Cement Manufacturing
Grinding and Milling
Evolutionary Modelling and Optimisation of Grinding Processes
Modelling a grinding circuit and using search algorithms to find operating points a person would not try.
- Duration
- 5 days
- Contact hours
- 30
- Level
- Intermediate
- Fee from
- On enquiry
Overview
Grinding circuits have many interacting variables and a cost surface with more than one minimum, which is exactly the problem evolutionary algorithms are good at. This week covers population models of breakage and classification, fitting them to plant data, and then the optimisation itself: defining an objective that reflects what the plant actually wants, encoding constraints honestly, and running genetic and related algorithms. The validation session is the important one, because a model that fits history and fails forward is worse than no model.
Who should attend
Process engineers with a quantitative background, modelling and simulation specialists, advanced process control engineers, and technical services staff.
What you will be able to do
- Build a population balance model of a grinding circuit.
- Fit model parameters to plant survey data.
- Define an objective function reflecting real plant economics.
- Apply evolutionary optimisation with constraints encoded.
- Validate a model against data it was not fitted to.
How it is taught
Short input, then work, across 5 days of roughly 6 teaching hours each. Exercises run on situations delegates bring rather than on a case study written for a different organisation. Cohorts are capped so that everybody is asked something.
Available as classroom, live online, in house.
Programme outline
Day 1
Modelling
- Population balance and breakage functions
- Classification models
- Circuit assembly and solution
Day 2
Fitting
- Survey data requirements
- Parameter estimation
- Over fitting, and recognising it
Day 3
The objective
- Energy, throughput and quality as competing aims
- Multi objective formulation
- Constraints that are real rather than assumed
Day 4
Optimisation
- Genetic and evolutionary algorithms
- Convergence, diversity and local minima
- Interpreting a solution set
Day 5
Validation
- Holdout data and forward testing
- Plant trials of a model recommendation
- When to stop trusting the model
What you leave with
- A fitted circuit model.
- An optimisation run with results interpreted.
- A validation report, including where the model fails.
Dates and fees
| Dates | Location | Format | Fee |
|---|---|---|---|
| 23 to 27 November 2026 | Cape Town, South Africa | Classroom | R15,000 |
| 30 November 2026 to 4 December 2026 | Nairobi, Kenya | Classroom | US$1,500 |
| 7 to 11 December 2026 | Dar es Salaam, Tanzania | Classroom | US$1,500 |
| 14 to 18 December 2026 | Dubai, United Arab Emirates | Classroom | US$1,500 |
Fees are per delegate and cover tuition, all programme materials and the certificate. Travel, accommodation and visas are arranged by the delegate. Registration closes two weeks before a programme starts.
Certificate
A GATI certificate recording what was attended, issued after the programme has run and attendance has been taken. Attending at least 80 per cent of the days earns a certificate of completion; attending less earns a certificate of attendance stating exactly what was attended. Every certificate carries a code that anybody can check against GATI's own record.
It records participation. and GATI does not claim one it does not hold.