Compressive strength prediction models of lightweight aggregate concretes using ultrasonic pulse velocity

Replacement of natural coarse aggregate with lightweight aggregate (LWA) offers not only the specific properties of concrete such as thermal, acoustic properties, or lighter weight concrete but is also dealing with wasted materials recycling and natural resources depletion. Ultrasonic pulse velocity (UPV) is closely correlated to both mechanical and physical properties of concrete and it is introduced and used to predict the compressive strength of concrete structures. This paper proposes two prediction models for compressive strength of lightweight aggregate concrete, a regression model and a back-propagation neural network (BPNN) incorporating UPV for different accuracy requirements. The regression model is a further generalised regression model for estimating compressive strength from UPV of lightweight aggregate concrete (LWAC) that can be adapted to different testing systems. The BPNN model, optimised by a genetic algorithm, is also employed for more accurate predictions compared to the regression model. To achieve this, a database which comprises a total of 603 sets of data from 26 different studies was compiled. In addition, the database also involves wide ranges of sizes of coarse aggregates (4 mm-40 mm) and LWAs (0.65 mm-30 mm), volume ratios of coarse aggregate to binder (0.53-9.66) and sand to aggregate (0 to 5.99), LWA volume fraction (0-100%), water to binder ratio (0.3-0.89), and curing time (1 day to 120 days). Statistical results indicate that regression model and BPNN model can produce a reasonable estimation of compressive strength for lightweight aggregate concrete but with different accuracy level. Hence, the two proposed models can be used to adapt to different expectations of accuracy in different situations. (c) 2021 Elsevier Ltd. All rights reserved.

Automated summary

This paper introduces two models for predicting the compressive strength of lightweight aggregate concrete, a regression model and a back-propagation neural network (BPNN). Statistical results show that both models can provide reasonable estimations of compressive strength, with the BPNN model offering more accurate predictions compared to the regression model.