pirão

Probabilistic Inference for Reliability Analysis from Observed data: Bayesian reliability estimation with Stan, with a local graphical interface. A sibling of farofa (simulation of repairable systems) and faultree (exact fault-tree analysis).

Pick a likelihood and, separately, a prior for each parameter; supply bounds and data. pirão writes the Stan program, compiles and caches it, samples it, and returns reliability at your mission times with 90% intervals, the reliability curve against the data's own Kaplan–Meier steps, and each parameter's posterior over the prior that was actually in force.

pip install pirao
pirao install-stan     # once: compiles CmdStan (a few minutes; needs a C++ compiler)
pirao gui              # opens the interface at http://localhost:8765

What it covers

• Likelihoods: exponential, Weibull, lognormal, gamma (time to failure, with right censoring); per-demand Bernoulli and binomial (demands).
• Priors: lognormal, normal, Student t, gamma, inverse gamma, exponential, Weibull, beta, half-normal, half-Cauchy, uniform, flat and Jeffreys, with any bounds. The interface shows the prior after the bounds and warns when they discard much of it.
• Relevance weights per row, for data from similar but not identical equipment (a fractional pseudo-posterior, labelled as such).
• Ready-made examples on public data (Proschan 1963 air conditioning; Lieblein and Zelen 1956 ball bearings) and on clearly illustrative numbers.
• Interface in Portuguese and English.
• Reproducible runs: a .zip with the specification, data, seed, Stan program and toolchain versions.

From the shell or Python

pirao examples                               # list the examples
pirao example relay-binomial --out relay     # spec.json, data.csv, sampler.json
pirao run relay/spec.json --data relay/data.csv --sampler relay/sampler.json --out relay/run

from pirao.core.examples import get_example
from pirao.core.results import summarize
from pirao.core.sample import run
from pirao.core.spec import ModelSpec, ParamSpec, PriorSpec

# 23 ball bearings, millions of revolutions to fatigue (Lieblein and Zelen 1956);
# rows are {"time": ..., "failure": 1 or 0}, one per unit.
rows = get_example("bearings-weibull")["rows"]

spec = ModelSpec(
    likelihood="weibull",
    mission_times=[25, 50],
    time_unit="Mrev",
    params={
        "shape": ParamSpec(lower=0.0, prior=PriorSpec(family="half_normal", hyper={"sigma": 2})),
        "scale": ParamSpec(lower=0.0, prior=PriorSpec(family="gamma", hyper={"shape": 2, "rate": 0.02})),
    },
)
print(summarize(run(spec, rows)).table[["mean", "q05", "q95"]])

More

The repository has the full README (data format, what the results mean, statistical caveats), the design notes, and a script that carries the posterior draws into farofa and faultree.

BSD 3-Clause licence. CmdStan (BSD 3-Clause) is downloaded by pirao install-stan, not redistributed.