What is Statistical process control (SPC)?

What is SPC (Statistical Process Control)?

SPC stands for Statistical Process Control — a method for monitoring and steering manufacturing processes in real time using statistical techniques. Instead of inspecting finished products and sorting out defective parts, SPC detects deviations while the process is running, before scrap is produced. That makes SPC preventive and process-oriented — the opposite of reactive final inspection.

The core tool of SPC is the control chart. It plots measured values chronologically and shows whether a process is running stably or whether systematic disturbances are creeping in. Walter A. Shewhart developed the first control chart in 1924 at Bell Laboratories, and his core idea still holds: every process varies. The decisive question is whether the variation is normal (random) or abnormal (systematic). You should only intervene when the variation is systematic.

In the automotive industry, SPC is effectively mandatory through standards such as IATF 16949 and the associated core tools (the AIAG SPC reference manual). But it is just as common in food production, plastics processing and pharmaceutical packaging, where SPC is used to demonstrate process capability to customers and auditors.

Common-cause vs. special-cause variation: the core principle

SPC distinguishes two types of process variation, and that distinction determines whether intervening helps or actually makes the process worse.

Common-cause variation is the natural scatter of a process: minor material differences between batches, slight temperature swings on the shop floor, normal tool wear over the course of a day. This variation is predictable and follows a statistical distribution (in most cases a normal distribution). As long as only common causes are at work, the process is considered stable or "in control".

Special-cause variation comes from exceptional influences: a worn bearing, a material change, an incorrect machine setting after a changeover, an untrained operator. These influences shift the process or widen its spread beyond the normal range.

The most common mistake in practice is over-adjusting. An operator sees a single value near the tolerance limit and re-sets the machine. If that value was only common-cause variation, the adjustment shifts the whole process and makes it less stable. Shewhart called this "tampering". The control chart prevents exactly this by showing whether an intervention is statistically justified or not.

Control charts: structure and function

A control chart consists of a center line (the process mean), an upper and lower control limit (UCL/LCL, at ±3 sigma) and, optionally, warning limits (at ±2 sigma). Measured values or sample statistics are plotted chronologically.

Important: control limits are not the same as tolerance (specification) limits. Tolerance limits define what the customer accepts. Control limits define what the process can deliver on its own. A capable process has control limits that sit comfortably inside the tolerance limits.

Control chart types at a glance

In discrete manufacturing the X̄-R chart dominates. In the process industries (chemicals, food) the individuals chart is more common, because batch samples often yield only one measurement per point in time.

Recognizing patterns: when is intervention needed?

A point outside the control limits is the most obvious signal. But there are subtler patterns that point to systematic disturbances long before a value breaches a limit. The most common rules (Western Electric / Nelson):

• 7 or more consecutive points on one side of the center line (a shift)
• 7 or more consecutive points steadily rising or falling (a trend)
• 2 out of 3 consecutive points beyond the 2-sigma warning limits
• 15 or more points within the 1-sigma zone (stratification — a sign of mixed sub-populations)

These rules are built into every SPC software and can be evaluated automatically. With manual, paper-based charting they are regularly missed, which dramatically reduces the value of SPC.

Process capability: understanding Cp and Cpk

A stable process is not automatically a capable process. Process capability means the process varies so little that its results reliably fall within the tolerance limits.

Cp (process capability index) compares the tolerance width with the process spread: Cp = (USL − LSL) / 6σ. A Cp of 1.0 means the process spread exactly fills the tolerance. At a Cp of 1.33 the spread uses only 75% of the tolerance, leaving a safety buffer.

Cpk (the critical capability index) additionally accounts for whether the process is centered. A process can have a Cp of 2.0 (very tight spread), but if the mean sits close to a tolerance limit, the Cpk is much lower. In practice Cpk is the more relevant figure, because processes are rarely perfectly centered.

What do these values mean in concrete terms? As a rule of thumb, a Cpk of 1.33 corresponds to roughly 63 defects per million (ppm), while a Cpk of 1.67 corresponds to only about 0.6 ppm. The jump from 1.33 to 1.67 therefore cuts the theoretically expected scrap by about two orders of magnitude — which is why safety-critical characteristics carry the higher requirement.

Typical Cpk requirements by industry

The difference between short-term capability (Cm/Cmk) and long-term capability (Pp/Ppk) reveals a lot about process stability. A process with Cmk = 2.0 at machine acceptance but Ppk = 1.1 in series shows that the machine can do it, but additional disturbances appear in series production (material scatter, thermal drift, tool wear across several shifts). Closing exactly that gap is the job of SPC.

SPC vs. final inspection: what SPC does not replace

SPC is process monitoring, not product inspection — and the two are frequently confused. SPC ensures the process stays stable and capable. A 100% final inspection ensures no defective part reaches the customer.

In practice the two approaches complement each other. SPC reduces the defect rate so far that final inspection finds barely any scrap. But it does not replace final inspection — especially not for safety-critical characteristics. A Tier-1 supplier that has to prove to its OEM that no defective part was shipped needs both.

A closely related term is SQC (Statistical Quality Control): SQC is the umbrella term for all statistical quality methods, including acceptance sampling; SPC is the process-accompanying part of it. The relationship to Six Sigma is similar. SPC is a tool within Six Sigma (specifically in the Control phase of DMAIC), but Six Sigma reaches far beyond it: problem definition, measurement-system analysis, root-cause analysis, improvement and control. SPC provides the ongoing monitoring; Six Sigma provides the improvement process.

Common mistakes when introducing SPC

Mistake 1: Keeping control charts on paper and ignoring the patterns. Paper-based SPC records individual values, but no one evaluates the patterns systematically. A trend across 7 points is missed on paper; in software it is flagged automatically. At SYMESTIC customer Klocke (pharmaceutical packaging), moving from manual to automated capture was a key lever: quantities and downtimes are recorded via DI gateways and evaluated without manual intermediate steps (see the case below).

Mistake 2: Confusing control limits with tolerance limits. A chart's control limits are calculated from the actual process spread (±3σ), not from the drawing tolerances. Anyone who uses tolerance limits as control limits reacts far too late — only to scrap rather than to process shifts. Conversely, calculating control limits for a process that is not capable (Cpk < 1.0) yields control limits that lie outside the tolerance. In that case the process must be improved first, before SPC can be used meaningfully.

Mistake 3: Starting SPC without a measurement-system analysis (MSA). If the measurement system itself causes 30% of the observed variation, the control chart does not reflect the process but largely the gauge. A Gage R&R study is mandatory before SPC data is interpreted. The AIAG guideline: the measurement system should contribute less than 10% of the tolerance width as variation (Gage R&R < 10%). Between 10 and 30% it is conditionally acceptable; above 30% it is unusable for SPC.

Mistake 4: Monitoring too many characteristics at once. 50 control charts per line sounds thorough but is unmanageable in practice. Quality engineers recommend starting with the 3 to 5 characteristics that have the biggest impact on scrap, customer complaints or function. The FMEA provides the prioritization: characteristics with a high risk priority number (RPN) or a "CC/SC" marking first.

SPC in the MES: automation instead of paper charts

SPC reaches its full potential when process data is not captured manually on paper but pulled automatically from the machine. An MES with integrated SPC software — like the cloud-native MES from SYMESTIC — handles three tasks that are barely feasible manually.

First: real-time capture. Process parameters (temperatures, pressures, forces, cycle times) are read directly from the PLC or via OPC UA and converted into control charts automatically. Manual entry errors disappear. The data is available to the second, instead of once per shift.

Second: automatic pattern detection. The system continuously checks the Nelson rules and triggers an alarm on a violation — before the control limit is even breached. At SYMESTIC customer Neoperl (assembly automation), PLC alarms are automatically correlated with downtimes and quality defects. That enables a root-cause analysis that would take days with manual capture.

Third: cross-site comparison. At Meleghy Automotive, also a SYMESTIC customer, OEE capture runs at the key process steps across 6 plants (Wilnsdorf, Gera, Bernsbach, Reinsdorf, Brandýs/CZ, Miskolc/HU). Quality figures and process data are synchronized with the ERP via a bidirectional SAP interface (ABAP IDoc). That creates the basis to compare process capability across plants and roll out best practices from the strongest plant to the others.

The decisive part is linking SPC data with machine and operational data. When a control chart shows a process shift, the quality engineer must be able to see immediately: which batch was running at that moment? Which operator was at the machine? Which machine parameters were set? Without that link, the control chart is an alarm system without root-cause clarity. This linkage is exactly why SPC inside an MES like SYMESTIC delivers far more than an isolated, standalone SPC tool.

In practice: Klocke runs SPC with SYMESTIC in a regulated environment

The Klocke Group, an international contract manufacturer in the pharmaceutical, cosmetics and food-supplement sectors, uses SPC with SYMESTIC in its GMP-regulated packaging operations. The entry point was a blister-packaging line at the Weingarten site.

The technical setup: quantities and downtimes are captured via DI gateways. All machines are connected through digital I/O devices, without the need for a LAN infrastructure. The data flows one-way through a file interface into the Navision ERP: order status and master data come from the ERP, while machine cycles and downtimes are assigned to the production orders.

Within 3 weeks the solution was scaled to all lines at the site. The results: 7 additional hours of production time per week, a 12% improvement in output and 8% higher availability. In a regulated environment it is especially relevant that all process data is documented and traceable automatically, which considerably simplifies the proof required for GMP audits.

SPC vs. predictive quality: boundaries and interplay

SPC is based on univariate statistics: one characteristic is observed over time and deviations are detected. That works well for processes with a few dominant influencing variables.

Complex processes with dozens of parameters (injection molding with 20+ process parameters, forming processes with material scatter, temperature, speed, lubrication) overwhelm classical SPC. This is where multivariate methods and predictive-quality approaches come in: machine-learning models that learn from historical process data which parameter combinations lead to scrap, and predict deviations before they become visible on a single control chart.

The two approaches do not replace each other. SPC remains the foundation: transparent, traceable, standard-compliant, auditable. Predictive quality extends SPC where univariate analysis is not enough. In practice most companies start with classical SPC and add predictive methods once the data foundation is in place and the organization is statistically mature enough.

SPC with SYMESTIC: from the paper chart to automated process monitoring

SYMESTIC is a cloud-native MES for mid-sized manufacturers. The integrated SPC function pulls measured values automatically from the PLC and OPC UA, runs control charts in real time, checks continuously against the Nelson rules and links every process shift directly to order, machine and operational data. Commissioning takes hours instead of months — with no servers of your own and no major IT project.

Frequently asked questions about SPC

What is SPC in manufacturing?
SPC (Statistical Process Control) is a method for monitoring manufacturing processes using statistical techniques. It detects process deviations in real time via control charts and prevents scrap before it occurs. Unlike final inspection, SPC is preventive and process-oriented.

What does the acronym SPC stand for?
SPC stands for "Statistical Process Control" — the statistical monitoring and steering of a manufacturing process while it runs.

What is the difference between Cp and Cpk?
Cp measures the ratio of tolerance width to process spread, without considering where the process is located. Cpk additionally accounts for how centered the process is within the tolerance. A process can have a high Cp, but if the mean is off-center, the Cpk comes out lower. In practice Cpk is the more relevant figure.

Which Cpk value is good enough?
In automotive (IATF 16949), Cpk ≥ 1.33 is the minimum requirement for existing processes. Safety-critical characteristics require Cpk ≥ 1.67 to 2.0. For new launches, Ppk ≥ 1.67 is required based on initial-sampling data.

What types of control charts are there?
There are charts for continuous (variable) characteristics — X̄-R, X̄-s and the individuals chart (I-MR) — and charts for attribute (countable) characteristics such as the p chart (defective proportion) and the c chart (defect count). In discrete series production the X̄-R chart is the standard; in the process industries the individuals chart is common.

What is the difference between SPC and SQC?
SQC (Statistical Quality Control) is the umbrella term for all statistical quality methods, including sampling and acceptance inspection. SPC is the process-accompanying part of it: the ongoing monitoring of the process via control charts while it produces.

Do I need SPC software, or is Excel enough?
Excel works as a starting point for a few characteristics. As soon as you monitor more than 5 to 10 characteristics, Excel hits its limits: no automatic pattern detection (Nelson rules), no real-time alarms, no connection to machine data. SPC software inside an MES automates data capture, checks continuously for rule violations and links process data with order and machine data.

How are SPC and OEE related?
SPC directly improves the quality factor of OEE. Stable processes produce less scrap and rework, which raises First Pass Yield. Indirectly, SPC also improves availability, because fewer stoppages are caused by quality problems (sorting, rework, machine stops on scrap).