Artificial neural network and fuzzy neural integrated systems for geotechnical modeling

One of the most important tasks of solving geotechnical problem is the interpretation of data from measurements made at field or at lab. The treatments of these parameters can be sorted into three main categories: (i) classification the data; (ii) site characterization through the interpreting of data; (iii) geotechnical properties prediction and modeling. Recently, a new approach called artificial neural network has emerged to fulfill those tasks.; Another important aspect of risk and decision analysis with geotechnical problems involves the consideration of many uncertain variables. There are basically two types of uncertainty (Casti, 1990): (i) ignorance, including measurement error, indecision about the mathematical form of the model and including stochasticity, spatial variation, and individual heterogeneity. Ignorance and variability are fundamentally different. The variability (statistical) uncertainty associated with these variables can be handled through probability and statistical theory. On the other hand, ignorance is subjective (non-random type) and can not be translated into probability in the same way. These often involve fuzzy information, i.e. information which is vague, imprecise, qualitative, linguistic, or incomplete.; Since the fuzzy set theory (FST) and artificial neural networks (ANN) are both numerical model-free estimators. They can share the ability to improve the intelligence of systems working in uncertain, imprecision and noisy environments. This suggests that we may combine fuzzy logic control and decision systems with artificial neural networks. Two different fuzzy neural network models have been developed in this study: (i) ANFISA—Adaptive Network-based Fuzzy Inference Systems for Approximation. (ii) FANN—Fuzzy Artificial Neural Network. Fundamentally, ANFISA is about taking a fuzzy inference system (FIS) and tuning it with a backpropagation algorithm based on some collection of input-output data. This allows the fuzzy inference systems to team. On the other hand, FANN model fuzzifies all the network parameters within the artificial neural network architecture, which will let the system be capable of processing fuzzy numbers.; The proposed artificial neural network model and two types of fuzzy neural integrated models are used for four different geotechnical applications, which are (i) Liquefaction-Induced Horizontal Ground Displacement Prediction. (ii) Liquefaction Potential Assessment. (iii) Axial Load Capacity of Piles Prediction. (iv) Uplift Capacity of Suction Caissons Prediction. Large historic databases have been used for each application for developing the models. (Abstract shortened by UMI.)...