**2. Experimental work**

#### **2.1 Experimental facility**

The experimental work was conducted on a large scale at the WWTP. The municipal wastewater treatment works are designed for an average dry weather flow (ADWF) capacity of 24 Ml/d, with the current flow of about 14 M/l/d, mostly of

domestic origin. All inflow is pumped from the catchment to the inlet works, consisting of screening, degritting, and flow division. The plant is equipped with two biological reactors, which are arranged in the University of Cape Town (UCT) process configuration for enhanced biological phosphorus removal (EBPR). The mixed liquor is thereafter distributed to four secondary settling tanks (SSTs). The SST overflow gravitates to maturation ponds followed by Ultraviolet (UV) disinfection, and the underflow gravitates to the return activated sludge (RAS) pump station where it is pumped back to the anoxic zone in the reactors. Two variable-speed-driven pumps (duty/standby) pump the waste-activated sludge (WAS) directly out of each reactor's final zone to the sludge dewatering building, where the WAS is thickened in two linear screens (Siemens Gravity table GTN 2500), followed by dewatering in two belt filter presses (Siemens BPF 2000 CMF Optima S11 Cascade). The filtrate and wash water from the thickening/dewatering trains gravitate back to the RAS division box where it is mixed with the RAS before entering the anoxic zones of the bioreactors. The pressed sludge (cake) is discharged via a centerless screw conveyor to a reinforced concrete hopper, from which sludge is discharged to trucks for removal from site for beneficial agricultural reuse application. A schematic diagram of the WWTP is given in **Figure 2**.

The loading on the belt presses can be controlled according to solids or flow loading. Control using solid loading is the preferred operation and is managed based on the output from the total suspended solids (TSSs) meter in the reactor aeration

**Figure 2.** *Schematic diagram of the WWTP.*

basin. The design parameters of the Linear Screens require a minimum WAS feed of 4500 mg/l mixed liquor suspended solids (MLSSs) from the reactors. Treated effluent is reused with an automated in-line filtration stage and is available as process water in the dewatering building, both for belt washing and polymer makeup purposes. The sludge-conditioning polymer (Flopam FO4650) is provided in 25kg bags in powder form, and powder is pneumatically transferred and wetted for blending and storage in two makeup tanks. This polymer solution flows into a sump from where it is withdrawn by the polymer dosing pumps and dosed in-line into the WAS before it is fed to the linear screens. The dosing rate is set at the supervisory control and data acquisition (SCADA), based on analysis data and visual observations. The dewatering performance is guided by a minimum cake solid content of 14% and a solid recovery of 95%, respectively, which is affected by the ageing infrastructure that is due for a major refurbishment.

#### **2.2 Experimental design**

In this work, a Box-Behnken response surface experimental design (BBD with five factors at three levels was used to examine the effect of process operating parameters on the BFP performance measures. The process conditions were polymer concentration (%) [**A**], polymer dosing rate (m<sup>3</sup> /hr) [**B**], sludge feed rate (m<sup>3</sup> /hr) [**C**], filter belt speed (Hz) [**D**], and linear screen speed (Hz) [**E**]. The BFP performance measures were sludge cake, filtrate suspended solids, and solid capture, and these were selected as responses with the addition of the yield stress. The range of the operating parameters investigated was chosen based on the preliminary trials. The ranges and levels of the selected operating parameters are presented in **Table 1**. The final design matrix consisted of 31 experiments including five center points. Due to difficulty in changing the polymer concentration according to the normal randomization of results, a grouped approach was selected to run a polymer concentration per day (The polymer concentration was defined as "hard to change" in the Design-Expert V11 software). Random repeats and intermediate conditions were selected by the Design-Expert software to ensure randomization.

#### **2.3 Experimental procedure**

The complete Box-Behnken design parameters are provided in **Table 2**. The trials were run over a period of 6 days to allow for the required polymer concentration to be


#### **Table 1.**

*Factors in Box-Behnken experimental design.*


#### **Table 2.**

*Complete Box-Behnken design parameters.*

prepared for each day in the plant at large scale. Each experiment was run for 30 min to reach steady state before samples were collected. The responses were measured three times, and the average value was recorded.

The response variables measured were the sludge cake solids, the filtrate suspended solids, solid capture, and the yield stress. The solids were measured using the standard oven drying technique for the cake, filtrate suspended solids, and the feed. The percentage solid capture (recovery) was calculated using the Eq. (3):

$$\text{Solid capture } (\%) = \frac{\text{solids in the feed} - \text{solids in the filter}}{\text{solids in the feed}} \times 100\tag{3}$$

A Paar-Physica MCR-51 rheometer fitted with a parallel-plate geometry, with a measuring cell CP50 of 50 mm diameter and a gap of 1 mm, was used for determining the sludge yield stress. The rheological tests were conducted in the controlled shear rate mode with the shear rate being the control parameter. Rheological measurements were performed by decreasing the shear rate linearly from 700 s�<sup>1</sup> to 100 s�<sup>1</sup> to exclude data with secondary flows at high shear rates and low torque at low shear rates. Thirty data points were collected, and each point was measured for 5 seconds. This protocol was repeated three times for each sludge sample to ensure reproducibility of the measurements. The data were then fitted with the Bingham model to obtain the yield stress.

The sludge samples tested were collected on the linear screen just after rake number 6. Before this point, the sludge proved to be difficult to measure due to dilute nature with rapid separation of sludge flocs and water as the shear force was applied. Due to the nature of sludge, there is some initial structure that is broken down when sheared several times. The sample was run several times to obtain representative sets of curves from repeat runs of three fresh samples. The average of the yield stress of the highest and lowest curve was determined for each run as a comparative result. It is recommended that the structure of the flocs be interrogated in future in subsequent test runs using oscillation tests. For this work to determine the comparative trends, the analysis was deemed sufficient. A typical rheogram is shown in **Figure 3**.

### **3. Results and discussion**

#### **3.1 Feed, filtrate, and cake solids**

The objective of these trial runs was to determine the operational factors of the belt filter press that contributes to efficient dewatering, indicated by the filtrate suspended solids and the percentage of solids in the cake. The rheological parameter describing the floc type produced by the specific polymer dosage rate, polymer concentration, the sludge feed rate, the belt speed, and the linear screen speed, with specific emphasis on determining the interaction that significantly affects the performance of the belt filter press. The factorial trial matrix with the responses is given in the Appendix. A summary of the feed solids concentration, the cake solids, and the filtrate suspended solids is shown in **Figure 4**. Cake solids of more than 14% were only obtained for one out of 31 runs.

#### **3.2 Relationship between FSS and cake solids and yield stress**

A power-law relationship was found between the filtrate suspended solids and the yield stress for the values obtained over the full range of the experimental results. Yield stress values between 1 and 5 Pa are sufficient to obtain FSS of between 0.9 and

**Figure 3.** *Typical rheogram.*

## **Figure 4.**

*Feed, filtrate, and cake solids results for all trial runs.*

0.5 g/L. **Figure 5** is useful to demonstrate the relationship between the yield stress and the FSS, but it does not give an indication of the process conditions required to obtain these yield stress values. The FSS was less than 0.5 g/L for 58% of the trial run

**Figure 5.** *Filtrate suspended solids as a function of yield stress.*

**Figure 6.**

*Cake solids as a function of yield stress.*

conditions. In contrast, there was no relationship found between the cake solids and the yield stress as shown in **Figure 6**.

#### **3.3 The effect of process conditions on belt filter press performance**

#### *3.3.1 Filtrate suspended solids*

The effect of process conditions on the filtrate suspended solids was investigated to determine the optimum process conditions. A split-plot design was used because of the number of variables as it also enabled the identification of process interactions that would normally not be seen when only one parameter was changed at a time. The filtrate suspended solids were measured at the outlet of the belt filter press. A sample was collected 20 minutes after the conditions were changed. The online FSS sensor was not responsive for the low readings obtained, and hence, a sample was collected,

and the standard drying test was conducted as well as filtration through 0.45μm Millipore filter.

#### *3.3.1.1 Analysis of variance*

The analysis of variance (ANOVA) is presented in **Table 3**. A quadratic model was selected without any transformation, with a step-wise backward selection, a p-value criterion, and an alpha value of 0.1. However, based on the Box-Cox analysis, an inverse square root transformation was required. P-values less than 0.0500 indicate that model terms are significant. Values greater than 0.1000 indicate that the model terms are not significant. Only the sludge feed (C) and the belt speed (D) were significant with p-values equal to or less than 0.05. The polymer concentration (a), polymer dosing, and linear screen speed (E) were not significant, with a p-value of 0.99, 0.40, and 0.32, respectively. However, the interaction between polymer concentration (a) and sludge feed and belt speed and linear screen speed as well as B<sup>2</sup> was significant. In this case C, aC, aD, aE, CD, DE, B<sup>2</sup> are significant model terms with a, B, and E added to maintain hierarchy. The final fit statistics obtained is showing an R<sup>2</sup> value of 0.7415 and adjusted R<sup>2</sup> of 0.5919 and a C.V. of 13.42%. These could be improved if Run 4 is excluded, but we are certain of the results even though it seems out of range, it is expected from such low polymer dosing.

The final equation obtained to determine the contribution of each factor is given in Eq. (4) (coded), and the equation for the prediction of filtrate suspended solids is given in Eq. (5) (actual).

$$1/\sqrt{\text{FSS}} = 1.62 + 0.0007\text{a} + 0.0323\text{B} - 0.1352\text{C} + 0.0647\text{D} + 0.0371\text{E} + 0.1122\text{ a}\text{C} \quad \text{(4)}$$

$$+ 0.1420\text{aD} - 0.1042\text{DE} - 0.1920\text{B}^2$$

$$1/\sqrt{\text{FSS}} = -6.245 - 38.61\text{a} + 20.00\text{B} - 0.080\text{C} + 0.138\text{D} + 0.215\text{E} + 0.261\text{a}\text{C} \quad \text{(5)}$$




#### *3.3.1.2 Evaluation of the filtrate suspended model*

The normal probability plot of the residuals is approximately linear, which shows that the model fits the data well as shown in **Figure 7**.

#### *3.3.2 Cake solids*

The resulting cake solids as a function of process conditions were investigated. The cake solids ranged from 10.39% to 14.26% over the full range of conditions used during the factorial trial. For the cake solids, it was found that an interaction exists between the polymer concentration and the sludge feed rate. Based on the statistical analysis (**Table 4**), the linear screen speed is a significant model term as well as the interaction between the polymer concentration and sludge feed, both of which had pvalues less than 0.05. The terms a and C and B were added to maintain hierarchy. The statistical analysis of the model showed that the model was acceptable based on the linear plot of normal plot of the residuals, Box-Cox plot confirmed a Lambda of 1 and


#### **Table 4.** *ANOVA for cake solids.*

that no further transform of the data was necessary. It is interesting to note that the belt speed is not a statistical significant operational factor in the prediction of the cake solids. Although the diagnostics are good, the Adj.R2 of 22.9% shows a very poor fit.
