UAnalytics 0.28.0

UAnalytics

.NET bindings for the u-analytics statistical process control and reliability engine.

Features

• Control charts: X̄-R, X̄-S, I-MR for variables; p, np, c, u and Laney p′ for attributes, with Phase I baselines for Phase II limits
• Run rules: applied to any series against limits you supply, or to a chart's own
• Process capability: Cp, Cpk, Pp, Ppk, Cpm, and a percentile method for non-normal data
• Measurement systems analysis: Gage R&R by the Average & Range method and by ANOVA, each returning the charts the method plots
• Reliability: Weibull fitting by maximum likelihood (WeibullMle) and median-rank regression (WeibullMrr); WeibullReliability — MTBF, R(t), h(t) and B-lives, with confidence intervals when given an MLE fit (WeibullReliability(fit, …)) or its Covariance
• Non-normal capability: BoxCoxCapability, and SigmaToPpm / PpmToSigma on the 1.5σ-shift convention
• Event-time trend: whether one unit's events (failures of a repairable system, incidents) arrive at a changing rate — LaplaceTrendTest, MilHdbk189Test, and the power-law process (Crow-AMSAA) fit PowerLawProcessFit. Pass the end of observation, or null when it stopped at the last event
• Cumulative counters: CounterIncrements — how much a meter or counter advanced between readings, each step Normal, Declared (a reset or replacement you pass as CounterDiscontinuity), Wrap (through the register's modulus) or Unexplained (no increase guessed) — and CounterTotals over intervals
• Hypothesis tests: t (one-sample, Welch, paired), Mann-Whitney, Wilcoxon, Jarque-Bera, Shapiro-Wilk, Anderson-Darling, Mann-Kendall, ANOVA, Kruskal-Wallis, Levene, Bartlett, χ² goodness of fit and independence, Fisher's exact; Bonferroni and Benjamini-Hochberg adjustment
• Stationarity: AdfTest (augmented Dickey-Fuller)
• Change detection: changepoints in one series or several aligned ones (DetectChangepointsMulti), Cusum and Ewma charts for small sustained shifts, period estimation, spectral residual scoring
• Correlation and regression: correlation matrices, simple linear regression
• Distribution fitting: best-fit selection across candidate distributions

Installation

dotnet add package UAnalytics

Usage

using UAnalytics;

using var analytics = new AnalyticsClient();

XbarRChartResult chart = analytics.XbarRChart(
[
    [10.1, 10.3, 9.8],
    [10.0, 10.2, 10.1],
    [9.9,  10.4, 10.2],
]);

Console.WriteLine($"UCL {chart.XbarUcl:F3}, sigma {chart.SigmaHat:F3}, in control: {chart.InControl}");

TrendTestResult trend = analytics.MilHdbk189Test([10, 19, 27, 34, 40, 45, 49, 52, 54, 55], end: null);
if (trend.Direction == TrendDirection.Increasing && trend.PValue < 0.05)
    Console.WriteLine("Failures are coming more often.");

Every method returns a record (XbarRChartResult, CapabilityResult, TrendTestResult, …) whose properties are the engine's result fields. A value the inputs do not define — a short-term index without a within-subgroup sigma, a t ratio of an exact fit — is null on a nullable property. Closed vocabularies are enums: RunRule (the run tests a chart applies and reports), RunRuleSet (the named sets — Shewhart, the default; WesternElectric; Nelson), TrendDirection, SigmaSource, GageStatus, and the options AdfModel, PeltCost and CorrelationMethod.

Input that cannot be analysed is refused rather than approximated: AnalyticsException carries a stable ErrorCode — the error body's code (insufficient_data, parameter_out_of_range, value_not_finite, …), the Parameter it is about, the Index of the offending element, the whole error body in Details and the native status in Status. For parameter_out_of_range, Details holds got, the inclusive bounds min / max (null on a side that is open or unbounded) and, only on an open side, exclusive_min / exclusive_max. Every code and the fields it carries are listed in the crate README — the WebAssembly and .NET bindings refuse with the same codes, and GetCodes() returns them as data — Codes(Refusals, Violations), each refusal with the fields beside it — from the table the engine refuses from.

Trimming and NativeAOT

The client uses no reflection: requests are built as JSON nodes and results are read through source-generated serialization, and the package is marked IsAotCompatible. It runs unchanged in trimmed and NativeAOT applications, and in .NET file-based apps.

Platforms

The package carries the native library for win-x64, linux-x64 and linux-arm64 (glibc 2.39 or later), osx-x64 and osx-arm64; no separate install is needed.

License

MIT