Data

Tools

Indexes

Applications

Experiments

Awesome

An open API service indexing awesome lists of open source software.

https://github.com/pabloagn/flux

An open source, declarative digital twin engine for electrochemical processes, written in Rust.
https://github.com/pabloagn/flux

chemical-engineering declarative-programming digital-twin digital-twins nix nix-flake process-engineering rust thermodynamics

Last synced: 17 days ago
JSON representation

An open source, declarative digital twin engine for electrochemical processes, written in Rust.

Awesome Lists containing this project

README 

          








Industrial-Grade Chlor-Alkali Digital Twin Platform






───────  ⏣  ───────












[![Last Commit](https://img.shields.io/github/last-commit/pabloagn/flux?style=for-the-badge&logo=git&logoColor=white&color=7AA89F&labelColor=000000&label=LAST%20COMMIT)](https://github.com/pabloagn/flux/commits/main) [![License](https://img.shields.io/github/license/pabloagn/flux?style=for-the-badge&color=7AA89F&labelColor=000000)](https://github.com/pabloagn/flux/blob/main/LICENSE)






## Executive Summary



FLUX is a production-ready digital twin platform for chlor-alkali electrolysis plants, processing 1M+ events/second from 180+ electrolytic cells. The system provides real-time plant control, predictive maintenance, and operational optimization while maintaining SIL-3 safety standards.



## Industry Context



The global chlor-alkali market represents an $80+ billion industry, with membrane cell technology dominating 65% of worldwide production capacity. Energy costs constitute 40-50% of total production costs, while membrane replacement represents 30-40% of operational expenses. Current industrial control systems lack the real-time optimization and predictive capabilities needed to address these economic pressures.



## Business Problems Addressed



### Membrane Degradation Management



**Challenge**: Membrane lifetime varies from 2-7 years with replacement costs of $500-1000/m². Current efficiency drops from 96% to 85% over membrane lifetime, with voltage increases of 0.2-0.5V. Unplanned membrane failures cause 5-10 days of downtime, costing millions in lost production.



**FLUX Solution**: Real-time membrane health monitoring using Butler-Volmer kinetics and ML-based degradation models. Predictive maintenance scheduling reduces unplanned downtime by 80% and extends membrane life by 15-20% through optimized operating conditions.



### Brine Quality Impact



**Challenge**: Impurities (Ca²⁺, Mg²⁺, SO₄²⁻) cause membrane fouling, requiring <20 ppb for optimal operation. Poor brine quality reduces current efficiency by 2-5%, directly impacting profitability.



**FLUX Solution**: Temporal correlation between upstream brine quality and downstream cell performance using GlassFlow's streaming joins. Accounts for 2-4 hour residence times, enabling proactive process adjustments before efficiency degradation occurs.



### Current Distribution Optimization



**Challenge**: 5-10% variation in individual cell voltages within electrolyzers causes hotspots that reduce membrane lifetime by up to 50%. Manual redistribution is typically done quarterly, missing optimization opportunities.



**FLUX Solution**: Continuous current distribution monitoring across all 180 cells with automatic rebalancing recommendations. Real-time optimization algorithms minimize cell-to-cell variations to <2%, significantly extending equipment life.



### Energy Cost Management



**Challenge**: Electricity represents 40-50% of production costs with volatile spot pricing. Load flexibility is limited by product demand, and startup/shutdown cycles damage membranes, constraining optimization options.



**FLUX Solution**: Dynamic load optimization integrating spot price forecasts, demand predictions, and equipment constraints. Operates within 85-95% turndown capability while maintaining product quality, achieving 5-10% energy cost reduction.



## Plant Configuration



### Physical Layout



The reference implementation models a world-scale chlor-alkali facility:



- **Capacity**: 1000 MT/day Cl₂ (365,000 MT/year)

- **Configuration**: 3 electrolyzer units × 100 MW each

- **Cell Count**: 180 cells (3 units × 3 stacks × 20 cells)

- **Technology**: Modern membrane cells (DuPont N2030 or equivalent)

- **Integration**: Complete brine treatment and product handling systems



### Operating Parameters



| Parameter | Value | Impact on Digital Twin |

|-----------|-------|------------------------|

| Current Density | 4-6 kA/m² | Determines production rate and efficiency |

| Cell Temperature | 85-90°C | Critical for membrane life and efficiency |

| Pressure Differential | 100-300 mbar | Safety-critical parameter for membrane integrity |

| Brine Concentration | 300 g/L inlet, 200 g/L outlet | Mass balance and efficiency calculations |

| NaOH Product | 32% concentration | Product quality control |



## Value Proposition



### Quantifiable Benefits



- **Efficiency Improvement**: 2-3% increase in current efficiency = $8-12M annual savings

- **Membrane Life Extension**: 15-20% longer life = $3-5M reduced replacement costs

- **Energy Optimization**: 5-10% reduction = $10-20M annual savings

- **Downtime Reduction**: 80% fewer unplanned shutdowns = $15-25M recovered production

- **Total Annual Value**: $36-62M for a 1000 MT/day facility



### Competitive Advantages



- **Latency**: <10ms control loop vs. 100-500ms in traditional DCS

- **Scale**: Handles 1M+ events/sec vs. 10-100K in conventional systems

- **Intelligence**: Embedded ML models vs. rule-based control

- **Safety**: SIL-3 certified with cryptographic command verification

- **Flexibility**: Cloud-native architecture vs. monolithic legacy systems



## System Architecture



### Data Pipeline Architecture



The platform implements a lambda architecture with specialized pipelines optimized for different operational requirements:



#### High-Speed Control Pipeline (Latency: <10ms)

```

Sensors → TSN Network → OPC UA → Kafka → QuestDB → Control TUI

                                    ↓

                             ScyllaDB (State Store)

```



#### Analytics Pipeline (Latency: 1-5s)

```

Sensors → Kafka → GlassFlow → ClickHouse → Analytics Dashboard

                     ↓

               ML Microservices → Predictions

```



#### Audit & Compliance Pipeline

```

All Events → Apache Pulsar (Cryptographic Signing) → TimescaleDB

                    ↓

               Immutable Audit Trail

```



### Core Components



#### Real-Time Data Layer

- **QuestDB**: Ingests 1M+ events/sec with nanosecond precision timestamps

- **ScyllaDB**: Maintains complete plant state with 10μs p99 latency

- **Redis**: Hot cache for control loops requiring <1ms access



#### Historical & Analytics Layer

- **ClickHouse**: Columnar storage for multi-year historical data

- **GlassFlow**: Stream processing for temporal joins and deduplication

- **TimescaleDB**: Regulatory compliance and audit trail



#### Control Systems

- **OPC UA Server**: Industrial protocol interface with TSN support

- **Safety Interlock System**: SIL-3 rated safety logic with 10ms override capability

- **Redundancy**: 3-node etcd cluster for leader election, automatic failover <2s



#### Machine Learning Platform

- **TorchServe**: Membrane degradation and efficiency prediction models

- **NVIDIA Triton**: GPU-accelerated electrochemical calculations

- **MLflow**: Model versioning and experiment tracking

- **Feast**: Real-time feature store for ML pipelines



### Safety Architecture



#### Write-Back Security

- Ed25519 cryptographic signatures on all control commands

- Challenge-response authentication for command authorization

- Dual-path verification through separate validation service

- Hardware watchdog timers on control-enabled instances



#### Failsafe Mechanisms

- Dead-man's switch: Automatic control disengagement on TUI disconnect

- SIS override: Hardware safety system can veto any software command

- Command expiry: Control commands auto-expire after 100ms

- State rollback: Automatic reversion to safe state on anomaly detection



### Data Flow Specifications



#### Message Distribution (1M+ events/sec)

- **High-frequency (600K/sec)**: Voltage, current @ 100Hz per cell

- **Medium-frequency (300K/sec)**: Pressure, electrolyte levels @ 10Hz

- **Low-frequency (100K/sec)**: Temperature, chemical composition @ 1Hz

- **Derived calculations**: KPIs, efficiency metrics, predictions



#### Time Synchronization

- PTP (IEEE 1588) grandmaster with GPS reference

- Sub-microsecond synchronization across all nodes

- Automatic clock drift compensation

- NTP fallback for non-critical systems



## Electrochemical Process Model



### Fundamental Reactions



The chlor-alkali process operates via the following half-reactions:



**Anode (Chlorine Evolution)**:

$$2Cl^- → Cl_2 + 2e^- \quad E^0 = 1.36V$$



**Cathode (Hydrogen Evolution)**:

$$2H_2O + 2e^- → H_2 + 2OH^- \quad E^0 = -0.83V$$



**Overall Cell Reaction**:

$$2NaCl + 2H_2O → Cl_2 + H_2 + 2NaOH \quad E^0_{cell} = 2.19V$$



### Production Calculations



Based on Faraday's laws of electrolysis:



$$m = \frac{M \cdot I \cdot t \cdot \eta}{n \cdot F}$$



Where:

- m = mass of product (kg)

- M = molar mass (kg/mol): Cl₂ = 0.0709, H₂ = 0.00202, NaOH = 0.040

- I = current (A)

- t = time (s)

- η = current efficiency (0.94-0.96)

- n = electrons transferred (2 for Cl₂ and H₂)

- F = Faraday constant (96,485 C/mol)



### Cell Voltage Model



$$V_{cell} = E^0 + \eta_{act} + \eta_{ohm} + \eta_{conc}$$



Where:

- E^0 = 2.19V (thermodynamic potential at 85°C)

- η_act = Activation overpotential (Butler-Volmer kinetics)

- η_ohm = Ohmic losses (membrane + electrolyte resistance)

- η_conc = Concentration overpotential



$$\eta_{act} = \frac{RT}{αnF} \ln\left(\frac{i}{i_0}\right)$$



### Energy Efficiency Metrics



**Specific Energy Consumption**:

$$SEC = \frac{\int V(t) \cdot I(t) \, dt}{m_{Cl_2}} \quad \text{[kWh/MT]}$$



Target: <2,450 kWh/MT Cl₂



**Current Efficiency**:

$$CE = \frac{m_{actual}}{m_{theoretical}} \times 100\%$$



Target: >94%



## Performance Specifications



### System Requirements

- **Throughput**: 1M+ events/second sustained

- **Control Latency**: <10ms sensor-to-actuator

- **Analytics Latency**: <5s for complex queries

- **Data Retention**: 24hr hot (QuestDB), 10yr cold (ClickHouse)

- **Availability**: 99.99% uptime (52 minutes downtime/year)



### Hardware Requirements



#### Minimum Production Deployment

- **Control Nodes**: 3x servers with TSN NICs, PTP support

- **Database Cluster**: 5x nodes (QuestDB: 2, ClickHouse: 3)

- **ML Compute**: 2x GPU nodes (NVIDIA A100 or equivalent)

- **State Store**: 3x NVMe nodes for ScyllaDB

- **Memory**: 512GB RAM total across cluster

- **Storage**: 100TB SSD for hot data, 1PB HDD for cold



### Network Architecture

- **Control Network**: TSN-enabled switches, redundant paths

- **Data Network**: 100Gbps backbone, RDMA support

- **Synchronization**: Dedicated PTP VLAN

- **Security**: Air-gapped control network, DMZ for external access



## Deployment



### Prerequisites

```bash

# System requirements

- Kubernetes 1.28+ with GPU operator

- Nix with flakes enabled

- Docker 24.0+

- etcd 3.5+

- PTP-enabled network hardware

```



### Quick Start

```bash

# Initialize Nix environment

nix develop



# Deploy infrastructure

flux-setup                          # Create directory structure

flux-deploy-core                    # Deploy Kafka, QuestDB, ClickHouse

flux-deploy-ml                       # Deploy ML platform

flux-deploy-control                  # Deploy OPC UA, SIS interface



# Start simulation

nix develop .#simulator

python -m flux.simulation --mode=closed-loop --cells=180



# Launch control TUI (requires authorization)

export FLUX_CONTROL_KEY=/path/to/key.pem

nix develop .#tui

cargo run --release --features=control



# Verify pipelines

flux-health-check --all

```



### Production Deployment

```bash

# Deploy with Terraform

cd infra/terraform

terraform init

terraform plan -var-file=production.tfvars

terraform apply



# Initialize Kubernetes cluster

kubectl apply -f infra/kubernetes/namespaces.yaml

kubectl apply -f infra/kubernetes/crds/

helm install flux ./charts/flux -f values.production.yaml



# Configure PTP

ptp4l -i eth0 -m -H -s /etc/ptp4l.conf

phc2sys -s eth0 -c CLOCK_REALTIME -w -m

```



## Development Status



### Completed

- Process simulation engine with electrochemical models

- Kafka topic architecture for 1M+ events/sec

- Basic TUI with real-time visualization

- Nix flake configuration



### In Progress

- [ ] QuestDB pipeline implementation

- [ ] ScyllaDB state management

- [ ] OPC UA server integration

- [ ] ML microservices deployment

- [ ] SIS interface implementation

- [ ] Cryptographic command signing



### Planned

- [ ] Full closed-loop control implementation

- [ ] Advanced predictive maintenance models

- [ ] Energy optimization with grid integration

- [ ] Multi-plant coordination

- [ ] Mobile operator interface

- [ ] AR/VR visualization



## Safety & Compliance



### Standards Compliance

- **IEC 61511**: Functional safety for process industries

- **ISA-84**: Safety instrumented systems

- **IEC 62443**: Industrial network security

- **ISO 27001**: Information security management

- **NIST 800-82**: Industrial control system security



### Safety Integrity Level

- Control loops: SIL-2 certified

- Emergency shutdown: SIL-3 certified

- Alarm management: ISA 18.2 compliant



## Testing



### Simulation Modes

```bash

# Open-loop simulation (sensor data only)

python -m flux.simulation --mode=open-loop



# Closed-loop simulation (with control responses)

python -m flux.simulation --mode=closed-loop



# Fault injection testing

python -m flux.simulation --mode=fault --scenario=membrane-rupture



# Load testing (verify 1M events/sec)

flux-load-test --rate=1000000 --duration=3600

```



### Verification

```bash

# Verify data flow

flux-trace --event-id= --show-path



# Check latencies

flux-latency-monitor --percentile=99



# Validate calculations

flux-verify-calculations --reference=aspen-plus

```



## Contributing



This is a reference architecture for industrial control systems. Contributions must maintain safety-critical standards and include comprehensive testing.



## License



Proprietary - Industrial Reference Architecture



## Support



For production deployments, [contact me](https://pabloagn.com/contact).