Quote of the day

A person who never made a mistake never tried anything new.

- Albert Einstein

Contact

Hybrid Physics-Data Models for Energy Efficiency Optimization in Heavy Process Industries

Open Industry & Manufacturing

Hybrid Physics-Data Models for Energy Efficiency Optimization in Heavy Process Industries

Hybrid Physics-Data Models for Energy Efficiency Optimization in Heavy Process Industries

Description

Reduce the specific energy consumption of energy-intensive industrial processes by combining physical conservation laws with data-driven learning and constraint-aware operating recommendations.

Details

Context and Problem Statement

Heavy process industries are energy intensive and often rely on expert-derived operating setpoints. Purely data-driven models may extrapolate poorly and violate physical consistency, while first-principles models may not capture fouling, degradation, or feedstock variability.

Research Question

Does a hybrid model constrained by mass and energy balances generalize better outside observed operating conditions than a purely data-driven model, and can it support safe setpoint recommendations?

Proposed Approach

Enforce conservation equations in a hybrid process model, learn residual dynamics from data, optimize setpoints under quality and safety constraints, and use sensitivity analysis to identify robust energy-saving levers.

Expected Contribution

A quantitative demonstration of the extrapolation benefit of physical constraints and an operational setpoint-optimization framework.

Expected Prototype

An operator advisory tool displaying expected energy savings, active constraints, and predictive uncertainty.

Datasets

Historical plant supervisory data, Tennessee Eastman, and laboratory quality analyses.

Challenges

Data synchronization, sensor drift, limited ability to experiment on live plants, and energy-quality trade-offs.

Research Question

Do physical conservation constraints improve out-of-domain generalization sufficiently to support reliable industrial process setpoint recommendations?

Innovation

The project evaluates out-of-domain generalization explicitly rather than optimizing only historical fit.

Expected Deliverable

A hybrid model validated on historical data, a constrained setpoint-optimization module, and a quantified energy-saving assessment.

Technologies

Physics-Informed Machine Learning System Identification and Process Balances Constrained Optimization Sensitivity Analysis Industrial Supervisory Systems

Required Skills

  • Process Engineering and Thermodynamics
  • Machine Learning and Optimization
  • Industrial Data Processing
  • Collaboration with Plant Operations Teams

Datasets

  • Historical process data from an industrial partner
  • Tennessee Eastman Process benchmark
  • Laboratory product-quality analyses

Morocco & Africa Relevance

Process industries are major energy consumers in Morocco, especially phosphate chemistry and construction materials, so even small efficiency gains can have substantial impact.