Computers today:
- One CPU can run more than 100s pipelines
- Fast SSD, terabytes of DDR4 RAM
- Size of a tablet or phone, <$1000
MPC software yesterday:
- Written in 1980-90s in Fortran IV/77
- Embarrassingly serial, single threaded
- Compact computations by minimal use of features
- VAX/VMS(16 MHz/16 MB) computer of the 90s was used, today a Raspberry Pi is 100 times powerful
- Linear approximations used with use of SQP
- No state estimation, or MV setpoints
- Simple single variable first order filter on outputs for plant-model mismatch
- Never been updated for any innovative Process Control feature or novel mathematics
- Tuning has been a 'business opportunity'
- Database for MPC is not transactional nor relational, but RDBMS infrastructure is used
- GUI and DB developers are not MPC implementers, hence huge software development cycle and cost
MPC of the Today:
- Use the future computer hardware (teraFLOPS)
- No approximation or compact computation
- Develop and use parallel mathematics in MHEMPC
- Combine all possible features required by an MPC, like flexible choice of objective function, nonlinear constraints, better state estimation and modeling
- Replacement of current ailing MPC installations can be justified by extra annualized profits by MHEMPC, it's maintenance shall be free
- Encrypted operation historical data passed over internet for tuning and monitoring of MHEMPC
A small example:
- A nonlinear CSTR having a simple 1st order endothermic reaction simulated
- Interface simulating an OPC server
- Data is collected by client controller, computes inputs(PID), and sends back to plant
- Choice of Openloop operation
- User interface, process trends and how to use the demo. Demonstrates expertise
