Simulation Meets
Reality

Doxazo Consulting closes the gap between process simulators and real plant performance. We pair rigorous chemical process simulation baselines with residual machine learning to deliver yield predictions that are physically grounded, validated against real plant data, and fully explainable.

Multiple
Models Benchmarked
Chemical
Process
Simulators
SHAP
Explainable Corrections
Validated
Performance Metrics
How It Works

The Correction Framework

01
Simulation Baseline

We build and validate a chemical process simulation across your plant's operating range, giving every corrected prediction a thermodynamically rigorous physics foundation.

Process Simulators Sensitivity Analysis
02
Plant Data Integration

Real operating data from your facility is sourced, cleaned and merged with the simulated outputs so every operating condition has a matched pair of simulated and observed yield.

Data Integration Plant Records Pandas
03
Residual Analysis

We compute and characterise the systematic gap between the simulated yields and real plant yields, confirming that the deviation is condition dependent and worth correcting before any model is trained.

Residual Error Diagnostics Validation
04
Residual Model Training

Traditional ML models such as Linear Regression, Gradient Boosting and XGBoost are trained and benchmarked against the residual, with cross validation and hyperparameter tuning selecting the strongest correction model.

XGBoost Gradient Boosting GridSearchCV
05
SHAP Interpretability

SHAP analysis is applied to the best model to attribute the correction to specific operating conditions, so your engineers can see which variables the simulator underrepresents.

SHAP Explainability Feature Attribution
06
Corrected Yield Deployment

The physics baseline and the residual correction are combined into a single corrected yield prediction, benchmarked against the standalone simulator on RMSE, MAE and R squared.

Hybrid Modelling Model Benchmarking Deployment
Applications

Where The Framework Applies

Residual ML · Case Study
HDPE Pyrolysis to Liquid Fuel
High density polyethylene (HDPE) pyrolysis yields are corrected against real plant data, addressing the systematic gap left by lumped reaction kinetics in Aspen HYSYS.
Simulator Correction · Refining
Catalytic Cracking Units
Catalyst deactivation and nonlinear loading effects are rarely captured in steady state simulators. Residual correction recovers the accuracy lost to these simplifications.
Thermochemical Conversion
Biomass Pyrolysis Systems
Feedstock variability and multi stage cracking reactions make biomass pyrolysis a strong candidate for the same physics baseline plus residual correction structure.
Refining · Hydroprocessing
Hydrocracking and Reforming Units
Complex, multi stage reaction networks in hydroprocessing units create the same simulator to plant gap that the residual framework is built to close.
Waste Valorisation
Waste to Energy Conversion
Post consumer feedstock introduces contaminants and moisture that steady state simulators cannot represent. Residual correction adapts predictions to real feedstock behaviour.
Any Simulated Process
Your Process, If a Simulator Exists
The framework needs only a validated simulator output, real plant data and a set of operating condition features, so it extends to any chemical process with a known simulation gap.
Who We Are

About the Framework

We sit at the intersection of engineering domain knowledge and modern artificial intelligence, a combination that is uncommon and genuinely valuable when correcting simulator predictions against real plant behaviour.

We specialise in one thing: closing the gap between what a process simulator predicts and what a plant actually produces. The chemical process simulator provides the physics baseline. A residual machine learning model captures the deviations the simulator cannot represent. Every corrected prediction comes with a clear, SHAP driven explanation of the operating conditions behind it.

We take a rigorous, evidence based approach to every engagement. Our simulations are validated, our models are cross validated, and our corrections are benchmarked against the standalone simulator before they are ever handed to your team.

Chemical Process Simulation
Residual Machine Learning
XGBoost & Gradient Boosting
Linear Regression
SHAP Explainability
Hybrid Physics-Informed Modelling
Sensitivity Analysis
GridSearchCV Hyperparameter Tuning
Cross Validation
Chemical Process Engineering
Predictive Modelling
Model Benchmarking
Get In Touch

Let Us Work Together

If your process simulator is drifting from real plant performance, we would like to hear about it. Whether you need a new chemical process simulation baseline, a residual correction model or a full physics informed framework, send us a message and we will respond promptly.