Numerical Modeling for Prediction of Compression Index From Soil Index Properties R. Nesamatha 1 and Dr. P.d. Arumairaj 2 1 Department of Geotechnical Engineering, Government College of Technology, Coimbatore-13. nesagce28@gmail.com Associate Professor, Department of Civil Engineering (Soils), Government College of Technology, Coimbatore-13. arumairaj@gmail.com ABSTRACT Prediction of compression index from the laboratory test was time consuming and laborious. The compression index of clay was determined using oedometer test. In this research, oedometer test was conducted on remoulded clay collected from various places in Coimbatore district. Laboratory tests namely liquid limit test, plastic limit test, oedometer test have been conducted to arrive the correlations. In this study, an attempt was made to correlate compression index (C c ) with liquid limit (W L ) and plasticity index (I P ) using MS EXCEL s trend line concept. The compression index was predicted also with the help of regression analysis and MATLAB-Artificial Neural Network (ANN).Compression index predicted from the above software was compared.ann model was given significant result. KEYWORDS: Artificial Neural Network, clay, liquid limit, plastic limit, free swell, compaction, plasticity index, compression index, correlation INTRODUCTION The rapid urban and industrial development poses an increasing demand for land. Due to this space constrain, buildings were to be constructed on unsuitable grounds such as soft and expansive soils. An apparent volume change was noted when the dry clayey soil comes in contact with water. Due to the intrusion of water, the soil experiences alternate shrinkage and swelling. Unequal settlement of structures were the major problems to deal with an expansive soils were having large void ratio and water content, low bearing capacity and high compressibility. Compressibility was related to settlement of the foundation or structures. If the compressibility was high, settlement will also be high. Settlement was calculated from the compression index. Compression index was a clay dependent parameter computed from the odeometer test.the process of consolidation test take longer durations. So it was beneficial if the value of compression index can be related with index properties such as liquid limit and plasticity index. Atterberg s limits of clay also indirectly reflect the clay - 4369 -
Vol. 20 [2015], Bund. 12 4370 content and clay type. Reliable correlations between the engineering and index properties of soils will reduce the work load of a soil investigation program, in case of urgency. Study Area was Coimbatore city, covered with black cotton soils.. In this work, four locations namely Government College of Technology (GCT Campus), Rathinapuri, SITRA, Government Polytechnic College (GPT), Veerakeralam in the city were identified for soil sampling and laboratory analysis.since clay in these areas has maximum swelling potential. Samples collected from the four locations were analyzed for its grain size distribution, Atterberg s limit, Standard Proctor Compaction, Consolidation and Differential Free Swell. An attempt was made to establish a correlation between compression index Vs plasticity index and compression index Vs liquid limit. In these investigation the observed value of the compression index compared with predicted value of the compression index using REGRESSION ANALYSIS and MATLAB-Artificial Neural Network LITERATURE REVIEW Amith Nath and S.S Dedalal presented paper on role of plasticity index in predicting compression behavior of clays. Clay samples were collected from Purulia, Bankura, midnapur in West Bengal along with commercially available bentonite, kaolinite and riverside sand passing 0.425 mm sieve having specific gravity 2.63 and effective size D 10 =0.16mm as the source of non plastic material. Artificially mixed soil samples were prepared using the above soils in varying proportions to get 50 numbers of different plastic limit and other index properties. A correlation was achieved to determine the compression ratio in terms of liquid limit, plastic limit,void ratio. Arpan Laskar and Sujit Kumar Pal (2008) carried out a detailed study on Geotechnical Characteristics of two different soils and their mixture and relationships between parameters. The two soil samples collected from NIT Agartala campus and Howrah were investigated and the particle size distribution of NITA sample was determined. Then NITA sample was replaced with different proportions of the Howrah soil. The mixture of both the soils (Mixed Soil) was also investigated to study the variations in properties and to establish correlations of soil parameters. Slamet Widodo and Abdelazim Ibrahim collected 20 samples in Supadio Airport in Pondianak, Indonesia.The soil specimens were tested in the laboratory and proposed three different equations to estimate the compression index of the soils. The summary equations of the literature was given below in the table 1 Table 1: Literature Summary Equation Author C c '=0.0021W L +0.0587 Amith Nath and S.S Dedalal C c '=0.0888 e 0 +0.0525 Amith Nath and S.S Dedalal C c '=0.0025 I P +0.0866 Amith Nath and S.S Dedalal PI = 0.7785 (LL -18.623) Arpan Laskar and Sujit Kumar Pal OMC = 0.43 (PI + 30) Arpan Laskar and Sujit Kumar Pal Cc = 0.0046(LL 1.39) Arpan Laskar and Sujit Kumar Pal
Vol. 20 [2015], Bund. 12 4371 Cc = 0.0058 (PI+13.776) C c =0.5217(e o -0.20) C c =0.5217(W n +11.57) C c =0.5217(W L -1.30) Arpan Laskar and Sujit Kumar Pal Slamet Widodo and Abdelazim Ibrahim Slamet Widodo and Abdelazim Ibrahim Slamet Widodo and Abdelazim Ibrahim COLLECTION OF SAMPLES The disturbed soil samples were collected in plastic bag from the depth of 0.5 1.0 m from four locations in coimbatore city viz., Government College of Technology (GCT campus), Rathinapuri, SITRA, Government Polytechnic College(GPT), Veerakeralam in Coimbatore. The samples were noted as S1, S2, S3,S4 and S5. Here, S indicates soil sample and the number refers the order of sample taken. The latitude and longitude of the location was shown in the table 2 below. Table 2: Locations of sample collection Sample no Location latitude Longitude S1 GCT (campus) 11 1 '3.69" 76 56 '3.375" S2 Rathinapuri, 11 1 '30.25" 76 57' 46.89" S3 SITRA 11 2 '5.69" 77 1' 56.76" S4 GPT campus 11 1' 44.91" 77 1' 54.63" S20 Veerakeralam 11 00 ' 27.48" 76 54' 38.51" EXPERIMENTAL STUDIES The laboratory tests conducted to find various physical and engineering properties of soil samples. All the tests were carried out as per Indian Standard code of Practice. The following laboratory tests were conducted for all soil samples. Specific gravity (G), grain size analysis, Atterberg s limits (liquid limit (W L ), plastic limit (W P ) and standard Proctor compaction Test (optimum moisture content (OMC)) and maximum dry density (MDD), consolidation characteristics (compression index (Cc) were evaluated in accordance with ASTM standards.
Vol. 20 [2015], Bund. 12 4372 SAMPLE PREPARATION FOR OEDOMETER TEST The specimen for Oedometer test was prepared in optimum moisture content and maximum dry density. Different static load was applied to the soil. The soil was compressed due to removal of water. The reduced volume or changing height was taken from the LVDT. The prepared clay sample was shown in figure 1 below. Figure 1: sample preparation for oedometer test The results obtained from the above tests were shown in the table 4. Sample no G Gravel sand Table 4: Soil properties for four samples Silt clay Soil type W l () W P I P Swell Free S1 2.84 0.2 20.8 33.97 45.03 CH 72 20.1 51.9 100 22 1.492 0.074 S2 2.82 0.4 14.8 14.4 70.4 CH 77.8 22.3 55.5 100 24 1.481 0.096 S3 2.77 0.1 19.3 23.6 57 CH 66.2 20.0 46.2 80 20 1.592 0.062 S4 2.76 0.3 30.41 29.19 40.10 CH 74.3 20 54.3 85 22 1.625 0.086 S5 2.82 1.1 33.5 24.85 40.55 CH 67.48 20.7 46.78 70 22 1.598 0.067 Omc Mdd g/cc C c Compression index was determined from the above test. The value of primary compression index was different for different type of soil shown in table 3 below. Table 3: Compression index for different types of soil Type of soil Compression index Dense sand 0.0005-0.01 Loose sand 0.025-0.05 Firm clay 0.03-0.06 Stiff clay 0.06-0.15 Medium soft clay 0.15-1.0 Organic soil 1.0-4.5 Rock 0
Vol. 20 [2015], Bund. 12 4373 REGRESSION ANALYSIS The observed value of liquid limit, plasticity index and compression index from the test was used as a input for the regression model.the following graph to be formed using MS EXCEL s trend line concept. Figure 2: Observed Compression Index ( Cc ) Vs liquid limit A relationship between compression index and liquid limit was arrived based on the above plot. C c =0.002W L -0.127.. (1) Figure 3: Observed Compression Index ( Cc ) Vs Plasticity Index A relationship between compression index and plasticity index was arrived based on the above plot. C c =0.002I P -0.02 (2) Using MS EXCEL with REGRESSION ANALYSIS tool Compression index was predicted from the equations (1) and (2) shown in below.
Vol. 20 [2015], Bund. 12 4374 Table 5: Predicted compression index from Regression model Observed value Predicted value from regression analysis 0.062 0.061685 0.067 0.065345 0.074 0.7827 0.086 0.84846 0.096 0.094854 Figure 4: predicted and observed compression index from regression analysis using PI Table 6: Predicted compression index from Regression model Observed value Predicted value from regression analysis 0.062 0.062248 0.067 0.064054 0.074 0.80003 0.086 0.087479 0.096 0.091217
Vol. 20 [2015], Bund. 12 4375 Figure 5: predicted and observed compression index from regression analysis using W L ARTIFICIAL NEURAL NETWORK Mat Lab was the main source of ANN. ANN Model was created using NN TOOL in MATLAB in terms of Liquid limit () and Plasticity index. An artificial neural network was created by simulating a network of model neurons in a computer. Neurons were the processing unit. It receives input from a number of other units or external sources, weighs each input and adds them up.the output changes from 0 to1 sometimes the neurons forming layer and organizing the work was called as Multi layer Perception Network. It has three parts as input layer, hidden layer and output layer. Each neuron in one layer connected to each neuron in another layer but no connection between the neurons in the same layer. The following mechanisms to be followed in the ANN learning, training, testing, back propagation and validation. The observed value of the liquid limit and plasticity index from the test result was taken as an input and the above mechanism was processed in the hidden layer and output was in the form of compression index. The observed liquid limit and observed plastic limit is taken as the input value for ANN model, observed compression index value is taken as a target, and the hidden layer was trained and tested. After training the following window was opened as shown in below. Figure 6: output after training The correlation coefficient value was 0.99, so the accuracy of ANN model was very higher and more reliable. The following window was appeared in the computer after finishing the training output.
Vol. 20 [2015], Bund. 12 4376 Figure 7: model window for ANN Figure 8: Output window Four samples of experimental data was entered into the above window as a input, the compression index was predicted as a output. The predicted compression index values were shown in below table 7 and 8 Table 7: Predicted compression index fromann model Observed value Predicted value from ANN analysis 0.062 0.0698 0.067 0.0712 0.074 0.074 0.086 0.086 0.096 0.096
Vol. 20 [2015], Bund. 12 4377 Figure 9: predicted and observed compression index from ANN using W L Table 8: Predicted compression index fromann model Observed value Predicted value from ANN analysis 0.062 0.057 0.067 0.067 0.074 0.0698 0.086 0.086 0.096 0.096 Figure 10: predicted and observed compression index from ANN using PI The result obtained from the test result was compared to the results of following authors. Table 8: Values of cc from laboratory test using some equations Sample no Arpan laskar Slamet widodo Test result S1 0.325 0.38 1.2 0.074 S2 0.35 0.401 1.305 0.096 S3 0.298 0.347 1.107 0.062 S4 0.335 0.395 1.245 0.086 S5 0.304 0.351 1.129 0.067 max 0.35 0.347 1.107 0.096 min 0.298 0.401 1.305 0.062 avg 0.322 0.378 1.197 0.077
Vol. 20 [2015], Bund. 12 4378 From the table that average value of C c for Coimbatore soil was 0.077 and classified as stiff clay in ranging from 0.06 to 0.15. CONCLUSION Based on the REGRESSION ANALYSIS and ANN modeling, Compression index was predicted. Regression analysis was given correlation coefficient value R 2 = 0.969 (W L as a input value ) value R 2 =0.91(I P as a input value ) ANN model had 5 inputs, 15 hidden layers and 1 output. It gives the maximum correlation coefficient value R 2 =0.966(W L as a input value ) value R 2 =0.983(I P as a input value ) compared to regression analysis ANN was used to predict the compression index of the soil. Liquid limit and plasticity index was an input parameter. REFERENCES [1] Nath, A. and Dalal, S. S. (2004), The role of plasticity index in predicting compression behavior of clays, Electronic Journal of Geotechnical Engineering, Vol. 9, 2004-Bundle. [2] Zeki Gunduz and Hasan Arman, possible relationships between compression and recompression indeces of low plasticity clay soil, Arabian journal for science and engineering, Vol 32, Number 2B. [3] A.M.Mustapha and M.Alhassan, Compression index correlation that best fits clay deposits in Nigeria IOSR Journal of Engineering, Vol 3 issue 11. [4]A. Sridharan and H.B.Nagaraj, Compressibility behavior of remolded, fine grained soils and correlation with index properties, Canadian Geotechnical journal, june2000; 37,3;Proquest Science journals page 712. [5] Amardeep Singh And Sahid Noor, Soil compression index prediction model for fine grained soils International Journal of Innovations in Engineering and Technology, Vol 1 Issue 4 Dec 2012. 2015 ejge