Modeling NOM removal by softening in a surface

1008

©iWA Pubiishing 2013 Water Science & Technology | 67.5 | 2013

Modeling NOM removal by softening in a surface water treatment piant Joâo M. L. Dias, Rui Oliveira and Michaei Semmens

ABSTRACT Phenomenological models and hybrid phenomenological-chemometric models were developed to predict natural organic matter (NOM) removal based on the real water treatment data from the city of Minneapolis over a 3 year period. The analysis of the modeling results showed that the phenomenological model was able to capture the major variations of NOM removal but it tended to over predict the NOM removal in independent data sets. These results could be significantly improved by the hybrid model, which was less biased and much more accurate than the phenomenological model. The phenomenological model parameters showed low statistical confidence because the available data, collected in real water treatment conditions, was not sufficiently informative to identify the complex model structure. By comparison, the hybrid modeling method enabled a more reliable discrimination of the most important factors affecting NOM removal.

Joäo M. L. Dias Rui Oliveira REQUIMTE, Systems Biology & Engineering Group, DQ/FCT, Universidade Nova de Lisboa, Campus Caparica, Portugal Michael Semmens (corresponding author) Dept of Civii Engineering, 500 Pilisbury Drive SE, universiîy of Minnesota, Minneapolis, MN 55455, USA E-mail: semmeOOiaumn.edu

The finai hybrid model was implemented in an Excel spreadsheet and can be easily used for NOM removal prediction and the control of chemical dosing. Key words | chemical dosage, hybrid modeling, natural organic matter (NOM) removal, partial leastsquares (PLS), phenomenological modeling, water treatment plant

INTRODUCTION Ntimerous studies have been conducted over the years to determine the effectiveness of lime softening and various coagulants on natural organic matter (NOM) removal from natural waters during water treatment. Most of these studies, especially those which have attempted to determine the factors which affect water treatment performance, and the mechanisms of removal, have been conducted with synthetic surface waters that can be reproduced consistently. For example, waters constituted with well characterized commercial humic and fulvic acids, or NOM extracts from surface waters, have been studied (Semmens and Field 1980; Liao & Randtke 1986; Hundt & Omelia 1988; Bose & Reckhow 1998; Lin et al. 2005;). These waters allow the efficacy of different coagulant types, combinations, dosages and pH conditions to be tested on the same water quality. Typically, the experiments have been conducted in jar tests with one parameter (pH, alkalinity, etc.) being manipulated to determine its effect. Unfortunately, treatment plants that treat surface waters experience significant variations in water quality throughout the year, and water quality parameters do not change in isolation, rather, the water doi: 10.2166/wst.2013.621

quality varies with multiple parameters changing simultaneously, often in a linlied way. The Minneapolis Water Treatment Plant, which draws its water from the Mississippi River at Fridley, is a classic example of an old surface water treatment plant. The quality of the Mississippi varies dramatically during each year. For the year 2007 the reported inñuent hardness concentrations ranged from just over 100-270 mg/L with an average hardness of 195 mg/L. The magnesium concentration varied from 2 to 110 mg/L with an average of 49 mg/L. The NOM concentrations ranged from about 5-17 mg/L, with an average of 8.7 mg/L as dissolved organic carbon (DOC). The temperature of the source water varied widely over the course of a normal year. During the winter the water temperattires fell close to 0 °C and during the summer the temperatures rose to values greater than 25 C. Similar variability was observed in the source water quality during both 2006 and 2008. fn the spring and early fall periods, snow melt and heavy rain caused high flows in the river and during these periods lower hardness and alkalinities were

1009

J. M. L. Dias et al. \ A hybrid modei to predict NOM remcvai during iime softening

meastired; however, NOM concentration and color clearly increased. In addition, it is likely that signiflcant temperature change, and seasonal variations also alter tbe character of the NOM in the water. These effects are difflcult, if not impossible, to capture in the typical jar tests conducted on extracted NOM. In addition, if jar tests are conducted on real source waters, then the timing of sample collection, and changing water quality, may modify the observed NOM removal behavior. The Minneapolis Water Treatment Plant was built as a lime softening plant in the early 1900s. It has been modifled over the years and in 2005 ultraflltration membranes were added to the treatment train. The plant is equipped to add potassium permanganate, ferric chloride, alum, high calcium slaked lime and powdered activated carbon (PAC) to optimize process performance. The dosages and the locations of addition can be manipulated and sludge can be recycled to enhance process performance (see Figure 1 for details). Good NOM removal during softening pre-treatment is

Water Science & Technoiogy | 67.5 | 2013

essential for control of tastes, odors and disinfection byproducts, and to avoid membrane fouling. In this paper we explore chemical dosage and water quality data collected by the city of Minneapolis over a 3 year period to determine if we can identify relationships which describe NOM removal performance and also to provide a practical tool to assist operators in determining the chemical dosages for treatment. To do this we explore a hybrid modeling approach which incorporates a simple phenomenological model with a statistical analysis using Partial Least-Squares (PLS).

MATERIALS AND METHODS Water pre-treatment data A data set comprising daily and weekly independent measurements of water quality and pre-treatment parameters

Spaulding predpitators (softening cones)

® Recarbonation Chambers

Mixing chambers

Sample point DOC/UVA

'.il Coagulation/ sedimentation

Granular media filters (T) Potassium Permanganate (KMnO4) (2) Powdered Activated Carbon (PAC) + Calcium Carbonate (CaCOs) (3) Lime (CaO) + Ferric or Aluminixim Sulfate (Al2(SO4)3 ( J ) Carbon Dioxide (CO2) (5) Powdered Activated Carbon (PAC) (6) Ferric Cbloride (FeCl3) and Chlorine (CI2) (7~) Ammonia (NH3) Figure 1 I Schematic of the process treatment train at the Minneapoiis Water works softening piant at Fridiey.

Mixing chambers

1010

¡. M. L. Dias eí al. \ A hybrid modei to predict NOiVl removai during iime softening

(828 records) and DOC and UVA (107 records), respectively, covering a 3 year operation period, namely 2006, 2007 and 2008, was available for PLS analysis. In order to use all available data, a linear interpolation in DOC and UVA was assumed over each week. To gain more confidence about the final results, the data collected in 2007 and 2008 were used for model calibration (586 records) while the 2006 data were used for model validation (242 records). Phenomenological NOM removal model The phenomenological model was built on the following main simplifying assumpfion: The precipitated solids, coagulants and PAC, adsorh NOM and approach equilibrium (or fractional equilibrium) with the final NOM concentration in the water (Liao & Randtke 1986). Lin et al. (2005) also assumed that the formation of the NOM-calcite complex can be defined by a stability constant, which incorporates both the total calcium carbonate concentration precipitated, M, and the free Ca^+ concentrafion. Fe and Al coagulants, which usually remove NOM best at low pH, have been shown to dramatically improve NOM removal during lime softening in the Minneapolis plant and so we assume their mechanism of removal is by improving adsorption. • For simplicity, we assume that each chemical follows a linear adsorpfion isotherm model (Liao & Randtke 1986). • Only a certain fraction, F, of the NOM can be removed by lime softening and/or coagulation. • Reactor ldnefics and the impact of process design on the effecfiveness of the chemicals are ignored; i.e. the physical mixing conditions and reactor effects are constant throughout the study.

Water Science & Technology | 67.5 | 2013

calcium and magnesium precipitated in the softening process, which depends upon lime dose and initial hardness. The value of [Ca^^], the final calcium concentration at the conclusion of softening, was determined from the excess lime concentration employed during softening. Since several investigators have observed greater NOM removals when specific ultraviolet absorpfion (SUVA) is high (Archer & Singer 2006), it is likely that F is related to SUVA in some fashion. Several authors have noted that the more aromatic and hydrophobic organics absorb most ultraviolet light and these compounds are better removed by chemical pre-treatment. Here, we define a fracfion F as related to SUVA by the following proporfionality in which Ä is a constant:

In this analysis, no specific mechanism is included for the action of permanganate because the mechanisms are not sufficiently lmown. Rather, the approach is to allow its significance to be detected by the stafistical approaches used in conjuncfion with the phenomenological model, as described later in the hybrid modeling section. If one considers the addifion of a single dosing chemical, then Equation (1) can be reformulated in a simpler form to yield the respecfive NOM removal fraction. For example, the fracfion of NOM removal due to PAC dosing can be obtained from Equation (1) by making [Ca+^] = DFe = DAi = O. This procedure resulted in the following four equations for the calculation of NOM removal due to the addition of a single dosing chemical: (3)

l-t-O-

The equations expressing the NOM removal as a funcfion of the different chemical dosages (DpAc, öpe and DAI) can be combined and the total NOM removal can be written as follows:

.,

K' M- [Ca++]

^' 1+K'M-

[Ca+^

(4)

(5)

o - [NOM] = O • DpAc• [NOM] + M • K'• [Ca+ {[NOM] - ( l - F ) - [ N O M ] o } K"'DM

{[NOM]-(l-F).[NOM]o}

(2)

SUVA

1 + K'" • 'AI D

(1)

vwth [NOM]o and [NOM] the inifial and final concentrafions of NOM in the treated water. The terms K', K", K"\