Study of Corrosion Inhibition Potentials of Eichhornia crassipes Leaves Extract on Mild Steel in Acidic Medium using Artificial Neural Network

methodologies some of the Prediction of corrosion behavior of steel in acidic environments is an essential step towards optimizing the design of equipment in any industrial setting. An artificial neural network (ANN) may be used as a reliable modeling method for simulating and predicting the corrosion behaviour. The present study has been conducted to investigate the corrosion inhibition potentials of Eichhornia crassipes (water hyacinth) leaves extract for mild steel in acidic media and to establish an appropriate ANN model for predicting corrosion behavior of mild steel in H SO 2 4 inhibited by Eichhornia crassipes . The experimental procedure employed weight loss method for corrosion rate measurements. Results have shown that Eichhornia crassipes is an effective inhibitor for corrosion inhibition of mild steel in acidic medium. A Levenberg-Marquardt (LM) ANN with single hidden layer having five neurons was employed to simulate the corrosion behaviour. The neural network was trained using the experimental corrosion database. Finally, validity of the proposed model was tested using standard statistical parameters. Results indicate that the trained ANN model is robust for predicting corrosion behaviour of mild steel in acidic media. Abstract


Study of Corrosion Inhibition Potentials of
According to Zhenyu et al., (2007) the mathematical units are conventionally constructed with three layers, i.e. input, hidden and output layer. The hidden layer usually contains one or several layers, and the number of neurons in each layer can be di? erent. The behaviour of an ANN depends on both the weights and the input-output function (transfer function) that is specified for the units. According to Mohanraj et al., (2012) transfer function consist of algebraic equations which are either linear or non linear. The most commonly used transfer functions are log-sigmoid and tangent sigmoid. The log sigmoid transfer function is used when all the values in input and output layer are positive while tangent sigmoid is used when there are negative values in the input or output layer.
The methodology for implementing ANN comprises four steps (Dou et al., 2007) which include Collection and pre-processing of the experimental data, Training the ANN, and optimization of its configuration, evaluating the performance of the ANN and returning to stage 2 if the performance is not satisfactory, and using the trained ANN for simulation or prediction.
Many articles have revealed instances where ANN has been applied in predicting corrosion in engineering systems.
The application of neural network methods to model the pitting corrosion behavior of a stainless steel as a function of solution composition and temperature was presented by Cottis et al., (1999). The ANN prediction exhibited reasonable correlation with data for simple one-and two-component solutions. The use of simulated data to test the neural network method in conditions similar to those being modeled was suggested as one method of obtaining a better assurance of the applicability and performance of the method.
A neural network can provide considerable information on the corrosion behavior of resistant metal alloys. Kamrunnahar and Urquidi-Macdonald (2011) developed an algorithm to predict behavior of corrosion resistant metal alloys using a supervised neural network method as a data mining tool. The data mining results were categorized and prioritized to certain parameters (i.e. pH, temperature, time of exposure, electrolyte composition, metal composition, etc.) and helped understand the synergetic effects of the parameters and variables on corrosion behavior. The results were in good agreement with the experimental values.
Prediction of low-carbon steel, copper and aluminum corrosion rates according to environmental parameters in the area of São Luis -Maranhão in Brazil using ANN model was developed by Elaine et al., (2009) The area along the ''702 -São Luis II -Presidente Dutra" 500 kV transmission line, located in an equatorial region, was employed for the purpose. A specific methodology was developed to determine the local corrosivity rate for these metals. Five atmospheric corrosion stations (ACS) were installed along the 702 transmission line in an extension of 200 km. Along with the meteorological data, local pollutants were collected and analyzed during a period of two years. In the same period, specimens were exposed to the atmosphere and periodically collected for corrosion evaluation. The obtained results indicated that the neural network can be used as a good corrosion estimator. Pintos et al., (2000) also presented an ANN-based solution methodology for modeling atmospheric corrosion processes from observed experimental values. The prediction of the corrosion rate of carbon steel in Iberoamerican Corrosion Map (MICAT) Project used seventy-two test sites in fourteen countries throughout Iberoamerica. The ANN model exhibited superior performance in terms of goodness of sum of square errors and residual distributions when compared against a classical regression model also developed in the context of the study. The methodology indicated that the tool can be used in the modeling of corrosion phenomenon from experimental data. Duo et al., (2007), while studying the amount of ultimate strength reduction of locally corroded plates, used ANN method to derive a formula to predict behaviour. It was found out that the proposed formulae accurately predicted the ultimate strength of locally corroded plates under uniaxial in-plane compression. Transverse location of pit corrosion was also an important factor in determining the amount of strength reduction.
Birbils et al., (2011) studied the corrosion rate and yield strength of magnesium-rare earth (Mg-RE) alloys using ANN. The dataset was provided by the additions of Ce, La and Nd to Mg in binary, ternary and quaternary combinations up to 6 wt. %. It was observed that the yield strength increased with RE additions, and the corrosion rates also systematically increased. The work permitted an understanding of Mg-RE alloy performance, which can be exploited in Mg alloy design for predictable combinations of strength and corrosion resistance.
The present study has been conducted to study and establish the inhibition potentials of Eichhornia crassipes for mild steel in acidic medium and to use neural network to simulate and predict the corrosion behavior of mild steel in this inhibited acidic medium. In the neural network simulation, specimen conditions, inhibitor conditions, and operation conditions, were used as the inputs of the network while corrosion rate was gained as the output. Finally, validity of the proposed model was tested using standard statistical parameters.

Preparation of water hyacinth extract
Water hyacinth leaves were obtained from the River Benue, shade-dried, crushed and ground into powder. Active ingredients were extracted using Soxhlet Extraction method. At each extraction cycle, sixty grams (60g) of the powdered water hyacinth were measured and extracted against 450 ml methanol for 2 hours. The resulting solution was then concentrated until the liquid was completely evaporated. This solid extract was used to study the corrosion inhibition properties and to prepare the required concentrations of the water hyacinth.

Specimen preparation
Mild steel specimens having nominal composition of 0.16% C, 0.30% Mn, 0.21% Si, 0.04% P, 0.03%S, and 99.257 Fe balance were used. Corrosion coupons were cut into 4 x 3 x 0.2cm dimensions for weight loss m e a s u r e m e n t s . T h e c o u p o n s w e r e mechanically abraded with 220, 400, 800 and 1000 grades of emery papers, degreased with acetone and rinsed with distilled water.

Weight loss measurements.
Experiments were performed at room temperature with different concentrations of the water hyacinth extract. The concentration range of the water hyacinth extract employed was varied from 0 -15 g/l and the electrolyte used was 600 ml of 1M H SO for each 2 4 experiment. Immersion time for the weight loss experiment was 6 hours. The results of the weight loss experiments are the mean of three runs, each with a fresh specimen and 600 ml of fresh acid solution. Corrosion rate was calculated from weight loss of the coupons at room temperature at various concentrations and immersion times, using the relation in Fontana (1998) givrn in equation (1) (1) Where, Äw is weight loss of mild steel Where, coupon in kg;D is density of mild steel 3 in kg/cm ; A is the Surface area of mild steel in 2 cm ; and T, the immersion time in hours The inhibition efficiency (IE %) and surface coverage (2 ) at each concentration for each coupon was calculated from the result obtained from the weight loss experiment as presented in equations 2 and 3 respectively. (2)

2
(3) Where, W and W are the weight loss of the 1 2 mild steel in the presence and absence of inhibitor respectively.

Neural Network Modeling
The input data captured in the artificial neural network was broadly grouped into three namely, specimen dimensions, inhibitor condition, and operating conditions. This resulted in a total number of nine inputs parameters as shown in Table 1 inhibited by Eichhornia crassipes leaves extract at various concentrations was taken as the output. Corrosion rate was computed using equation (1). Since the best neural network's architecture and learning algorithm are not known in advance, a trial and error approach was used to find the best network characteristics for matching the particular input/output relationship. These were investigated, using MATLAB 7.9.0 (r20096), for one layered network 9 The linear function f(x) = x) was used between the hidden and output layer. Pre-processing of the input parameters was carried out before training of the neural network. Inhibitors concentration and specimen dimensions were pre-determined and presented to the network, while exposure time and temperature were scaled using equation (5) (Dragen, 2010) (5) Where; -current input value, maximum input value and -minimum input value.

Results and Discussion
After a series of repeated trials using different neural network architectures namely; one layered network 9 [1] 1, 9 [5]  crassipe leaves extract. It can also be observed from the figure that the higher the concentration of inhibitor, the lower the corrosion rate. As is observed, the ANN model may well predict the effect of inhibitor concentration on the corrosion rate of mild steel in H SO . The ANN model was also 2 4 applied to simulate the effect of inhibitor efficiency on the corrosion rate. The results of variation of corrosion rate with inhibitor efficiency are presented in Fig. 4 while those of effect of weight loss on corrosion rate are presented in Fig. 5. These figures argue that corrosion rate decreases with increase in both inhibitor efficiency and weight loss. As is also observed, the simulated corrosion rates may well tract their corresponding measured values. It also indicates that the neural model has performed well as reported by other researchers, Dragen (2010), Cottis et al., (1999), Kamrunnahar and Urquidi-Macdonald (2010).

Assessment of the proposed model
To have a more detailed quantitative view regarding the predictability of the model the root mean square error (RMSE) and Nashscurtclisse efficiency (NSE) were used to evaluate the model (Mohanty, et al., 2013).
These are presented in equations 6&7 respectively: Where; d is the desired output for p performance and O is the network output. p (7) Where; E is the experimental value, P , the i i predicted while is the mean of experimental values. Table 2 shows the values of the predicted and computed (targeted) corrosion rates together with their respective relative errors. It can be observed that values of corrosion rate predicted using the ANN model deviated by amounts less than half a percent from their measured values and NSE values of close to unity indicating that they are, indeed in good agreement with the targeted values. Accuracy of the prediction was also determined by considering scatter of predicted and measured values around regression lines as shown in crassipe leaves extract.

Conclusion
In this study, potentials of Eichhornia crassipe (water hyacinth) as corrosion inhibitor for mild steel in acidic media was established while procedure for modeling the influence of relevant factors such as specimen dimension, inhibitor conditions, and operating conditions using artificial neural network has been suggested. It is shown that Eichhornia crassipe gave 98% inhibition efficiency and that the computer-based model provided may help in predicting corrosion behavior. The best results in prediction of the corrosion