Essay Example on Lithology rock types mapping from downhole Geophysical









Introduction Lithology rock types mapping from downhole geophysical log acquired over various tectonic setting around the world is of importance to understand the crustal heterogeneity evolution and tectonics of the area More recently neural network learning in Bayesian framework BNN has been extensively used for the well log data Maiti and Tiwari 2009 2010a b Ojha and Maiti 2016 BNN offers a platform where many shortcomings of ANN can be avoided including over fitting The objective of the present study is to use BNN HMC supervised classification scheme for classification of lithology rock types for large and non simulated new ocean drilling data of Integrated Ocean Drilling Program IODP Expedition 323 project Study area Site U1339D of IODP expedition is used as study area The site is dominated by the mixture of three components biogenic mainly diatom frustules with varying proportions of nannofossils foraminifers silicoflagellates and radiolarians volcaniclastic mainly fine ash and siliciclastic clay to pebble sized clasts Takahashi et al 2011a b Wehrmann et al 2011 Detailed information about the study area downhole data can be found in Takahashi et al 2011a b and Wehrmann et al 2011 Here in this study sonic P wave velocity density density porosity electrical resistivity and gamma ray logs from the IODP hole number U1339D have been used for the analysis of rock type classification Figure 1 Location map of the site U1339 of IODP Expedition 323 drilling and coring sites shown by yellow circle Theory and Methods The BNN is a special configuration of an ANN where the network s weight update is followed by a Bayesian learning Bishop 1995 Ultimately the BNN emulates the functional relationship between an input well log and an output space domain lithology class from a finite data set by adjusting network parameters weight and biases 

The objective function is defined as 1 where and are respectively the target and the BNN output at each node in the output layer Often regularization is incorporated to smooth the error function as follows 2 where and are hyper parameters which control synaptic weight values In the BNN first a posteriori probability distribution pdf of weight is assumed as a Gaussian Then the likelihood function is derived Assuming that the noise distribution in the target data is Gaussian a posteriori pdf of the target is expressed using a prior pdf and likelihood function 3 where is the likelihood term is a normalization factor scaling 4 Here is predicted data when i e weight is the most probable weight and is the posterior pdf of weight vector Results and discussions The BNN HMC analysis applied to the downhole data reveals the detail distribution of lithology against depth below sea floor at the site U1339D For the purpose of training and calibration of the BNNs model total data 632 was divided via Matlab based code Maiti and Tiwari 2010 The training data set consists of 316 number of pairs validation set has 158 pairs and test set composed of 158 samples The data are normalized so that they fall in the range between 1 to 1 The nonlinear sigmoid function is used for establishing the nonlinear mapping between input and output domain The network topology is with one input layer with five input nodes and one hidden layer with nine hidden nodes and one output layer with three output nodes is employed for the present analysis From the present analysis it is confirmed that there are three types of lithology dominating over the entire depth interval from 87 15 meters below seafloor mbsf to 180 87 mbsf of site U1339D

The diatom rich mixed siliciclastic biogenic diatom silt Diatom ooze and ash layers occurs cyclically Figure 2 Down hole data of density density porosity gamma ray p wave velocity Vp and resistivity of site U1339D of IODP Expedition 323 and Bayesian neural networks optimized by Hybrid Monte Carlo BNN HMC classification of lithology Color bar shows the probability of occurring particular lithology Conclusions An approach of supervised HMC BNN classification method employed here makes the analysis of the rock type robust and controlled The present approach is found to be effective for giving the complete descriptions of sediment distribution against the depth at site U1339D Three litho types associated with Diatom silt Diatom ooze Ash layers are clearly distinguished by HMC BNN approach which could be useful for analysis of other tectonically complex area of oceanic setting around the world References Bishop C M 1995 Neural networks for pattern recognition Oxford University Press Maiti S Tiwari R K 2009 A hybrid Monte Carlo method based artificial neural networks approach for sediment boundaries identification A case study from the KTB bore hole Pure and Applied Geophys 166 2059 2090 Maiti S Tiwari R K 2010a Automatic discriminations among geophysical signals via the Bayesian neural networks approach Geophys 75 1 E67 E78 Ojha M and Maiti S 2016 Sediment classification using neural networks an example from the site U1344A of IODP Expedition 323 in the Bering Sea Deep Sea Research Part II Topical Studies in Oceanography 125 126 202 213 Takahashi K Ravelo C Zarikian A the IODP Expedition 323 Scientists 2011a Proceedings of the Integrated Ocean Drilling Program 323 1 53 doi 10 2204 iodp proc 323 101 2011

Write and Proofread Your Essay
With Noplag Writing Assistance App

Plagiarism Checker

Spell Checker

Virtual Writing Assistant

Grammar Checker

Citation Assistance

Smart Online Editor

Start Writing Now

Start Writing like a PRO