Skip to contents

Monte Carlo Uncertainty on Risk Priority Numbers

Source: R/risk_monte_carlo.R
risk_monte_carlo.Rd

An FMEA scores each risk with single integer values for probability, impact, and detectability, but those assessments are judgements with uncertainty. This simulation treats each score as uncertain, draws many possible registers, and reports the resulting distribution of Risk Priority Numbers, both per risk and for the register as a whole. It turns a single RPN into a range and a probability of being critical, which is what a risk review actually wants to know.

Usage

risk_monte_carlo(
  register,
  n = 10000,
  spread = 1,
  config = risk_config_default(),
  seed = NULL
)

Arguments

register

A risk_register from create_risk_register(), or a data.frame with risk_id, probability, impact, and detectability.

n

Number of Monte Carlo iterations. Default 10000.

spread

Standard deviation of the per-component uncertainty, in scale points. Default 1. Use 0 for a deterministic result.

config

A risk_config from risk_config_default(), used for the critical threshold.

seed

Optional integer seed. When supplied the random state is set for the call and restored on exit, so the result is reproducible without disturbing the caller's RNG stream.

Value

A list of class "risk_monte_carlo" with n, spread, crit_threshold, point_total (the deterministic register RPN), total (named numeric summary of the total RPN distribution), total_draws (the simulated totals, for plotting), and per_risk (a data.frame with risk_id, rpn, mc_mean, mc_sd, p05, p50, p95, and prob_critical).

Details

For each draw, every FMEA component is sampled from a normal distribution centred on its assessed value with standard deviation spread (in scale points), then rounded and clamped to the 1–5 scale. spread = 0 reproduces the deterministic RPN. The per-draw RPN is probability * impact * detectability, and the register total is their sum. The critical threshold is the lower bound of the top RPN band in config.

Examples

risks <- data.frame(
  risk_id = c("R001", "R002", "R003"),
  description = c("Missing SDTM vars", "Unmapped derivations", "Define drift"),
  probability = c(4, 3, 2), impact = c(5, 4, 3), detectability = c(2, 3, 4)
)
rr <- suppressMessages(create_risk_register(risks))
mc <- risk_monte_carlo(rr, n = 2000, seed = 1)
mc$total
#>  mean    sd   p05   p50   p95   min   max 
#>  96.1  32.0  50.0  92.0 156.0  21.0 230.0 
mc$per_risk
#>   risk_id rpn mc_mean mc_sd p05 p50 p95 prob_critical
#> 1    R002  36    35.1  19.8   9  32  75         0.048
#> 2    R001  40    36.9  19.2  12  34  75         0.031
#> 3    R003  24    24.1  15.5   6  20  50         0.011