Inferensys

Glossary

Statistical Process Control (SPC)

A quality control methodology that uses statistical methods to monitor a process, distinguish between common and special cause variation, and signal when corrective action is needed.
Operations room with a large monitor wall for system visibility and control.
QUALITY METHODOLOGY

What is Statistical Process Control (SPC)?

A foundational quality control methodology using statistical methods to monitor, control, and improve a process through the identification and reduction of variation.

Statistical Process Control (SPC) is a quality control methodology that applies statistical methods to monitor a process, distinguish between common cause variation (inherent noise) and special cause variation (assignable events), and signal when corrective action is needed to prevent defective output. It provides a data-driven framework for achieving process stability and capability.

SPC relies on control charts, which plot process data over time against calculated control limits. When a data point falls outside these limits or exhibits non-random patterns, the process is deemed statistically out of control, triggering a root cause investigation. This methodology is foundational to modern adaptive process control loops, where automated SPC triggers feed directly into AI-driven optimization systems.

Statistical Process Control

Core Components of SPC

Statistical Process Control (SPC) is a quality control methodology that uses statistical methods to monitor a process, distinguish between common and special cause variation, and signal when corrective action is needed.

01

Control Charts

The primary visual tool of SPC. A control chart plots process data over time against a calculated centerline and statistically derived upper and lower control limits. These limits are not specification limits; they represent the voice of the process. Data points falling within the limits indicate common cause variation inherent to the process. Points outside the limits, or non-random patterns within them, signal a special cause variation that warrants investigation and corrective action.

  • X-bar and R charts: Monitor the mean and range of subgroups for variable data.
  • p-charts and c-charts: Track the proportion or count of defective units for attribute data.
  • Individuals and Moving Range (I-MR) charts: Used when only a single measurement per time period is available.
02

Common Cause vs. Special Cause Variation

The foundational distinction in SPC. Common cause variation is the natural, inherent variability in a process due to the cumulative effect of many small, unavoidable factors. It is predictable within statistical bounds and requires a fundamental process change to reduce. Special cause variation arises from specific, identifiable sources external to the stable system, such as a broken tool, a new batch of raw material, or an untrained operator. It is unpredictable and makes the process unstable.

  • Common cause: Requires management action to improve the system.
  • Special cause: Requires local action to find and eliminate the specific assignable cause.
  • Confusing the two leads to tampering, which increases variability.
03

Process Capability Analysis

A set of indices that quantify how well a statistically stable process output conforms to customer specification limits. This analysis is only valid after a process is in a state of statistical control. Key indices include Cp, which measures the potential capability if the process were perfectly centered, and Cpk, which accounts for both spread and centering relative to the nearest specification limit. A Cpk of 1.33 is a common benchmark for a capable process.

  • Cp: (USL - LSL) / 6σ, measuring potential capability.
  • Cpk: min[(USL - μ) / 3σ, (μ - LSL) / 3σ], measuring actual capability.
  • Pp and Ppk: Analogous indices calculated using overall long-term standard deviation, not within-subgroup variation.
04

Western Electric Rules

A set of decision rules for interpreting patterns on a control chart to detect non-random, out-of-control conditions even when all points fall within the control limits. These rules increase the sensitivity of the chart to small process shifts. A single point outside the 3-sigma limit is the most basic rule, but others include patterns like a run of seven consecutive points on one side of the centerline, or a trend of six points steadily increasing or decreasing.

  • Rule 1: One point beyond the 3σ limit.
  • Rule 2: Nine points in a row on one side of the centerline.
  • Rule 4: A run of 14 points alternating up and down, indicating systematic over-adjustment.
05

Rational Subgrouping

The strategy for collecting samples to construct control charts. A rational subgroup is a sample selected such that the variation within a subgroup is primarily due to common causes, while the variation between subgroups can capture special causes. Proper subgrouping is critical for the chart to function correctly. Subgroups are typically collected consecutively over a short time interval to minimize within-group variability.

  • Within-subgroup variation: Used to calculate control limits; represents short-term common cause noise.
  • Between-subgroup variation: Used to detect shifts in the process mean over time.
  • Incorrect subgrouping can produce control limits that are too wide or too narrow, rendering the chart useless.
06

Pre-Control

A simplified, statistically less powerful alternative to full control charts, used for quick setup validation and ongoing monitoring of highly capable processes. The specification range is divided into zones: a green zone in the center, yellow zones at the edges, and red zones outside the spec limits. Simple rules based on consecutive part measurements determine if the process is set up correctly and remains stable, without requiring plotted charts or calculations.

  • Green Zone: The middle half of the specification range.
  • Yellow Zone: The remaining area within the specification limits.
  • Rule: To qualify a setup, five consecutive parts must fall in the green zone. For monitoring, two consecutive yellow parts signal an adjustment.
STATISTICAL PROCESS CONTROL

Frequently Asked Questions About SPC

Clear, technically precise answers to the most common questions about Statistical Process Control, its mechanisms, and its role in modern software-defined manufacturing.

Statistical Process Control (SPC) is a quality control methodology that uses statistical methods to monitor a process, distinguish between common cause variation (inherent noise) and special cause variation (assignable events), and signal when corrective action is needed. It works by continuously sampling process outputs, plotting them on a control chart with statistically calculated upper and lower control limits, and applying run rules to detect non-random patterns. When a data point falls outside these limits or violates a run rule, the process is deemed unstable, triggering an investigation to identify and eliminate the assignable cause before defective output is produced. This shifts quality assurance from post-hoc inspection to real-time prevention.

VARIATION TAXONOMY

Common Cause vs. Special Cause Variation

A comparative analysis of the two fundamental categories of process variation defined by Walter Shewhart, distinguishing inherent system noise from assignable external disruptions.

FeatureCommon Cause VariationSpecial Cause Variation

Origin

Inherent to the process design

External to the system or a specific event

Predictability

Stable and predictable within control limits

Unpredictable and sporadic

Statistical Distribution

Normal (Gaussian) distribution

Non-random, skewed, or bimodal patterns

Frequency

Chronic; always present

Sporadic; intermittent

Impact Magnitude

Typically small, consistent variance

Potentially large, disruptive shifts

Corrective Action

Requires fundamental process redesign

Requires immediate local investigation

Control Chart Signal

Points within ±3 sigma limits

Points outside limits or non-random runs

Management Responsibility

System-level management intervention

Local operator or engineering correction

Prasad Kumkar

About the author

Prasad Kumkar

CEO & MD, Inference Systems

Prasad Kumkar is the CEO & MD of Inference Systems and writes about AI systems architecture, LLM infrastructure, model serving, evaluation, and production deployment. Over 5+ years, he has worked across computer vision models, L5 autonomous vehicle systems, and LLM research, with a focus on taking complex AI ideas into real-world engineering systems.

His work and writing cover AI systems, large language models, AI agents, multimodal systems, autonomous systems, inference optimization, RAG, evaluation, and production AI engineering.