Cogeneration power plants offer diverse operating challenges because of the variety of units involved, including processes for electricity generation, steam production, and steam and chilled water distribution for both heating and cooling local buildings and heating and cooling unit operations. The plants must meet demands for electrical energy and heating and cooling, while also adhering to constraints imposed by operating costs, environmental emissions and plant safety.

The cogeneration power plant on the Storrs campus of the University of Connecticut (UConn Cogen) has a capacity of 25 MW of electricity and 200,000 lbs/hr of steam. The electric demand of the campus is met with water-cooled electric generators on three 7.5 MW gas turbines and one 5 MW steam turbine. Three heat recovery steam generators (HRSGs) located after the gas turbine generators each contain 600 psi steam systems and 125 psi steam systems. The 600 psi steam is used to run the steam turbine generator, while the 125 psi steam is used for campus heat in the winter and running chillers for campus cooling in the summer.

For additional steam production, gas fired duct burners can be lit in each HRSG and five auxiliary boilers are used as needed. NOx stack emissions are controlled with a vaporized liquid ammonia injection system. To provide cooling, the plant contains three steam driven chillers, two electric chillers, and two gas turbine chillers. Cooling towers on the plant roof provide cooled water for both the generators and the chillers.

The cogeneration facility is capable of producing more electrical energy than the campus uses, but state regulations will not allow excess electricity to be exported onto the power grid. So the challenge of keeping up with changing electrical, heating and cooling demands due to the New England climate and academic cycle is heightened because the plant must operate as efficiently as possible without producing excess electricity.

The UConn Cogen plant contains approximately 150 control loops to regulate process operation. These must be properly designed and tuned to maintain production within the changing operational constraints mentioned. To determine where and how to improve plant operation, controller performance assessment (CPA) tools are being applied. Plant data is continuously being recorded to an OSI Process Information data historian. Using the process variable (PV) and the controller output (CO) signals available during regular operation, CPA benchmarks have been developed to indicate which control loops require attention.

Many benchmarks compare current performance to desired performance, but do not indicate the nature of the controller misbehavior when it has been determined that performance is not as desired [1]. Published methods with diagnostic capabilities for classifying the full range of controller behavior, from sluggish to aggressive, require that the data be tracked for isolated disturbance events [2, 3].

In assessing the performance of the UConn Cogen power plant, a new CPA method was applied that is capable of diagnosing the full range of tuning-related controller behaviors without isolating disturbances. The new method is a novel automated application of the autocorrelation function, ACF, which uses regulatory data and does not require a priori information about the process. The relationship of the ACF to the closed loop transfer function has been established by Box and Jenkins [4] and Harris [5]. Several researchers [1, 6, 7] have noted that a plot of the ACF can be used as a first estimate of controller performance and ACF plots are included as a tool in most CPA software packages.

The work proposed for presentation uses the ACF as an automated process monitoring tool. The characteristic shapes of the ACF are extracted with a “pattern recognition” model and a new performance metric, the Relative Damping Index (RDI), is computed from the model parameters. The RDI provides an easy-to-interpret measure of the nature of the controller behavior from sluggish to aggressive. The new ACF analysis also provides a means for determining how to correct proportional and integral (PI) controller tuning when it is indicated that the controller performance is unacceptable.

The pattern recognition model is a general second order underdamped time domain form. The model provides a period of oscillation,τ_n, and a damping factor, ξ, for classifying the predicted disturbance rejection response of the controller. A method based on an initial estimate of τ_n and confidence intervals is presented as a means for selecting the length of the ACF to use in the least-squares model fit.

A user-defined range of acceptable controller performance based on ξ is combined with the computed ξ to calculate the RDI. RDI = (ξ_act – ξ_agg)*(ξ_slug – ξ_act)^-1, where ξ_act is the damping factor computed from the ACF, ξ_agg is the limit of acceptable aggressive behavior, and ξ_slug is the limit of acceptable sluggish behavior. The RDI is positive when ξ_act is within the user-defined range. When the controller response is too aggressive, the RDI is negative and less than one, and when the response is too sluggish, the RDI is negative and greater than one. In this work, methods are included which provide a means of making an informed selection of the acceptable performance range.

The final contribution of the new CPA method is a guide for correcting controller tuning when it has been determined that the performance is not acceptable. A PI tuning map method is provided as a means to choose new controller gain and integral time parameters. The map provides a visual guide with reference points established by ξ and τ_n.

Control loops throughout UConn Cogen were analyzed in this study. By identifying underperforming loops (both too sluggish and overly aggressive), the controller tuning could be corrected and the performance improved. The improvements are verified not only with the ACF approach, but also by isolating disturbances to show the change in disturbance rejection, showing reduced variance in the process variable and controller output, and by testimonial from operators working with the process daily.

[1] Jelali, M. (2006). "An overview of control performance assessment technology and industrial applications." Control Engineering Practice 14(5):441-466.

[2] Salsbury, T. I. (2005). "A practical method for assessing the performance of control loops subject to random load changes." Journal of Process Control 15(4): 393-405.

[3] Visioli, A. (2006). "Method for proportional-integral controller tuning assessment." Industrial and Engineering Chemistry Research 45(8): 2741-2747.

[4] Box, G. and G. Jenkins (1976). Time Series Analysis: Forecasting and Control. San Francisco, Holden-Day.

[5] Harris, T. J. (1989). "Assessment of control loop performance." Canadian Journal of Chemical Engineering 67(5): 856-861.

[6] Shah, S. L., R. Patwardhan, et al. (2001). "Multivariate controller performance analysis: Methods, applications and challenges." CPC-6 98: 187-219.

[7] Hugo, A. (2001). "Process controller performance assessment." Hydrocarbon Processing 80(4): 85-90