Risk-oriented quality control model for as-built and acceptance documentation at the stages of construction, reconstruction, and commissioning using methods of statistical sampling and stratification

Abstract

The article presents a scientifically grounded model of risk-oriented quality control for as-built and acceptance documentation at the stages of construction, reconstruction, and commissioning, using methods of statistical sampling and stratification. The aim of the study is to develop a mathematical and organizational framework that allows replacing labor-intensive total document control with a selective yet statistically valid control based on assessing the risk of missing critical defects. The empirical base consists of datasets of as built and acceptance documentation from 42 investment and construction projects of various functional purposes for the period 2018-2023 (18,450 storage units), including concealed and critical works certificates, as-built drawings, laboratory test reports, and material passports; the data were extracted from digital archives and information systems using text mining technologies. The methodology relies on stratified sampling with optimal allocation according to Neyman, Pearson’s chi-square criterion and Fisher’s exact test, a hypergeometric distribution for estimating the probability of defect omission, as well as an adaptation of the FMEA method for ranking types of nonconformities by risk priority index. As a result, documentation strata were formed according to their influence on structural and fire safety, engineering networks, architectural solutions, and landscaping, for each of which statistical characteristics of defectiveness and recommended sampling ratios were determined; an inverse relationship between the criticality index and the average share of defective documents was demonstrated, and an inflection point of the risk function was identified around a 15% sampling rate for structurally significant documentation. A comparative experiment showed that the risk-oriented model allows reducing labor costs for control by 69,06% compared to full inspection while maintaining 89,4% defect detection efficiency and a controlled residual risk of about 10,6%, mostly associated with low-hazard nonconformities. The prospects for integrating the developed model into digital construction management platforms, common data environments, and BIM systems are discussed, enabling automated formation of control plans and a transition toward predictive analysis of as-built documentation quality.