GFIS | Green Fuel Intelligence System

BITS Pilani WILP M.Tech AIML Dissertation PIML Engine Active

Research Context

Mechanistic models like ADM1 or CFD-DEM accurately simulate anaerobic digestion and gasification, but require massive computational resources (days of computing for seconds of simulation). GFIS solves this via a Digital Twin. By utilizing a hybrid XGBoost + LSTM architecture bounded by Physics-Informed loss functions, the digital twin decouples physical time from computational time, granting operators the ability to instantaneously forecast a complex 48-hour operational window.

⏱ Temporal Stride / Accelerated Horizon Control

Adjust the simulation clock. High speeds bypass physical constraints by utilizing the auto-regressive PIML inference engine.

Speed: 1.0x (Real-time)
T+00:00:00
1x (SCADA Sync) 100x 1,000x 10,000x (Full Forecast)

Control Room Scenario Link

Process Variable Coordinator

Values can be imported from the Model Control Room or adjusted here during the 48-hour run. The simulator recalculates VFA/ALK risk, methane quality, yield and warning behavior from these operating conditions.

Back to Control Room
Imported state Standalone simulator mode
Expected VFA pressure Stable tendency
Digital twin meaning Control decisions alter simulated plant behavior.
Simulation Memory 0 records
Time Action State / Effect

Virtual Plant State LIVE

MSW Buffer
48h Inventory Limit
Gasifier
Ambient
NiFeCa Cat. 99.7%
AD Reactor
Bio-Stable
CH4 Yield 0.00

Hybrid Inference Architecture

ADM1 Data Generation
Synthetic bounding of edge cases.
XGBoost + LSTM Stack
[Non-Linear Features] [Temporal Inertia]
Physics Constraints (PIML)
Ltot = LMSE + λ·max(0, Ŷ - Ymax)2
Syngas Ratio
0.00 H2/CO
Optimal: 1.32
VFA/ALK (Soft Sensor)
0.12
Predicts Acidogenesis
Equivalence Ratio
0.35 ER
Air/MSW Base
Methane Quality
0.0 % CH4
LSTM Projection
~/GFIS/inference.log
GFIS Engine standing by...

1. Gasification: Catalyst Coking Curve

Tracking non-linear degradation of Ni45Fe15Ca40 to empirical 83% limit over 48h.

2. AD: Methane Yield & Soft Sensor Warning

AI prediction of CH4 output mapping against VFA/ALK accumulation (collapse boundary > 0.4).