Volume no :2, Issue no: 1, July (2010)

CONCRETE STRENGTH ESTIMATION USING RELEVANCE VECTOR MACHINES

Author's: Jale Tezcan, Qiang Cheng and Arif Cekic
Pages: [61] - [76]
Received Date: May 26, 2010
Submitted by:

Abstract

Concrete is a highly versatile structural material that has been used for many centuries. Recently, there has been a marked increase in the use of supplementary cementitious materials and chemical admixtures. This increase, while facilitating the adjustment of concrete properties for specific uses, has also complicated the task of estimating the relevant performance measures from the mix ingredients. For structural engineering purposes, the most relevant parameter is the 28-day compressive strength. An ideal strength estimation method should not only allow the use of different sets of predictive variables, but also account for prediction uncertainty. This paper proposes a probabilistic approach for the estimation of the 28-day compressive strength of concrete, based on the state-of the-art relevance vector machine (RVM). An RVM belongs to the class of sparse kernel classifiers, which are powerful tools in classification and regression. Recently, the support vector machine (SVM), a sparse kernel model with a functional form identical to that of the RVM, has already proved successful in modelling concrete behavior. The RVM-based approach proposed in this paper offers several advantages over the ones based on SVM. First, by using a probabilistic kernel, an RVM provides information about prediction uncertainty. Second, compared to an SVM, an RVM uses fewer kernel functions for comparable generalization performance, providing a sparser representation. Third, the RVM model parameters are automatically determined from the training set, unlike the SVM algorithm, where the selection of model parameters typically involves cross-validation. A demonstrative application comparing the two approaches is presented. The results from this study confirm the advantages of the proposed method and demonstrate its effectiveness.

Keywords

concrete compressive strength, mix design, relevance vector machines.