Self Compacting Concrete Presentation 2020 FYDP (1).pptx

Self Compacting Concrete

1. Development of Nonlinear Regression and Artificial Neural Network Models for Predicting the Compressive Strength of Self-Compacting Concrete Project Advisor: Prof. Dr. Asif Hameed Department of Civil Engineering University of Engineering and Technology, Lahore Presented by: Tayyab Afzal 2020-CIV-73 Qalb E Abbas 2020-CIV-76 Saif Ullah 2020-CIV-108 Muhammad Amir 2021-P/2020-CIV-110

2. 4 INTRODUCTION Self Compacting Concrete (SCC) Defined as: “Concrete that is able to flow and consolidate under its own weight, completely fill the formwork even in the presence of dense reinforcement, while maintaining homogeneity and without the need for any additional compaction”.

3. 4 INTRODUCTION Why SCC is Preferred over NVC? No vibration. No noise Faster construction Ease of placement Durable & reliable structures

4. 4 MIX DESIGN METHOD OF SCC Empirical Design Method Compressive Strength Method Close Aggregate Packing Method Statistical Factorial Model Rheology of paste model Fix coarse and fine aggregate first, and then obtain self- compactability by adjusting W/B and SP dosage. For a given set of required properties, make the best estimation of the mixture proportions, and then carry out trial mixes to prove. Based on ACI 211.1 method for proportioning conventional concrete. Use GGBS in SCC based on the strength requirements and consider the efficiency of GGBS. Use densified mixture design algorithm, derived from maximum density theory and excess paste theory. For this purpose, a term Packing factor is used. (Factor to adjust the relative content of aggregate and paste). Mainly based on the void content and the blocking criteria Statistical Factorial model are used to derive design charts which correlate input mix design variables to output material properties, mainly consisting of measurements of fresh state properties as well as compressive strength. The rheology of paste matrix largely dictated the segregation resistance and workability of fresh concrete It is proposed that minimum paste yield stress and viscosity must be exceeded to avoid segregation under both static and dynamic conditions.

5. 4 PROBLEM STATEMENT The compressive strength of Self-Consolidating Concrete (SCC) is influenced by a multitude of factors such as the type of cementitious materials used, the water-to-binder ratio, the dosage of superplasticizer, and the characteristics of the aggregate. These factors often exhibit nonlinear relationships with the strength properties, and there is not any standard mix design method for SCC. Normally it is prepared with empirical methods. The properties of SCC can vary significantly depending on the materials used, the proportions in the mix, the conditions under which it is cured, and the environmental factors it is exposed to. Therefore, it is crucial to develop predictive models using Nonlinear Regression Analysis and ANN that can reliably account for these variations.

6. 4 OBJECTIVES OF STUDY To perform statistical analysis to investigate the influence of concrete constituents on the compressive strength and workability of self-compacted concrete. To develop on nonlinear regression and artificial neural network models for predicting the compressive strength and slump flow of SCC. To identify the most significant parameters in SCC mix design that impact compressive strength. Two data sets are considered to develop predictive models for the compressive strength and workability of SCC. The first data set is from laboratory experiments by an MSc student at UET Lahore (Khalid,2016), and the second is from published research papers. Scope of work

7. 4 INTRODUCTION Principles of SCC Mixture D esign Increase paste flow Increase paste viscosity to prevent segregation Potentially minimizing the risk of blockage. Lower the heat of hydration, enhance stability Coarse aggregate is small sized and of limited quantity HRWRs Viscosity modifying Agents Mineral admixtures

8. 4 INTRODUCTION Comparison Between Composition of SCC and NVC Air Water Powder (Cement +Mineral Fillers) Fine Aggregate Coarse Aggregate Composition of NVC Composition of SCC Air Water Cement Fine Aggregate Coarse Aggregate Self compact ability is achieved by: Reducing w/cm ratio Increasing powder contents Reducing Coarse Aggregate Using HRWRA & VMA Principal Mineral Fillers fly ash, silica fume, limestone, marble powder, ground granulated blast furnace slag ( ggbs ), bagasse etc.

9. 4 INTRODUCTION Fresh Properties of SCC (EFNARC, 2005) Characteristic Test method Measured Value EFNARC, 2005 ACI 237R-07 Measure Class 1-Flow ability / Filling ability to flow freely and fill all spaces in complicated formwork. Slump Flow Test Total spread Slump Flow SF-1, SF-2, SF-3 Flow distance Visual stability of the mixture 2-Viscosity / Flow ability : Measure of speed of flow of SCC T 500 V-Funnel Flow time Flow time VS1& VS VF1& VF22 Rate of flow 3-Passing ability The ability to flow through congested reinforcement without blocking L-Box test J-Ring test Passing ratio. Step height, total flow PA1 & PA2 Flow rate and distance Flow rate 4-Segregation Resistance Ability to remain homogeneous, no bleeding, no segregation Sieve segregation V-Funnel test after 5 min-seg regation resistance Percent laitance SR1 & SR2 Segregation of aggregates

10. Nonlinear regression is a form of regression analysis in which data is fit to a model and then expressed as a mathematical function . NLR Models: Pol ynomial Exponential Logarithmic Uses This model is used for more complicated data sets . INTRODUCTION Non Linear R egression (NLR)

11. 4 INTRODUCTION (Key definitions) Artificial Intelligence: It is simulation of human intelligence processes by machines, including learning, reasoning and self correction Machine Language: Is the study that uses statistical methods enabling machines to improve with experience. Neural Network: It composed of interconnected nodes that process information to recognize patterns and relationships in data Deep Learning: Is the study that makes use of Neural Networks imitate functionality just like a human brain.

12. 4 INTRODUCTION What is Artificial Neural N etwork (ANN) The concept of neural network is inspired by human brain. ANNs are set of algorithms that involve learning from data to make predictions. Batch size effects the accuracy of predicted variable Why to use ANNs Non linearity in data Non normality in data Robust against noise, outliers and small data set also.

13. 4 INTRODUCTION Types of Neural Networks Basis on the data flows from the input node to the output node. 1. Feedforward neural networks Forward direction of flow of information. Once moves forward, never turns back. No feedback is given. 2. Backpropagation algorithm Target output is provided. Error is calculated Backpropagates till error is minimized

14. 4 INTRODUCTION Multilayer Perceptron (MLP) It works like human brain, each neuron is connected to next neuron , that’s why its multilayer. Used for non linearly separable data sets. It consist of input layer, output layer and two or more hidden layer. It contains different non linear activation functions. Sigmoid Hyperbolic tangent ReLU

15. INTRODUCTION Activation Functions for MLP Sigmoid Function It is used for binary classification Value varies from 0 to 1. Hyperbolic tangent function Data is centralized along zero. Its value varies from -1 to 1. Problem of vanishing gradient. Rectified linear unit function ( ReLU) Its values ranges from o to max. ( 0,max) No problem of vanishing gradient.

16. INTRODUCTION Radial Basis Function (RBF) RBF value depends only on the distance between the input and some fixed point, either the origin or some other fixed point, called a center. It contains input layer, output layer and only one hidden layer Non linearly separable data. Radial basis activation functions Uses: It is used to process high-dimensional data, have quick training and testing times.

17. 4 INTRODUCTION Training of the Multilayer P erceptron The learning (training) process of a neural network is an iterative process in which the calculations are carried out forward and backward through each layer in the network until the loss function is minimized. Divided into three main parts: Forward propagation (Forward pass) Calculation of the loss function Backward propagation (Backward pass/Backpropagation)

18. 4 INTRODUCTION Comparison Between M ultilayer P erceptron and Radial Basis F unction MLP RBF Multiple hidden layers Only one hidden layer Used for non-linear data Used for non-linear and high-dimensional data. Weighted sum of inputs is calculated and transferred to hidden layer and activation function is applied. Directly transfers data from input layer to hidden layer and activation function is applied. Data is transferred from one neuron to next neuron and they are inter connected to form a network. Its value depends on distance between input and origin or other fixed point called center.

19. 4 Independent Variables Depended Variables (CS) Stastical AnalysisAnalysis Coarse Aggregates Water binder raio Cement Fly Ash Superplastizers Fine Aggregates Data Splitting Randomly Training Data (70%) Analysis Testing Data (30%) Developing Models NLR ANN Performance evaluation criteria R2, RMSE, MAE, SI Analysis Sensitivity Analysis for Independent Variables Using ANNAnalysis Data Collection and Analysis METHODOLOGY (DATA ANALYSIS CHART )

20. 4 METHODOLOGY ( Phase-I) ( Javed Khalid,2016) Mix Proportions Sr. No. Mixture ID PC SF FA Water Fine Aggregate Coarse Aggregate HRWRA VWA kg/m 3 kg/m 3 kg/m 3 kg/m 3 kg/m 3 kg/m 3 kg/m 3 kg/m 3 1 C 475 FA 0.0% SF 0.0% 475 - - 178.62 954 830 2.00% 1.50% 2 C 380 FA 20.0% SF 0.0% 380 - 95.0 175.20 934 830 2.00% 1.50% 3 C 356 FA 25.0% SF 0.0% 356 - 119.0 178.10 919 830 2.00% 1.50% 4 C 333 FA 30.0% SF 0.0% 333 - 143.0 181.00 904 830 2.00% 1.50% 5 C 309 FA 35.0% SF 0.0% 309 - 166.0 184.00 888 830 2.00% 1.50% 6 C 428 FA 0.0% SF 10.0% 428 47.5 - 183.60 890 830 2.00% 1.25% 7 C 416 FA 0.0% SF 12.5% 416 59.4 - 188.60 864 830 2.00% 1.00% 8 C 404 FA 0.0% SF 15.0% 404 71.3 - 193.60 838 830 2.00% 1.00% 9 C 392 FA 0.0% SF 17.5% 392 83.1 - 198.60 812 830 2.00% 1.00% 10 C 380 FA 0.0% SF 20.0% 380 95.0 - 203.70 786 830 2.00% 0.75% 11 C 392 FA 10.0% SF 7.5% 392 36.6 47.5 184.40 886 830 2.00% 1.25% 12 C 368 FA 15.0% SF 7.5% 368 36.6 71.3 187.30 871 830 2.50% 1.00% 13 C 344 FA 20.0% SF 7.5% 344 36.6 95.0 190.20 856 830 3.00% 0.75% 14 C 321 FA 25.0% SF 7.5% 321 36.6 117.8 193.10 841 830 3.00% 0.50% 15 C 297 FA 30.0% SF 7.5% 297 36.6 142.5 196.00 826 830 4.00% 0.25%

21. Mix Proportions Sr. No. Mixture ID PC SF FA Water Fine Aggregate Coarse Aggregate HRWRA VWA kg/m 3 kg/m 3 kg/m 3 kg/m 3 kg/m 3 kg/m 3 kg/m 3 kg/m 3 16 C 380 FA 10.0% SF 10.0% 380 47.5 47.5 189.50 860 830 2.50% 1.25% 17 C 356 FA 15.0% SF 10.0% 356 47.5 71.5 192.40 845 830 2.75% 1.00% 18 C 333 FA 20.0% SF 10.0% 333 47.5 95.0 195.30 830 830 3.00% 0.75% 19 C 309 FA 25.0% SF 10.0% 309 47.5 118.8 198.20 815 830 3.25% 0.50% 20 C 285 FA 30.0% SF 10.0% 285 47.5 142.5 201.20 799 830 4.00% 0.25% 21 C 368 FA 10.0% SF 12.5% 368 59.4 47.5 194.50 834 830 2.50% 1.25% 22 C 344 FA 15.0% SF 12.5% 344 59.4 71.3 197.40 819 830 2.75% 1.00% 23 C 321 FA 20.0% SF 12.5% 321 59.4 95.0 200.30 804 830 3.00% 0.50% 24 C 297 FA 25.0% SF 12.5% 297 59.4 117.8 203.20 789 830 3.50% 0.25% 25 C 273 FA 30.0% SF 12.5% 273 59.4 142.5 206.20 773 830 4.00% 0.25% 26 C 356 FA 10.0% SF 15.0% 356 71.3 47.5 199.50 808 830 2.50% 1.25% 27 C 333 FA 15.0% SF 15.0% 333 71.3 71.3 202.40 793 830 2.75% 1.00% 28 C 309 FA 20.0% SF 15.0% 309 71.3 95.0 205.30 778 830 3.00% 0.50% METHODOLOGY ( Phase-I) ( Javed Khalid,2016)

22. 4 Descriptive Statistics Variables N Min. Max Mean Std. Deviation Variance Skewness Kurtosis Cement (kg/m 3 ) 28.00 273.00 475.00 352.11 46.86 2195.43 0.56 0.31 Silica Fumes (kg/m 3 ) 28.00 1.00 95.00 46.17 25.82 666.47 -0.51 -0.22 Fly Ash (kg/m 3 ) 28.00 1.00 166.00 77.37 51.59 2661.82 -0.19 -1.06 Water (kg/m 3 ) 28.00 175.20 206.20 192.91 8.81 77.59 -0.39 -0.84 Fine Agg . (kg/m 3 ) 28.00 773.00 954.00 843.43 48.39 2341.66 0.59 -0.37 Coarse Agg . (kg/m 3 ) 28.00 830.00 830.00 830.00 0.00 0.00     HRWR (%) 28.00 0.02 0.04 0.03 0.01 0.00 0.77 -0.33 VMA (%) 28.00 0.00 0.02 0.01 0.00 0.00 -0.25 -1.12 CS ( Mpa ) 28.00 21.51 45.55 29.81 5.69 32.42 1.05 1.39 Slump (cm) 28.00 52.50 74.80 61.50 5.35 28.64 0.34 -0.03 STATISTICAL EVALUATION ( Phase-I)

23. 4 Box P lot Indicating Ranges of Concrete C onstituents STATISTICAL EVALUATION ( Phase-I)

24. 4 STATISTICAL EVALUATION ( Phase-I) Relationship of CS With Other Concrete Constituents

25. 4 STATISTICAL EVALUATION ( Phase-I) Relationship of Slump flow With O ther Concrete Constituents

26. 4 F requency of Occurrence of Compressive S trength STATISTICAL EVALUATION ( Phase-I) 222 22-28 29-35 36-42 43-48

27. 4 STATISTICAL EVALUATION (Phase-I) Correlation Heat M ap with Compressive Strength

28. 4 DEVELOPMENT OF NLR MODEL ( Phase-I) Predictive Model/Equation for CS (NLR) CS = 2.8x10 -11 x (PC) -1.404 x ( SF) 0.035 x ( FAsh ) -0.081 x (W) 3.712 x (FA) 2.884 x (CA) -0.420 x (HRWR) 0.07 x ( VMA) 0.073 + (-1.143X10 -9 ) x (PC) 0.999 x ( SF) 0.956 x ( FAsh ) 1 x (W) 0.998 x (F.A) 1.001 x (CA) 0.008 x (HRWR) 0.973 SC = Compressive strength ( Mpa ) CA = Coarse Aggregate (Kg/m 3 ) FA = Fine Aggregate ( Kg/m 3 ) W = Water ( Kg/m 3 ) Fash = Fly Ash ( Kg/m 3 ) PC = Portland Cement ( Kg/m 3 ) SF = Silica Fumes (%) HRWR = High range water reducing admixture (%) W/B = Water to binder ratio

29. 4 DEVELOPMENT OF NLR MODEL ( Phase-I) Predictive Model/Equation for Slump Flow (NLR) Slump =2.057 x (PC) 0.572 + 5.958 x (SF) 0.206 + 0.015 x (FA) 1.32 + (-32.145) x (W) -0.091 + 0.01 x (FA) -5.675 + 0.026 x (CA) -5.544 + 970.519 x (HRWR) 1.403 + -1.57E-06 x (VMA) -2.397 Slump = Diameter of slump flow (cm) CA = Coarse Aggregate (Kg/m 3 ) FA = Fine Aggregate ( Kg/m 3 ) W = Water ( Kg/m 3 ) Fash = Fly Ash ( Kg/m 3 ) PC = Portland Cement ( Kg/m 3 ) SF = Silica Fumes (%) HRWR = High range water reducing admixture (%) W/B = Water to binder ratio

30. 4 DEVELOPMENT OF ANN MODEL ( Phase- I ) Predictive Model/Equation for CS (ANN) CS = + + ………. + + threshold   Where β ’ = Values obtained from matrix Node = weight in hidden layer Synaptic Weight = Strength of interconnected neurons. Threshold = Bias term (add flexibility to data) Values in the matrix are the synaptic weights in input layer CS = + + 0.324    

31. 4 DEVELOPMENT OF ANN MODEL ( Phase-I) Predictive Model/Equation for Slump (ANN) Slump = + + ………. + + threshold   Slump= + + (-0.114)     Where β ’ = Values obtained from matrix Node = weight in hidden layer Synaptic Weight = Strength of interconnected neurons. Threshold = Bias term (add flexibility to data) Values in the matrix are the synaptic weights in input layer

32. 4 RESULTS AND DISCUSSIONS ( Phase-I) Comparison Between M easured and Predicted CS Sr. # Mixture ID Measured CS 28 Days NLR Prediction Residual ANN Prediction Residual Mpa Mpa Mpa Mpa Mpa 1 C 475 FA 0 SF 0 27.8 27.4 0.4 27.5 0.3 2 C 380 FA 20 SF 0 22.5 21.6 0.9 21.6 0.9 3 C 356 FA 25 SF 0 23.9 24.2 -0.3 22.1 1.8 4 C 333 FA 30 SF 0 25.5 26.6 -1.1 23.8 1.7 5 C 309 FA 35 SF 0 28.9 28.8 0.1 28.8 0.1 6 C 428 FA 0 SF 10 30.7 29.7 1.0 30.0 0.7 7 C 416 FA 0 SF 12.5 33.0 33.5 -0.5 31.7 1.3 8 C 404 FA 0 SF 15 34.7 37.3 -2.6 36.6 1.8 9 C 392 FA 0 SF 17.5 45.6 41.2 4.3 42.2 3.4 10 C 380 FA 0 SF 20 42.9 45.2 -2.3 46.2 3.2 11 C 392 FA 10 SF 7.5 23.3 23.3 0.0 23.9 0.6 12 C 368 FA 15 SF 7.5 25.0 26.7 -1.7 26.3 1.3 13 C 344 FA 20 SF 7.5 30.5 29.5 1.0 30.3 0.2 14 C 321 FA 25 SF 7.5 36.8 33.3 3.5 35.2 1.6

33. 4 15 C 297 FA 30 SF 7.5 34.0 35.0 -1.0 35.5 1.5 16 C 380 FA 10 SF 10 24.7 24.6 0.1 23.3 1.5 17 C 356 FA 15 SF 10 27.0 26.5 0.6 26.7 0.3 18 C 333 FA 20 SF 10 30.0 29.0 1.0 31.4 1.4 19 C 309 FA 25 SF 10 31.0 32.2 -1.2 31.8 0.8 20 C 285 FA 30 SF 10 31.6 33.6 -2.0 31.9 0.3 21 C 368 FA 10 SF 12.5 25.0 25.1 -0.2 23.3 1.7 22 C 344 FA 15 SF 12.5 26.4 26.5 -0.1 26.7 0.2 23 C 321 FA 20 SF 12.5 27.0 28.6 -1.6 31.6 4.5 24 C 297 FA 25 SF 12.5 29.8 30.1 -0.3 31.6 1.8 25 C 273 FA 30 SF 12.5 32.3 32.4 -0.1 31.4 0.8 26 C 356 FA 10 SF 15 21.5 26.3 -4.8 23.0 1.5 27 C 333 FA 15 SF 15 29.2 27.2 2.0 28.8 0.3 28 C 309 FA 20 SF 15 34.2 29.3 4.9 32.2 2.0 Sr. # Mixture ID Measured CS 28 Days NLR Prediction Residual ANN Prediction Residual Mpa Mpa Mpa Mpa Mpa RESULTS AND DISCUSSIONS ( Phase-I) Comparison Between M easured and Predicted CS

34. 4 Comparison B etween M easured and Predicted Slump Values Sr. # Mixture ID Slump flow NLR Prediction Residual ANN Prediction Residual c m c m c m c m c m 1 C 475 FA 0 SF 0 59.5 59.61 0.11 61.54 2.04 2 C 380 FA 20 SF 0 54.0 57.24 3.24 59.95 5.95 3 C 356 FA 25 SF 0 58.0 57.11 0.89 59.89 1.89 4 C 333 FA 30 SF 0 59.5 57.15 2.35 59.87 0.37 5 C 309 FA 35 SF 0 55.0 57.05 2.05 59.86 4.86 6 C 428 FA 0 SF 10 62.0 62.83 0.83 61.69 0.31 7 C 416 FA 0 SF 12.5 57.0 62.39 5.39 61.84 4.84 8 C 404 FA 0 SF 15 6 2.5 61.9 0.6 61.82 0.68 9 C 392 FA 0 SF 17.5 64.0 61.33 2.67 60.41 3.59 10 C 380 FA 0 SF 20 64.5 62.29 2.21 60.82 3.68 11 C 392 FA 10 SF 7.5 64.5 62.88 1.62 60.84 3.66 12 C 368 FA 15 SF 7.5 65.0 64.09 0.91 62.74 2.26 13 C 344 FA 20 SF 7.5 65.5 62.7 2.8 61.34 4.16 14 C 321 FA 25 SF 7.5 55.5 63.11 7.61 60.36 4 .86 RESULTS AND DISCUSSIONS ( Phase-I)

35. 4 15 C 297 FA 30 SF 7.5 67.5 63.35 4.15 64.36 3 .14 16 C 380 FA 10 SF 10 63.3 61.99 1.31 61.33 1.97 17 C 356 FA 15 SF 10 63.0 62.2 0.8 61.32 1.68 18 C 333 FA 20 SF 10 53.8 58.27 4 .47 55.51 1 .71 19 C 309 FA 25 SF 10 59.6 62.99 3.39 62.66 3.06 20 C 285 FA 30 SF 10 64.5 61.43 3.07 61.3 3.2 21 C 368 FA 10 SF 12.5 58.3 61.7 3.4 61.32 3.02 22 C 344 FA 15 SF 12.5 56.0 61.65 5.65 57.49 1.49 23 C 321 FA 20 SF 12.5 66 61.31 4.69 61.79 4.21 24 C 297 FA 25 SF 12.5 70.5 65.57 4.93 66.95 3 .55 25 C 273 FA 30 SF 12.5 74.8 68.44 5.36 69.7 5.1 26 C 356 FA 10 SF 15 60 62.44 2.44 61.35 1.35 27 C 333 FA 15 SF 15 67 63.59 3.41 62.68 4.32 28 C 309 FA 20 SF 15 61.3 61.37 0.07 62.2 0.9 Sr. # Mixture ID Slump flow NLR Prediction Residual ANN Prediction Residual c m c m c m c m c m RESULTS AND DISCUSSIONS ( Phase-I) Comparison B etween M easured and Predicted Slump Values

36. 4 RESULTS AND DISCUSSIONS ( Phase-I) Validation Parameters for Compressive S trength Parameters Equations Developed Models Range Best Values NLR ANN R 2 0.75 0.913 0-1 1 RMSE 3.292 1.694 0 - ∞ MAE 2.26 1.35 0 - ∞ SI 0.11 0.057 < 0.1 Excellent 0.1-0.2 Good 0.2-0.3 Fair > 0.3 Poor

37. Parameters Equations Developed Models Range Best Values NLR ANN R 2 0.71 0.856 0-1 1 RMSE 3.532 1.75 0 - ∞ MAE 2.43 1.2 0 - ∞ SI 0.15 0.062 < 0.1 Excellent 0.1-0.2 Good 0.2-0.3 Fair > 0.3 Poor RESULTS AND DISCUSSIONS ( Phase-I) Validation Parameters for Slump F low V alue

38. 4 METHODOLOGY ( Phase-II) Abstract of Constituent Materials Sr. No. Fine Aggregate Coarse Aggregate Cement Fly Ash Water W/B Superplasticizer Slump flow CS kg/m3 kg/m3 kg/m3 kg/m3 kg/m3 kg/m3 kg/m3 kg/m3 kg/m3 1 886.77 843.55 471.43 1.00 165.00 0.35 7.07 69.00 43.00 2 886.77 843.55 471.43 1.00 165.00 0.35 7.07 70.50 42.00 3 886.77 843.55 471.43 1.00 165.00 0.35 7.07 72.50 41.00 4 886.77 843.55 471.43 1.00 165.00 0.35 7.07 74.00 40.00 5 913.00 837.00 290.00 100.00 175.50 0.45 2.32 43.40 42.70 6 478.00 837.00 250.00 261.00 281.05 0.55 1.25 70.50 17.00 7 910.00 837.00 210.00 100.00 201.50 0.65 1.68 57.50 19.10 8 742.00 837.00 250.00 160.00 225.50 0.55 2.00 62.50 24.10 9 786.00 837.00 210.00 220.00 193.50 0.45 1.68 55.50 26.70 10 709.00 837.00 290.00 100.00 253.50 0.65 0.58 62.30 26.60 11 625.00 837.00 290.00 220.00 229.50 0.45 0.58 34.50 32.90 12 742.00 837.00 250.00 160.00 225.50 0.55 1.25 60.50 26.00 13 742.00 837.00 250.00 160.00 225.50 0.55 1.25 62.50 28.50

39. 4 METHODOLOGY ( Phase-II) Abstract of Constituent Materials Sr. No. Fine Aggregate Coarse Aggregate Cement Fly Ash Water W/B Superplasticizer Slump flow CS kg/m3 kg/m3 kg/m3 kg/m3 kg/m3 kg/m3 kg/m3 MPa MPa 14 742.00 837.00 250.00 160.00 225.50 0.55 1.25 60.50 26.40 15 739.00 837.00 250.00 160.00 225.50 0.55 0.00 41.90 27.30 16 594.00 837.00 317.00 160.00 262.35 0.55 1.59 69.70 29.10 17 562.00 837.00 210.00 220.00 279.50 0.65 0.42 73.70 10.20 18 742.00 837.00 250.00 160.00 225.50 0.55 1.25 60.00 25.30 19 919.00 837.00 250.00 160.00 155.80 0.38 1.25 20.00 36.30 20 746.00 837.00 250.00 160.00 225.50 0.55 2.50 79.00 26.70 21 566.00 837.00 250.00 160.00 295.20 0.72 1.25 88.00 11.00 22 891.00 837.00 183.00 160.00 188.65 0.55 0.92 36.10 22.10 23 925.00 750.00 340.00 180.00 200.00 0.38 1.00 69.00 49.00 24 925.00 750.00 340.00 180.00 220.00 0.42 1.00 69.00 39.00 25 925.00 750.00 340.00 190.00 230.00 0.43 1.00 69.00 46.00 26 925.00 700.00 340.00 190.00 220.00 0.42 1.00 69.00 45.00

40. 4 Descriptive Statistics Variables N Min. Max. Mean Std. Deviation Variance Skewness Kurtosis Cement (kg/m 3 ) 289 35 541 399.9 102.31 10469.73 0.243 -0.142 CS (MPa) 289 5 50 35 9.39 88.29 -1.463 -1.281 FA (kg/m 3 ) 289 100 1207 809.4 124.6 1554.2 0.592 0.131 Slump flow (cm) 289 100 520 313 8.21 74.48 0.492 0.131 SP (kg/m3) 289 28 19.58 17.73 31.15 W/B Ratio 289 0.22 0.77 36.97 0.0938 91.9 0.509 0.077 CA (kg/m 3 ) 289 561 1118 839.4 7.84 17777.80 0.21 -0.047 Fly Ash ( kg/m 3 ) 289 1 525 135.14 5.04 7439.49 0.59 1.53 Water (kg/m 3 ) 289 36.36 319 197.35 36.85 1358.5 0.088 1.98 STATISTICAL EVALUATION ( Phase-II)

41. 4 STATISTICAL EVALUATION ( Phase-II) Box P lot Indicating Ranges of Concrete C onstituents

42. 4 STATISTICAL EVALUATION ( Phase-II) Relationship of CS With Other Concrete Constituents

43. 4 STATISTICAL EVALUATION ( Phase-II) Relationship of Slump Flow With Other Concrete Constituents

44. 4 STATISTICAL EVALUATION ( Phase-II) 4-10 10-15 15-20 20-25 25-30 30-36 36-42 42-47 47-52 F requency of Occurrence of Compressive S trength

45. 4 STATISTICAL EVALUATION ( Phase-II) Correlation Heat M ap with CS and Slump Flow

46. 4 DEVELOPMENT OF NLR MODEL ( Phase-II) Predictive Model/Equation for CS (NLR) CS = 0.001 x FA 1.362 + 0.061 x CA 0.791 + 3.457x C 0.209 + (-9.245) x Fash 0.051 + 4.475 x W 0.227 + -42.015 x W/B 2.355 + 1.314 x SP 0.548 SC = Compressive strength ( Mpa ) CA = Coarse Aggregate (Kg/m 3 ) FA = Fine Aggregate (Kg/m 3 ) W = Water (Kg/m 3 ) Fash = Fly Ash (Kg/m 3 ) PC = Portland Cement (Kg/m 3 ) SF = Silica Fumes (Kg/m 3 ) W/B = Water to binder ratio

47. 4 DEVELOPMENT OF NLR MODEL ( Phase-II) Predictive Model/Equation for Slump Flow (NLR) Slump = 0.043x FA 0.839 + 0.045 x CA 0.893 + 13.942x C 0.153 + (5.974) x Fash 0.231 + 16.982 x W 0.159 + (-120.947) W/B 170.061 + (-29.916) x SP- 0.293 Slump = Diameter of slump flow (cm) CA= Coarse Aggregate (Kg/m 3 ) FA = Fine Aggregate (Kg/m 3 ) W = Water (Kg/m 3 ) Fash = Fly Ash (Kg/m 3 ) PC = Portland Cement (Kg/m 3 ) SF = Silica Fumes (Kg/m 3 ) W/B = Water to binder ratio

48. 4 DEVELOPMENT OF ANN MODEL ( Phase-II) Predictive Model/Equation for CS (ANN) CS = + + ………. + + threshold   CS = + + (-0.686)   Where β ’ = Values obtained from matrix Node = weight in hidden layer Synaptic Weight = Strength of interconnected neurons. Threshold = Bias term (add flexibility to data) Values in the matrix are the synaptic weights in input layer  

49. 4 DEVELOPMENT OF ANN MODEL ( Phase-II) Predictive Model/Equation for Slump (ANN) Slump = + + ………. + + threshold   Slump = + + + 1.158   Where β ’ = Values obtained from matrix Node = weight in hidden layer Synaptic Weight = Strength of interconnected neurons. Threshold = Bias term (add flexibility to data) Values in the matrix are the synaptic weights in input layer  

50. 4 RESULTS AND DISCUSSIONS ( Phase-II) Comparison Between M easured and Predicted CS Sr. # Measured CS 28 Days NLR Prediction Residual ANN Prediction Residual Mpa Mpa Mpa Mpa Mpa 1 43.00 42.20 0.80 38.30 4.70 2 42.00 42.20 0.20 38.30 3.70 3 41.00 42.20 1.20 38.30 2.70 4 40.00 42.20 2.20 38.30 1.70 5 42.70 34.89 7.81 42.00 0.70 6 17.00 24.09 7.09 25.33 8.33 7 19.10 24.62 5.52 24.62 5.52 8 24.10 27.68 3.58 28.16 4.06 9 26.70 31.30 4.60 31.67 4.97 10 26.60 22.00 4.60 20.14 6.46 11 32.90 29.48 3.42 34.71 1.81 12 26.00 27.27 1.27 27.63 1.63 13 28.50 27.27 1.23 27.63 0.87 14 26.40 27.27 0.87 27.63 1.23

51. 4 Comparison Between M easured and Predicted Slump Flow Sr. # Slump flow NLR Prediction Residual ANN Prediction Residual cm cm cm cm cm 1 69.00 6 5.06 3.94 70.39 1.39 2 70.50 6 5.06 5.44 70.39 0.11 3 72.50 65.06 7.44 70.39 2.11 4 74.00 6 5.06 18.94 70.39 3.61 5 43.40 5 1.98 8.58 52.74 9.34 6 70.50 60.26 10.24 69.21 1.29 7 57.50 60.13 2.63 56.02 1.48 8 62.50 61.81 0.69 65.74 3.24 9 55.50 62.93 7.43 57.24 1.74 10 62.30 54.07 8.23 65.17 2.87 11 6 4.50 57.15 7.35 56.88 7.35 12 60.50 59.54 0.96 62.05 1.55 13 62.50 59.54 2.96 62.05 0.45 14 60.50 59.54 0.96 62.05 1.55 RESULTS AND DISCUSSIONS ( Phase-II)

52. 4 RESULTS AND DISCUSSIONS ( Phase-II) Parameters Equations Developed Models Range Best Values NLR ANN R 2 0.64 0.76 0-1 1 RMSE 4.28 2.35 0 - ∞ MAE 2.98 1.71 0 - ∞ SI 0.25 0.073 < 0.1 Excellent 0.1-0.2 Good 0.2-0.3 Fair > 0.3 Poor Validation Parameters for Compressive S trength

53. RESULTS AND DISCUSSIONS ( Phase-II) Parameters Equations Developed Models Range Best Values NLR ANN R 2 0.595 0.69 0-1 1 RMSE 4.58 2.75 0 - ∞ MAE 3.21 1.91 0 - ∞ SI 0.27 0.1 < 0.1 Excellent 0.1-0.2 Good 0.2-0.3 Fair > 0.3 Poor Validation Parameters for Slump F low

54. CONCLUSION Developed models for ANNs outperformed the NLR model with higher R 2 value of 0.913 and lower RMSE (1.694), MAE (1.35) and SI (0.057) vales. ANNs can significantly save computational effort as it simplify the test protocol required to optimize a given mixture by reducing the number of trial batches to achieve mixture variables. Predicted results align well with experimental values in all phases: training, testing, and validation. Normally, determining compressive strength takes 28 days but ANN model can predict the compressive strength value quickly and accurately .

55. 4 Thank You