Introduction

What is Metacontrol ?

The primary objective of the Metacontrol methodology is to facilitate the implementation of the Self-Optimizing Control (SOC) concept in industrial processes via software analysis, in a comprehensive user interface.

The SOC concept is used to guide a decision on how to design the control structure of a given process.

By definition from the main author of the methodology, Dr. Sigurd Skogestad from NTNU:

“Self-optimizing control is when one can achieve an acceptable loss with constant setpoint values for the controlled variables without the need to re-optimize when disturbances occur.”

Fig. 1 Self-Optimizing Control fundamental concept: The pursue of a control structure based on constant setpoint policy, capable of minimizing the loss to an acceptable magnitude, when compared with the reoptimized process every time that a disturbance occur (Real Time Optimization).

The self-optimizing control structure selection problem has a combinatorial nature: Generally in an industrial process, there are dozens (even thousands!) of variables, and a set of available degrees of freedom that can be consumed by a subset of the possible candidate controlled variables. There are mainly two ways to solve this problem:

• “Brute-force” Approach:

  Each possible control structure is evaluated, one at a time. Depending on the number of the degrees of freedom and available measurements, that can take literally, forever.

• Local (linear) methods:

  Based on quadratic approximation of the objective function using Taylor series expansions, Local methods were developed by Dr. Sigurd Skogestad and his collaborators to quickly “pre-screen” the most promising subsets of controlled variables. Metacontrol is based on these mathematical formulations, with a neat User Interface.

In order to use the Local methods, it is necessary to obtain high-order data with respect to the process gradients and the objective function hessians. In order to do this, Metacontrol uses powerful machine learning formulations (Kriging Interpolators) to obtain such data in with robustness.

Last but not least, other decisions such as which type of controllers to use or how to tune them is a responsibility of classical control design, whose concepts are not implemented (yet) in this software.

Important

The basic idea behind Metacontrol is to tell you what control structure you should implement, not how you should implement.

Nonetheless, Metacontrol is a congregation of methodologies such as Surrogate modeling via Kriging metamodels, Black-box process optimization and SOC, In order to determine process optimal operating point, high-order data obtainment to use SOC mathematical formulations and consequently, generating SOC-based control structures. All of that within a comprehensive User Interface, allowing the control structure designer/engineer/scientist (hey, that’s you) to keep his/hers focus only on synthesizing the control structure, rather than wasting hours “jumping” between several software environments, such as: Process simulators (Aspen Plus) and numerical packages (MATLAB, Python, Microsoft Excel, etc.). This tool is made by engineers that struggled with this (us), in order to try to solve such struggle for the scientific community.

Installation

Currently, there are two ways to install the software: via binaries or source code.

Prerequisites:

1. Windows OS;

2. AspenTech Aspen Plus installed;

Installing from binaries (.exe)

This is the most straightforward way to install Metacontrol. You just need to download and install the desired version (generally, our most recent and stable version) from the repository releases page.

This is the recommended option for most engineers and scientists that just want to study SOC-Based Control Structure selection and are not interested in programming details. You will only install our software and will be good to go.

Installing from source

Metacontrol as stated before, is fully open-source. Want to inspect or change our code? Just follow the steps below!

Additional prerequisites:

• A conda installation from either Anaconda or Miniconda;

Steps:

1. Create the virtual environment.

   You will need to create a conda environment exclusively for Metacontrol and install the Python interpreter via the following conda prompt command:

   conda create -n your_env_name -c conda-forge python
   

   The argument your_env_name can be changed to whichever valid name you like. We suggest to keep it simple to remember (simply calling metacontrol or mtc will suffice), since you will need it activated whenever you run the Metacontrol application.

2. Then activate the environment via:

   conda activate your_env_name
   

3. Install the optimization package.

   Now you will need to install the Python IpOpt optimization package required. This step is crucial, so we recommend you follow the package installation instructions accordingly.

4. Then you will need to install the base packages via the commands:

   conda install -c conda-forge pywin32 pandas sklearn simplejson matplotlib scipy
   
   pip install py-expression-eval
   
   conda install -c felipes21 pydace surropt pysoc
   

5. Download the Metacontrol source code.

   From the repository and unzip it to your folder of preference.

6. Run the application:

   You will need to have a conda prompt with the environment you created in the previous steps activated. Then you just type the command:

   python path/to/mainwindow.py
   

   The argument path/to/ is just the path to folder you unzipped. Your can either change the current directory via the cd command and running python mainwindow.py, or type the full path as mentioned above.