Vol. 21 No. 3 (2026): Reliability: Theory & Applications

From Control to Catastrophe: A Common Thread in Modern Applied Statistics

This issue of Reliability: Theory & Applications presents a rare collection of studies that, at first glance, belong to incommensurable subject domains — microgrid power engineering, reliability theory, actuarial modeling, and organizational risk assessment at explosive facilities. Yet a careful reading reveals a common methodological axis running through all five papers: the problem of justified model selection under uncertainty and the price of that choice.

The opening paper by Belhouchet et al. sets the tone for the entire collection. Comparing three microgrid control strategies — classical droop control, model predictive control (MPC), and fuzzy logic control (FLC) — the authors move beyond a simple comparison of technical indicators. The application of two-way analysis of variance (ANOVA) and a regression model for predicting energy cost transforms an engineering experiment into a full-fledged statistical study. The key finding — that MPC achieves the best absolute performance while FLC demonstrates practically comparable robustness at substantially lower computational cost — reads as an invitation to discuss where the boundary between optimality and expediency lies. It is precisely this question that becomes the leitmotif of the issue. The papers on distribution theory — the new extended exponential Rayleigh (NEER) distribution by
Kumar et al., the exponentiated Gumbel power function (EGuPF) distribution by Uwadi et al., the induced Xgamma distribution by Ruidas et al., and the resilience distribution based on the Marshall–Olkin framework (RD) by Shameera and Bindu — form a second line of meaning. All four works are devoted to extending classical distributions in order to enhance their flexibility: capturing J-shaped, bathtub-shaped, and non-monotonic hazard functions, modeling right-skewed data, and accounting for the "resilience" effects of a system. What they share is meticulous attention to estimation procedures: maximum likelihood, least squares, maximum product of spacings, and their comparison by mean squared error in simulation studies. Notably, the paper by Ruidas et al. includes a dedicated section proving model identifiability — an issue that often remains behind the scenes in applied literature, yet without which any parameter estimate loses its meaning. 

The paper by Gvozdev and Ovsyanik transports us to an entirely different plane — organizational risks at explosion- and fire-hazardous enterprises. The "conditional bit" measure proposed by the authors represents an original attempt to quantify the contribution of qualitatively described organizational causes to a realized hazardous event. The use of an additive weighted model and an interval scale, as well as Kendall's coefficient of concordance for assessing expert agreement, demonstrates how a rigorous statistical apparatus can be adapted to weakly structured subject domains. This work is an important reminder that the reliability of technical systems is not reducible to equipment failures; it is to a significant extent determined by the quality of management and the human factor.

The issue closes with a short but conceptually important note by Schabe, in which the Akaike information criterion is applied to compare parametric models with artificial intelligence models. The author's conclusion is disheartening for adherents of "black boxes": as sample size grows, the information loss of AI models increases as O(n log n), whereas a parametric model, if correct, retains a linear dependence. This does not mean that AI is useless — rather, it points to its proper place in the research process: as a tool for preliminary exploration, not for final inference.

The common thread uniting all the papers can be formulated as follows: every model is a compromise between accuracy and complexity, and statistics is the discipline that makes this compromise explicit. Whether it is the choice between MPC and FLC in a microgrid, between MLE and MPS in parameter estimation, or between a parametric and a non-parametric approach — everywhere it is a matter of the same thing: the need for a justified, verifiable, and reproducible choice.

The editorial board hopes that the materials presented here will resonate both with reliability theory specialists and with researchers working at the intersection of statistics, engineering, and risk management. Interdisciplinarity is not a fashionable slogan but a necessary condition for solving the problems that the modern technosphere poses to us.

Editorial Board