ANN and DOE Analysis of Corrosion Resistance Inhibitor for Mild Steel Structures in Iraq

Modelling of the effect of newly developed mild steel (MS) corrosion inhibitor in Iraq was investigated using artificial neural network (ANN) and Response Surface Methodology Design of Experiment (RSM-DOE) methods. The most significant parameters among the parameters studied and the optimum coating conditions was also investigated. Weight loss method (WLM) as well as Scanning electron microscope (SEM) were used in the experimental work to obtain data for modelling. The inhibitor used was made in the center’s laboratories called N-(3-Nitrobenzylidene)-2-aminobenzothiazole. The MS specimens were tested for different immersion times and corrosive solution temperatures. Different concentrations of the inhibitor from of 0, to 1000 mg/L were used in the study. The results showed that within the concentrations studied, the corrosion inhibition performance increased with increasing N-(3-Nitrobenzylidene)-2-aminobenzothiazole concentration. The ANN model proposed with the Gaussian activation function was accurate for both testing and validation up to 99%. The RSM method used indicated that comparing time and concentration alone, inhibitor concentration was more significant than the immersion time in the corrosive solution. On the other hand, the effect of temperature and time were opposite to one another. While increasing time of immersion increased corrosion rate, temperature effect was the opposite.


Introduction
Degradation of materials by corrosion represents a challenging task to chemical and materials science engineers.This phenomenon still needs to be clearly and fully understood and characterized.This puts heavy pressure on materials scientists, researchers and industries to try to find solution for this disaster causing phenomena that causes the industrial components to be life-limited (Guthrie et al., 2010).
Several studies in the literature tried to shed light on this serious phenomenon (Melchers et al., 2010, Han and Song, 2008, Melchers, 2006, Xue et al., 2001, Melchers, 2009, Jones, 1996, Eiman and Sameh, 2012).Several factors were tried and suggested to counter the effect of this reaction including proper selection of materials, improved service and maintenance operation and provide certain coating materials to prevent or reduce the effect of corrosion, hence, improve its service life.
Seawater is the most aggressive working environment that causes corrosion.However, due to its variability, it is difficult to simulate its effect on corrosion on the laboratory scale (Mo¨ller. Et al., 2006).Researchers reported that the rate of corrosion increased four times when put for 3 weeks in a 3.5 % NaCl solution than in natural seawater (Siddique et al., 2011).
In Iraq, one of the significant source of corrosion (especially to oil industry) is the higher-than-normal levels of sulfur and water in crude oil and its products.This makes the working environment in oil industry more aggressive to piping materials.Moreover, most of the oil reserve and supply in Iraq is in the city of Basra.This city lies at the gulf shore.This makes an additional factor for corrosion to its components.Organic inhibitors offer good performance against corrosion.
They have several advantages including being economical and more practical to be used to reduce/prevent loss of materials as a result of corrosion (Porcayo-Calderón et al., 2015, Finšgar andMilošev, 2010).
Due to experimental limitations, it becomes necessary to develop other alternatives to better understand corrosion phenomena, reduce time, the number of experiments, as well as control the process.One such alternative is the use of mathematical tools.ANN, GA DOE models represent a good option to describe corrosion behavior (Chen et al., 2016, Ndukwe andAnyakwo, 2017).
ANN is based on the biological functions of the brain where connections of neurons form a network.The prediction performance depends on a learning stage and corresponds to the correlation of the inputs and outputs of the model (Khaled and Mobarak, 2012, Khaled, 2010, Millán-Ocampo et al., 2018).Several researchers used ANN in the prediction of corrosion inhibition in pipeline steel (Colorado-Garrido et al., 2009, Efimov et al, 2011, Hernández et al, 2009, Komijani et al., 2017).
Based on the above, this research was undertaken in the University of Technology, Energy and Renewable Energies Technologies Center to try to find solution for this problem.The main aim is to try to find a locally available, cheap, efficient and non-toxic corrosion inhibitor for mild steel.This inhibitor should not react or interfere with the material from which the targeted component is made of.

Synthesis of the "N-(3-Nitrobenzylidene)-2-aminobenzothiazole"
The inhibitor was synthesized following the steps shown in Figure (1).The raw chemical components were procured from Sigma Aldrich.The reflux of the amino compound namely 2-aminobenzothiazole was done with carbonyl compound namely 3-nitrobenzaldehyde in molar ratio 1:1 in the presence of ethanol as solvent.The final product was then tested for its chemical and physical properties.

Weight Loss Method
In this method, MS samples were immersed in 1 M HCL with/without the inhibitor for comparison.The study was divided into two phases.The first phase aimed at finding the best inhibitor concentration.
The inhibitor concentrations were varied for each test as follows (25, 100, 250, 500, 1000 mg/L) for the period of 1, 5, and 10 hrs and the performance was tested using weight loss technique.This part of the experiment was done at atmospheric conditions of about 303 K.In the second phase, the best performing concentration was chosen, and the temperature effect on its performance was studied.
The temperature was also varied between 303 to 333 K at an increment of 10 K.The samples were then removed, washed, dried and weighed.The inhibition efficiency (IE) was estimated using the following equation: Where o is the weight of the MS without inhibitor while 1 is the weight of the MS with inhibitor in gm.

Artificial Neural Networks
The data consisted of three inputs: Temperature (K), Time (hr) and inhibitor concentration (mg/L).The outputs for each model were represented by the amount of material corroded and inhibition efficiency.SAS Jmp® software was used for the development of the models, evaluating different combinations of activation functions and the number of neurons was increased until the best correlation between input and output variables was achieved.
The training process was proposed to minimize the prediction error of the ANN through the different connections between weights and biases.It was possible using the Gaussian function.The outputs are produced using data from neurons in the input and hidden layers, and the bias, summation and activation functions.
The net output of each layer is calculated by summation function.The summation function used in this study is given in the equation ( 2) below The activation function provides a curvilinear match between the input and output layers.Besides, it determines the output of the cell by processing the net input to the cell.
There are several activation functions that can be used for training.In this study, Gaussian function showed the best accuracy with least hidden neurons.The Gaussian transfer function of the ANN model in this study is given in the equation (3) below: = (3) Where "x" is a linear combination of the x variables.
In the output layer, the output of the network is generated by the data being processed by the hidden layer and transferred to the external world.The significant advantages of ANN are their learning ability and the freedom to use different learning algorithms.And that for the inhibition efficiency is given by equation ( 5) : Inhibition Efficiency (%) = (-336.403883795337) -1272.56684882916 * H1 + 6.97300787360302 *H2 -51.9271483590782 * H3 + 1520.03348903796 *H4 + 157.173511293188 * H5 (5) Where : As for the effect of temperature and time on weight loss (gm), the model predicted is shown below in equation ( 6 The mathematical model predicted by the algorithm has the following accuracy features: The results of the ANN model for both testing and validation is shown below in Figures (4 a and b).The last stage is to use this data obtained in the design of experiment (Response Surface Methodology) to find the most significant factor.This was done using Minitab R18.This is shown by what is called Pareto chart.These charts are used to identify the most frequent factors affecting certain output, the most common causes of this type of output.Pareto charts can help to focus improvement efforts on areas where the largest gains can be made.
Main effects plots are used to the differences between level means for one or more factors.There is a main effect when different levels of a factor affect the response differently.A main effects plot graphs the response mean for each factor level connected by a line.
The Pareto and main effect charts for all factors of this study are shown below in Figures (5).and inhibition efficiency compared with immersion time.
Figures also showed no significant interaction between those two variables (AB), and that concentration has some sort of squared effect on preventing inhibitor.
The main effect clearly shows that both factors affect the rate of corrosion, but to different levels.There is clear significant effect for inhibitor concentration over the immersion time.
Regarding inhibition efficiency, main effect figures clearly show that there is an optimum value for both factors that gives best efficiency.This at around 0.75 g/L concentration with 7.5 hrs immersion rate.
Based on the figures, it is clear that corrosive solution temperature showed most significant effect on corrosion (indicated by weight loss) compared with immersion time.
The figure also showed no significant interaction between those two variables (AB), and that both variables have some sort of second order effect on preventing corrosion.
The main effect clearly shows that both factors affect the rate of corrosion, but to different levels.There is clear significant effect for solution temperature over the immersion time.

Conclusion
An experimental and numerical analysis of the corrosion inhibition performance for local materials made in Iraq.
The effect of concentration of inhibitor coating as well as the corrosive solution temperature and time of exposure to such environment was studied.
Further, mathematical modelling was made with the aim of finding the most significant parameter on corrosion rate.
Results showed that corrosive solution temperature as well as the inhibitor concentration were most influential to corrosion phenomena.Exposure time was effective to lesser extent.

Table 1 .
Model specifications