Search

The unconfined compressive strength (UCS) of rocks is one of the most important properties used in rock science and engineering practices. Thus, many researchers attempt to develop computational intelligence models for estimating the UCS of different types of rocks. This study develops and compares eight models to estimate the rock UCS. Six hybrid ANNs were constructed using artificial bee colony, ant colony optimization, antlion optimizer, imperialist competitive algorithm, shuffled complex evolution, and teaching learning-based optimization algorithms. Additionally, two regression-based computational models, namely multivariate adaptive regression splines (MARS) and Gaussian process regression (GPR), were utilized. The experimental datasets at different locations in Malaysia and Iran, including the UCS and other rock indices, were collected and used in this work. Experimental findings indicate that the performances of all models are adequate for UCS estimation, with R2 values scattered in the range of 0.87 to 0.97 and 0.94 to 0.98 in the training and testing phases, respectively. However, the developed MARS model was ranked high compared to other models, including GPR and hybrid ANNs, and the RMSE of the MARS model was found to be 2.27 MPa. Overall, the developed MARS model outperformed other employed models based on the R2 and RMSE criteria and can be utilized precisely for the UCS estimate of rocks. [ABSTRACT FROM AUTHOR]

Bridges are playing a major role in the socio-economic development of any country over the world. Suspension bridges are one of the most sensitive structures to various external influences and loads. Therefore, the need for structural monitoring system, maintenance and deformation prediction for these types of structures is important and vital. Time of observations for the purpose of structural deformation can vary from a few hours, days to several months or even years. This paper investigates the performance of several soft computing techniques for point displacement modeling using GNSS technique during the process of monitoring the structural deformation of suspension highway bridge, taking into consideration the effect of wind, temperature, humidity and traffic loads during the operational and short-term measurements. Due to the availability of a large amount of positions data generated from GNSS data positions for monitoring the deformation of such structure, artificial neural networks (ANNs) and adaptive neuro-fuzzy inference system (ANFIS) should be chosen. One of the main objectives of this paper is to investigate the optimum predictive soft computing model for processing GNSS positions and points displacement prediction. Several mathematical models and two cases of data amount (66.67% and 50% of all available data) for dynamic and kinematic state are applied and compared for prediction of suspension bridge displacement with confidence interval with a probability ρ = 0.95, Δ = ± 2σ. The resulting point displacement values by applying ANNs and ANFIS, which used a confidence interval with a probability of ρ = 0.95, Δ = ± 2σ when using 66.67% of all data, are more accurate and reliable than any other applied methods, and therefore, ANNs and ANFIS can provide a significant improvement of understanding and predicting the structure deformation values where conventional mathematical modeling techniques were not as accurate or capable especially in dynamic prediction of displacements. [ABSTRACT FROM AUTHOR]

Bridges are playing a major role in the socioeconomic development of any country over the world. Suspension bridges are one of the most sensitive structures to various external influences and loads. Therefore, the need for structural monitoring system, maintenance, and deformation prediction for these types of structures is important and vital. Time of observations for the purpose of structural deformation can vary from a few hours, days to several months, or even years. This paper investigates an integrated monitoring system using GNSS observations for studying the deformation behavior and points displacements prediction for suspension highway bridge, taking into consideration the effect of wind, temperature, humidity, and traffic loads during the operational and short-term measurements. Due to the complexity of the mathematical processing of large GNSS monitoring data for obtaining reliable results, adequate model of several alternatives should be chosen. One of the main objectives of this paper is to investigate the optimum predictive soft computing model for processing GNSS observations and points displacement prediction. Several mathematical models and two cases of data amount (66.67% and 50% of all available data) for dynamic and kinematic state are applied and compared for prediction of suspension bridge displacement with confidence interval with a probability ρ = 0.95, Δ = ±2σ. The resulting point displacement values by applying ANNs and ANFIS, which used a confidence interval with a probability of ρ = 0.95, Δ = ± 2σ when using 66.67% of all data, are more accurate and reliable than any other applied methods; therefore, ANNs and ANFIS can provide a significant improvement of understanding and predicting the structure deformation values where conventional mathematical modeling techniques were not as accurate or capable especially in dynamic prediction of displacements. [ABSTRACT FROM AUTHOR]

This study presents multi input–single output (MISO) soft computing models for forecasting the significant wave height (Hs) of lakes and investigates the impact of input variables for designing the prediction models. Wind direction and speed, air and water temperature changes, wave period and direction, and sea level pressure variables are used as input parameters for the model design to predict the Hs of Michigan Lake. The M5 regression tree (M5T) and robust regression (RR) are used to study the effectiveness of input parameters that impact Hs prediction model. Furthermore, four MISO soft computing models, namely regression (RR and M5T), artificial neural network (ANN), and least square support vector machine (LSSVM), are evaluated and compared to estimate the best model for predicting the Hs. The results show that the wave's period variable has high impact for prediction of lake Hs values, while the other variables have the same impacts on the prediction accuracy of Hs. The comparison between the developed models shows that the four model performances are acceptable in the training stage, while the MISO-ANN model gives better precision for the prediction of Hs of the lake in the testing stage. [ABSTRACT FROM AUTHOR]