Pile foundations are typically employed when top-soil layers are unstable and incapable of bearing super-structural pressures. Accurately modeling pile behavior is crucial for ensuring optimal structural and serviceability performance. However, traditional methods such as pregnancy testing, while highly accurate, are expensive and time-consuming. Consequently, various approaches have been developed to predict load settlement behavior, including using artificial neural networks (ANNs). ANNs offer the advantage of accurately replicating substrate behavior's nonlinear and intricate relationship without requiring prior formulation.This research aims to employ artificial neural network (ANN) modeling techniques to simulate the load-settlement relationship of drilled piles. The primary aims of this study are threefold: firstly, to assess the effectiveness of the generated ANN model by comparing its results with experimental pile load test data; secondly, to establish a validation method for ANN models; and thirdly, to conduct a sensitivity analysis to identify the significant input factors that influence the model outputs. In addition, this study undertakes a comprehensive review of prior research on using artificial neural networks for predicting pile behavior. Evaluating efficiency measurement indicators demonstrates exceptional performance, particularly concerning the agreement between the predicted and measured pile settlement. The correlation coefficient (R) and coefficient of determination (R^2) indicate a strong correlation between the predicted and measured values, with values of 0.965 and 0.938, respectively. The root mean squared error (RMSE) is 0.051, indicating a small deviation between the predicted and actual values. The mean percentage error (MPE) is 11%, and the mean absolute percentage error (MAPE) is 21.83%.
Due to the expansion of industrial operations globally in recent years, waste output has risen. So these wastes must be reduced by recycling and reusing to achieve environmentally friendly buildings and find various alternative materials in critical cases. The statistical indicators are used as practical study including Multiple linear regression (MLR) and artificial neural network (ANN) models. The study's goals were to assess the effectiveness of granite waste (GW) as a replacement for cement, sand, plastic, and binder in specific building applications and the relationships between MLR and ANN approaches. Results show the efficiency of adding granite waste to some construction stages and replacing it with cement in the mixture and examining its strength, it gave excellent results in addition to good results for its use as a binder in cement mortar, while the results were weak when used as a substitute for sand and plastic in insulator because it's classified as fine sand, Therefore, it cannot be used as a substitute for sand in the construction. The statistical models give an effective indicator to use GW as an alternative material ( binder and cement) based on the coefficient of correlation (R2) for the two models MLR and ANN equal to 83.4 % and 80 % respectively.
In the current article, an experimental investigation has been implemented of flow and heat transfer characteristics in a parabolic trough solar collector (PTSC) using both nano-fluids and artificial neural networks modeling. Water was used as a standard working fluid in order to compare with two different types of nano-fluid namely, nano-CuO /H2O and nano-TiO2/ H2O, both with a volume concentration of 0.02. The performance of the PTSC system was eval-uated using three main indicators: outlet water temperature, useful energy and thermal efficiency under the influence of mass flowrate ranging from 30 to 80 Lt/hr. In parallel, an artificial neural network (ANN) has been proposed to predict the thermal efficiency of PTSC depending on the experimental re-sults. An Artificial Neural Network (ANN) model consists of four inputs, one output parameter and two hidden layers, two neural network models (4-2-2-1) and (4-9-9-1) were built. The experimental results show that CuO/ H2O and TiO2/H2O have higher thermal performance than water. Overall, it was veri-fied that the maximum increase in thermal efficiency of TiO2/H2O and CuO/H2O compared to water was 7.12% and 19.2%, respectively. On the oth-er hand, the results of the model 4-9-9-1 of ANN provide a higher reliability and accuracy for predicting the Thermal efficiency than the model 4-2-2-1. The results revealed that the agreement in the thermal efficiency between the ANN analysis and the experimental results about of 91% and RMSE 3.951 for 4-9-9-1 and 86% and RMSE 5.278 for 4-2-21.
The goal of the current investigation is to construct an artificial neural network (ANN) to estimate the ultimate capacity of the composite columns consisting of a rectangular steel tube filled with concrete (RSTFC) under concentric loads. The experimental results of (222) samples collected from previous researches were used in constructing the proposed network. Totally (45) specimens were randomly chosen for network testing while the remaining (177) speci-mens were used to train the network. The information used to create the ANN model is ar-ranged into (6) variables represents the different dimensions and properties of the RSTFC col-umns. Based on the input information, a formulated network was used to estimate the columns' ultimate capacity. Results obtained from the formulated network, available laboratory tests, and Eurocode 4 and AISC equations were compared. The network values were closer to the laboratory values than the calculated values according to the specifications of the mentioned codes. It has been shown that the formulated ANN model has a high ability to estimate the RCFST ultimate capacity under concentric loads
The Artificial Neural Network (ANN) and numerical methods are used widely for modeling andpredict the performance of manufacturing technologies. In this paper, the influence of millingparameters (spindle speed (rpm), feed rate (mm/min) and tool diameter (mm)) on material removalrate were studied based on Taguchi design of experiments method using (L16) orthogonalarray with 3 factor and 4 levels and Neural Network technique with two hidden layers and neurons.The experimental data were tested with analysis of variance and artificial neural networkmodel has been proposed to predict the responses. Analysis of variance result shows that tooldiameters were the most significant factors that effect on material removal rate. The predictedresults show a good agreement between experimental and predicted values with mean squarederror equal to (0.000001), (0.00003025), (0.002601) and (0.006889) respectively, which produceflexibility to the manufacturing industries to select the best setting based on applications.
The feasibility of using an Artificial Neural Network (ANN) for controlling time- varying dynamical system is presented. The direct adjusting of neural controller by direct adaptive control (DAC) is available, by using the error between output of plant and desired input. The finite recurrent back propagation (FRBP) is used in the learning process, because the ability of this method to capture the nonlinearly and overcome the problem of time varying system. Hybrid controller structure used in this paper, where the parameters of classical controller are adjusted with time at specified freezing points for time varying dynamical system, and summed the outputs of two controllers and enter to the plant, identify of system by ANN to get the optimal initial condition for neuro controller. A single channel for Spacecraft model is used as an example in this paper, satisfactory results are obtained, which explain the ability of recurrent neural network (RNN) to identify time varying dynamical system and overcome for all its problem and explain the ability of this structure of hybrid neuro controller to use with time varying dynamical system.
Severe three phase trips are simulated on four arbitrary locations of an uncontrolled power system transmission lines. The responses of three measurable state variables of the system (rotor speed, stator direct axis current, and stator quadrature - axis current) are recorded, and suitable ANNs are trained to detect and locate the positions of the corresponding trips. The paper proves that this method is quick, active and accurate to diagnose and find the locations of that kind of trips.