Logistic regression analysis for confidence level estimates of additive percentage at high service temperatures

• Published in: The international journal of pavement engineering Vol. 22; no. 7; pp. 912 - 926
• Main Authors: Babazadeh, Abbas, Jafari, Mohammad
• Format: Journal Article
• Language: English
• Published: Abingdon Taylor & Francis 07.06.2021; Taylor & Francis LLC
• Subjects: Additives; Asphalt; Butadiene; Confidence intervals; confidence level; High temperature; Laboratories; logistic regression; Parameters; Pavements; Performance grading; Regression analysis; repeatability and reproducibility conditions; Reproducibility; required additive percentage; Statistical analysis; Statistical methods; Statistical models; Styrenes; Temperature; Thin films
• ISSN: 1029-8436; 1477-268X
• DOI: 10.1080/10298436.2019.1652827

A new method is proposed to consider confidence level in estimating the required percentage of an additive, Styrene butadiene styrene (SBS) or polyphosphoric acid (PPA), to modify a PG 58-22 asphalt binder considering high temperature (HT) performance grade. The Superpave HT performance grading (PG) test was performed on repetitive rolling thin film oven (RTFO) samples to capture the variability in the measurement of the rutting parameter G*/sinδ. For each case of SBS and PPA additives, the logistic regression analysis is used to establish a statistical model that relates additive percentage and service temperature to the confidence level that the rutting parameter of a sample is equal to 2.2 kPa or greater. It is shown that using the integer grades of the Superpave system largely overestimates the required additive percentage when the pavement temperature is far below the next higher grade. The paper suggests an alternative method based on the fitted logistic models, which can offer far less additive when the confidence level is taken a few per cent lower than 100%. It is also shown that performing the test in two laboratories, compared to a single laboratory, yields more significant variation in the measurements. In this case, Superpave grades method cannot ensure the high reliability of estimation when the pavement temperature is very close to the next higher grade, while the suggested method provides estimations at very strong confidence levels.