SeminarTopics.co.in

Analysis of Durability Of High Performance Concrete Using Artificial Neural Networks


Published on Feb 14, 2016

Abstract

The essence of high performance concrete (HPC) emphasizes three main characteristics namely active mineral additives like fly ash, silica fume & super plasticizers apart from three basic ingredients i.e.; cement, aggregate & water in conventional concrete. They have been incorporated to make highly workable ,high strength and durable concrete. HPC design seems more complicated due to more number of ingredients.

Maintaining a low water -binder ratio with adequate workability is more complicated. Traditionally expert civil engineers can produce HPC mix proportions by using empirical results from previous research plus their experience to achieve required performance. But the number of components in the making of concrete has gone up to 10. This makes the empirical methods insufficient as the number of properties to be investigated has gone up as well.

Supervised Learning

During the training session of a neural network, an input stimulus is applied that results in an output response. The response is compared with the target response. If the actual response differs from the target response, the neural network generates an error signal, which is then used to calculate the adjustment that should be made to the network synaptic weights so that the actual output matches the target output. In other words, the error is minimized, possibly to zero. The two commonly used learning techniques are the Delta rule and Gradient Descend rule.

The Delta rule is based on the idea of continuous adjustments of the value of the weights such that the difference of the error (delta) between the desired output value and the actual output value of a processing element is reduced. This is also known as the Widrow-Hoff learning rule.

In the Gradient Descend rule, the values of the weights are adjusted by an amount proportional to the first derivative (the gradient) of the error between the desired output value and the actual output value of a processing element, with respect to the value of the weight.

Experimental Program And Data Collection

The first step in developing the network is to obtain good and reliable training and testing examples. To obtain the data for developing the neural network models, a database of high strength and durable concrete is produced by collecting the data sets from experiments by Parichatprecha combined with data sets from previous researches . The influence of using different pozzolanic materials, cement content, and water-to-binder ( W / B ) ratios on the durability of concrete was experimentally investigated by measuring the charge passed of concrete in accordance with ASTM C1202-97.

The workability of concrete expressed in terms of slump was kept constant by varying the dosage of superplasticizer based on poly-carboxylic ether (PCE). Two types of pozzolanic material were used, namely pulverized fly ash and a combination of pulverized fly ash and condensed silica fume. The cementitous materials were varied from 400-550 kg/m 3 with W / B ranging from 0.3 to 0.4. Control specimens without pozzolanic materials of concrete were also cast and tested for comparison. ASTM C1202-97 Rapid Chloride Permeability Test (RCPT) was used in this experimental program for hardened concrete.

Analysis of Durability Of High Performance Concrete Using Artificial Neural Networks

 

This test method covers the determination of the electrical conductance of concrete to provide a rapid indication of its resistance to the penetration of chloride ions. After 28 days’ curing, cylindrical specimens of 100 mm diameter and 200 mm length were cut to 50 mm thick on each end. These specimens were saturated in water for 18 ± 2 h until fully saturated and then allowed to surface dry in air for at least 1 h. Next the specimens were placed on suitable silicon and complete coating of all surfaces was ensured. One side of the cell contained 3.0% NaCl solution and the other 0.3 M NaOH solution.

The current (ampere-seconds) was recorded at 30-min intervals during a testing period of 6 h. Based on the charge that passed through the sample, a qualitative rating was made of the concrete’s permeability, as shown in Table 2 in accordance with ASTM C1202-97. A total of 30 mixes were made and the specimens were tested for their charge passed over a duration of 6 h.

To expand the prediction range of the model built with the experimental data, 56 concrete mixtures and their test results were culled from previous researches . The 28-day compressive strength of all data is in the range of 30–120 MPa. Of these, the ANNs model is developed, trained and tested by using a total of 86 data sets. Table 1 illustrates the general details of the concrete evaluation in this study. The data used in ANNs model are arranged in a format of eight input parameters which include OPC, F, SF, W, SP, CA, FA, and W/B ratio. To test the reliability and accuracy of the models, 20% of the 86 data sets were randomly selected as test sets, while the remaining 70 samples were used to train the network.

The output of the model is the total charge passed in accordance with ASTM C1202 or AASHTO T277. The input and output of a typical neural network is in the range of 0–1. The use of the higher number is not desirable as the networks are generally simulated on a computer and this can create floating-point overflow problems . Therefore, setting the input and output in the range of 0–1 is essential to normalize their values to suit the network’s functioning. In this study, x/xmax normalization technique was applied for transforming the input and output values remaining in the range of 0–1.

NEURAL NETWORKS FOR MODELLING DURABILITY OF HPC

The electrical conductivity of concrete is determined by both pore structure and the chemistry of the pore solution, which are dependent on the dosage of cement, water, SP, fine aggregate, coarse aggregate and type and dosage of pozzolanic materials. The ANNs model developed in this study has eight neurons in the input layer, one hidden layer, and an output layer as shown in Fig. 4. The selection of the number of nodes in the hidden layer is the most challenging part in the total network development process. Unfortunately, there are no fixed guidelines available for this purpose and hence this has to be done by the trial-and-error method.

In this study, the neural networks were developed and performed under MATLAB programming. The learning algorithm used in the study was gradient descent with adaptive learning rate back-propagation, a network training function that updates weight and bias values according to gradient descent with adaptive learning rate . The error incurred during the learning process was expressed in terms of mean-squared-error (MSE).

After a number of trials as shows in Fig. 5, the best network architecture and parameters that minimize the MSE error of training data were selected as follows:

• 8 input units;

• 1 hidden layer;

• 25 hidden units;

• 1 output unit;

• Activation function = sigmoidal function;

• Learning rate = 0.1

• Learning cycles = 10,000.

CONCLUSIONS

The following conclusions were drawn from the above case study :

• The statistical test results indicate that the models are reliable, accurate, and illustrate how ANNs can be used to efficiently predict the durability of HPC.

• Based on the simulated total charge passed model built using trained neural networks, the optimum content of various ingredients of HPC were obtained.

• Although the capability of the proposed network is limited to the data located within the available range of training data in the database, the available range of the system could be easily expanded by retraining the neural networks with additional data from trial mixes.

• The capability of ANNs to predict complicated computations has been explored in the casestudy. This feature of ANNs can be extended to various fields of Civil Engineering to solve the problem of result forecasting.