Keywords: neural networks, modelling, prediction, concrete strength, metakaolin, fly ash.

As the most widely used and versatile construction material in the world, concrete has many advantages in comparison with other construction materials, including low cost, adaptability and applicability under many conditions. Over the last forty years, there has been growing interest in the use of by-products to partially replace Portland cement (PC) in concrete to assist the conservation of natural resources. One of the most widely used materials is fly ash (FA), which is industrial by-product from bituminous coal-fired power stations. In recent years the utilisation of calcined clay in the form of metakaolin (MK) as a pozzolanic addition for concrete has received considerable interest in durability enhancement. The use of FA and MK improves the properties of concrete.

The investigation into optimising the contributory effects of FA and MK on concrete strength by introducing them in combination has been carried out by Bai et al [1]. Data obtained on PC-FA-MK well reflected the effect of various replacement levels of FA and MK on the strength. It would be more useful to develop reliable predictive models based on the data to help designers to select FA-MK combinations in performance terms in a reliable but economic manner.

Artificial neural networks (ANN) are being used to solve a wide variety of problems in civil and structural engineering [2,3]. Unlike expert systems [4], ANNs are ideally suited for such problems in dealing with data as an artificial neural network can learn, and therefore can be trained to find solutions, classify data, and forecast future events.

This paper presents neural network models for the prediction of strength of concrete incorporating FA and MK. All developed neural networks and their evaluation statistics are summarised in Table 61.1. In Table 61.1, all R-values confirm good network models and fit between the target values and output values. The values of Avg. Abs., Max. abs. and RMS indicate clearly that the developed neural nets are reliable. Neural network NN-S can be used as a universal model for predicting compressive strength at any age with high accuracy. However, other networks could give even better prediction if prediction of strength at specific age at 28, 90 days and 1 year only is needed.

On the basis of the models developed, iso-strength maps were plotted using trained neural networks. This makes it possible for the designer to produce mixtures with various blend compositions for a given range of strength specification. In addition, the models for predicting long-term strength with or without early strength results were recommended to save time and cost for construction contractors.

1

Bai, J., Sabir, B.B., Wild, S., Kinuthia, "Strength development in concrete incorporating PFA and Metakaolin", Magazine of Concrete Research, 2000, 52, No.3, 153-162, ISSN0024-9831.

2

Bai, J., Wild, S., Ware, J.A., Sabir, B.B. "Using neural networks to predict workability of concrete incorporating metakaolin and fly ash", Advances in Engineering Software. doi:10.1016/S0965-9978(03)00102-9

3

Flood, I. "Similating the construction process using neural networks", Pro. of the 7th International Symposium on Automation and Robotics in Construction, ISARC, Bristol Polytechnic, Bristol, UK, 1990.

4

Byars, E.A., Bai, J., Dhir, R.K. CONEX: "An expert system for concrete mix design and durability estimation", Proc. Int. Conf. on Information Technology in Civil and Structural Engineering Design ITCSED' 96, 1996, 137-142, Inverleith Spottiswoode, Edit. I.A.M.Macleod, B.Kumar and A.Retik, Edinburgh, 1996.

purchase the full-text of this paper (price £20)

go to the previous paper
go to the next paper
return to the table of contents
return to the book description
purchase this book (price £82 +P&P)