Concrete Strength Prediction Using Artificial Neural Network Machine Learning Algorithm

Abstract

Compressive strength of concrete is an important parameter in the design of reinforced concrete structure and it is also used by industry to show the performance of the concrete. However, it is difficult to predict compressive strength of concrete since it is affected by many factors such as age, ratio of water to cement conditions of curing and compaction. This study is to predict the concrete strength using artificial neural network machine learning algorithms. The dataset of this study will be collected from the published literature that consist 191 rows of and focused on the parameter that will affect the compressive strength of the concrete. The measurable parameter for input variable consists of water, cement, fine aggregate, coarse aggregate, various types of admixtures and et cetera. Besides that, from this parameter it will extend into a few variables such as types of aggregate and type of sand. Compressive strength categorized as an output variable. Artificial neural network is one of the algorithms used in machine learning for modeling the data. An artificial neural network model will be developed for predicting the compressive strength of concrete. The modeling framework can be carried out by using KNIME software without writing a programming code. Dataset will split into 2 partitions: 80% for training set and 20% for testing set. Some input variables are removed due to high correlation to each other to achieve the best possible performance. The result shown 5 hidden layer and 14 hidden neurons per layers performed the best neural network architecture in this study. The R-squared value for the generated model is 0.838, it is considered high and quite good for the model. The graph shown predicted strength obtained from the model is quite similar for actual strength. It means that the model has a reliable prediction capability. With the help of this model, it can be used to verify if the designed proportion of the design mix can fulfil the requirement for target strength and do not required to produce a lot of concrete sample in lab.