Skip to content

Latest commit

 

History

History
55 lines (47 loc) · 3.05 KB

File metadata and controls

55 lines (47 loc) · 3.05 KB

Project Roadmap - BioProcess-Twin Hub

Table of Contents

  1. Phase 0: Foundation & Documentation
  2. Phase 1: Mechanistic Engine (ALBATwin)
  3. Phase 2: Synthetic Data Generation
  4. Phase 3: Hybrid Modeling (PINN)
  5. Phase 4: User-Ready Product

Phase 0: Foundation & Documentation

Goal: Establish the theoretical and architectural groundwork.

  • Requirement Analysis: Define PRD and User Stories.
  • Architecture Design: Modular system design (Architecture.md).
  • Mathematical Specification: Transcribe ALBA model equations (Math_Model.md).
  • Environment Setup: Initialize uv project, pyproject.toml, and git repository.

Phase 1: Mechanistic Engine (ALBATwin)

Goal: Build a robust Python simulator that reproduces the results of Casagli et al. (2021).

  • Core Implementation:
    • Implement ReactorConfig and StateVector.
    • Implement BioKinetics (Stoichiometry + Rates).
    • Implement HydroChemistry (pH solver).
  • Solver Integration: Connect components with scipy.integrate.
  • Testing (TDD):
    • Unit tests for individual kinetic functions.
    • Mass balance verification tests (Critical).
  • Validation: Compare simulation outputs with paper benchmarks (Figure 2 & 3 of Casagli 2021).

Phase 2: Synthetic Data Generation

Goal: Use the validated engine to create a "Ground Truth" dataset for AI training.

  • Scenario Definition: Define ranges for Light, Temperature, and Influent Loads.
  • Batch Simulation: Run parallel simulations covering the operational space.
  • Data Processing: Clean, normalize, and structure the output into training tensors.
  • Noise Injection: Add Gaussian noise to simulate sensor inaccuracy (optional, for robustness testing).

Phase 3: Hybrid Modeling (PINN)

Goal: Develop a Physics-Informed Neural Network that learns from data while respecting physical laws.

  • Network Architecture: Design the neural network (Input: Env/State -> Output: dX/dt or Next State).
  • Loss Function Engineering:
    • $L_{total} = L_{data} + \lambda \cdot L_{physics}$
    • Implement $L_{physics}$ using the residual of the Phase 1 differential equations.
  • Training: Train the model using the synthetic dataset.
  • Evaluation: Compare PINN performance vs. Pure Data-Driven models vs. Mechanistic Model.

Phase 4: User-Ready Product

Goal: Deploy the tool for end-users (researchers/operators).

  • Dashboard: Build a Streamlit app for visualization and interaction.
  • Dockerization: Create a Dockerfile for easy deployment.
  • Documentation: User manual and API reference.
  • Publication: Prepare technical report or paper on the Hybrid Twin implementation.

Last updated: March 10, 2026 by Anibal Rojo