Multi-variable control loop assessment

A method of identifying at least one new disturbance variable in a process loop represented by a disturbance model using cross-correlation of the loop or disturbance model residue and signals not used by the disturbance model to improve the disturbance model and the minimum achievable variation by eliminating the influence of the said at least one disturbance variable comprising:

• applying an orthogonal decomposition method to find orthogonal latent variables that correlate with said loop model residue and which account for successively decreasing variations in said loop model residue, and associating respective weightings between said plurality of known disturbances and each of said latent variables;
• calculating a disturbance model residue which is orthogonal to said latent variables and represents primarily loop variation from said unknown disturbances by subtracting said successive orthogonal latent variables from said loop model residue;
• performing time series analysis on said disturbance model residual error using said model loop delay to decompose said disturbance model residue into a controllable component and an uncontrollable component;
• calculating a minimum achievable loop variation based on selective removal of respective ones of said latent variables by calculating the sum of uncontrollable components of the disturbance model residue and the remaining latent variables;
• calculating a loop performance index as a ratio of standard deviation in said measured loop output to said calculated minimum achievable loop variation;
• and
• performing loop diagnostics in the event said loop performance index exceeds a predetermined amount.

Apparatus for assessing a process having at least one measured loop input, one measured loop output and a plurality of known and unknown disturbances, comprising:

• a loop modeling block for generating a predicted model of said process using parameter estimation, including predicting model loop delay;
• a model residue block for calculating a loop model residue between said loop measured output and said predicted model output;
• a disturbance modeling block for applying an orthogonal decomposition method to find orthogonal latent variables that correlate with said loop model residue and which account for successively decreasing variations in said loop model residue, and associating respective weightings between said plurality of known disturbances and each of said latent variables;
• a model residue block for calculating a disturbance model residue which is orthogonal to said latent variables and represents primarily loop variation from said unknown disturbances by subtracting said successive orthogonal latent variables from said loop model residue;
• a time series analysis block for performing time series analysis on said disturbance model residual error using said model loop delay to decompose said disturbance model residue into a controllable component and an uncontrollable component;
• a block for calculating a minimum achievable loop variation based on selective removal of respective ones of said latent variables by calculating the sum of uncontrollable components of the disturbance model residue and the remaining latent variables;
• and
• a further block for calculating a loop performance index as a ratio of standard deviation in said measured loop output to said calculated minimum achievable loop variation and in the event said loop performance index exceeds a predetermined amount then performing loop diagnostics.

Apparatus for identifying at least one new disturbance variable in a process loop represented by a disturbance model using cross-correlation of the loop or disturbance model residue and signals not used by the disturbance model to improve the disturbance model and the minimum achievable variation by eliminating the influence of the said at least one disturbance variable comprising:

• a disturbance modeling block for applying an orthogonal decomposition method to find orthogonal latent variables that correlate with said loop model residue and which account for successively decreasing variations in said loop model residue, and associating respective weightings between said plurality of known disturbances and each of said latent variables;
• a model residue block for calculating a disturbance model residue which is orthogonal to said latent variables and represents primarily loop variation from said unknown disturbances by subtracting said successive orthogonal latent variables from said loop model residue;
• a time series analysis block for performing a time series analysis on said disturbance model residual error using said model loop delay to decompose said disturbance model residue into a controllable component and an uncontrollable component;
• a block for calculating a minimum achievable loop variation based on selective removal of respective ones of said latent variables by calculating the sum of uncontrollable components of the disturbance model residue and the remaining latent variables;
• and
• a further block for calculating a loop performance index as a ratio of standard deviation in said measured loop output to said calculated minimum achievable loop variation and in the event said loop performance index exceeds a predetermined amount then performing loop diagnostics.