Simulation-Assisted Statistical Process Control
Also known as: Simulation-based SPC, Monte Carlo SPC, SA-SPC, Simulation-integrated SPC
Simulation-assisted statistical process control (SA-SPC) combines computer simulation — typically Monte Carlo or discrete-event simulation — with classical SPC methods to design, test, and calibrate control charts and monitoring schemes before or alongside deployment on a real production process. Rather than relying solely on closed-form analytical assumptions, SA-SPC uses simulated data to evaluate chart performance under realistic, often non-normal process conditions.
Read the full method
Sign in with a free account to read this section.
Method map
The neighbourhood of related methods — select a node to explore.
When to use it
Use SA-SPC when standard analytical SPC assumptions are unlikely to hold: non-normal process distributions, small Phase I sample sizes causing parameter estimation uncertainty, autocorrelated observations, or complex multi-stage processes where closed-form ARL expressions do not exist. It is also the right choice when designing a novel chart type or comparing competing schemes before committing to a monitoring strategy. Do NOT use SA-SPC as a substitute for collecting sufficient real process data — simulation calibrates the scheme but cannot replace Phase I baseline data from the actual process. Avoid when the process is well-characterised, truly normal, and standard Shewhart chart analytical limits are known to be adequate.
Strengths & limitations
- Evaluates SPC scheme performance under realistic non-normal, autocorrelated, or multivariate process conditions where analytical ARL expressions do not exist.
- Allows direct comparison of multiple chart designs (Shewhart, CUSUM, EWMA) on equal footing before deployment.
- Quantifies the impact of parameter estimation error — a known weakness of standard Phase II charts — by simulating estimation from small Phase I samples.
- Can model complex multi-stage or batch manufacturing processes that resist closed-form analysis.
- Produces empirical ARL distributions rather than just point estimates, giving a fuller picture of chart performance variability.
- Quality of results depends entirely on the accuracy of the input simulation model; a poorly fitted process model yields misleading ARL estimates.
- Computationally intensive: achieving low Monte Carlo error on ARL estimates typically requires tens of thousands of replications per scenario.
- Requires statistical and simulation modelling expertise beyond standard SPC practice.
- Simulation results may not transfer to the real process if the process changes after the simulation study was conducted.
Frequently asked
How many simulation replications are enough to estimate ARL reliably?
A common rule of thumb is at least 10,000 replications per scenario to keep the coefficient of variation of the ARL estimate below about 1–2%. For high-ARL0 scenarios (e.g., ARL0 around 370) the variance of run length is large, so 50,000 or more replications may be needed for acceptable precision. Always report the simulation standard error alongside the ARL estimate.
Can SA-SPC be used for multivariate processes?
Yes — this is one of its main strengths. Analytical ARL expressions for multivariate charts such as Hotelling's T-squared are complex and often intractable under non-normality or estimation error. Simulation allows straightforward evaluation of multivariate chart performance by generating correlated multivariate observations from the fitted process distribution.
What software is typically used?
R (particularly the spc, MSQC, and qcc packages), Python, MATLAB, and Arena or Simio for discrete-event simulation are all commonly used. For Monte Carlo ARL estimation, purpose-built R scripts using random-number generation from fitted distributions are the most common research-grade approach.
Is SA-SPC the same as using software to draw control charts?
No. Drawing control charts in software such as Minitab or SPC software simply applies standard formulae to real data. SA-SPC specifically means using stochastic simulation to design, evaluate, and calibrate the monitoring scheme — typically before or in parallel with real deployment — rather than mechanically applying a standard chart.
When should I prefer analytical ARL formulas over simulation?
Prefer analytical solutions when the process is well-characterised as normal and sample sizes are large enough that parameter estimation error is negligible. Analytical solutions are exact and require no computational effort. SA-SPC adds value specifically when these conditions break down — non-normality, small samples, autocorrelation, or novel chart designs lacking closed-form solutions.
Sources
- Montgomery, D. C. (2009). Introduction to Statistical Quality Control (6th ed.). Wiley. ISBN: 978-0470169926
- Jensen, W. A., Jones-Farmer, L. A., Champ, C. W., & Woodall, W. H. (2006). Effects of parameter estimation on control chart properties: A literature review. Journal of Quality Technology, 38(4), 349–364. link ↗
How to cite this page
ScholarGate. (2026, June 3). Simulation-Assisted Statistical Process Control. ScholarGate. https://scholargate.app/en/experimental-design/simulation-assisted-statistical-process-control
Which method?
Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.
- Control chartExperimental design↔ compare
- Design of experimentsExperimental design↔ compare
- MONTE-CARLO-SIMULATIONDecision-making↔ compare
- Process Capability AnalysisStatistics↔ compare
- Six Sigma DMAICQuality Management↔ compare
- Statistical Process ControlExperimental design↔ compare