Quality prediction for concrete manufacturing

Abstract The problem of extracting information from several sources of information is a very important issue in intelligent systems. In the field of manufacturing concrete — one of the most common construction materials — in Hong Kong, this problem is quite common. There is no direct formulation of concrete mix for specified properties, and all of the mixes are designed by experience and subject to quality inconsistency due to many possible mixing variations. This paper describes an application of neural network techniques to the acquisition of qualitative knowledge during the production of concrete. It shows the capabilities of the developed model for the analysis and representation of production data and prediction of the quality of concrete under different mixing formulations. The simulation results indicate that the neural network's prediction is generally superior to those of conventional methods which often require time-consuming trial mixes for verifying the specified properties before mass production for use.