Modeling of Dezincing of Galvanized Steel Scrap

Ezaz Ahmed and Abhijit Ray^*

Department of Biotechnology, Raipur Institute of Technology, Raipur

ABSTRACT

Dezincing of galvanized steel scrap was investigated by preferential dissolution of the galvanized coatings in a variety of acidic environments comprising dilute and moderate concentrations of Hydrochloric, Sulfuric and Nitric Acids. The data thus generated was used to develop an Artificial Neural Network model to estimate the extent of dezincing as a function of thickness and concentration of acids. It was found that the model developed has very good accuracy level.

Key words: Dezincing, galvanized steel, Artificial Neural Network.

INTRODUCTION:

Bastar plateau is having good agro climatic situation for maize production Good quality steel scarp is always a preferable and important source of raw material for steel industry. However recent study shows that there will soon be a shortage of good quality steel scrap and scrap from steel processing industry given the high rate of ferrous consumption. Like in conventional basic oxygen steel making ferrous acts both as a source of Iron and as a cooling medium. Hence, there is a need to generate usable scrap from steel-intensive consumer goods and redundant investment goods. Every year large amount of cold rolled steel is used as a raw material in automobile industry in most of developed and developing nations. The resultant scrap is basically a mixture of single sided and double sided galvanized or zinc-coated steel. Only a small percentage of the scrap is zinc free. In India some other industrial and domestic application of galvanized steel includes construction of pipes, water tanks, storage vessels, transportation case linings fences, roofing sheets, buckets, etc. The large volume of scrap generated from these applications together with scrap from disused or redundant impose serious problems to corrosion and material engineers, particularly in the underdeveloped countries, where, due to poor maintenance and inadequate material reclamation schemes, most of these scraps are usually dumped and allowed to waste away in the open areas, causing serious environmental and safety hazards.

Methods of Recycling Of Galvanized Steel Scrap

Survey of literature shows that most of chemical recycling techniques for galvanized steel involves use of alkaline dezincing[1] i.e. use of hot caustic soda (NaOH) to form sodium zincate(Na2ZnO2) and (NaHZnO2). The zinc is recovered as a dendritic powder while the zinc-free steel is readily utilized by the foundries. In this process dissolution occurs at a measurable rate only if the zinc is in electrical contact with a material of lower hydrogen over potential such as platinum, iron and nickel or if a strong oxidizing agent is added. However, acid–dezincing technique fig(1) is duly recommended as an efficient and economically viable alternative for dezincing of galvanized steel scrap[2]. The very low acid concentrations usually involved makes the scheme rather safe and the dezincing reaction occurs spontaneously with no need for heating or use of external electric power. The important reason behind the use of dilute acids is to completely dissolve the zinc coating with little or no inherent attack on underlying steel substrate which is ultimately recycled from various end uses while the zinc is recovered from solution either in the pure form or as a specific metal salt dependent upon the type of acid used for the stripping.

Artificial Neural Network

An Artificial Neural Network is an information-processing paradigm that is inspired by the way brain process information. It is composed of a large number of highly interconnected processing elements (neurons) working in unison to solve the specific problems [3]. Knowledge is internally presented by the value of the weights and the topology of connections. ANN learns from examples, a certain set of input-output mapping by optimizing weights on the branches that link the nodes of the ANN. Once the structure of input output space is learned, novel input pattern can be classified. These networks can learn and adapt themselves to input from the actual process allowing representation of complex engineering system, which are difficult to model either with traditional model based, or engineering knowledge based expert system. ANN performs successfully where other methods do not succeed. Basically most application of Neural Network falls into the following categories like prediction, classification, data association, nonlinear mapping, pattern recognition etc. ANN has been shown to be extremely suited to model highly complex and nonlinear phenomena. Most common for chemical engineering application is MLP, which is a Multilayer feed forward network. MLP is trained using back propagation algorithm suggested by Rumelhart [4]. Back propagation is generalized for Window-Hoff learning rule to multiple layer networks and nonlinear differentiable transfer functions. A schematic of the MLP network with two hidden layer is as shown in fig 2.

A – Galvanized steel scrap B – Zinc Solution

C - Washer D - Dezinced Steel

E – Spent Acid F – Zinc Recovery

G – Zinc Salt

Fig 1. Schematic illustration of the acid chemical recycling of galvanized steel scrap

Various applications of ANN are, modeling of distillation column [4], modeling of heat exchanger [5], and hybrid first principles model for fixed bed reactor [6]

Experimental Procedure

Galvanized steel sheets of various thickness viz 0.45mm, 0.65mm, & 1.15 mm were taken. These were cut into small rectangular pieces of 4x4cm prior to dezincing in 0.05N, 0.1N, 0.25N of 100ml HCl, H₂SO₄ and HNO₃and the respective dezincing time were carefully recorded. Each specimen were carefully weighed, clamped to a specimen holder and suspended vertically downward in a vessel containing appropriate acid medium. Total immersion of the pieces was ensured. The specimen were consecutively corroded for 2min., 3min., 5min., 10min., 1hr., 2hr., and 5hr., (or more) after which they were thoroughly rinsed with distilled water, dried in a steam of hot air and weighed. The weight loss were then determined by the difference in weights before and after the experiment and curve between weight loss and time were plotted for galvanized steel in specific acid environment.

Thickness I

Fig 2. A schematic of artificial neural network with two hidden layer

Method of data generation

The data thus generated by the experiments were divided into two parts training data set and test data set. Training data set were used to train the neural network model whereas test set was used to test the accuracy of prediction of the trained network. Different types of network architecture were tried to find the best model for dezincing of galvanized steel.

Table 1- Weight loss versus time measurements for galvanized steel scrap under varying acidic medium Specimen thickness – 0.45mm

Time (Min)	Weight Loss (gm)
	HCl			H2SO4			HNO3
	0.05 N	0.1N	0.25N	0.05N	0.1N	0.25N	0.05N	0.1N	0.25 N
5	0.02	0.02	0.1	0.001	0.01	0.01	0.01	0.02	0.04
10	0.02	0.03	0.11	0.005	0.03	0.04	0.03	0.03	0.18
15	0.025	0.04	0.12	0.005	0.03	0.05	0.05	0.07	0.26
20	0.025	0.04	0.14	0.01	0.03	0.14	0.07	0.12	0.32
30	0.03	0.05	0.16	0.03	0.03	0.41	0.08	0.16	0.4
60	0.04	0.06	0.185	0.1	0.32	0.41	0.1	0.17	0.41
120	0.05	0.08	0.19	0.105	0.325	0.42	0.12	0.2	0.43
180	0.05	0.09	0.19	0.12	0.33	0.43	0.145	0.24	0.49
240	0.055	0.1	0.195	0.12	0.34	0.43	0.15	0.25	0.55
300	0.06	0.11	0.2	0.12	0.34	0.45	0.15	0.28	0.61

Specimen thickness – 0.65mm

Time (Min)	Weight Loss (gm)
	HCl			H2SO4			HNO3
	0.05N	0.1N	0.25N	0.05N	0.1N	0.25 N	0.05N	0.1N	0.25 N
5	0.005	0.005	0.01	0.01	0.02	0.05	0.01	0.04	0.05
10	0.005	0.005	0.02	0.02	0.08	0.15	0.02	0.06	0.09
15	0.01	0.01	0.06	0.08	0.19	0.29	0.04	0.1	0.18
20	0.01	0.01	0.11	0.11	0.26	0.29	0.06	0.14	0.24
30	0.02	0.07	0.21	0.14	0.27	0.29	0.1	0.18	0.29
60	0.07	0.15	0.25	0.21	0.28	0.29	0.12	0.21	0.39
120	0.11	0.19	0.25	0.29	0.28	0.31	0.15	0.23	0.51
180	0.13	0.21	0.25	0.3	0.31	0.36	0.16	0.24	0.63
270	0.14	0.22	0.27	0.31	0.33	0.41	0.16	0.24	0.66
300	0.15	0.23	0.275	0.31	0.33	0.49	0.17	0.25	0.67

Network Topology

Number of Neurons				Sum of square of errors
I Layer	II Layer	III Layer	IV Layer
1	0	0	0	4x10^-3
10	1	0	0	1.8x10^-5
4	12	1	0	7.5x10^-6
3	3	3	1	8.1x10^-6

Specimen thickness 0.45mm

fig.3

Specimen thickness 0.45mm

fig.4

specimen thickness 0.65mm

fig 5

RESULTS AND DISCUSSION:

Dezincing of Galvanized Steel Scrap:

Experiments were performed to analyze the dezincing of galvanized steel scrap. The data obtained are shown in the Tables 1 and Table 2. Experiments were carried out at room temperature and atmospheric pressure, with various thicknesses of the specimen using different concentrations of HCl, HNO₃ and H₂SO_4.The data thus generated were used to prepare graph (Fig 3 to 8) between time and weight loss using concentration of acid as a parameter. Fig 3 to 8 shows that HNO₃ gives the maximum recovery of zinc in minimum time for all the concentrations. From Fig 3 and 6, it is clear that recovery of zinc using HCl increases with increase in thickness of the plate, as the amount of zinc recovered for 0.45 mm thickness is 3 hrs, however same amount is recovered in 1hr for the plate of thickness 0.65mm.Fig 3,4 and 5 shows that the slope of the curve is high during the first 45 min then it almost flattens, it means maximum amount of zinc is recovered during this period and the amount of zinc recovered after this period is not appreciable. The recovery of zinc for acids 0.1N HNO3 and H₂SO₄ is nearly same as that for 0.25NHCl.Fig 4, 5, 7 and 8 shows that for acids H₂SO₄ and HNO₃ the recovery of zinc is initially slow and it increases with progress of time

ANN Modeling

Fig 9 and 10 compares the actual vs predicted values of dezincing of galvanized steel scrap for various thickness. Various topologies of network model had been tried and the final value of performance function was recorded in the Table 3. The optimum number of neurons for different topology, except for single layer, had been obtained by trial and error procedure. As can be seen form Fig 9 and 10, the network topology with 3 and 4 layer predicts the value with minimum error. It was observed during the modeling that increase in number of layer increases the prediction time and cause the over fitting of data.

CONCLUSION:

The ANN model has been developed for dezincing of galvanized steel scrap. From the figures it is found that a neural network model with 3 or 4 layers of neurons models the dezincing process very successfully and accurately. It can be said that if past data are available related to dezincing, and then it can be used for the training of neural network model that would predict the recovery of zinc as a function of time and thickness of steel scrap plate with high level of accuracy.

REFERENCES:

[1] Ijomah M.N.C & Ijomah A.I., “Chemical Recycling Of Galvanized Steel Scrap”, Indian J.Chem. Tech. Vol 10 (3), pp 159-165, 2003

[2] Anderson J.A., “An Introduction to neural networks” Prentice Hall of India Pvt. Ltd., Delhi, 1999.

[3]Rumelhart D.E.. “Bach propagation Training Algorithm”, Developed in 1986 at MIT.

[4] Baratti R, Vacca G & Servida A., “Hydrocarbon Process”, June (1995) 35

[5] Mandavgane S.A., et.al. Chem Engg. World, 39(3), pp 75, 2003

[6] Zhou, X., Qi, H., and Yuan, W., “Hybrid First Principles Model for Fixed Bed Reactor”, Chem Eng. Sci., vol-54,pp. 2521-2526,(1999)

Received on 10.12.2009

Accepted on 28.12.2009

© A &V Publication all right reserved

Research J. Science and Tech. 1(3): Nov. Dec. 2009: 113-114