You are on page 1of 12

contents

Monitoring and Optimizing PID Loop Performance


Randy Burch Advisor Advanced Process Control Software Development Aspen Technology. Inc. Houston, TX 77077 KEYWORDS
Regulatory Control, PID Loop Performance, PID Tuning and Simulation, Sustained Performance

ABSTRACT
Implementation of a Distributed Control System (DCS) provides for significant regulatory control benefits through improved control capabilities, increased system reliability, and in reducing overall maintenance costs. The DCS control and economic benefits are commonly supplemented through commissioning of advanced control applications, model-based predictive controllers, unit optimization, and enterprise planning systems. With the benefits of these various advanced control implementations largely realized in many process plants and with growing pressures on profitability, however, companies are extremely focused on improving performance and optimizing their daily operations. This focus has resulted in a recent shift to emphasizing maintenance of existing applications to sustain economic benefits. This paper will present the benefits derived specifically from one of these maintenance efforts: monitoring and optimizing PID loop performance. Starting from a discussion of the case for investing in regulatory controls, the presentation will then define the challenges and opportunities related to loop maintenance, as well as the performance metrics required for successful optimization. A strategy for integrating long-term PID loop maintenance and performance software also includes process model identification, and loop simulation and tuning optimization to achieve desired control objectives.

INTRODUCTION
The typical DCS system consists of an integrated system architecture of application modules that distribute process control intelligence across the communications network, through a centralized control room with shared operator interface (consoles), and configurable control function components. This modular architecture reduces overall wiring cost and complexity, and promotes scalability for control system expansion. Significant regulatory control benefits are realized through better closedloop control performance, the ability to implement more sophisticated control schemes, and to provide for integrated operator decision support. After implementation, DCS economic benefits are achieved by reducing overall and maintenance costs through integration and use of common of hardware control components, and system related training

Copyright 2004 by ISA The Instrumentation, Systems and Automation Society. Presented at the ISA 2004, 5-7 October 2004, Reliant Center Houston, Texas, www.isa.org

expenses. Recent networking and communications technology advancements have also resulted in significant cost reductions and increased reliability. Current industry practice includes supplementing the DCS control and economic benefits through commissioning of advanced control applications, model-based predictive controllers, unit optimization, and enterprise planning systems (Figure 1).

Figure 1- Benefits of Control and Optimization With the benefits of these various advanced control implementations largely realized in many process plants and with growing pressures on profitability, companies are extremely focused on improving performance and optimizing their daily operations. This has resulted in a recent emphasis shift on maintenance of existing applications to sustain economic benefits.

INVESTING IN REGULATORY CONTROLS MAINTENANCE


Basic instrumentation system (measurements and PID loops) performance has a direct and significant impact on the performance of advanced process controls (APC) and the level of economic benefits they can generate. Even in process plants where no advanced control strategies have been implemented, maintaining optimum PID control loop performance is important for stable operations, reliable response to typical upsets, and efficient unit operation. Poor regulatory control performance can increase operator workload, promote highly variable and unreliable operation, impact plant safety, and in some cases, result in unit shutdown. In applications where multivariable model-based control and/or unit optimization is implemented, the ability to derive economic benefit from continuous operation at unit constraints is dependent on having optimum regulatory controls in place. Since typical benefits of an advanced control application range

Copyright 2004 by ISA The Instrumentation, Systems and Automation Society. Presented at the ISA 2004, 5-7 October 2004, Reliant Center Houston, Texas, www.isa.org

from 3-5% of operating costs, protecting even a 1% decrease in benefits justifies performance monitoring to prevent degradation of the regulatory controls. Optimum regulatory controls are required to allow advanced control applications to:

Operate close to unit constraints Closely track optimization system or planning and scheduling operation targets Move smoothly to new setpoint targets

Changes in process unit operating strategy, feed rate and quality changes, equipment revamps, and APC project pre-testing and commissioning activities are other opportunities where additional benefits in time-savings can be achieved by having control loop assessment software to assist in PID performance evaluation and loop tuning efforts.

LOOP MAINTENANCE CHALLENGES AND NEW OPPORTUNITES


Since maintaining optimum regulatory controls is essential and important for providing consistent, safe process operation and securing economic benefits, why have companies historically neglected them? One reason is that regulatory controls tend to get overlooked once higher-level advanced control and optimization strategies have been implemented and commissioned. Several other factors have also contributed, including the following:

Maintaining regulatory control performance for several hundred PID loops can be very timeconsuming and technically intensive for the process control engineer or instrument technician. Managing data collection and information transfer between the DCS and historical databases is not often straight-forward and can be subject to throughput limitations. Extracting and diagnosing control-relevant information for assessing control loop performance is often difficult.

Recent development in identification technology and increasing computation capabilities offer several new opportunities for overcoming the challenges associated with continuous monitoring and assessment of PID control loops:

Recent studies in control loop performance monitoring and assessment technology have been implemented in industrial applications. Real-time information systems are now widely used in process plants, and a large volume of historical operating data is available for analyses. Web based software makes it easy to access this information, and allows for remote access. Significant advances in computation processing power and communications technology have resulted in increased system performance and lower hardware costs.

AN INTEGRATED SOFTWARE SOLUTION


An integrated regulatory control monitoring software package should take advantage of the recent controller performance assessment technology advancements, and hardware/system architecture improvements. PID loop monitoring by itself does not entirely solve the full problem, and it is essential

Copyright 2004 by ISA The Instrumentation, Systems and Automation Society. Presented at the ISA 2004, 5-7 October 2004, Reliant Center Houston, Texas, www.isa.org

that the workflow automation process for this integrated maintenance solution promotes efficiency and coordination between plant operators, control engineers, and supervisors (Figure 2). Several key functions within this integrated strategy are required for sustaining optimum regulatory loop performance:

Collect, trend and extract historical PID loop information at appropriate frequencies to analyze control loop performance Provide display and reporting functions which easily identify deficient control loop performance that can be used by both operators and instrument technicians Include analyses and performance metrics for the process control engineer or instrument technician to validate control loop issues Correct poor PID control or tune to work-around mechanical limitations using process model identification, PID loop tuning, and simulation features Extract historical information and perform PID analysis on regularly scheduled maintenance intervals, and/or on an as-needed basis

Validate control performance

Identify deficient PID loop control

Control Loop Monitoring


Resolve control or mechanical issues Analyze control loop issues

Online Offline

Although there is overlap of work-flow processes, the control loop monitoring tasks can typically be classified into online and offline functions

Figure 2 Integrated PID Loop Monitoring Processes

ONLINE MONITORING OF CONTROL PERFORMANCE


The intent of the online mode and user interface is to compute and display real-time control loop performance metrics, to detect a change in the performance of a controller or group of controllers, and to indicate and prioritize control loops that need attention or have potential issues. HISTORICAL DATA COLLECTION At the base level of an integrated regulatory loop monitoring system is the distributed control system hardware. While most DCS vendors include historical database capabilities in their hardware offerings, typically a separate information database/historian server is employed for long-term storage of historical information. Some key functions for the historical database server (and application software) include data compression for reducing disk requirements and full un-compression for visualization of information, graphical display and reporting/trending tools, and client/server architecture to allow remote monitoring.

Copyright 2004 by ISA The Instrumentation, Systems and Automation Society. Presented at the ISA 2004, 5-7 October 2004, Reliant Center Houston, Texas, www.isa.org

Usually the information database/historian has the best overall view of the performance of the various control systems in a manufacturing facility. To take advantage of this situation, the regulatory control performance monitoring system should be integrated to work as a part of the historical information database system. The database historian server and integrated control loop monitoring software should include capabilities to do the following:

Access a historical database record structure that includes configurable scan rates, as appropriate for various PID loop parameters Auto-schedule specified control loops to enable high frequency data collection for a specified interval, and later revert to a lower collection frequency to balance communications loading or system performance issues Auto-perform control loop analyses and performance metrics calculations for scheduled PID loops on completion of user-specified data collection intervals Provide an efficient and easily maintainable user interface for configuration of the PID loop sustained monitoring operations Configure auto-scheduled or exception-based control loop performance reporting, including email messaging

CONTROLLER PERFORMANCE ASSESSMENT Recent research and development efforts in the area of primary control loop performance assessment have been targeted at reducing the maintenance burden. While there are many methods of evaluating the performance of controllers based on known process models in the design stage, new techniques have been introduced for monitoring control loop performance. A control objective of minimizing process variance over time makes good sense from a statistical analysis viewpoint; therefore, a minimum variance controller is commonly used to benchmark control loop performance. Harris provided a method for estimation of performance assessment based on minimum variance under feedback control using routine closed-loop operating data. This technique has attracted significant interest and has been further developed by many researchers. According to Harriss approach, a control loop performance index assessment is typically based on a known process dead time, and indicates the control loop performance in response to random disturbances, or the ability to return to setpoint within a specified interval. In practice, where an accurate estimation of the process dead time is often unknown, a number of extended horizon performance index values are calculated over a range of assumed dead times (Figure 3). While the performance index provides a good assessment of overall controller performance, other PID analyses can be used to interpret whether deficient controller performance is a result of improper tuning parameters, mechanical problems, process disturbances, etc.
ADDITIONAL CONTROLLER PERFORMANCE METRICS

An impulse response function plot represents the closed-loop impulse response to an assumed white noise type disturbance. This function is a direct measure of how the PID controller is performing in rejecting disturbances or in tracking of setpoint changes. In practice, the impulse response can be

Copyright 2004 by ISA The Instrumentation, Systems and Automation Society. Presented at the ISA 2004, 5-7 October 2004, Reliant Center Houston, Texas, www.isa.org

considered as an approximate disturbance finite impulse response (FIR) model of the closed PID loop. Therefore, the traditional characteristics of a PID loop; overshoot, damping-ratio, settling time, oscillation, can be used to check the performance intuitively.
Typically, a reference set of data is collected for normal operation with the previous controller tuning to compare performance of the controller after tuning adjustments. Note the significant improvement in the TC872 [07dec03] performance index after PID control loop retuning. A performance index value near or equal to 1.0 is optimum. Control loops with sluggish tuning, as in the pre-tuned TC872 [19nov03] analysis, and control loops with excessive oscillation produce suboptimum results.

Figure 3- PID Performance Index Plot The auto-correlation analysis plot ot the control error (PV-SP) is an approximate measure of how close the existing controller is to its ideal best achievable closed-loop minimum variance performance condition, or how predictable the error is over the time horizon of interest. A plot of auto-correlation versus time should quickly decay to zero after the process delay, and coefficient values should lie between 95% confidence limits, if the controller has achieved minimum variance (Figure 4).
Auto-Correlation Error Plots are also typically used as an indication of the PID Loop tuning quality. A slowly decaying autocorrelation function over time implies an under-tuned or sluggish loop as shown in the TC872 [19nov03] analysis example. Improved response for reducing error in the re-tuned TC872 [07dec03] control response, is indicative of better PID control as also illustrated in the previous performance index plot.

Figure 4 PID Auto-Correlation Error Plot


Copyright 2004 by ISA The Instrumentation, Systems and Automation Society. Presented at the ISA 2004, 5-7 October 2004, Reliant Center Houston, Texas, www.isa.org

The closed loop frequency response of a PID loop is another measure often used to check whether oscillations exist in a loop (Figure 5). The analysis is based on the closed-loop disturbance model, assuming the input is white noise, ouput is the PV response under control, and the setpoint is help constant. The closed-loop frequency response is often plotted together with the minimum variance control response as a benchmark to check the possibility of performance improvement through controller re-tuning.
Large deviation in the low frequency range typically indicates lack of integral action or weak proportional gain. Large deviations in the high frequency range are typically an indication of significant measurement noise. Significant peaks in the medium frequency range, as shown in the FC813 [19nov03] analysis, typically indicate an aggressive or overtuned controller, or the presence of oscillatory disturbances. After retuning, note the removal of this peak in the FC813 [07dec03] response.

Figure 5 PID Closed Loop Frequency Response Plot Several other loop status and statistical metrics that can provide useful information for assessing controller performance include:

An in-service value is calculated for percentage of time the control loop is in auto or cascade mode. PID controllers that spend a significant amount of time in manual mode are usually indicative of deficient control behavior or mechanical issues. Controller output saturation is the percentage of time during the specified analysis period where the control valve output is saturated or at output limits, comprising the ability of the PID loop to maintain setpoint. Standard deviation of process measurement and controller output. Large deviations in the process measurement may be indicative of a noise, control valve problems, or an overaggressively tuned controller. An oscillation index is an estimated percentage of the control error that is made up of oscillations (or periodic cycles) rather than random disturbances. If oscillatory control loop behavior is noted, then the cycle period can be estimated from the closed-loop frequency response plot.

ANALYSIS REPORTING INTERFACE A web-based interface will be employed for connectivity to the historian/database server and controller performance monitoring system. The functions for auto-scheduled PID loop data collection,

Copyright 2004 by ISA The Instrumentation, Systems and Automation Society. Presented at the ISA 2004, 5-7 October 2004, Reliant Center Houston, Texas, www.isa.org

management of data collection frequencies, and auto-run of control performance analyses on a continuous recurring interval can be thought of as the online mode of the performance monitoring system. Primary users of the performance monitoring interface include maintenance technicians and support engineers whose are interested in the performance of the plant primary controls. The display content should clearly provide the following information:

What loops are not working properly? Have any PID tuning or configuration changes been made? Are there any significant changes in control loop performance?

An example PID Analysis Results web display included below supports several useful features for easily assessing deficient control loop performance including: sorting capability on various PID analysis metrics (columns), color highlighting of performance metrics that fall below user-specified limits, a textual description of possible issues with control performance in addition to the calculated performance metrics values, and inclusion of user-assignable loop priority (Figure 6).

Figure 6 - Example PID performance calculations display The ability to implement performance indices that are mutually consistent is important. It is also critical that specific loop performance limits are configurable, allowing analysis results to be normalized, such that the control performance metrics can be assessed on a common basis. For example, a flow loop with fast process dynamics will have a different set of criteria for measuring good control behavior than would an integrating level control loop. Importing and capturing the DCS tuning-related information in the online web interface is also useful for offline trouble-shooting of control issues. Access to primary control loop instrument range limits, current PID tuning coefficients, and controller configuration parameters is more convenient through web-based displays, rather than using the DCS console. Historian/database server capabilities can be utilized and displayed via the web interface for long-term validation (trending) of performance metrics, in providing for clear visual indication of control loop degradation. Applying statisticical metrics to the PID analysis results also offers another mechanism to easily detect control performance issues. Again, it is important that the integrated online maintenance reporting interface takes advantage of available performance metrics to identify problems, quickly

Copyright 2004 by ISA The Instrumentation, Systems and Automation Society. Presented at the ISA 2004, 5-7 October 2004, Reliant Center Houston, Texas, www.isa.org

diagnose potential causes, and prioritize maintenance activities. A major benefit of the automated monitoring of hundreds of PID loops is in reducing the investigation and troubleshooting time for hopefully a small subset of loops whose performance is suspect.

OFFLINE RESOLUTION OF CONTROL ISSUES


While monitoring loop performance is necessary, it is still not sufficient for maintaining optimum regulatory controls. Additional tools are needed to help analyze control performance issues, efficiently resolve issues when possible, or provide adequate tuning work-arounds to relieve mechanical issues. The performance monitoring system includes an offline mode for maintenance technicians and process control engineers to perform the following functions:

Troubleshoot deficient regulatory control performance Configure controller performance limits as appropriate to detect when a controllers performance has changed Correct poor PID tuning

The first offline troubleshooting function is employed after a controllers performance has been identified as suspect. In this case, the various performance metrics results may be sufficient for isolating control issues, or examination of additional historical data sets and possible step testing of the control loop may be required to confirm the cause of deficient control behavior. A second offline configuration function is necessary to define the performance metrics for acceptable control behavior. The function requires knowledge of the physical process, loop design and control objectives including economics, and on some basis of historical operating data. The calculations for the performance metrics will necessarily need to be performed over a specified interval of history for the effective results. Once the loop is considered a candidate for improvement, the next step is to resolve the identified control performance issues, if possible through loop re-tuning or maintenance/repair of mechanical issues. Important offline analysis functions for the correction of deficient loop control include process model identification, determination of optimum tuning parameters, and simulation of user-specified tuning (prior to implementation) for correction of deficient loop control. PROCESS MODEL IDENTIFICATION Obtaining an accurate process model is essential for tuning and simulation of control loops. Traditional methods for identifying a process response models typically include conducting appropriate experiments, such as open loop steps. This methodology typically works well, and the use of first-order and second-order model concepts for gain, dead time, and time constant are understood by most control engineers. The downside of an open loop step testing is that it can be disruptive to normal plant operation, requires operator attention, and can create potentially unsafe conditions. Recently developed subspace model identification technology has become a popular tool for the identification of state-space models that have been used in multivariable predictive control

Copyright 2004 by ISA The Instrumentation, Systems and Automation Society. Presented at the ISA 2004, 5-7 October 2004, Reliant Center Houston, Texas, www.isa.org

applications. New subspace model identification techniques for PID loop model identification using closed loop data have also been implemented in industrial applications. A significant advantage of employing subspace model identification is that process models can be identified while under closed loop operation, relieving concerns for operator supervision and safety issues. Calculated results from the subspace model identification include optimum low-order model reduction for process gain, deadtime, and appropriate time constants (Figure 7).

Model identification results include calculation of open-loop process gain and trends for the open-loop process model response and controller output response.

The closed-loop PV response to a unit step change is obtained for closed-loop data.

Figure 7 Model Identification Plot It is an important step to visually confirm the model identification with the observed raw data plots for the history interval of interest. Other enhancements in the model identification software support use of open-loop, closed-loop, or a combination of open and closed loop data, auto-slicing of bad data measurements, and identification of integrating processes (ramp variables). OPTIMIZE TUNING FOR CONTROL OBJECTIVES Many control loops in the typical process plant that can be easily tuned from trial-and-error or practical experience. A number of complex loops however that typically require more than these basics, often demanding a combination of skills including an understanding of process dynamics, control theory, process engineering, loop interactions, and familiarity with economic and operating objectives. Even for a control loop-tuning expert who may possess many of these attributes, this task can be very time consuming. There are many well-known tuning methods: Internal Model Controller (IMC), Ziegler-Nichols, Cohen-Coon, etc. in addition to thousands of other published articles for tuning PID loops. With few exceptions, these methods typically fall short in some instances. More importantly, these methods do

Copyright 2004 by ISA The Instrumentation, Systems and Automation Society. Presented at the ISA 2004, 5-7 October 2004, Reliant Center Houston, Texas, www.isa.org

not typically provide the ability to tune for deliberate control behavior or may not match the specific DCS vendor algorithm. Ideally, an optimization algorithm is implemented to best-fit or minimize the least-squared control error for both setpoint change and disturbance or load change, in the determination of the PID tuning coefficients. Inclusion of weighting factors in the optimization algorithm then allows the user to make tuning behavior trade-offs:

Tighter control for setpoint change vs. load disturbance Suppression of controller output to limit size of initial output move Model uncertainty factor to increase robustness vs. aggressive control

Including the ability to user-specify tuning values (and simulate), or to override the identified process model parameters, is useful for what-if type tuning scenarios, or when the user may want to impose a model mismatch to examine the resulting control behavior for a given set of tuning coefficients. These trade-offs in control behavior can also be manipulated to work-around mechanical issues.
PID LOOP SIMULATION

The PID Simulated Response plot provides several trends of the PV response and PID output for an initial unit setpoint move, and after one steady-state period, for an injected load disturbance. These displayed trends include the setpoint step move, and simulated controller responses/outputs using the calculated (or optimized), user-specified, and current PID tuning coefficient terms (Figure 8). Optimization of loop performance is based on a simulation of the PID loop, using exact DCS vendor equations.
The upper trend includes the initial PV response to unit setpoint move for one time to steady-state, and second PV response to load disturbance on output of one-half unit. The different curves correspond to use of current tuning, optimized tuning, and userspecified tuning coefficients. The lower trend indicates the appropriate PID controller output response for the setpoint move and load disturbance.

Figure 8 - PID Simulated Response

Copyright 2004 by ISA The Instrumentation, Systems and Automation Society. Presented at the ISA 2004, 5-7 October 2004, Reliant Center Houston, Texas, www.isa.org

Iterating on adjustments for PID tuning trade-offs as described in the previous section is efficient, and allows the user to visually shape the desired process control response. Again, it is critical to validate the simulation plot using the current implemented PID tuning coefficients against the actual closedloop raw data. If this validation step fails, then the identified process model is not accurate for providing correct tuning adjustments.

CONCLUSIONS
There is a continuous need to increase production efficiency and maximize economic benefits for any given process plant. Typically, hundreds of PID loops are employed at the basic regulatory level for controlling the process. In the case where advanced control strategies are employed to derive additional economic benefits, these applications are dependent on an optimum base level of regulatory controls. Good regulatory control system performance is essential in order to maintain safe operation, maintain product quality and minimize operating cost. An important objective of evaluating the PID loop control performance is to define the methods and implementation strategy that will reduce a large amount of process information to a few easily interpreted metrics numbers that provide a clear picture of overall control loop performance. To maintain desired PID loop operation, a sustained control performance software offering should include features for historical data collection, controller performance metrics, process modeling and loop tuning/simulation capability, and utilize an efficient user-interface for display and reporting functions. Integration of these functions provides an opportunity to validate PID control performance, quickly identify control loop deficiencies, prioritize efforts in resolution of control issues, and distribute some of the efforts associated with this activity on a continuous basis for sustaining optimum regulatory control. Although economic benefits derived from implementing a performance monitoring strategy may be difficult to quantify on a per-loop basis, often a single control issue can manifest into contributing to poor overall unit or process performance. The ability to provide sustained regulatory control performance is easily justified through improvements in online time, resource reductions, improved safety, and control effectiveness or quality of process financial performance.

REFERENCES
Thomas J. Harris, Assessment of Control Loop Performance Canadian Journal of Chemical Engineering 67 October 1989, pages 856-861

Copyright 2004 by ISA The Instrumentation, Systems and Automation Society. Presented at the ISA 2004, 5-7 October 2004, Reliant Center Houston, Texas, www.isa.org

You might also like