Mild steel corrosion

ARTIFICIAL NEURAL NETWORK AS A PREDICTION TOOL FOR MILD STEEL CORROSION

Abstract

Mild steel is one of the major construction materials, which is extensively used in Chemical and allied industries for the handling of acid, alkali and salt solutions. The corrosion behavior of Mild Steel in Hydro Chloric Acid is investigated by Mass Loss measurements method. The Rate of Corrosion for mild steel at various aqueous environments by varying Chloride ion concentration (0.1N to 0.75 N), pH (0.12 to 1) and Temperature (290K to 333K) is modeled by means of Artificial Neural Network. Fifty values of Rate of Corrosion with 3 neurons at the input are used to model the corrosion behavior of Mild Steel, using Back Propagation Algorithm Neural Network approach. The applicability of the developed Model is verified by comparing the computed results and the experimental results obtained in this study. It is found that the empirical model developed by using Neural Network seemed to have a high prediction capability of the Rate of Corrosion.

Key words: Mild Steel, Rate of Corrosion, Neural Network, pH, Temperature, Chloride Concentration.

1. Introduction

Mild steel finds application in many industries due to its easy availability, ease of fabrication, low cost and good tensile strength besides various other desirable properties. It suffers from severe Corrosion when it comes in contact with Chloride solutions, during acid cleaning, transportation of acid, de scaling, storage of acids and other chemical processes [1-3]. The Corrosion of Mild Steel is of fundamental academic and industrial concern that has received a considerable amount of attention. Chloride ions are the major cause for the Corrosion of steel regarding Rate of Corrosion determined when the steel is in contact with Chloride solution. Hydro Chloric Acid is the most difficult of the common acids to handle from the standpoints of Corrosion and materials of constructions. Extreme care is required in the selection of materials to handle the acid by itself, even in relatively dilute concentrations or in process solutions containing appreciable amount of Hydro Chloric Acid. This acid is very corrosive to most of the common metals and alloys [4]. Various experimental studies have been conducted considering the different parameters. The effect of temperature on a chemical reaction of practical and theoretical important [5] like most chemical reactions, the Rate of Corrosion of Mild Steel increases with temperature especially in media in which evolution of hydrogen accompanies corrosion, e.g; during Corrosion of steel in acids. The most significant environmental conditions, which influence the Corrosion behavior of steels, are the Chloride ion concentration, Temperature and pH. Reliable prediction of the Rate of Corrosion behavior is the fundamental step towards effective control of Corrosion. At first thought, this may appear easy due to the fact that extensive Corrosion literature is available. However, this may not be easy, since the variability of the Corrosion process may be involved. For example, a set of Corrosion data collected in particular environment cannot be considered for all changes in environmental parameters. Also real world Corrosion never seems to involve the same conditions as laboratory tests. The problem is not lack of information, but lack of techniques to made use of that information. Artificial Neural Network (ANN) technique [6] is suited for the problem that involves non linear interpolation. Ability to learn by example makes Neural Network flexible and powerful [7]. In recent past, ANN has been successfully applied to model various Corrosion behaviors [8-11].

In this study, a Neural Network model using Back Propagation Algorithm (BPA) has been used to predict the Rate of Corrosion of Mild Steel with three input neurons such as pH, Temperature and Chloride concentration. Out of Fifty data set, thirty data set of Rate of Corrosion of Mild Steel (by Mass Loss measurement) has been taken for training set. Twenty data is used for testing pattern. A good correlation between experimental and predicted data is obtained, which shows a high prediction capability of Neural Network.

2. Experimental

2.1 Materials

Mild Steel specimens are cut to size of 5 x 1 cm from Mild Steel sheets having the percentage composition: Fe = 99.686, Ni =0.013, Cr=0.043, S=0.014, P=0.009, Si=0.007, Mn=0.196 and C = 0.017. The surfaces of specimens are polished with emery papers ranging from 110 to 410 grades. The Hydro Chloride Acid with various concentrations of 0.1N, 0.15N, 0.2N, 0.25N, 0.3N, 0.35N, 0.4N, 0.5N and 0.75N with pH from (0.12- 1) are prepared. All the solutions are prepared with AR grade chemicals in double distilled water.

2.2 Methods

2.2.1 Mass Loss measurement

Mass Loss measurements are performed as per American Society for Testing and Materials (ASTM) method [12] to determine the Rate of Corrosion. The polished Mild Steels are initially weighed in an electronic balance. After that the specimens are suspended with the help of thread and glass rod in 100 ml beaker containing Hydro Chloric Acid having concentration range from 0.1N to 0.75N with pH (0.12-1) at different temperature varied from 290K to 333K. The specimens are removed after 4 hours exposure period. They are dried and reweighed to determine the Rate of Corrosion.

3. Neural Network Modeling

3.1 Overview of Neural Network

Neural Network is a functional abstraction of the biological neural structures of the central nervous system. The fundamental unit of ANN is the processing element, also called an Artificial neuron or simply a neuron. Some neurons interact with the real world to receive input and some provide the real world with the output. Rests of the neurons remain hidden. A neuron basically contains three main components namely weight, bias, and activation function. A typical neuron can be seen in Figure 1. Multi layer perception (MLP) is the basic and commonly used ANN model. There are at least three main layers in MLP, which are input, output, and hidden layers. Each neuron in the input layer is connected to hidden neurons in the hidden layer, and there are no connections among the units of the same layer. The number of neurons in each layer may vary depending on the problem. The neurons are expanded by a threshold factor and a logistic sigmoid function expressed as output = (1+ e-input)-1 is employed for the activation function [13, 14].

3.2 Explicit Neural Network formulation for Rate of Corrosion

In this study, the Back Propagation Algorithm is used for training the Neural Network. The Neural Network Architecture that proved able to solve the problem has three input neurons, one output neuron and one hidden layer with five processing elements (neurons) respectively. In the input data used, the temperature lies in the interval 290K to 333k, the Chloride concentration varies from 0.1N to 0.75N while the pH values are in the range of 0.12 to 1. The Rate of Corrosion measured by Mass Loss measurements method is used as output. Out of 50 data set, 30 data set are used to train the NN while the remaining is used for testing. Since test data set are not used for training, it essentially verified the ability of any ANN model to associate and generalize a true physical response, which is unknown to the network.

3.3 Training Pattern

The training for learning a set of data is performed with weight (connection strength), transfer function, and biases. When the training pattern is given, the network produces some output based on the current state of its synaptic weights. This output is compared to the known output, and a mean squared error is calculated. The error value is then 'back propagated' through the network and small changes are made to the weights in each layer. The weight changes are calculated to reduce the error signal. The cycle is repeated until the desired output is reached. The learning process, or training, forms the interconnection between neurons and is accomplished by known inputs and outputs, and presenting these to the ANN in some ordered manner. Due to the interconnection, signals are sent from the input layer to the output layer through the hidden layer. The intensity of the transmitted signal is determined by the weight of the interconnections. It is used to properly obtain the model by iteratively adjusting the values of interconnections between the neurons while the sum of squared residuals between calculated and expected values are minimized (15, 16). Prior to the execution of the Training process of Neural Network, the Input and Output parameters were normalized in the range of (-0.95; 0.95) via in order to acquire accurate results.

The Back Propagation Algorithm (or) the generalized delta rule uses sigmoid activation function f(x) = 1/ (1+e-x). The BPA is a supervised learning algorithm that aims at reducing overall system error to a minimum. In this learning procedure, an initial weight vectors w0 is updated, wi (k+1) = wi (k) + µ (Ti-Oi)f'(wixi)xi

Where, wi = The weight matrix associated with ith neuron; xi = Input of the ith neuron; Oi = Actual output of the ith neuron; Ti = Target output of the ith neuron, µ is the learning rate parameter. The Training data set is given in Table 1.

The error between calculated results and expected results are decreased with increasing epochs and training for learning is finished within a target convergence. One hidden layer is found to be adequate for the present problem. Neurons in the hidden layer more than 12 are not tried in order to avoid overfitting. It could be observed that ANN model with 10 hidden neurons produced best performance and is considered to be the optimal configuration for the present problem. An iterative search for the optimum learning rate and momentum is done in Table 2. A suitable learning rate and momentum can prevent the network from being trapped in local minimum error surface. From table 2, the best learning rate and momentum are found to be 0.6 and 0.7 respectively with a hidden layer for various numbers of epochs.

3.4 Validation of the proposed NN model

In this part of the study, the developed NN model is verified through the relevant data obtained from experimental. For this, 9 data are selected randomly as seen in Table 3. Then, these data are evaluated using the developed model. To demonstrate the robustness and generalization capability of the model, the statistical test results of validation data set are also given in Table 4. It is observed that there is a significantly high compromise between the experimental and the developed model indicating the powerful performance of NN Model.

3.5 Testing Pattern

It has to be noted that having a NN learn, a training set perfectly is not the main issue, but having the network provide correct answers for data it has never seen before is therefore more NN model requires a testing phase to check the effectiveness or accuracy of the network, once it has been trained [17,18]. To make sure that the network has not just memorized the training data but really extracted the general features of the problem, some new examples, not included in the training set, are presented to the NN. If the performance of the network on this test set is satisfactory, the network can be assumed properly trained and is ready to be used. During the Testing (external validation), the simulated model is used to compute the Rate of Corrosion of Mild Steel samples for various pH and Chloride Concentration at the Temperatures of 318K and 333K, which are not involved in training pattern. These results are plotted in a linear regression pattern (Computed versus Experimental Results) which is shown in Figure 2 and Figure 3 for the temperatures 318K and 333K respectively.

4. Results and Discussion

4.1 Performance of the developed model

A high prediction capability is achieved for testing data set even though it was not employed in the training of the NN. Therefore the NN appears to have a high generalization capability. The overall performances of Validation data set and testing data set are evaluated via mean absolute percentage error and the correlation coefficient r which is shown in Table 5. As seen in Table 5, high correlation coefficient 0.99 is achieved for both Validation and testing data sets. Moreover the model provided highly reasonable mean absolute percentage errors 2% and 3% for the Validation and testing data sets respectively. The results of testing phase in Figure 2 and 3 indicated that the NN was reasonably high capable of generalizing between the input variables and the output response.

4.2 Effect of Chloride content

Mass Loss measurement has been used in the present study to determine the Rate of Corrosion. The variation of Corrosion Rate of Mild Steel with Chloride ion concentration in acidic environment is shown in figure 4, which shows that Rate of Corrosion increases with increase in Chloride concentration at 290K temperature. It is because of Chloride ions are highly aggressive for Mild steel due to very high solubility of iron Chloride. In this study, the range of Rate of Corrosion varied from 1.396 mmpy to 507.91mmpy by varying Chloride concentration from 0.1N to 0.75N and temperature from 290K to 333K. It has been suggested that Corrosion occurred as a result of adsorption of aggressive anions on film followed by penetration of this film under the influence of an electrostatic field [19].

4.3 Effect of pH and temperature

Figure 5 shows the dependence of Corrosion on the pH at 290K temperature. It could be observed that the Rate of Corrosion increases with decrease in pH. This strong influence of pH on the Rate of Corrosion of mild steel could be attributed to the result of acceleration of Cathodic reaction due to high concentration of hydrogen ions.

Figure 6 shows that Corrosion increases with increase in temperature (290K, 303K, 310K, 318K and 333K) at 0.75N Chloride concentration. The results indicate the possibility of a temperature induced change in the Mild Steel surface. The defect structure of semiconductor Fe-Cr-Ni alloys may change from P-type to N-type with temperature. It has been argued that N-type films could be more susceptible to Corrosion initiation than P-type films due to the existence of Oxygen vacancies. This may enhance the transport of Chloride ions through the oxide lattice [20, 21]. It could be observed that almost a linear relationship has been obtained between Rate of Corrosion and Temperature. The gradient of the Rate of Corrosion versus Temperature curves varied between 1.39 Mmpy to 507 mmpy over the whole range of Chloride Concentration and pH. The combined influence of Chloride concentration and pH on Rate of Corrosion at the Temperatures 290K, 303K and 310K is shown in Figure 7, Figure 8 and Figure 9 respectively. It could be observed that Rate of Corrosion increases with the decrease in pH, and increase in Chloride concentration.

The above analysis suggests that the developed ANN model can efficiently simulate the Intricate inter relationship between the Rate of Corrosion and Various environmental parameters viz. Chloride Concentration, pH and Temperature. The model in turn would help to predict the Rate of Corrosion of Mild Steel as a function of the above environmental parameters with a high degree of accuracy and reliability.

5. Conclusion

Mild Steel is subjected to Mass Loss measurement tests in various aqueous environments by varying Chloride Concentration (0.1N to 0.75N), pH (0.12 to 1) and Temperature 290K to 333K. A three layer Neural Network Model with Back Propagation Algorithm is employed to predict the Rate of Corrosion of Mild Steel with three input parameters (pH, Chloride Concentration and Temperature). The developed Model is fast and is able to produce an output that has minimum error. On modeling with Neural Network, a good correlation between experimental and predicted data is obtained. The Correlation Coefficient of the Validation and Test data is 0.998 and 0.99 respectively, which reflects the excellent predictability of the Model. Besides, it is seen that by increasing Chloride concentration and Temperature, and decreasing pH are found to increase the Rate of Corrosion.

References:

1. Refaey S A M, Appl Surf Sci., 2005, 240(1-4), 396-404.

2. Quraishi M A and Sharma H K, J Appl Electrochem., 2005, 35(1), 33-39.

3. Ashassi-Sorkhabi A, Shaabani B and Seifzadeh D, Appl Surf Sci., 2005, 239(2), 154-164.

4.Int. J. Electrochem. Sci., 2 (2007) 278 - 284Inhibition of Acid Corrosion of Mild Steel by Pyridoxal andPyridoxol Hydrochlorides A. O. James b , N. C. Oforka b , Olusegun K. Abiola *a 5. Int. J. Electrochem. Sci., 2 (2007) 996 - 1017, Temperature Effects on the Corrosion Inhibition of Mild Steel in Acidic Solutions by Aqueous Extract of Fenugreek Leaves

Ehteram A. Noor.

6.Sha W,Edwards KL.The use of arti .cial neural networks in materials science based research.Mater Design 2007;28:1747.

7.Yilmaz Muharrem,Metin Ertunc H.The prediction of mechanical behavior for steel wires and cord materials using neural networks.Mater Design 2007;28:599.

8.Cottis RA,Qing Li,Owen G,Gartland SJ,Helliwell IA,Turega M.Neural network methods for corrosion data reduction.Mater Design 1999;20:169.

9.Smets HMG,Bogaerts WFL.Deriving corrosion knowledge from case histories: the neural network approach.Mater Design 1992;13:149.

10.Malinov S,Sha W.Application of arti .cial neural networks for modeling correlations in titanium alloys.Mater Sci Eng A 2004;365:202.

11.Parthiban Thirumalai,Ravi R,Parthiban GT,Srinivasan S,Ramakrishnan KR, Raghavan M.Neural network analysis for corrosion of steel in concrete.Corros Sci 2005;47:625.

12. E-Journal of Chemistry, 2009, 6(4), 1003-1008, Inhibition of Mild Steel Corrosion in 1N H2 SO4 Mediumby Acid Extract of Nyctanthes arbortristis Leaves R.SARATHA and V.G.VASUDHA

13.Mandal Sumantra,Sivaprasad PV,Venugopal S.Capability of a feed-forward arti .cial neural network to predict the constitutive .ow behavior of as cast 304 stainless steel under hot deformation.J Eng Mater Technol 2007;129: 242.

14.Mandal Sumantra,Sivaprasad PV,Venugopal S,Murthy KPN.Arti .cial neural network modeling to evaluate and predict the deformation behavior of stainless steel type AISI 304L during hot torsion.Appl Soft Comput 2009;9:237.

15. Ramana, K.V.S; Anitha, T; Sumantra mandal; Kaliappan, S; Shaikh, H; Siva prasad, P.V; Dayal, R.K; Khatak, H.S; Materials and design, Elsevier, 2009, 30, 3770-3775.

16. Kartalopoulos, S.V; Understanding Neural Networks and Fuzzy Logic- Basic Concepts and Applications, Prentice Hall, New Delhi. 2000.

17. H.M.G. Smets and W.F.L. Bogaerts, SCC analysis of austenitic stainless steels in chloride-bearing water by neural network techniques, Corrosion (1992) 618~23.

18.Ji T,Lin T,Lin X.A concrete mix proportion design algorithm based on arti .cial neural networks.Cement Concrete Res 2006;36:1399 -408.

19.Hoar TP, Mears D, Rothwell G, The relationships between anodic passivity, brightening and pitting , Corros Sci 1965; 5:279.

20. Manning PE, Duquette DJ. The effect of temperature (25-289°C)on pit initiation in single phase and duplex 304L stainless steels in 100ppm Chloride concentration. Corros Sci 1980;20:597.

21. Bianchi G, Cerquetti A, Mazza F, Torchio S. International conference on localized corrosion, Stachie RW, Brown BF, Kruger J, Agrawal A, editors, NACE, Houston TX, 1974. p.399.

Please be aware that the free essay that you were just reading was not written by us. This essay, and all of the others available to view on the website, were provided to us by students in exchange for services that we offer. This relationship helps our students to get an even better deal while also contributing to the biggest free essay resource in the UK!