Alper Sezer

Soil clustering by fuzzy c-means algorithm

In this study, hard k-means and fuzzy c-means algorithms are utilized for the classification of fine grained soils in terms of shear strength and plasticity index parameters. In order to collect data, several laboratory tests are performed on 120 undisturbed soil samples, which are obtained from Antalya region. Additionally, for the evaluation of the generalization ability of clustering analysis, 20 fine grained soil samples collected from the other regions of Turkey are also classified using the same clustering algorithms. Fuzzy c-means algorithm exhibited better clustering performance over hard k-means classifier. As expected, clustering analysis produced worse outcomes for soils collected from different regions than those of obtained from a specific region. In addition to its precise classification ability, fuzzy c-means approach is also capable of handling the uncertainty existing in soil parameters. As a result, fuzzy c-means clustering can be successfully applied to classify regional fine grained soils on the basis of shear strength and plasticity index parameters.

Models for Prediction of Daily Mean Indoor Temperature and Relative Humidity: Education Building in Izmir, Turkey

In this research, several models were developed to forecast the daily mean indoor temperature (IT) and relative humidity values in an education building in Izmir, Turkey. The city is located at a hot–humid climatic region. In order to forecast the IT and internal relative humidity (IRH) parameters in the building, a number of artificial neural networks (ANN) models were trained and tested with a dataset including outdoor climatic conditions, day of year and indoor thermal comfort parameters. The indoor thermal comfort parameters, namely, IT and IRH values between 6 June and 21 September 2009 were collected via HOBO data logger. Fraction of variance (R2) and root-mean squared error values calculated by the use of the outputs of different ANN architectures were compared. Moreover, several multiple regression models were developed to question their performance in comparison with those of ANNs. The results showed that an ANN model trained with inconsiderable amount of data was successful in the prediction of IT and IRH parameters in education buildings. It should be emphasized that this model can be benefited in the prediction of indoor thermal comfort conditions, energy requirements, and heating, ventilating and air conditioning system size.

Simple models for the estimation of shearing resistance angle of uniform sands

Angle of shearing resistance is the key for the strength analysis of soils, since this parameter is commonly used for the description of shear strength of a soil. Many factors including soil mineralogy, particle shape, grain size distribution, void ratio, organic content, as well as water existence are effective on this parameter. Use of shear box tests for the determination of angle of shearing resistance is prevalent in geotechnical engineering practice, since it is rather successful in identification of shear strength of granular media. However, for cases in which shear box tests exhibit unreliable outcomes, alternative methods for the determination of this parameter could be beneficial. In this investigation, a number of nonlinear multiple regression, adaptive neuro-fuzzy inference systems, and artificial neural network (ANN) models are employed for the estimation of estimating the angle of shearing resistance of uniform sands by means of several grain size distribution, particle shape, and density parameters. Data including results of 132 shear box tests, particle shape identifiers, and grain size distribution parameters on uniform sands are used in the models. In training sessions, results of 104 tests are selected randomly and the results of remaining 28 tests are considered for testing sessions. The results revealed that the performance of a simple ANN architecture is sufficient for pre-evaluation of shearing resistance angle of uniform sands with the help of selected parameters. Since generalization of these models necessitates vast amount of experiments, great care should be dedicated on the assessment of similarity of training as well as testing data.

EFFECT OF FRACTAL DIMENSION ON THE STRAIN BEHAVIOR OF PARTICULATE MEDIA

In this study, the influence of several fractal identifiers of granular materials on dynamic behavior of a flexible pavement structure as a particulate stratum is considered. Using experimental results and numerical methods as well, 15 different grain-shaped sands obtained from 5 different sources were analyzed as pavement base course materials. Image analyses were carried out by use of a stereomicroscope on 15 different samples to obtain quantitative particle shape information. Furthermore, triaxial compression tests were conducted to determine stress–strain and shear strength parameters of sands. Additionally, the dynamic response of the particulate media to standard traffic loads was computed using finite element modeling (FEM) technique. Using area-perimeter, line divider and box counting methods, over a hundred grains for each sand type were subjected to fractal analysis. Relationships among fractal dimension descriptors and dynamic strain levels were established for assessment of importance of shape descriptors of sands at various scales on the dynamic behavior. In this context, the advantage of fractal geometry concept to describe irregular and fractured shapes was used to characterize the sands used as base course materials. Results indicated that fractal identifiers can be preferred to analyze the effect of shape properties of sands on dynamic behavior of pavement base layers.

Evaluation and use of clustering algorithms for standard penetration test data classification

Abstract The standard penetration test (SPT) is the most common test conducted in the field, and it is used to determine in situ properties of different soils. Although it is a matter of debate, these tests are also used for the determination of the consistency of fine-grained soils, whereby the test results can also be utilized to establish numerous empirical correlations to predict the strength of soils in the field. In this study, unsupervised clustering algorithms were employed to classify the SPT standard penetration resistance value (SPT-N) in the field. In this scope, shear strength and liquidity index parameters were used to classify the SPT-N values by taking the classification system of Terzaghi and Peck (1967) into consideration. The results showed that the input parameters were successful for classifying the SPT-N value to an acceptable degree of strength attribute. Therefore, in cases where the SPT tests are unreliable or could not be performed, laboratory tests on undisturbed specimens can give valuable information regarding the consistency and SPT-N value of the soil specimen under investigation. Data in this study is based on several tests that were conducted in a region; nevertheless, it is advised that the results of this study should be evaluated using global data.

Prediction of shear development in clean sands by use of particle shape information and artificial neural networks

Research highlights? Artificial neural networks are used for simulation of shear box tests conducted on clean sands. ? Results of shear box tests are supported with particle size and shape information. ? Performance of SCG learning algorithm produced acceptable outcomes in stress-strain modeling of clean sands. Particle shape is one of the most important factors affecting the shear strength of granular soils. Regarding to the knowledge that the grain size distribution is more effective on strength characteristics of soils in comparison with the particle shape information, clean sands of similar grain size distributions and diverse particle shapes are disposed. Afterwards, shear box tests are employed on these sands to obtain the stress-strain relationships, and resulting internal friction angles. For the simulation of results, artificial neural networks (ANN) of eight architectures using three different learning algorithms are constituted. The results revealed that the network with two hidden layers utilizing Levenberg-Marquardt learning algorithm is the most successful alternative. Nevertheless, on account of the possible improvements on the database and the learning duration, scaled conjugate algorithm should be preferred, which yields mathematically congruent curves, in comparison with the experimental values. Finally, it can be underlined that, use of ANN for simulation of shear development in granular soils is promising, if the inputs and output parameters are correctly determined.

Prediction of Impact Resistance Properties of Concrete Using Radial Basis Function Networks

This study presents an investigation of the prediction of impact resistance of steel-fiber-reinforced concrete and ordinary concrete specimens. In the experimental part of this study, parameters identifying impact resistance of various concrete mixtures were determined using an impact test machine, in accordance with ACI Committee 544. For this aim, concrete specimens containing three different aggregates (basalt, limestone and natural aggregate) were cured in water at 20 ◦C for 28 days. After curing impact resistance tests were performed on specimens having compressive strength values between 20 and 50 MPa, to determine the blows to initial crack and failure. The specimens were also subjected to splitting tensile strength and ultrasonic pulse velocity tests. Initially, using blows to initial crack and failure, many attempts were made to classify the impact resistance of different types of concrete in terms of the origin of used aggregate, strength properties or ultrasonic pulse velocity, however, this made no sense. The specimens could only be classified in terms of steel fiber presence. Therefore, radial basis function network, which belongs to another kind of unsupervised classifier network, was used to estimate the two abovementioned impact resistance parameters. In this scope, independent from aggregate origin used in preparation of specimens, compressive strength, splitting tensile strength and ultrasonic pulse velocity of the specimens were used to predict the impact resistance parameters of the concrete specimens. The results revealed that three listed parameters can be used for estimation of blows to formation of initial crack and failure. Scatter plots, root mean square error and absolute value of average residual parameters were used to verify the errors in predictions, which were very low, compared with the uncertainty in test and ambiguity of data in hand.