Neural Network Modeling Research Papers

The importance of recurrent top-down synaptic connections for the anticipation of dynamic emotions

2025, Neural Networks

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of... more

descriptionView Paper arrow_downwardDownload

Neural network prediction model for fine particulate matter (PM2.5) on the US–Mexico border in El Paso (Texas) and Ciudad Juárez (Chihuahua)

by Joaquín Ordieres

2024, Environmental Modelling & Software

The daily average PM 2.5 concentration forecast is a leading component nowadays in air quality research, which is necessary to perform in order to assess the impact of air on the health and welfare of every living being. The present work... more

Fig. 1. U.S.—Mexican border region, as defined under the La Paz Agreement.

Fig. 2. Urban air quality monitoring network on the US—Mexico border in El Paso (Texas) and Ciudad Juarez, Chihuahua (Mexico).

Typically, a neural network is composed of a set of neurons laid out in layers. Commonly, those layers are classified as input layer, hidden layers and output layer. with complex problems, particularly in the fields of association, classification and prediction. Many re- searchers have shown that neural networks can solve almost any problem more efficiently than the traditional modeling and statistical methods (Hornik, et al., 1989; Masters, 1993). In this paper, we compare three neuronal network architectures in order to obtain the best possible efficiency in the outcomes of Multilayer Perceptron (MLP), Square Multilayer Perceptron (SMLP), and Radial Basis Function (RBF) analyses.

Fig. 5. Artificial neural network structure. Some neural networks do not have hidden layers and are used as more linear statistical techniques. These networks (with input and output layers only) are useful in many linear or semi-linear applications, but in general, it is difficult to get accurate results in non- linear problems (McCullagh and Nelder, 1989). We understand that particulate matter is clearly a non-linear

Fig. 7. MAE index evolution as a function of the number of hidden nodes for the different variances assessed. Another is the Mean Absolute Error (MAE), de- fined as:

Fig. 6. Radial Basis Function structure.

Fig. 8. Daily times series of the measured and predicted concentrations of PM>.s.

Fig. 9. Scatter plots of the measured and predicted concentrations of PM) s.

Fig. 10. Histograms of the residuals for the different models assessed. A short time PM>.5 prediction model has been built taking into account a large number of samples from The MLP network provides acceptable predictions, in spite of the difficult environmental conditions of the location (i.e. even though PM, data were considered, samples still show hard peaks of inmissions seasonally, mainly due to the common presence of dus storms in this area). The SMLP network shows a very similar behavior, although a few more neurons in the hid necessary to obtain the same error found case. This fact has a direct influence on t den layer are in the former he sample size necessary for the right training, as previously mentioned. Finally, the RBF network shows the best the shortest training times and best st behavior, with ability. These results suggest that the widely used MLP should be replaced by the more convenient RBF net work.

Primary NAAQS: the levels of air quality that the EPA judges necessary, with an adequate margin of safety, to protect the public health. Secondary NAAQS: the level of air quality that the EPA judges necessary to protect the public welfare from any known or anticipated adverse effects. National Ambient Air Quality Standard (NAAQS) for the US Table |

Input variables used in the forecast models Table 2

descriptionView Paper arrow_downwardDownload

Colour mismatch in compounding of plastics: processing issues and rheological effects

by Jamal Alsadi

2023

Demand for plastic products is continuously growing with population increase and higher living standards. Imparting colours to plastics is a combination of art and science and plays a key role in production of attractive products for a... more

Demand for plastic products is continuously growing with population increase and higher living standards. Imparting colours to plastics is a combination of art and science and plays a key role in production of attractive products for a variety of consumer demands. It requires a good understanding of the processing parameters and formulation of compositions including colour pigments and their dispersion. This thesis presents experimental observations and statistical analysis to investigate scientific reasons for colour mismatches in compounded plastics. Material processing issues were investigated. Six methods were used to study and improve the understanding of colour matching, colour stability, and consistency of compounded plastic materials in order to minimize wastage. The first was Data Mining, in which historical data was analyzed to select formulations, particularly for pigments and polycarbonate blends of compounded plastics that were known to suffer from colour mismatch. The second method involved the use of historical data and an artificial neural network (ANN) to predict the resultants colours based upon pigment formulations. In the third method, a parametric study was utilized to investigate issues in the dispersion of pigments due to the effects of processing conditions. Temperature and feed rate were found to have major effects on colour deviations. The fourth method involved studying the compounding process for blends of two polycarbonate resins (PC) of different melt flow index (MFI). It was also found that the PC composition of 30-70 wt.% was superior in terms of colour matching. The fifth method studied effects of rheology. The viscosity data collected allows predictions of viscosity changes with varying temperatures, which had a direct impact on polymer colour changes during processing of blends. The sixth method looked at the effects of processing parameters on dispersion, pigment size distribution (PSD), and the morphology of pigments in polycarbonate compounds. In general, the total colour difference decreased when processing parameters were increased. De-agglomeration occurred in zones of high shear and was found to significantly increase the number of particles and lower colour differences. I would like to dedicate this work for my parents, who passed away during my study. My parents were the most beautiful things I ever had. I attribute my success in life to the moral, intellectual, and physical education I received from them. I would also like to show gratitude to my family for their support and encouragement throughout my research and higher education. This support has been vital, and I shall never forget it. Much love and thanks are extended to my friends and others who have lent their support and encouragement through this endeavour. I would like to express my thanks to my supervisor Professor Ghaus Rizvi for his patient guidance, encouragement and useful critiques of this research work. I also want to extend my deepest appreciation to him for giving me every opportunity to enhance my graduate experience from introducing me to an industrial organization facility, to allowing me to attend conferences, seminars and workshops across North America and North Africa to present my work. I am thankful to my supervisory committee members Dr. Hossam Kishawy and Dr. Yuping He for their guidance and advice. I am particularly grateful for the assistance given, not only for their technical knowledge, but also for showing me how to differentiate the important things and keep the most basic questions in mind. I would also like to express great appreciation to the people who helped and supported me during the preparation of this thesis. I also acknowledge the support and guidance by Professor Amir Hossein Behravesh. I would also like to take the time in expressing great thanks to my colleagues in our Mechanical Engineering Lab, Dr. Usman Saeed and PhD students Shahid and Rasel for their kind collaboration and efforts. Major funding for this research was graciously by SABIC IP and Natural Sciences and Engineering Research Council (NSERC) of Canada .I would like to extend my gratitude for support provided by the staff at SABIC IP Cobourg, Canada for their assistance in the experiments, lab tests: Daniel Ross. Joseph Wembley, James Price, and Richard Clarke. This journey iv would not have been possible without their help and support. The material samples used throughout the experimental process were generously provided SABIC IP.

Table 4.1. Composition of compounding material (g3)

Table 4.4. Processing conditions with speed variation Table 4.3. Processing conditions with temperature variation

Table 6.2. Pigments requiring the highest percentage of adjustments

Figure 2.1. Incident and reflection of light on an opaque material [5] in Figure 2.1.

Figure 2.2. Scattering, absorption, transparency, and translucency of incident light [7] 2.2 Factors of Colours (The Triad) will appear white; and if the visible light is absorbed by the pigment, then the object appears

Figure 2.4. Two sets of observer functions with difference in field of vision [14]

Figure 2.5. Schematic of a digital simulation of a visual colourimeter — spectrocolourimeter [13] 7 Colour Space and Tristimulus Values

Figure 2.6. Three-dimensional colour space of hue, lightness, and saturation [10] pure hue to black; and (c) tones, the range from a pure hue to any gray. We can measure a colour quantitatively by creating scales for the hue, lightness, anc

Figure 2.7. Colour in the CIE 1976 (L*a*b*) colour space [20]

Table 4.6. Parameters and experimental design level used The desired colour output in terms of CIE L*, a*, b* values was L*= 69.59, a*= 1.42

Table 4.5. Processing condition with feed rate variation

Table 5.1. Regression model (3-level factor design)

Figure 6.30. Storage modulus of Blend 100%-R1, 30%, 0% at the temperature of 255°C: (a) without additives and pigments, and (b) with additives and pigments fell into a narrow band at the higher frequencies.

Figure 6.31. Complex viscosity at a temperature of 255°C: (a) without additives and pigments, and (b) with additives and pigments

CMC (Colour Measurement Committee of the Society of Dyes and Colourists of Great Britain): Organization that developed and published in 1988 a more logical, ellipse-based

Figure 3.1. (a) Architectural graph of a multilayer perception (MLP) with one hidden layer. (b) Basic concepts of the CNN-supervised learning algorithm [51]

Figure 4.1. Coperion Co-Rotating twin screw extruder Figure 4.2. Injection Molding Machine KM-100

Figure 4.3. (a) Haake Minilab system (Thermo Fisher Scientific), and (b) schematic of the design [115]

Figure 4.4. Ares-G2 rotational rheometer: (a) photo, and (b) schematic diagram for the parallel plate (TA- Orchestrator software) [116]

Figure 4.5. Tinius Olsen MP600M Et. Plastometer Melt [117] Figure 4.6. X-rite-7000A-Spectrophotometer [118] according to ASTM D1238.

Figure 4.8. (SEM), Joel 5500 LV [120] Figure 4.7. Keyence-Digital-Microscope VHX-1000 [119]

Figure 4.10. X-ray Micro-CT-scanner-Sky Scan 1172 [122] Figure 4.9. Microtrac $3500 Particle Size Analyzer (PSA) [121]

Figure 4.11. Schematic diagrams of process methods of plastics Table 4.2. Composition of R1-R2 (polycarbonate resins) blends

Figure 4.12. Experimental design of 3 level 3 factor colour. The additional centre point was added to estimate the experimental error.

In the equations above, E, is an actual result, Ep is a predicted result, E,, is a mear value, and N is the number of patterns. The regression coefficient multiple-R* compares t

Figure 6.1. Overall flow chart for colour mismatch analysis of the compounded plastic in the flow chart in Figure 6.1.

Figure 6.2. The six grades with the highest rate of adjustment The table shows the fourteen grades containing Pl that had the highest rate of adjustment. Six of these grades were selected for additional statistical and experimental investigations to study the effects of the formulations on the colour shift. The sample grades are illustrated in Fisure 6.2.

Figure 6.4. Comparison of (a) RMS, (b) mean percentage error, and (c) regression of different neural network algorithms

Figure 6.5. Comparison of NN output and experimental output for colour tristimulus values L*, a*, and b* of training data

Figure 6.6. Comparison of experimental and neural network DE values in ascending order maximum value of 4.2, as shown in Figure 6.6.

Figure 6.7. Comparison of (a) regression and (b) MSE for different algorithms cycle GDM depends on two training parameters: the learning rate and momentum coefficient. terms of regression and MSE were out of the desirable range, as shown in Figure 6.7. The

Figure 6.8. Network training cycle for LM algorithm Table 6.5. R-Squared and Mean Square Error for 16 neurons in hidden layer

Figure 6.9. Comparison of experimental and neural network results for L*, a*: (a) training data and (b) test data

Figure 6.10. Contours of DE with respect to contribution of (a) DL* and Da* and (b) DL* and Db* value of DE was obtained with DL* = 0.973, Da* = 0.0013, and Db* = 0.024.

Figure 6.11. Effect of feed rate on dE* Figure 6.12. Effect of feed rate on b* level of speed.

Figure 6.13. Effect of screw speed on dE*

Table 6.8. Possible solutions for each grade

Figure 6.15. Interaction and overlay plot

Figure 6.16. Effect of temperature on opaque colour (dE*)

Figure 6.18. Effect of speed on opaque colour (dE*) Figure 6.17. Effect of feed rate on opaque colour (dE*) at higher feed rates and at the centre level of speed.

Figure 6.19. Effect of temperature on colour output (dE*) 6.4 Response Surface Methods

interactions are presented in Figure 6.20.

experimental error. The model appears valid for the prediction of colour values. represents the coded unit of parameters and Y-axis represents the desirability. The plot

Figure 6.24. Contour plot for L* along temperature and feed rate

Figure 6.25. 3D surface graph flagged with optimal desirability fitness of optimal solutions as the response mean values are within 95% CIs.

Figure 6.26. 2D contour graph flagged with optimal desirability Graphical Optimization (Overlay plot). Solutions suggested by the numerical

Figure 6.27. Graphical optimization and sweet spot flagged for a desired solution desired set of process conditions that truly satisfies all the constraints.

Figure 6.28. Variation in (a) melt flow index vs. polycarbonate resin R2, wt. %, and (b) pressure difference with respect to polycarbonate resin R2, wt. % with and without additives

Figure 6.29. Strain sweep test for blend R1-R2 (50%-50%) at 10Hz

Figure 6.32. Complex viscosity behaviour vs. frequency for all blends of R1-30%: without and with addition of additives at investigated temperatures along with the corresponding Carreua models

Figure 6.33. Shear viscosity for Blend R1-30% at temperatures of 230, 255, and 280°C: (a) without additives, and (b) with additives magnitude of shear viscosity is quite comparable to that of complex viscosity. temperatures are depicted in Figure 6.33, without and with the addition of additives. The

y Rheological Behaviour of the Blend Compounded at SABIC (SB) in all of the blends. ‘iscosity coincided with the linear regime of shear viscosity. Similar behaviour was observed Initially, the composites were rheologically characterized using an Ares-G2

Figure 6.35. Strain sweep test for grade PC1 content of 30 wt. % at 10 Hz

Figure 6.36. Storage (a) and loss (b) modulus at 230, 255, and 280°C

the examined temperatures of 230, 255, and 280°C), as shown in Figure 6.37. measured at temperatures of 230, 255 and 280°C display a shear thinning effect as

280°C. The increase in temperature and frequency shows a proportional relation: decreasing

Figure 6.40. Tan 6 for R1 30% blends at 255°C Figure 6.39. Complex viscosity for R1 30% blends at 255°C

Figure 6.41. Effect of temperature on colour output with respect to dE* (DE) value (dE*) was obtained at the higher temperatures namely 255°C and 280°C with dE*=0.3.

Figure 6.42. Effect of rheological parameters at temperature 230, 255, and 280°C on colour reases the number of particles (frequency) dispersed, ultimately reducing the colour shift

Table 6.14. Carreau model parameters at reference temperature of 255°C 6 Effect of Processing Parameters on Dispersion

Figure 6.44. CIELAB colour differences for the samples produced at temperatures of 230, 255, and 280°C 255°C; both had better dispersion than those processed at 230°C.

Figure 6.45. Pigment size distribution in the blends produced at the examined temperatures

Figure 6.46. CIELAB Colour differences at various feed rates: 20, 25, and 30 kg/hr parameter samples Effect of feed rate on Pigment size distribution

Figure 6.47. Pigment size distribution at the examined feed rates

Figure 6.48. CIELAB colour differences for the samples processed at 700, 750, and 800 rpm Effect of Speed on Pigment Size Distribution

Figure 6.49. Pigment size distribution at the examined screw speeds

Figure 6.50. Average particle size for four pigments using wet analysis with varying ultrasound time at 30W 6.7.2 SEM Characterization

Figure 6.51. SEM micrograph of black (a), white (b), yellow (c), and red (d) pigments

of particles had a size of 0.8 microns at 280°C in comparison to 38% at 230°C.

Figure 6.53. (Clockwise from top left) Micrograph at 5000X of the blend R1-30% at 230°C, 20 kg/hr, 700 rpm, and 800 rpm (colour chips) temperatures, feed rates and screw speed.

Figure 6.54. DOM-Particle distribution for chip sample processed at 750rpm (left), and microtome sample processed at 280°C (right) micron chip samples, cut by a microtome.

Figure 6.55. DOM-agglomerations for (a) 700rpm, (b) 800 rpm, (c) 230°C, and (d) 30 kg/hr; samples magnified to 2000X, measured by 50-micron chip, and cut by microtome

Figure 6.56. SEM micrograph of polycarbonate grade compound at 230, 255, and 280°C agglomerations than that of the sample, produced at higher parameters.

Figure 6.57. SEM micrograph of polycarbonate grade compound at 20, 25, and 30 kg/hr

Figure 6.58. SEM micrograph of polycarbonate grade compound at 700, 750, and 800 rpm

Figure 6.59. Micro CT scanner micrographs of the blend processed at 750 rpm, 255°C, and 25 kg/hr: (a) agglomerations, (b) particle shape, and (c, d) pigment distribution

colour variations in different compounding processes.

Figure 6.61. Viscosity (a) and particle size distribution (b) for the blends processed in two different extruders (ML and SB) at a temperature of 255°C

Figure 6.62. Experimental viscosity models WOA (ML) and WA (SB)

constant regime of complex viscosity coincided with that of shear viscosity. mtaining 30 wt. % of resin R1 processed at temperatures of 230, 255, and 280°C. The

Figure 6.64. Particle diameter distribution for (a) temperature, (b) feed rate, (c) speed, and (d) for 230°C temperature layers at thickness A (90 um), B (70 um), and C (50 um)

Figure A—1. Comparison of NN and experimental output for colour tristimulus values L*, a*, b*: validation data

Figure A—2. Comparison of NN and experimental output for colour tristimulus values L*, a*, b*: test data

Figure B—3. Mixing two colours additively leads to a lighter colour; if red, green, and blue light are mixed equally together at full power, you get white light (from WWW.MICA.EDU) yellow. If red, green, and blue light are mixed equally together at full power, you get whit

Table 6.1. Demonstrating the percentage of adjustment of the lots run in 2009

Figure B—7. Hue changes by moving position on the colour circle (from Hyper Physics) distinguish red from green, blue from yellow, etc. Munsell defined five principal hues (red,

Table 6.3. Percentage of adjustment of the lines caused by red pigments Red Pigment and the Polymer Grades. SABIC IP produces many grades of PC.

The table shows the fourteen grades containing Pl that had the highest rate of Table 6.4. Percentage of adjustment of grades formulated with red pigments

Table 6.6. Colour formulation of the three grades mismatch. It indicates that for all the three grades, the minimum deviation from the target

Table 6.7. ANOVA for each response (A=Temp, B=Speed, C=FeedRate, D=Grade)

Table 6.9. Compounding material for the opaque grade

Table 6.10. Analysis of variance for L*, a*, and b* The R-Squared value indicates that the components usually would explain about 78% of the variability in the responses whereas about 22% of the variability in the responses

Table 6.11. Comparison of experimental and predicted values

Table 6.13. Fitting parameters of Carreau model, PC blend (ML) with pigment and additives (WA), and without pigment and additives (WOA) Table 6.12. Carreau model parameters for SB at 230, 255, and 280°C

Master Curves at Reference Temperature of 255°C

Table 6.15. Effect of processing parameters on colour in terms of tristimulus values

Table A-1. The complete dataset used for training (70%), validation (15%), and testing (15%); R=resin, A=additives, P=pigments

Table A—2. Weights between the input layer and hidden layer for L*, a*, and b*

Figure B—6. Cultural variations in the meaning of colour (http://www.swiss-miss.com/2005/08/page/2) Cultural Variations in the Meaning of Colours: See figure below D65: The CIE standard illuminant that represents a colour temperature of 6504K. This is the

descriptionView Paper arrow_downwardDownload

Simulation of bleaching of soda pulp from Hesperaloe funifera by polynomial and neural fuzzy models

by Alejandro Rodríguez

2023, Scientific Research and Essays

Influence of variables [soda, (0.5 to 3.0%), hydrogen peroxide (1.0 to 10.0%) and time (1 to 5 h)] in the bleaching of soda pulp of Hesperaloe funifera, on the properties of bleached pulps, was studied. Polynomial and neural fuzzy models... more

descriptionView Paper arrow_downwardDownload

ANNMD-Artificial Neural Network Model Developer

by Jure Smrekar

2022

This booklet presents a tool for Artificial Neural Network (ANN) modelling called the Artificial Neural Network Model Developer (ANNMD). The ANNMD is a generic tool for system modelling with ANNs and its development is based on... more

descriptionView Paper arrow_downwardDownload

Non-Destructive Analysis of Kraft Pulp by Nir Spectroscopy

by Anna Illa

2022

A fast and convenient analytical technique is needed for quality assessment and precision management in pulp production. The main objective of this study was to evaluate the ability of NIR spectroscopy to predict diverse pulp properties.... more

descriptionView Paper arrow_downwardDownload

Modeling hydrogen peroxide bleaching of soda pulp from oil-palm empty fruit bunches

by M. Roncero

2021

The influence of the variables soda (0.5-3.0%), hydrogen peroxide (1.0-6.0%) and time (1-5 h) in the bleaching of soda pulp of empty fruit bunches (EFB) from oil-palm, on the properties of bleached pulps, was studied. Polynomial and... more

descriptionView Paper arrow_downwardDownload

Refiner Production of Sodium Perborate and Hydrogen Peroxide Bleached Mechanical Pulp: Characterization and Physicomechanical Properties

by Gaye Cakal

2021, Cellulose Chemistry and Technology

Paper can be manufactured without bleaching, but high quality white paper grades, which require bleaching, are in demand as well. In paper mills, oxidizing or reducing chemicals are utilized to bleach the pulp. In this study, the... more

descriptionView Paper arrow_downwardDownload

Comparison of reinforced oxygen delignification methods for old corrugated board (OCC) fibres

by emrah peşman

2021, Drewno

The present work focuses on the effects of sodium percarbonate and hydrogen peroxide as reinforcing agents for the oxygen delignification of old corrugated container paper (OCC). The chemical and crystalline structure, physical and... more

descriptionView Paper arrow_downwardDownload

The Effects of Sodium Percarbonate and Perborate Usage on Pulping and Flotation Deinking Instead of Hydrogen Peroxide

by emrah peşman

2021, BioResources

The main objective of this study was to evaluate the potential of sodium percarbonate and sodium perborate utilization during repulping of old news and magazine paper mixture. A series of experiments were performed to determine the... more

descriptionView Paper arrow_downwardDownload

Using artificial neural networks in estimating wood resistance

by Maderas. Ciencia y Tecnologia

2018

The purpose of this research was to evaluate the potential of Artificial Neural Networks in estimating the properties of wood resistance. In order to do so, a hybrid of eucalyptus (Eucalyptus urograndis) planted in the Northern Region of... more

descriptionView Paper arrow_downwardDownload

Simulation of bleaching of soda pulp from Hesperaloe funifera by polynomial and neural fuzzy models

by alejandro rodriguez

2017, Scientific Research and Essays

Influence of variables [soda, (0.5 to 3.0%), hydrogen peroxide (1.0 to 10.0%) and time (1 to 5 h)] in the bleaching of soda pulp of Hesperaloe funifera, on the properties of bleached pulps, was studied. Polynomial and neural fuzzy models... more

descriptionView Paper arrow_downwardDownload

Modeling hydrogen peroxide bleaching of soda pulp from oil-palm empty fruit bunches

by alejandro rodriguez

2016

The influence of the variables soda (0.5-3.0%), hydrogen peroxide (1.0-6.0%) and time (1-5 h) in the bleaching of soda pulp of empty fruit bunches (EFB) from oil-palm, on the properties of bleached pulps, was studied. Polynomial and... more

descriptionView Paper arrow_downwardDownload

DESORPTION OF METAL IONS FROM KRAFT PULPS. PART 1. CHELATION OF HARDWOOD AND SOFTWOOD KRAFT PULP WITH EDTA

by Kim Granholm and

2015, Bioresources

Chelation of unbleached and oxygen bleached hardwood and softwood kraft pulps with EDTA was studied. The main focus was on the desorption of magnesium, manganese, and iron due to their impact in TCF-bleaching. Desorption of other metal... more

EDTA, ethylenediaminetetraacetic acid, is a multidentate chelating agent with 6 coordinating sites (Fig. 1), and it forms 1:1 complexes with most metal ions (Ringbom 1963). EDTA contains both -COOH and —NH; as the functional groups and forms a ring structured chelate around the metal ion.

In Fig. 2 the ay-coefficients of iron (IID) and magnesium(II) are plotted as function of pH for EDTA, DTPA, and DCTA, ie. the chelating agents that are commonly used for desorption of metal ions in pulping processes. The ay-coefficients presented in Figs. 2 and 3 are based on the equilibrium constants given by Ringbom (1963) at 20 °C or 25 °C. The side reactions of the metal ions with hydroxide ions are included in the given am-coefficients. The ligand concentrations are assumed to be 0.01 M. As the trivalent iron has harmful effects on the pulping process, the Ofe-coefficient should be as high as possible in order to enable an effective desorption of iron(II). All the three chelating agents show quite similar curves (Fig. 2a) but DCTA has, compared with EDTA and DTPA, a slightly better ability to complex with iron(II) at pH below 9. Chelation of iron(III) is very effective with all three chelating agents in the alkaline and neutral pH regions. However, still at pH 1 the of am-coefficients for DCTA and EDTA are as high as ca 10’. At pH values higher than 11 iron(III) starts to precipitate as Fe(OH); even in presence of the chelating agent.

Calculated am-values for some metal ions with 0.0027 M EDTA are shown in Fig. 3. Alkali metal ions, like Na, K, and Li that form very weak chelates with EDTA ‘Smith and Martell 1989) are not included in the figure, as their a values are close to 1. it can clearly be seen that the metal ions are chelated by EDTA with varying strengths. Fes Al™, and Cu’ form the most stable chelates, but also Zn", Fe*’, and Mn?" form rather strong chelates with EDTA. It is important to note the large difference in the chelating strength of EDTA for iron(II) and iron(II) within the entire pH range. This means that the oxidation state of iron in the chelating step of the pulping process is of sreat importance. A clear pH effect on the chelating power of the EDTA can also be seen in Fig. 3. The lower the pH is the lower is the chelating effect of EDTA. This is due to srotonation of EDTA.

Fig. 4. The remaining manganese concentrations in oxygen bleached hardwood and softwood kraft pulps after chelations with EDTA of different concentrations

Fig. 6. The remaining manganese and magnesium concentrations vs. time during chelation of oxygen bleached softwood kraft pulp (0.2% EDTA, pH 5) . In order to study the time required for the chelation reaction, pulp samples were taken from the pulp slurry every 5" minute during the chelation experiment, followed by an immediate thorough washing. The chelation was done with 0.2% EDTA at pH 5 on oxygen-bleached softwood pulp. The manganese and magnesium concentrations during this chelation experiment are shown in Fig. 6.

As can be seen in Fig. 7a the chelation of manganese from oxygen bleached hardwood pulp showed very clear differences in the desorption curves with and without EDTA. Without EDTA the removal is due to ion exchange with hydrogen ions. With EDTA a clear chelation effect can be observed in the whole pH range studied. Removal of iron is more problematic. Even with 0.4% EDTA only slightly more than 50% of iron was chelated (Fig. 7b). In removal of magnesium from oxygen bleached hardwood pulp only a slight chelating effect with EDTA was obtained in the pH region 5 to 10 (Fig. 7c). At pH <5 the main reaction is the ion exchange between magnesium and hydrogen ions. In chelation of manganese from oxygen bleached softwood pulps 0.4% EDTA was very effective (Fig. 8a). At pH < 7 the removal of manganese with EDTA was almost quantitative and more effective than from hardwood pulp. On the other hand, it was more

Table 2. Concentrations of Different lons in Softwood Kraft Pulps Before and After Chelation with 0.2% EDTA at pH 5, 7, and 10 Determined by the Semiquantitative ICP-MS Technique As can be seen in Table 1, the concentrations of several metal ions, especially sodium, potassium, and manganese, were much higher in the oxygen bleached hardwood kraft pulp than in the unbleached hardwood kraft pulp. One reason for this may be that he oxygen bleaching step increases the concentration of metal binding anionic groups in he pulp. Su et al. (2004) showed by potentiometric acid-base titrations that the concen- ration of carboxylic groups increased during oxygen delignification of kraft pulps. Table 2, however, indicates that in the softwood kraft pulp the concentration of the metal ions both in unbleached and oxygen bleached pulps were rather similar. Although there were big differences in the original metal ion concentrations between unbleached and oxygen bleached hardwood pulps, the concentrations of magnesium, manganese and iron in the unbleached and oxygen bleached pulps were rather similar after the chelations, indicating a rather effective chelation power of EDTA under the used experimental conditions. Hence, the chelation can successfully be done either before or after the oxygen delignification.

Table 3. Mg/Mn Ratios in Hardwood Kraft Pulps Chelated with 0.4% EDTA at Different pH Values — Chelation had only a very weak effect on the removal of magnesium from oxygen bleached hardwood pulp (Fig. 7c). Magnesium ions are also only weakly bound to oxygen bleached softwood pulps and are removed mainly by ion exchange with hydrogen ions. The small differences between the curves in Fig. 8c are obviously due to experimental errors and have no practical relevance. The a-coefficients given in Fig. 2b, calculated on theoretical basis, show that magnesium ions form a weak complex with EDTA at pH below ca 8, and in this pH range no chelation effect can hardly be obtained as was experimentally approved in Figs 7c and 8c. Lapierre et al. (1997) have shown that an acid wash (pH=1.5) removed efficiently manganese and magnesium. As we already have shown in Fig. 3, the a-Mn(EDTA) at pH | and a-Mg(EDTA) at pH 4 are ca. 1 and there is no chelation of these ions and the removal of manganese and magnesium at pH- values below 1 and 4. respectively. are entirely due to ion exchange with hydrogen ions.

Table 4. Mg/Mn Ratios in Softwood Kraft Pulps Chelated with 0.4% EDTA at Different oH Values The Mg/Mn ratio for the hardwood pulp, when no EDTA was used, was about 2, and for the softwood pulp 6-9 within the whole pH range, i.e. the pH has only a very small effect on the Mg/Mn concentration ratio in these cases. By using EDTA, manganese was removed, but EDTA had only a weak or even no effect on the removal of magnesium. EDTA improved the removal of manganese at all the tested pH values, but he best improvement was obtained at the neutral pH values (pH 5-8). The highest Mg/Mn ratio, after chelation with EDTA was 38, obtained at pH ca. 4 for the oxygen bleached hardwood kraft pulp (Table 3), and for the oxygen bleached softwood kraft pulp he highest Mg/Mn ratio was as high as 7300 at pH ca. 5 (Table 4). An obvious explanation for the much lower Mg/Mn concentration ratio obtained for the hardwood pulp is that manganese is much more strongly bound to the functional groups in this pulp han to the softwood pulp.

descriptionView Paper arrow_downwardDownload

Application of artificial neural networks for predicting tensile index and brightness in bleaching pulp

by Onur Tolga Okan and

2015

The purpose of this study was to develop artificial neural network (ANN) models for predicting the effects of wood species, sodium perborate tetrahydrate (SPBTH) ratio, time, and beating degree on tensile index and brightness in bleaching... more

descriptionView Paper arrow_downwardDownload

The Effects of Sodium Percarbonate and Perborate Usage on Pulping and Flotation Deinking Instead of Hydrogen Peroxide

by Evren Ersoy Kalyoncu

2015, BioResources

The main objective of this study was to evaluate the potential of sodium percarbonate and sodium perborate utilization during repulping of old news and magazine paper mixture. A series of experiments were performed to determine the... more

descriptionView Paper arrow_downwardDownload