming.

FAQ

How many points are needed before limits are valid? Twenty to twenty-five subgroups is the usual minimum, because the estimate of variation is unstable below that. Limits drawn from a handful of points move with every new reading and give false alarms.

Should the chart be paused during a trial run? Yes. A new process, a new stencil or a new paste lot should be qualified first, then charted. Charting during a known change mixes two processes on one set of limits and produces signals that mean nothing.

What is the difference between control and capability? Control describes whether the process is stable and predictable, and capability describes whether that stable process fits the tolerance. A process can be in control and still fail every board, which is why both studies are needed.