Investigation of chlorine wall decay in an old, decommissioned metallic pipe using a pipe section reactor

Chlorine disinfection is the most commonly used method in centralized water supply systems all over the world for ensuring microbiological safety of drinking water. One of its main advantages is that the residual free available chlorine, which is the sum of hypochlorite ions, hypochlorous acid and chlorine gas, acts as a secondary disinfectant. It provides a residual bactericidal effect that provides further drinking water protection in the pipe network. According to the World Health Organization standards, concentrations up to 5 mg/l of free chorine do not present a health risk for the consumers (WHO 2017). The WHO recommends that the concentration of free chlorine at the beginning of the water supply system is at least 0.5 mg/l after 30 minutes' contact of the disinfectant with water. At the end consumer, the recommended concentration is at least 0.2 mg/l; if, however, the concentration is above 0.3 mg/l some individuals can experience unpleasant smell and taste of water.

The maintenance of certain concentrations of free residual chlorine creates a challenge for the water operators, since its concentration is constantly changing through the water distribution network. The decay of free chlorine is associated with the occurrence of different chemical reactions in the bulk phase of the water and at the inner pipe walls (Rossman et al. 1994). The development of distribution network simulation models has proven to be a useful tool for predicting the concentration of residual chlorine in any point of the network at a given time. For the last 35 years, this field has been continuously augmented with the contributions of many researches, who strive to get more accurate results to be applied in drinking water supply management practice (Grayman 2018).

Chlorine wall decay is usually associated with the interaction of chlorine with existing biofilm, corrosion deposits and the pipeline material. Field approaches (Vasconcelos et al. 1997) and laboratory studies (Rossman et al. 2001; Digiano & Zhang 2005) have been used to investigate the kinetics of wall reactions. The lack of a standard method is due to the large variety of factors influencing the process. Some studies focus on the chemical and physico-chemical parameters of water (Kiene et al. 1998; Digiano & Zhang 2005; Rossman 2006); others focus on the age, material, diameter and roughness of the pipes (Zhang et al. 1992; Vasconcelos et al. 1997; Al-Jasser 2007) or on the hydraulic conditions inside the pipes (Digiano & Zhang 2005; Clark et al. 2010).

The present study aims to investigate the free chlorine decay kinetics in an old, decommissioned section of metallic pipe from the water distribution network of the city of Sofia. The study determines experimentally the overall decay rate (K) and bulk reaction coefficient (k_b) in laboratory conditions using tap water. The modelling of chlorine decay, for the specific case of the Sofia water system, is a necessary step for enhancing the control of disinfection, i.e. appropriate allocation of disinfection facilities and assurance of sufficient free residual chlorine in a network serving over 1.3 million consumers.

А laboratory pipe section reactor (PSR) was built, in accordance with the design of Digiano & Zhang (2005) (Figure 1).

An old iron-made pipe (internal diameter 150 mm, L = 300 mm), decommissioned from the water distribution network of Sofia city, was used for the experiments. It was put into operation in the 1980s and was taken out of service in 2019. About 50% of the water supply network in Sofia consists of such pipes with similar characteristics. The sealing plates and the inner cylinder (diameter 120 mm) are made of acrylic. Eight rectangular shaped stainless steel baffles, B × L = 20 × 125 mm, are attached to minimize the vortex in the annular space between the cylinder and test pipe. A propeller of diameter 100 mm and variable speed (up to 600 r·min⁻¹) creates a circulating flow through the annular space. To prevent random errors due to the occasional introduction of air into the PSR during manual sampling (Hua et al. 2017), the system was equipped with a top-up reservoir (Figure 1).

A series of experiments was made using tap water from the distribution system of Sofia city. Each experiment in the series included two parallel tests:

• Test 1: PSR test to determine the overall reaction coefficient (K);

• Test 2: Bottle test for evaluation of the bulk reaction coefficient (k_b).

The model water solution was prepared with tap water and sodium hypochlorite, mixed in a sterilized, prewashed with chlorine-demand-free-water (Andrew et al. 1998) 10 l brown glass bottle. The tap water used was from the ‘Losenetz reservoir’ of the Sofia city water distribution system, which is fed by two different water sources, which are chlorinated several times before mixing in the storage reservoir. The temperature of the solution and the initial chlorine concentration were measured before each set of experiments. The PSR was filled with model water and a sample was taken to define starting condition. Then the propeller was started, which time was considered to be the beginning of the overall chlorine decay test. Samples were taken from the PSR at random time intervals. The experiment ended when the chlorine concentration in the PSR dropped below 0.05 mg/l. In parallel with the PSR tests, bottle tests were also performed. The rest of the model solution was poured into several 100 ml sterilized brown glass bottles, which had been prewashed with chlorine-demand-free-water. Each bottle was completely filled and closed with a polypropylene cap to prevent contact between the water and air. Bottles were stored in a dark environment with constant temperature, close to the temperature in the PSR. Simultaneously with the samples from the PSR, a sample from one of the small brown bottles was taken. Additional bottle tests were also performed to investigate the bulk reactions in greater detail.

For chlorine concentration, a standard DPD colorimetric method is used with sensitivity of ±0.02 mg/l (HACH 2009). А thermometer with 0.1 °C reading intervals was used for temperature measurement. For other water parameters of concern (e.g. iron, manganese, nitrites, ammonia and TOC), standard laboratory methods (HACH 2007) were used.

The flow velocity in the PSR was measured by detecting the flow travel for a certain time. For this purpose, a small piece of material with density close to water was placed in the PSR and the time for 10 full circuits was measured.

A linear relationship between the motor rotation speed (N r·min⁻¹) and the average water flow velocity in the PSR (u_av m/s) was established. The velocity in the annular space, u m/s, was calculated based on the u_av and the cross-section area of the annular space for N = 200 ÷ 400 r·min⁻¹.

In total, 14 series of experiments were performed with constant propeller rotation speed in accordance with annular flow velocity of u = 0.39 m/s, and with different initial chlorine concentrations ranging from 0.30 to 1.80 mg/l. The average temperature of the tap water for all the tests varied within the range T = 11.3 – 13.9 °C. For every single series of experiments, the temperature amplitude was below 2.0 °C.

For an initial chlorine concentration C₀ ≤ 0.85 mg/l (Table 1, series 6–14) a general first order model (e.g. Equation (5)) describes very well the observed results. Since during the experiment the hydraulic conditions are constant, the temperature change is minor, and the annular flow velocity is constant; k_f can be considered a constant. In consequence, applying Equation (4.2), K_w1 should also be constant. The results show, however, that K_w1 varies in a relatively wide range – from 1.402 to 2.492 h⁻¹. This leads to the conclusion that a mass transfer limited model (e.g. Equation (4.2)) does not describe appropriately the experimental results. At higher initial concentrations (Table 1, series 1–5), by applying the general first order model (e.g. Equation (5)), the difference between the calculated and observed value reaches 0.24 mg/l (Figure 2(a)) and the average absolute error is ≥0.02 mg/l, which is the sensitivity level of the laboratory method for chlorine measurement.

In conclusion, all experiments show a non-linear relationship (e.g. first order reaction model) for the chlorine decay in the PSR. First order wall reactions are observed in previous studies for ductile iron pipes (Rossman et al. 2001; Digiano & Zhang 2005) and for steel pipes (Al-Jasser 2007). The obtained results also indicate that wall reactions have much greater impact than bulk reactions for the tested pipe fragment. A similar conclusion was also reached by Digiano & Zhang (2005) and Rossman (2006).

The obtained results suggest, however, that the general first order reaction model (Equation (5)), except for the mass transfer limited model (Equation (4.2)), can predict accurately enough the chlorine decay reaction in the tested pipe only if the initial chlorine concentration is ≤0.85 mg/l. It should be noted that the model used for bulk reactions is simple first order. The same model was used in previous studies (Kiene et al. 1998; Rossman et al. 2001; Hallam et al. 2002; Digiano & Zhang 2005; Al-Jasser 2007); however, no information was found regarding how changes in the initial chlorine concentration might have affected the accuracy of the obtained results. In this study for the Sofia water supply system, where different initial chlorine concentrations were used (Table 1, column 2), the authors suggest that a bulk reaction rate of different order might describe better the bulk chlorine decay.

Values of K_w1 are calculated using least-squares regression analysis (Table 1, column 9). For all the series, except one, the mean absolute error is ≤0.02 mg/l, which is comparable to the measurement error (Table 1, column 10). The average Nash-Sutcliffe efficiency coefficient is 0.997, which implies better convergence than the first order general model, where the average NSE is 0.982 (Figure 2(a)). It is important to mention that both zero and first order models describe very well the bulk reactions, as also noted by other researchers (Devarakonda et al. 2010).

An exponential decrease in the value of K_w1 was found as C₀ increases (Figure 2(b)), after applying Equation (6.2). Since the mass transfer coefficient k_f is constant, such dependence can only be explained by a change in the rate coefficient of wall reactions k_w1. A similar trend has been observed by other authors (Hallam et al. 2002; Rossman 2006). Chlorine reaction with iron containing materials results in oxidation of the ferrous ion (Fe²⁺) to the ferric ion (Fe³⁺), which latter precipitates as ferric hydroxide (Vieira et al. 2004). This creates a protective layer on the inner pipe-wall surface, which is the main source of Fe²⁺ ions (Digiano & Zhang 2005). Depending on the hydraulic conditions, shear stress can occur at the inner wall of the pipe. The corroded material (i.e. the ferric hydroxide) is sheared off, resulting in the exposure of additional sections of the metal pipe to chemical interaction (Furnass et al. 2013). This process leads to a change in the ratio between the fast and slow reacting with chlorine species. At low initial chlorine concentration, the reaction rate would be determined only by the concentration of the fast-acting agents, while at higher C₀ the slow-acting ones will also affect the process.

Four additional bottle tests were performed at temperature 13 – 14 °C to investigate the bulk reaction kinetics. The tap water quality analyses showed the presence of total manganese Mn = 0.008 mg/l, ammonium NH₄⁺ = 0.021 mg/l and total organic carbon TOC = 2 mg/l. The concentration of total iron and nitrite were below the detection limits, respectively <0.02 and <0.05 mg/l.

In all the tests, similar kinetics of chlorine decay was observed, which could be divided into two phases (Figure 3). The first phase, with a duration of about 5 hours, shows a significantly faster chlorine decay than the second phase. Simple and parallel first order models were applied to determine the bulk chlorine decay coefficients (Table 2). A regression and least squares regression analyses were performed accordingly. It can be concluded that the parallel first order model describes more accurately the experimental results (Figure 3 and Table 3, columns 1–6).

The values for the overall zero order bulk decay coefficient (K_b0=k_b0,I+k_b0,II) obtained during the first 3.5–4.8 hours of the bottle tests are in the range 0.046 – 0.068 mg/l.h⁻¹ (Table 3, columns 7–8). They correspond well to the obtained values for the bulk decay coefficient during the PSR series, which are in the range of 0.044–0.109 mg/l.h⁻¹ (Table 1, column 8).

In order to evaluate the first order wall reaction coefficient, k_w1, nine additional series of experiments were performed in the PSR with initial chlorine concentrations C₀ = 1.18; 0.87 and 0.46 mg/l (±0.02) and annular space velocities of u = 0.27; 0.50 and 0.62 m/s. The results were compared with the previous series of experiments with u = 0.39 m/s (Table 1). The overall wall reaction rate, K_w1, is calculated using Equation (6.2).

Previous studies concluded that with the increase of water velocities, chlorine wall demand also increases (Grayman et al. 2002; Doshi et al. 2003; Clark & Haught 2005). The results of the current study show however, that the initial chlorine concentration also significantly influences the chlorine wall decay process (Figure 4(a)).

The free chlorine decay in an old metallic pipe was investigated under laboratory conditions with tap water from the Sofia distribution system. The overall decay rates (K, K_w and K_b), the bulk reaction rate coefficient (k_b), the wall reaction rate coefficient (k_w) and the influence of the mass transfer process have been determined based on a series of experiments. The laboratory pipe section reactor, being a closed homogenous system, is suitable for the study of the kinetics of chlorine decay in water.

It can be concluded that a first order general model is applicable for predicting overall chlorine decay rates (K, K_w and K_b) at lower initial concentrations (0.3–0.85 mg/l) for the tested specific conditions from the Sofia water supply system; however, the model is not appropriate for describing the laboratory results at higher initial chlorine concentrations (0.88–1.8 mg/l).

With regard to the chlorine decay bulk reactions, a first order parallel model characterizes well the tested tap water. The proposed zero order parallel model, however, describes even better the laboratory results, having Nash-Sutcliffe efficiency coefficients over 0.99.

The first order wall reaction model is appropriate for predicting the wall reaction decay in the studied metallic pipe. The respective wall reaction coefficient (k_w1) has been determined for different initial chlorine concentrations (0.3–1.8 mg/l) and different hydraulic conditions (velocities of 0.27 m/s 0.39 m/s, 0.50 m/s and 0.62 m/s). There is a clear trend of decreasing k_w1 (from 0.03 to 0.008 m/h) with the increase of the initial chlorine concentration.

This investigation was made within the framework of Contract No. Д - 92/16 of CNIP at UACEG. The authors would like to thank also the Sofiyska Voda AD, part of Veolia Group, who contributed greatly to this research.