Agent Skills: Membrane System Design Skill

Expert skill for membrane filtration and separation system design including process selection, flux calculations, fouling analysis, and concentrate management.

UncategorizedID: a5c-ai/babysitter/membrane-system-design

Install this agent skill to your local

pnpm dlx add-skill https://github.com/a5c-ai/babysitter/tree/HEAD/plugins/babysitter/skills/babysit/process/specializations/domains/science/environmental-engineering/skills/membrane-system-design

Skill Files

Browse the full folder contents for membrane-system-design.

Download Skill

Loading file tree…

plugins/babysitter/skills/babysit/process/specializations/domains/science/environmental-engineering/skills/membrane-system-design/SKILL.md

Skill Metadata

Name
membrane-system-design
Description
Expert skill for membrane filtration and separation system design including process selection, flux calculations, fouling analysis, and concentrate management.

Membrane System Design Skill

Membrane filtration and separation system design for water and wastewater treatment applications.

Purpose

This skill provides comprehensive capabilities for designing membrane treatment systems, including process selection, flux and recovery calculations, fouling analysis, pretreatment requirements, and concentrate management planning.

Capabilities

Membrane Process Selection

  • Microfiltration (MF) applications
  • Ultrafiltration (UF) applications
  • Nanofiltration (NF) applications
  • Reverse Osmosis (RO) applications
  • Process selection criteria and decision matrix
  • Hybrid system configurations

Flux and Recovery Rate Calculations

  • Design flux determination
  • Temperature correction factors
  • Recovery rate optimization
  • Concentration polarization effects
  • Osmotic pressure calculations
  • Permeate quality estimation

Pretreatment Requirements Assessment

  • Feed water characterization
  • Silt Density Index (SDI) analysis
  • Modified Fouling Index (MFI) calculation
  • Pretreatment technology selection
  • Chemical conditioning requirements

Fouling Analysis and Mitigation

  • Fouling mechanism identification
  • Biofouling assessment
  • Scaling potential analysis
  • Colloidal fouling evaluation
  • Organic fouling characterization
  • Mitigation strategy development

Concentrate Management Planning

  • Concentrate characterization
  • Disposal options evaluation
  • Zero Liquid Discharge (ZLD) considerations
  • Brine concentration technologies
  • Regulatory compliance for disposal

CIP System Design

  • Clean-in-Place system configuration
  • Chemical cleaning protocols
  • Cleaning frequency optimization
  • Chemical compatibility assessment
  • Cleaning effectiveness monitoring

Energy Recovery Device Selection

  • Pressure exchanger sizing
  • Turbocharger systems
  • Energy recovery efficiency
  • Economic analysis
  • System integration

Membrane Pilot Testing Protocols

  • Pilot system design
  • Test protocol development
  • Data collection requirements
  • Performance metrics
  • Scale-up considerations

Prerequisites

Installation

pip install numpy scipy pandas matplotlib

Optional Dependencies

# For optimization
pip install scipy pymoo

# For visualization
pip install plotly seaborn

Usage Patterns

Membrane System Sizing

import numpy as np
from dataclasses import dataclass
from typing import Dict, List, Optional

@dataclass
class FeedWaterQuality:
    """Feed water quality parameters"""
    tds_mg_l: float
    temperature_c: float
    ph: float
    tss_mg_l: float
    toc_mg_l: float
    hardness_mg_l: float = 0  # as CaCO3
    silica_mg_l: float = 0
    sdi: float = 0  # Silt Density Index

@dataclass
class MembraneElement:
    """Membrane element specifications"""
    manufacturer: str
    model: str
    area_m2: float
    permeability_lmh_bar: float  # L/m2/hr/bar
    salt_rejection: float  # decimal
    max_recovery: float  # decimal
    min_concentrate_flow_m3_hr: float

class ROSystemDesign:
    """Reverse Osmosis system design"""

    def __init__(self, feed: FeedWaterQuality, element: MembraneElement):
        self.feed = feed
        self.element = element

    def osmotic_pressure(self, tds_mg_l: float) -> float:
        """Calculate osmotic pressure in bar"""
        # Simplified van't Hoff equation
        # pi = i * C * R * T
        # For typical water: pi (bar) ≈ 0.0385 * TDS (g/L)
        return 0.0385 * (tds_mg_l / 1000) * (273.15 + self.feed.temperature_c) / 298.15

    def temperature_correction_factor(self) -> float:
        """Temperature correction factor for flux"""
        # TCF = exp(2640 * (1/298 - 1/T))
        T_kelvin = 273.15 + self.feed.temperature_c
        return np.exp(2640 * (1/298.15 - 1/T_kelvin))

    def calculate_flux(self, feed_pressure_bar: float, recovery: float) -> float:
        """Calculate permeate flux in LMH"""
        # Average osmotic pressure (considering concentration polarization)
        avg_concentration_factor = 1 / (1 - recovery/2)
        avg_tds = self.feed.tds_mg_l * avg_concentration_factor
        pi_avg = self.osmotic_pressure(avg_tds)

        # Net driving pressure
        ndp = feed_pressure_bar - pi_avg - 1  # 1 bar backpressure

        # Flux with temperature correction
        tcf = self.temperature_correction_factor()
        flux = self.element.permeability_lmh_bar * ndp * tcf

        return flux

    def design_system(self, feed_flow_m3_hr: float, target_recovery: float,
                      feed_pressure_bar: float) -> Dict:
        """Design RO system for given requirements"""

        # Calculate flux
        flux = self.calculate_flux(feed_pressure_bar, target_recovery)

        # Permeate flow
        permeate_flow = feed_flow_m3_hr * target_recovery

        # Required membrane area
        required_area = (permeate_flow * 1000) / flux  # m2

        # Number of elements
        num_elements = np.ceil(required_area / self.element.area_m2)

        # Elements per vessel (typical 6-8)
        elements_per_vessel = 6
        num_vessels = np.ceil(num_elements / elements_per_vessel)
        actual_elements = num_vessels * elements_per_vessel

        # Concentrate flow and quality
        concentrate_flow = feed_flow_m3_hr * (1 - target_recovery)
        concentrate_tds = self.feed.tds_mg_l / (1 - target_recovery)

        # Permeate quality
        avg_passage = 1 - self.element.salt_rejection
        permeate_tds = self.feed.tds_mg_l * avg_passage * (1 + target_recovery)

        # Specific energy consumption estimate
        pump_efficiency = 0.80
        sec_kwh_m3 = (feed_pressure_bar * 100) / (36 * pump_efficiency * target_recovery)

        return {
            'design_flux_lmh': flux,
            'required_area_m2': required_area,
            'num_vessels': int(num_vessels),
            'elements_per_vessel': elements_per_vessel,
            'total_elements': int(actual_elements),
            'permeate_flow_m3_hr': permeate_flow,
            'concentrate_flow_m3_hr': concentrate_flow,
            'permeate_tds_mg_l': permeate_tds,
            'concentrate_tds_mg_l': concentrate_tds,
            'specific_energy_kwh_m3': sec_kwh_m3
        }

# Example usage
feed = FeedWaterQuality(
    tds_mg_l=2000,
    temperature_c=25,
    ph=7.5,
    tss_mg_l=5,
    toc_mg_l=3,
    sdi=3
)

element = MembraneElement(
    manufacturer='Example',
    model='BW30-400',
    area_m2=37.2,
    permeability_lmh_bar=3.5,
    salt_rejection=0.995,
    max_recovery=0.15,
    min_concentrate_flow_m3_hr=3.6
)

ro_system = ROSystemDesign(feed, element)
design = ro_system.design_system(
    feed_flow_m3_hr=100,
    target_recovery=0.75,
    feed_pressure_bar=15
)

print(f"Design flux: {design['design_flux_lmh']:.1f} LMH")
print(f"Number of vessels: {design['num_vessels']}")
print(f"Total elements: {design['total_elements']}")
print(f"Permeate TDS: {design['permeate_tds_mg_l']:.0f} mg/L")
print(f"Specific energy: {design['specific_energy_kwh_m3']:.2f} kWh/m3")

Scaling Potential Analysis

class ScalingAnalysis:
    """Membrane scaling potential analysis"""

    # Solubility product constants at 25C
    Ksp = {
        'CaCO3': 3.3e-9,
        'CaSO4': 4.9e-5,
        'BaSO4': 1.1e-10,
        'SrSO4': 3.4e-7,
        'SiO2': 120  # mg/L saturation
    }

    def __init__(self, water_quality: Dict):
        self.wq = water_quality

    def langelier_saturation_index(self, temperature_c: float,
                                   tds_mg_l: float) -> float:
        """Calculate Langelier Saturation Index for CaCO3"""
        pH = self.wq.get('pH', 7.5)
        Ca = self.wq.get('Ca_mg_l', 100)
        alkalinity = self.wq.get('alkalinity_mg_l', 100)

        # pHs calculation (simplified)
        pCa = -np.log10(Ca / 40080)  # Convert to mol/L
        pAlk = -np.log10(alkalinity / 50040)

        # Temperature and TDS corrections
        A = (np.log10(tds_mg_l) - 1) / 10
        B = -13.12 * np.log10(temperature_c + 273) + 34.55
        C = np.log10(Ca / 40.08) - 0.4
        D = np.log10(alkalinity / 50.04)

        pHs = (9.3 + A + B) - C - D
        lsi = pH - pHs

        return lsi

    def stiff_davis_index(self, ionic_strength: float) -> float:
        """Calculate Stiff-Davis Stability Index for high TDS waters"""
        pH = self.wq.get('pH', 7.5)
        Ca = self.wq.get('Ca_mg_l', 100)
        alkalinity = self.wq.get('alkalinity_mg_l', 100)

        # Activity coefficient correction
        K = 2.22e-14  # CaCO3 equilibrium constant

        pCa = -np.log10(Ca / 40080 * 0.4)  # With activity correction
        pAlk = -np.log10(alkalinity / 50040 * 0.4)
        pK = -np.log10(K)

        pHs = pK + pCa + pAlk
        sdi = pH - pHs

        return sdi

    def calcium_sulfate_saturation(self, recovery: float) -> float:
        """Calculate CaSO4 saturation ratio at given recovery"""
        Ca = self.wq.get('Ca_mg_l', 100)
        SO4 = self.wq.get('SO4_mg_l', 200)

        # Concentration factor
        cf = 1 / (1 - recovery)

        # Ion product in concentrate
        Ca_conc = (Ca / 40080) * cf  # mol/L
        SO4_conc = (SO4 / 96060) * cf  # mol/L
        ip = Ca_conc * SO4_conc

        # Saturation ratio
        sr = ip / self.Ksp['CaSO4']

        return sr

    def silica_saturation(self, recovery: float, temperature_c: float) -> float:
        """Calculate silica saturation ratio"""
        SiO2 = self.wq.get('SiO2_mg_l', 20)

        # Concentration factor
        cf = 1 / (1 - recovery)
        SiO2_conc = SiO2 * cf

        # Temperature-dependent saturation (simplified)
        saturation_limit = self.Ksp['SiO2'] + (temperature_c - 25) * 2

        return SiO2_conc / saturation_limit

    def analyze_scaling_potential(self, recovery: float,
                                  temperature_c: float = 25) -> Dict:
        """Complete scaling potential analysis"""
        tds_concentrate = self.wq.get('tds_mg_l', 1000) / (1 - recovery)

        results = {
            'recovery': recovery,
            'concentration_factor': 1 / (1 - recovery),
            'concentrate_tds_mg_l': tds_concentrate,
            'lsi': self.langelier_saturation_index(temperature_c, tds_concentrate),
            'caso4_saturation': self.calcium_sulfate_saturation(recovery),
            'silica_saturation': self.silica_saturation(recovery, temperature_c)
        }

        # Risk assessment
        results['caco3_risk'] = 'HIGH' if results['lsi'] > 0.5 else \
                               'MODERATE' if results['lsi'] > 0 else 'LOW'
        results['caso4_risk'] = 'HIGH' if results['caso4_saturation'] > 0.8 else \
                               'MODERATE' if results['caso4_saturation'] > 0.5 else 'LOW'
        results['silica_risk'] = 'HIGH' if results['silica_saturation'] > 0.8 else \
                                'MODERATE' if results['silica_saturation'] > 0.5 else 'LOW'

        return results

# Example usage
water_quality = {
    'pH': 7.8,
    'tds_mg_l': 2000,
    'Ca_mg_l': 150,
    'SO4_mg_l': 300,
    'alkalinity_mg_l': 180,
    'SiO2_mg_l': 25
}

scaling = ScalingAnalysis(water_quality)
results = scaling.analyze_scaling_potential(recovery=0.75, temperature_c=25)

print(f"Concentration factor: {results['concentration_factor']:.2f}x")
print(f"LSI: {results['lsi']:.2f} - CaCO3 risk: {results['caco3_risk']}")
print(f"CaSO4 saturation: {results['caso4_saturation']:.2f} - Risk: {results['caso4_risk']}")
print(f"Silica saturation: {results['silica_saturation']:.2f} - Risk: {results['silica_risk']}")

CIP Protocol Development

class CIPProtocol:
    """Clean-in-Place protocol development"""

    def __init__(self, membrane_type: str = 'polyamide'):
        self.membrane_type = membrane_type

        # Chemical compatibility
        self.ph_limits = {
            'polyamide': (2, 11),
            'cellulose_acetate': (4, 7),
            'polysulfone': (1, 13)
        }

        self.temperature_limit = {
            'polyamide': 45,
            'cellulose_acetate': 35,
            'polysulfone': 50
        }

    def recommend_cleaning_chemicals(self, fouling_type: str) -> List[Dict]:
        """Recommend cleaning chemicals based on fouling type"""
        recommendations = {
            'biofouling': [
                {'chemical': 'NaOH', 'concentration': '0.1%', 'ph': 12, 'temperature_c': 35},
                {'chemical': 'Biocide', 'concentration': 'Per manufacturer', 'ph': 7, 'temperature_c': 25}
            ],
            'organic': [
                {'chemical': 'NaOH', 'concentration': '0.1%', 'ph': 12, 'temperature_c': 35},
                {'chemical': 'Surfactant', 'concentration': '0.1%', 'ph': 10, 'temperature_c': 30}
            ],
            'colloidal': [
                {'chemical': 'NaOH', 'concentration': '0.1%', 'ph': 11, 'temperature_c': 30},
                {'chemical': 'EDTA', 'concentration': '1%', 'ph': 10, 'temperature_c': 30}
            ],
            'scale_calcium': [
                {'chemical': 'HCl', 'concentration': '0.2%', 'ph': 2, 'temperature_c': 25},
                {'chemical': 'Citric acid', 'concentration': '2%', 'ph': 3, 'temperature_c': 30}
            ],
            'scale_silica': [
                {'chemical': 'NaOH', 'concentration': '0.1%', 'ph': 11, 'temperature_c': 35}
            ]
        }

        return recommendations.get(fouling_type, [])

    def generate_cip_procedure(self, fouling_types: List[str],
                              system_volume_m3: float) -> Dict:
        """Generate complete CIP procedure"""
        ph_min, ph_max = self.ph_limits.get(self.membrane_type, (2, 12))
        temp_max = self.temperature_limit.get(self.membrane_type, 40)

        procedure = {
            'membrane_type': self.membrane_type,
            'ph_operating_range': f'{ph_min} - {ph_max}',
            'max_temperature_c': temp_max,
            'system_volume_m3': system_volume_m3,
            'steps': []
        }

        # Initial flush
        procedure['steps'].append({
            'step': 1,
            'action': 'Low pressure flush',
            'duration_min': 10,
            'flow_rate': 'High',
            'description': 'Flush with permeate water to remove loose deposits'
        })

        # Cleaning steps based on fouling types
        step_num = 2
        for fouling in fouling_types:
            chemicals = self.recommend_cleaning_chemicals(fouling)
            for chem in chemicals:
                if chem['temperature_c'] <= temp_max and ph_min <= chem['ph'] <= ph_max:
                    procedure['steps'].append({
                        'step': step_num,
                        'action': f'Chemical clean - {chem["chemical"]}',
                        'concentration': chem['concentration'],
                        'ph': chem['ph'],
                        'temperature_c': chem['temperature_c'],
                        'duration_min': 30,
                        'recirculation_time_min': 60,
                        'soak_time_min': 30,
                        'description': f'Target: {fouling} fouling removal'
                    })
                    step_num += 1

        # Final flush
        procedure['steps'].append({
            'step': step_num,
            'action': 'Final flush',
            'duration_min': 20,
            'description': 'Flush until pH and conductivity match feed water'
        })

        return procedure

# Example usage
cip = CIPProtocol(membrane_type='polyamide')
procedure = cip.generate_cip_procedure(
    fouling_types=['biofouling', 'scale_calcium'],
    system_volume_m3=5
)

print("CIP Procedure:")
for step in procedure['steps']:
    print(f"  Step {step['step']}: {step['action']}")

Usage Guidelines

When to Use This Skill

  • Membrane system design and specification
  • Pilot testing protocol development
  • Fouling diagnosis and mitigation
  • System optimization and troubleshooting
  • Pretreatment system design

Best Practices

  1. Characterize feed water thoroughly before design
  2. Use pilot testing for critical applications
  3. Design for maintainability including CIP access
  4. Monitor membrane performance trends regularly
  5. Maintain chemical dosing records for troubleshooting
  6. Plan for concentrate management early in design

Process Integration

  • WW-003: Membrane Treatment System Design (all phases)
  • WW-005: Water Reuse System Implementation (membrane phases)

Dependencies

  • numpy: Numerical calculations
  • scipy: Optimization routines

References

  • AWWA M46 "Reverse Osmosis and Nanofiltration"
  • Dow Water & Process Solutions Technical Manual
  • Hydranautics Technical Application Bulletins