Prediction of Compressive Strength of Concrete in Palestinian Governorates Using Machine Learning Techniques رسالة ماجستير

Abstract

In this thesis, Machine Learning (ML) techniques were used to predict the compressive strength of concrete in the Palestinian governorates. The datasets were collected from Palestinian laboratories and factories from seven Palestinian governorates, which consists of five subsets, and each sub dataset is related to a specific type of Palestinian concrete. The thesis work is divided into three phases: In the first phase, the process is divided into two parts, firstly; the implementation of clustering algorithms to the whole data of the Palestinian governorates. Secondly; implementation of clustering algorithms to each sub dataset that presents data in each governorate. The factors determining results showed that the Expectation- Maximization (EM) algorithm is completely identical to the Kohonen Self-Organizing Maps (KSOM) algorithm. The results from these two algorithms are similar, thus these two algorithms were used to determine the main factors that affect the concrete compressive strength (PCCS). The results obtained by using K-mean clustering algorithms show that they are more accurate prediction for improving the concrete compressive strength. The second part is the use of ML techniques to classify the compressive strength of concrete, where three methods were used: MLPNNs, Support Vector Machine (SVM), and Ensemble Algorithm. The accuracy results were 93.5%, 80.4% and 90.2% respectively for B200 concrete, and the classification results for B250 concrete were 90.0%, 66.5% and 75.5% respectively. For B300 concrete, the classification results were 93.3%, 68.3% and 79.2% respectively. The classification results were 90.6%, 83.3% and 85.6% respectively for B350 concrete, and the classification results were 90.0%, 80.6% and 78.6% respectively for B400 concrete. VI The results showed that the MLPNNs using Levenberg–Marquardt algorithm are the most accurate for each type of concrete. The classification models were applied on the dataset which was collected from Palestinian governorates laboratories after it removes other parameters and remains only factors that affect Palestinian Concrete Compressive Strength (PCCS) obtained from clustering algorithms. The new dataset was implemented on the classification models like MLPNNs, linear support vector machine, and Ensemble algorithm show the results are close to those obtained previous experiences that were implemented on pervious datasets and the accuracy results for the new dataset were 92.5%, 75.4% and 88.0% respectively. The final part depends on the use of machine learning techniques to predict the compressive strength of concrete using three different Artificial Neural Networks (ANNs) techniques; Multilayer Perceptron Neural Networks (MLPNNs), Radial Basis Function Neural Networks (RBFNNs), and Recurrent Neural Networks (RNNs). It is found that the ANNs Techniques are effective tools for predicting the Compressive Strength of concrete. The mean square error (MSE) results were obtained from these ANNs models were 0.0107, 0.0064, and 0.0012 respectively where the MLPNNs using Levenberg–Marquardt model produce the best prediction result.