Influência da distribuição de probabilidade dos dados no controle estatístico da qualidade de obras e serviços rodoviários

Abstract

This study analyzed the influence of probability distribution of data on the statistical quality control of road construction and maintenance services, focusing on asphalt binder content (%CAP) and percent passing the No. 200 sieve (No. 200). The assumption of data normality, as adopted in standards such as DNER-PRO 277/97, was investigated through statistical analysis and computer simulations. The methodology involved collecting data from three highway projects in Ceará (Obra 1, Obra 2, and Obra 3), performing normality tests (ShapiroWilk, Kolmogorov-Smirnov, and Anderson-Darling), and fitting probability distributions. Additionally, the Monte Carlo method was employed to simulate different scenarios and evaluate the impact of data distribution on service acceptance rates. The log-logistic distribution provided the best fit for the %CAP data, whereas the normal distribution proved inadequate in most cases. For No. 200, no single distribution best fitted the data from all three projects. The log-normal distribution was the most appropriate for Obra 1, while the normal distribution showed a better fit for Obra 2, and the logistic distribution was the best fit for Obra 3. The results demonstrated that assuming normality as a premise in acceptance criteria can lead to significant distortions. In Obra 2, for instance, the acceptance rate for %CAP varied by up to 18.0%, depending on the adopted distribution, showing a rejection bias, increasing the risk for the contractor beyond the estimated level. Thus, adopting more representative statistical distributions can enhance quality control in asphalt paving, reducing errors in service acceptance or rejection and minimizing technical and financial impacts. In light of these findings, it is recommended to review regulatory criteria and expand studies on the subject under different paving conditions.