Raw water from a dam reservoir was used to assess the cleaning efficiency of the membrane process. Water quality data throughout the operational period are detailed in Table S1. Although turbidity was low, the high concentration of TOC and the presence of Mn in the raw water raised concerns about potential organic fouling due to organic matter and inorganic fouling resulting from Mn in the ultrafiltration (UF) process. To address this complex water quality issue, a pilot plant was designed to evaluate the necessity of implementing a permanganate oxidation process as pretreatment to control incoming Mn and mitigate the organic load on the membrane.

The advanced water treatment pilot plant was composed of manganese oxidation process, UF, UV-hydrogen peroxide advanced oxidation process (AOP), and granular activated carbon process. However, to understand fouling characteristics, the same raw water was directly filtered using a mini-UF membrane to induce severe conditions. The mini-UF membrane module (S10N, Evoqua Water Technologies, Pittsburgh, PA, USA) had a pore size of 0.04 μm and was composed of hydrophobic polyvinylidene difluoride (PVDF). The UF membrane applied in the pilot plant had the same specifications.

A mini-UF module was detoured from the pilot plant and operated together on site for fiber analysis and cleaning studies, and it was removed from service after four months of operation without pretreatment or chemical cleaning. Fiber samples obtained from this module were cut and made into pencil modules for fiber analysis and laboratory cleaning. The cleaning protocols selected from the laboratory cleaning study were verified using the pencil module. The pencil module was manufactured using two fibers with 400 mm and 200 mm of total and free fiber lengths, respectively, providing a 0.0013 m² of membrane surface area per individual pencil module (Fig. 1a). Eighty fiber samples were extracted from the mini-module, and forty pencil modules were fabricated with two fibers from the pre-cut fiber samples. The tests conducted in this cleaning study included module inspection, pencil module preparation from fiber sampling, cleaning protocol screening with pencil modules, module analysis, and recovery flux tests.

Clean water permeability tests and post-clean fouling studies were conducted with reverse osmosis (RO) water in a bench-scale filtration unit (Fig. 1b). This unit was designed to filter at constant pressure, provide air and liquid backwashes, and operate with soaking CIP method. The filtrate flow was measured using two lab balances with a digital output capability. The pressure and temperature were measured using transducers, digitized, and logged in a programmable logic controller (PLC) system. The operating flux, membrane resistance, and permeability were calculated automatically using the PLC system. The clean water permeability of each pencil module was tested by filtering RO permeate for approximately 5 min. For post-clean fouling studies, a test feed with a turbidity of approximately 10 NTU was prepared by combining appropriate proportions of the test site feed water (drawn from South Creek) and test site RO permeate (UF-filtered South Creek water followed by RO filtration). Post-clean fouling performance of each pencil module was determined by filtering the test feed through the pencil module for approximately 4–8 h without backwashing. The cleaning efficiency was calculated as shown in Eq. (1) to evaluate the effects of chemical cleaning on permeability recovery with foulant removal.

During the cleaning process using pencil modules, the test membranes were immersed in selected cleaning agents under defined conditions. RO permeate was used as the diluent for all cleaning solutions. The chemicals and cleaning conditions used in the screening experiments are presented in Fig. 2. Citric acid at concentrations ranging from 0.1% to 2% was used as the primary acidic cleaning agent. Phosphoric acid (PA) was added to adjust the pH to a target value of 2.0. Sodium hypochlorite at concentrations between 500 and 1,500 mg/L was used as the main alkaline cleaning agent. The chemical concentration, contact time, and water temperature were varied to evaluate their effects on cleaning efficiency.

The contact times for each chemical were determined based on typical chemical cleaning durations (2–4 hours per cleaning agent) used in drinking water treatment plants that treat surface water. The concentration ranges for the two chemicals were also selected based on the ranges commonly used in on-site CIP and maintenance cleaning procedures. The cleaning temperatures were selected from within the recommended range for PVDF membranes (20–45°C). These temperatures not only influence the reaction kinetics of acidic and alkaline agents but also represent feasible and practical values for heating large volumes of water in field applications. Although PVDF membranes exhibit good thermal resistance, prolonged exposure to temperatures above 50°C can lead to pore structure deformation and membrane degradation under chemically aggressive conditions. Therefore, 20°C and 40°C were selected as representative temperatures for this study. Dual chemical cleaning using pencil modules was conducted under various combinations of conditions derived from the screening tests. The acid–alkaline cleaning sequence was adopted based on the chemical rationale and manufacturer guidelines for membrane operations during surface water treatment. Acid cleaning was prioritized to remove inorganic foulants, such as metal oxides, which can interfere with the efficacy of subsequent alkaline cleaning.

Scanning electron microscopy (SEM) and energy-dispersive X-ray spectroscopy (EDS) were used to identify and determine the elemental composition of membrane foulants [35]. Post-clean hydraulic performance tests were conducted with pencil modules after the selected cleaning chemicals. Net permeability recoveries were calculated for all trials. The full-module permeability and hydraulic performance were evaluated prior to the module cleaning study. Full-module cleaning, which is the most effective cleaning method, was identified in a pencil module cleaning study. Furthermore, the pilot-scale module was validated by post-clean hydraulic performance. Pencil modules were used to confirm the cleaning results for each test condition. The overall permeability recovery was then compared with the permeability of the new fibers. A total of twenty-three lab-scale chemical cleanings with pencil modules were completed in the screening trial, including duplicates.

The experimental results were reclassified according to the full factorial design and response surfaces were implemented using RStudio software (version 2024.04.2). The design aimed at revealing the impact of chemical concentration and temperature on the permeability of the membrane after cleaning. Four independent variables were chosen, including the concentration of CA (X₁), concentration of NaOCl (X₂), temperature (X₃), and contact time (X₄), to assess the effects of acidic and alkaline cleaning on the permeability of the cleaned module compared to the permeability of a new fiber (Y₁) and determine any changes in permeability after cleaning compared to that of the new fiber (Y₂) (Table 1). Since only experimental data were applied, X₂ had a level of 3 and the other independents had a level of 2. A preferable formulation was derived from mathematical computations. In addition, surface plots and contour lines were drawn to visualize the effect of each independence. Variance Inflation Factors (VIFs) were evaluated for each term to check for multicollinearity. VIFs less than 5 were considered acceptable and indicated that multicollinearity was not severe.

The module was screened, and the fibers appeared dark brown. Black solids were found in the module bundle near the top of the filtrate pot and were assumed to be manganese complexes. The filtrate and aeration pots remained in good condition. In addition to examining the full module, the fiber samples were removed from the module and analyzed via SEM (Fig. 3). Cross-sectional SEM images of the module revealed typical virgin structured fibers. Close-up images of the fibers did not show any signs of internal membrane fouling. Surface SEM images also revealed a thin, widespread fouling layer on the membrane. The close-up surface SEM images primarily revealed traces of inorganic matter on the surface.

The EDS analysis was performed on a very small section of the surface. No foulant was found in the internal wall structure of the fibers. High concentrations of organic materials (C) were present in all fiber samples. Small amounts of inorganic materials, mainly aluminosilicates and manganese oxides, were found at 3 of map on the fiber surfaces. EDS scan results are presented in Table S2.

Acid screening tests were performed to identify the most effective acidic cleaning condition. The effects of citric acid concentration, temperature, and chemical contact time on the efficacy of acidic cleaning were evaluated using these tests. A moderate increase in permeability recovery was observed after cleaning with 2.0% citric acid compared to that after cleaning with phosphoric acid (Fig. 4a). The average cleaning efficiency of phosphoric acid was only 12.4%. The improved average cleaning efficiencies with 0.1, 0.5, 1.0, and 2% citric acid and pH adjustment using phosphoric acid were 10.5, 12.2, 11.8, and 15.4%, respectively. The effects of temperature and chemical contact time on the citric acid cleaning efficiency were evaluated (Fig. 4b). All cleanings were conducted using 2.0% citric acid. The effect of increasing temperature from 20 to 40°C at a chemical contact time of 2 h on permeability recovery was observed, and the average cleaning efficiency was 14.7 and 19.7%. At a chemical contact time of 4 h, the average cleaning efficiencies were 13.7% at 20°C and 22.7% at 40°C. Extending the chemical contact time from 2 to 4 h had a minor effect on permeability recovery. However, noticeable improvements in permeability were observed at elevated temperatures. After four months of use, the mini-UF module showed the formation of the inorganic foulants containing manganese and aluminosilicate. Cleaning with citric acid achieved approximately 12% cleaning efficiency, which improved to 15% when the pH was adjusted to 2 using phosphoric acid. This indicates that citric acid effectively removed manganese ions through chelation while phosphoric acid-based optimization of the solution to pH 2 enhanced the dissolution of inorganic substances, such as manganese, thereby improving the cleaning efficiency. Overall permeability recovery was improved at elevated temperatures combined with extended chemical contact time; 2.0% citric acid at 40ºC and 4 h of contact time exhibited the optimum average permeability recovery.

Sodium hypochlorite screening trials were conducted to identify effective cleaning conditions. Fig. 4c shows the results of cleaning with sodium hypochlorite at different concentrations. The cleaning efficiency improved marginally with increase in NaOCl concentration. Sodium hypochlorite at 500, 1000, and 1500 mg/L showed cleaning efficiencies of 40.4, 40.5, and 46.4% based on 20 trials and a total chemical contact time of 2 h. When the contact time extended to 4 h, the cleaning efficiencies changed to 37.3, 39.3, and 44.3%. Significant improvement in permeability recovery was found when the solution temperature was increased from 20 to 40ºC, as the cleaning efficiency increased from 44.3 to 73.6% at the contact time of 4 h and NaOCl concentration of 1,500 mg/L. The raw water characterization analysis indicated that organic fouling was more dominant than inorganic fouling, which was further supported by the cleaning performance observed after alkaline cleaning. Sodium hypochlorite effectively solubilized organic matter through an oxidation-based mechanism and exhibited high cleaning efficiency under elevated concentrations and temperatures due to increased reaction rates. Further investigation into the characteristics of organic foulants is recommended through detailed analyses, such as analyses of the characteristics of organic compounds in the cleaning waste solution or FTIR analysis of the fouled and cleaned membranes. Sodium hypochlorite at a concentration of 1500 mg/L, temperature of 40ºC, and contact time of 4 h achieved the highest overall permeability recovery among all alkaline cleaning conditions tested.

Based on the screening trials of single step cleanings, further tests were conducted with several combinations of acidic and alkaline chemicals to compare cleaning efficacy of dual cleaning. Moreover, the effects of temperature and contact time of agents on the dual cleaning process were evaluated (Fig. 4d). Dual cleaning was performed using 10 post CIP pencil modules. Cleaning with sodium hypochlorite recovered most of the fiber permeability in all the test samples. The dual cleaning showed cleaning efficiency between 47.9 and 85.1%, while the second cleaning step with the alkali agents exhibited an efficiency between 35.4 and 61.7%. All dual cleans that included 1500 mg/L sodium hypochlorite showed excellent permeability recoveries, with the average cleaning efficiency of 76.9%. Dual clean with the optimal overall permeability recovery was established under high temperature and long contact time conditions: 2.0% citric acid, 40ºC, and 2–4 h of chemical contact time followed by 1500 mg/L sodium hypochlorite, 40ºC, and 4 h of chemical contact time achieved an average cleaning efficiency of 80.5%. On the optimal cases the contact time of the first step with CA had marginal impact on the permeability recovery.

The RSM analysis based on full factorial design was performed using RStudio (v. 2024.04.2). The results showed that all data sets, except one with the independent factor of CA concentration, were best suited to the two-factor interaction model (2FI) (Table 2). To assess the significance of the model, four statistical parameters were considered at a 99% confidence level: F-value, P-value, R², and adjusted R². All models showed significance, with P-values less than 0.01, and the experimental F-values were more remarkable than the F-critical values. All terms showed VIFs between 1.0 and 1.2, indicating that the models did not have severe multicollinearity. Additionally, all R² values were between 0.9 and 1.0, which revealed strong data fit between the experimental and predicted values [36].

Eq. (2) and (3) for Y1 showed that the agent concentration and temperature significantly affected the permeability of the cleaned module. The positive two-variable interactions of X2 and X3 (p<0.01, Table 2) indicated that an increase in these factors results in a higher alkaline cleaning efficiency (Fig. 5a, Fig. 5b). In addition, the positive linear interactions of X1 and X3 (p<0.01, Table 2) revealed that acidic cleaning could be improved with higher values of these factors (Fig. 5c, Fig. 5d).

Eq. (4) for Y₂ showed that the NaOCl concentration and temperature significantly affected the change in permeability, which was similar to Eq. (2). The positive two-variable interactions between X₂ and X₃ (p<0.001, Table 2), which are described in Eq. (4), are shown in Fig. 5e and Fig. 5f. Eq. (5) for Y₂ showed that contact time and temperature are significant factors in the change of permeability via acidic cleaning. The positive two-variable interactions of X₃ and X₄ (p<0.001, Table 2) showed that a longer contact time and higher temperature improved the acidic cleaning (Fig. 5g, Fig. 5h).

Therefore, higher concentrations, contact times, and temperatures would be considered optimal conditions for chemical cleaning in the range of the experiment. The optimal cleaning protocol derived from the model was as follows: 2.0% citric acid, 40ºC, and 4 h followed by 1500 mg/L sodium hypochlorite, 40ºC, and 4 h.

A pilot test using a mini-UF (S10) was conducted to perform filtration, dual chemical cleaning, and post-cleaning hydraulic performance validations. The feed water to this trial unit was a blend of surface water and UF filtrate water to achieve a turbidity of 10–15 NTU. The trial unit replicated the CIP process of the UF S10 unit, including multiple cycles of soaking and recirculation with the CIP solution to achieve the target contact time. The operating flux during CIP was set to 40 LMH. The dual CIP was performed with 2.0% citric acid at 40ºC and 4 h and 1500 mg/L sodium hypochlorite at 40ºC and 4 h. The received module permeability was 236 LMH/Bar@20°C, which was higher than that expected from a module after 4 months of service without any chemical clean. Dual CIP recovered the module permeability to 285 LMH/Bar@20ºC, which is within the typical range of the permeability of a new module (Fig. 6a).

During the post-clean performance test, the pilot trial unit was set to typical operating conditions, including a flux of 60 LMH with a backwash interval of 20 min. Raw feed and filtrate blend were used to validate the post-cleaning performance. The turbidity of the blended feed ranged between 10 and 15 NTU. The TOC content of the blended feed was approximately 10 mg/L. Dual CIP achieved good permeability recovery and performed well during the post-CIP hydraulic performance test. No post-clean hydraulic issues were observed in this module. CIP-UF reduced the initial membrane resistance (Eq. (6)) by more than 20% compared to fouled UF (Fig. 6b). The comparison of the increased rate of filtration resistance per filtration cycle showed a lower fouling rate of 28%. It was confirmed that physical cleaning is better at recovering reversible membrane fouling in the performance of CIP-UF. The analysis of the changing pattern of filtration resistance according to the operation cycle showed that the fouling rate of CIP-UF was 0.225/h, which was 24.5% lower than that of fouled UF at 0.297/h. Additionally, the TMP increased during the filtration mode and the total operation period was delayed, indicating that the dual cleaning protocol was effective in minimizing the effect of reversible fouling.

where J represents the permeate flux (m³/m²·s), represents the permeate viscosity (kPa·s), TMP represents the transmembrane pressure (kPa), R represents resistance (m⁻¹).