Method for Detecting Faults in an Industrial Process or Part thereof or for Controlling and / or Optimizing an Industrial Process or Part thereof

摘要

The invention relates, for example to a method for detecting faults in a flow net- work controlled by valves or in a part of it, wherein the flow network is nonlinear in regard to at least one process variable, wherein the method consists of at least following steps: - the structure of the flow network or a part of it is identified by a model equation or a set of model equations comprising linear functions of the process variables or variables derived therefrom, wherein some or all or none of the coefficients of the said linear functions can be fixed, or/and nonlinear functions, wherein said nonlinear functions may involve up to three process variables or variables derived therefrom, which are inputs of said nonlinear functions, wherein said non- linear functions may be involved to more than one model equation and wherein the nonlinear functions involved in model equation(s) are defined as weighted sums of products of basis functions of the inputs of the said nonlinear functions, wherein said weights of products are the coefficients defining the nonlinear functions, wherein a set of basis functions is selected for each input of said nonlinear function(s) before the identification of model equation(s), - imposing the conditions on some or all or none of the nonlinear functions involved in said model equation or set of model equations to monotonously in- crease or decrease with respect to some or all inputs of said nonlinear functions, and said conditions are converted to the constraints imposed on said coefficients defining the nonlinear functions, - imposing the conditions on some or all or none of the signs of the coefficients of the linear functions involved to said model equation(s) - collecting measured values for all process variables or variables derived there- from involved to the model equation(s) for achieving a set of data samples, wherein each data sample is a collection of values of all said process variables or variables derived therefrom at a fixed moment, - identifying the coefficients of linear functions of said model equation or said set of model equations and said coefficients defining the nonlinear functions, by minimizing the errors of the model equation(s) at every data sample while taking into account the flattening of said model equations(s), whereby the errors to be minimized in said model equation(s) are computed by substituting the values of the process variables or variables derived therefrom to the equation(s), wherein the flattening of equations is achieved by minimizing the sum of total curvatures of the said nonlinear functions taken with nonnegative weights, wherein said minimization of said errors of model equations together with said sum of total curvatures is performed under the said constraints imposed both on the coefficients defining said nonlinear functions which represent the monotonousness conditions for the said nonlinear functions with respect to their inputs and on the signs of the coefficients of the linear functions involved to said model equation(s), - selection of the fault detection thresholds for some or all said identified equation(s), - computing the errors of said identified model equations by substituting the cur- rent values of the process variables or variables derived therefrom to the said equation(s), - evaluating said errors of the identified model equations against the respective fault detection thresholds, and detecting faults in the flow network or in a part of it.