1 st Int’l Conf. on Recent Advances in Information Technology | RAIT-2012 | 978-1-4577-0697-4/12/$26.00 ©2012 IEEE BPNN and ANFIS Models for Prediction of Floor Bearing Characteristics of Weak Rock Foundations Sriram V. V. N. Malladi Department of Mechanical Engineering & Mining Machinery Engineering Indian School of Mines, Dhanbad, INDIA-826004 malladisriram@gmail.com Dheeraj Kumar Department of Mining Engineering, Indian School of Mines, Dhanbad, INDIA-826004 dheeraj@dkumar.org Abstract: Analysis of stability (mainly bearing strength and settlement) under a footing on regularly bedded, jointed and layered model rock mass is conducted using Back Propagating Neural Network (BPNN) and Adaptive Neuro Fuzzy Inference Systems (ANFIS). The inputs required for the modeling were imported from the laboratory results of the measurements carried out earlier [1]. Rock mass were modeled as elastic- plastic with Drucker- Prager failure criteria for plane strain condition. The results of the footing settlements and bearing strengths derived from BPNN and ANFIS models were compared with the footing settlements corresponding to the maximum applied bearing pressure on floor strata (for different sizes of footing plates and also under varying anisotropy conditions of floor strata) as obtained from the experimental results and FEM investigations and the bearing strengths obtained from the laboratory investigations. It is deduced that ANFIS model predicts accurately well vis-à-vis experimental results though the results predicted from BPNN model compares well with those of FEM analysis. Keywords: ANFIS, Neuro Fuzzy, ANN, Bearing strength, Settlement I. INTRODUCTION There is a general tendency to assume that any structure founded on bedrock will be totally safe against settlement and instability. Despite the apparently favorable stability conditions for structures founded on rock, there are instances of foundation failure too. Failure may include excessive settlement due to the presence of undetected weak seams, or deterioration of the rock with time, as well as collapse resulting from movement of block of rock in the foundation formed by continuous intersecting joints, bedding planes or faults. The discontinuous planes of cracks and joints in rock mass are considered as important factors influencing the local stabilities of rock foundations. The majority of foundation on rock are spread footings at the ground surface. Several studies were carried out widely for experimental determination and theoretical and statistical estimation of bearing capacity [2, 3, 4, and 5] involving laborious experimental setup and complex underlying physical process parameters thus resulting into conservative findings. Settlement prediction is one of the most challenging geotechnical engineering problems because a considerable level of uncertainty often affects it which in turn influences design. Several researchers [6, 7, 8, and 9] have predicted settlement using various uncertain parameters, analytical methods, regression analysis and simplified methods. The conventional plate loading test, punch test and other similar tests used for determination of bearing capacity in laboratory or field take long time and at the same time these tests results are quite varying due to the experimental limitations. In such situation, Artificial Intelligence is potentially useful, where the underlying physical process relationships are not fully understood and well-suited in modeling dynamic systems on a real-time basis, for predicting the floor bearing characteristics of weak rock foundations. In recent times, ANNs have been applied to many geotechnical engineering applications. For example, ANNs have been used in pile bearing capacity prediction [10]; stress-strain modeling of sands [11]. Zhou and Wu [12] have used these networks to interpret the site investigation. The settlements predicted using ANN tool were compared with the values predicted by three commonly used deterministic methods. The results indicated that ANNs are a promising method for predicting settlement of shallow foundations on cohesionless soils, as they perform better than the conventional methods. Several researchers, as mentioned above, have studied the different aspects in geotechnical engineering by using ANN modeling. However, the application of ANN in predicting the floor bearing characteristics of weak rock formations is relatively few. Thus, the present study developed a model on predicting the settlement and bearing strength of weak rock formations by using ANN programming. The main objective of this research finding is concerned with making more accurate predictions with the help of ANN and ANFIS models in respect of bearing strength and settlement and study how the predictions compare with the experimental and FEM analysis of bearing characteristics of weak floor rock foundations (massive, jointed and layered) of limited and varying thickness. The Neural Network toolbox and the Fuzzy Logic toolbox provided by the MATLAB 7.6 have been used in the present work.