Applying Offline Tools To Reduce On-Site Control Implementation Effort
18/06/2013 Honeywell Process Solutions
Productivity tools and work process changes to expedite advanced control projectsKen Allsford, Bhaskar Iyer and Aric Tomlins, Honeywell Process Solutions
The Practitioner's Challenge
Process control engineers are always challenged to be more effective to successfully commission new advanced regulatory control strategies and multivariable control applications. To address the challenge, the practitioner needs to continuously improve and apply understanding of both control technology in general and the process technology specifics relevant to each control project. This article discusses some of the newer productivity tools and work process approaches that are now becoming more widely deployed.
The development programs of control product software vendors continuously improve core and auxiliary products. Although one of the outcomes of these programs is a migration to "Product Suites" with user interfaces based on a work flow perspective, there is still a vast range of "designed-in" capability and functionality. Also, as knowledge and experience filters through the user community, the practical consequences of software enhancements are that improvements to product features enable capabilities that are beyond the original implementation rationale. These capabilities also have to be understood and internalized. Software product improvements are also obviously tied to particular releases. In combination, these factors lead to the challenge of minimizing the time to modify and update control project work processes to fully incorporate efficiencies facilitated by software product improvements.
As well as establishing effective general control project work processes, the practitioner is further challenged to be specifically efficient in regards to the execution of each and every project. Practically, this requires the practitioner to develop a deep understanding of the actual process unit(s) relevant to the control project. And this needs to happen over a relatively short duration. Early understanding helps to "filter" inputs such as comments from operations and to identify upfront any issues that may challenge project success. Early understanding also enables better definition of the control problem and, as such, selection of an appropriate solution from amongst the available options. These activities have a parallel in terms of the Front-End Engineering Design (FEED) phase typical for engineering projects in the process industries. On completion of the design and selection of the solution to the control problem, the project moves on to implementation. This corresponds to the detailed design, construction and commissioning phase typical for engineering projects in the process industries. A primary technical objective here is the "identification" of the relationships of the independent variable(s) (Manipulated and Disturbance Variables, MV and DV) to the dependent variable(s) (Controlled Variables, CV). A secondary technical objective is discovery of reasonable control application "tuning" parameters given the selected control product technology. During implementation, the practitioner is once again challenged to continue to be effective.
Established Approaches
Integration of the knowledge and experience of plant operations and engineering personnel into the selection and implementation of control problem solution has always been integral to the success of a controls project. The core aspect of this knowledge and experience can be viewed as 'tacit knowledge'. This is knowledge that is neither expressed nor declared openly but rather implied or simply understood and is often associated with intuition. As such the capture of this knowledge is not well-defined - control project work processes usually deploy both structured interviews and ad hoc discussions as tools to help in this "knowledge transfer" to the control project team. This knowledge transfer is primarily aimed at the higher level of process understanding and initial definition of the control problem solution (in terms of CV, MV and DV). It is usually done in conjunction with historical data review including visualization, regression analysis and use of general spreadsheet-oriented tools.
The goal of advanced process control is successful projects leading to competitive advantage for the organization through superior process operations. Roll-out was done by engineers, many of some renown and with a background in chemical engineering. As such, "toolkits" based on process engineering models were sometimes developed to support this activity. This was particularly the case for refining and other large-scale petrochemical processes. These toolkits were typically developed initially with the limited focus of providing output just for process control. Examples include toolkits for fractionators (based on distillation principles), delayed coking and platforming. The degree of technology "openness" and toolkit productization varies. Commercialization of such toolkits represents a challenge as often a toolkit is focused on a particular process unit and there are only so many installations of that process unit worldwide. Where available and maintained, these toolkits remain a key component of the control solution for certain processes. However the focus and narrative of control product software vendors relatively quickly moved towards data-centric modeling and efficiency tools that have broad applicability to all control projects.
The earliest advanced control software products included "control model" identification tools. The tools analyze plant data and a key output is statistically validated time-domain relationships between independent and dependent variables. Relationships of independent to dependent variables are generally sought in terms of a linear steady state "gain" and some definition of the time-dependent path to steady state. Generally, accuracy for the former is prioritized over accuracy for the latter as (relative) steady state gains are used for both control and economic optimization and can also be used in validating the viability of the selected control solution. In addition the tacit knowledge transfer between process operations and the control project team discussed above is often more successful for some general aspects of the dynamic relationship (such as dead time and first order time constant) than for steady state gain.
Initially these tools were not integrated with their input data generation and control practitioner organizations would emphasize their work practices that enable efficient collection of this input data with minimum explicit effort and disruption to the process unit. Nevertheless, in some circles the mind-set was that identification was a very costly process with long elapsed times. And this is not to mention the impact of changes to controller design as a consequence of revised requirements definition or increased understanding of the best solution to the control problem.
Control product vendors responded by integrating the data generation step with the identification step in real time. Indeed these tools have received significant write-up in the technical press, including many documented successes. The first generation of such tools required the process that is being identified to be operated in "open loop" whilst identification is performed by stepping the process. In practice this leads to the selection of small steps and identification on small sub-groups of independent and dependent variables at a time in order to minimally interfere with the process operational goals. As the tools have evolved, the current state of the art now allows what is termed automated closed-loop stepping and identification. This means that the controller simultaneously controls the plant during the testing and identification process. This leads to the largest reduction in effort for both project owner and project implementer. In turn, this simultaneous control aspect associated with the online identification requires a "Seed Model", which is an estimate of the time domain relationships for the independent and dependent variables that is sufficient for constraint control as a minimum.
Opportunities
These improvements to data-centric modeling and efficiency tools change the basis by which a control project organization can define itself as efficient in implementing control projects. Now the emphasis is on provision of the seed model and being able to understand and select to utilize as appropriate from the many continual additions to the control product software suite. In tandem, other technologies that can potentially be used to meet this challenge of requirements change for control project efficiency have also developed. Three such technologies that have evolved in capability and practice within the industrial engineering environment are the enhanced in-built functionality of generic process engineering modeling tools; knowledge management deployment; and use of virtualization technologies.
Application of Process Modeling Tools
What is the direction of process engineering modeling? At a high organizational level, the concept of "model life cycle" and its associated management is receiving increased attention given the expansion of both the role of dynamic modeling during engineering design and the use of high fidelity operator training simulators (OTS). This has impacted the development programs of simulation vendors. Another consequence is more frequent inclusion of a "process-matching" activity as part of the scope of specific simulation projects.
At the higher technical level, simulation vendors are providing better support for connectivity and data exchange, including OPC, for both reporting and other uses. At a more detail technical level, both the number of unit operations selectable from the operations palette and their predictive capability and reliability continues to increase. Whilst examples are dependent on the simulation product, for the former the inclusion of multivariable control operations from the major vendors can be referenced; whereas for the latter improved representation of the performance characteristics of rotating equipment can be referenced.
Historically, some organizations have had difficulty integrating the control project engineering discipline with process modeling disciplines. So it is worthwhile to note potential benefits from utilizing process models to develop seed models. These include:
-€¢ Process models can be run anywhere. Process model predictions engage users and give them "something to think about".
-€¢ In terms of "stepping", process models can be treated as exactly repeatable for steps of the size relevant to control model identification. A process model has to be stepped only once and the data captured for identification whereas the real plant must be stable and still subjected to multiple steps to reduce identification uncertainties arising from noise, unmeasured disturbances and real-life nonlinearities. Also, conceptually the model step size is not constrained on the high side whereas in the real plant the step size reflects a trade-off between obtaining sufficiently high signal/noise ratio and not creating a process operations problem. Finally, measured disturbance variables can be stepped in process models, whereas this may not be possible for the real plant.
-€¢ Dynamic process models generally run faster than real time and reporting can be configured to be visually engaging. As such they can be set up as a concept demonstration for those on the periphery of a control project implementation. For the control project team, they provide opportunities to investigate control structure and optimization strategy; calculate benefits pre-implementation; evaluate outputs from data-centric tools; evaluate "compromise tuning" through use as a simulator (real-world nonlinear model) for the controller; and explore the effect of unmeasured disturbances for the controller.
The first two points above are neutral with respect to whether the process model is steady state or dynamic, although conceptually more information can be obtained from the latter. For example a steady state process model will be able to provide insight into process gains but not, explicitly, process dynamics.
So having noted the benefits, what are the concerns? To generalize, concerns relate to "fear of the unknown", both in terms of development effort and practical validity for purpose. What about development effort? As a guideline, if the simulator application development does not involve "research" (i.e. it is essentially a well-defined configuration activity, such as a hydrocarbon separation column), then the effort estimate for a relevant process-matched model is similar to the estimate for more conventional step testing in the field based on use of data-centric modeling tools. Even if a suitable starting point may not be available, it may still be pragmatic to develop a process-matched model. This is especially so where step testing is a challenge, such as greenfield projects; offshore; and processes where either tight control is paramount or steady state operation rarely occurs for the duration relevant to step the process to obtain control models. This would be all the more so if the organization's strategy leans towards process-model centric. As a library of pre-existing process engineering models becomes available, extension to new scenarios can be expected to be both at reduced cost and improved reliability of the effort estimate.
Concerns about the validity of a model for a particular purpose reflect uncertainty in the concept (actually work process) of "process matching" of the model. Process matching and model validation is a broad subject and only a few aspects are discussed here. This skill develops from an understanding of the implemented key modeling decisions, which in turn relate to available options and model use objectives. The central focus of many processes is the reactor and if present this will likely form the start point for determining the usefulness of the process model for seed model generation. In-house expertise may be available to provide guidance. Moving beyond the reactor, if model predictions under process turndown conditions are of interest, then the approach to predict pressure profile and heat transfer performance under these conditions should be evaluated. Note that steady state models are often simpler in implementation concept than dynamic models. Pressure profile may be fixed and energy integration reflected through trivial balancing without considering equipment performance capabilities. For all models, boundary condition stream compositions and specifications needs to be understood and possibly modified prior to use. (Dynamic models are usually based on fixed pressure boundaries with flow calculated, although sometimes boundary flow may be fixed and pressure calculated. For the usual case of fixed pressure boundaries at the inlet to a dynamic process model, a real or artificial flow controller is usually configured to set the flow.)
How best to balance the potential benefits from utilizing process models whilst mitigating the concerns? This will depend on the specifics of an organization's continuously evolving hard and soft structures. But generally, a target will be more co-ordination and/or integration of the simulation and process control application teams. Specifically, model use can first be deployed, where judged appropriate, as part of the tool-set for control structure and optimization strategy design and for supporting implementation of more narrowly focused strategies. For the latter, in many organizations, tuning deployed for advanced regulatory control strategies within an OTS is used as the starting point during actual commissioning.
Application of Knowledge Management Deployment
Knowledge management deployment takes the form of restricted-access "wikis" and "repeatables" with various options for moderating postings. An objective is the leveraging of experience of individuals within the team to other members of the team. Through hosting of recordings, documents and directly posted notes of knowledge and experience, wikis are effective for internal transfer of tacit knowledge. This includes, as examples, suggestions for both generic learnings on the roles of and trade-offs for different tuning parameters and best practices for control of a particular process unit. Repeatables can be viewed as tools owned and managed by project teams rather than product development teams.
Due to the large pool of potential contributors, wikis are useful for tackling the challenge of reducing the time to update control project work processes to incorporate efficiencies facilitated by software product improvements. In addition to the organization's official Subject Matter Experts (SME), wiki-users can also identify other knowledgeable contacts to arrange short discussions by noting the major wiki contributors in any particular technical area.
In the "FEED" phase of a control project, sharing of knowledge on wikis helps to hone in on the controller design. The posting of "expectation matrices" for linear multivariable control; expected inferentials; transform expectations for linearization (column purity, valve positions as examples); and informal notes relevant to process unit control all help in this regard. Following on from design, sharing of repeatables can help expedite control project implementation. One example of a repeatable is pre-processed data model libraries for different expectation matrices (suitably edited to protect confidential information and documented with respect to engineering units, plant percentage loading, etc).
Application of Virtualization Technologies
Virtualization technologies enable an integrated system that formerly comprised two or more physical computing platforms to now run on a single computer through use of multiple virtual machines. For example, for greenfield projects, virtualization simplifies the use of OTS to develop and verify advanced control applications. Virtualization offers several other productivity opportunities for control projects, particularly associated with reducing hardware costs and minimizing software conflicts. Also, through having an exact copy of a site installation available, virtualization enables superior and more comprehensive support. Examples include using offline virtual machines for first verification of control configuration maintenance and maintaining snapshots of virtual machines for possible rollback during a software upgrade.
Seizing the Opportunities
The objective defined earlier is to identify a seed model that is at least sufficient for constraint control during closed loop control model identification for a control project. A final practical strategy combines these six strategies. Knowledge management is deployed within steps 3 and 5 whilst process models are deployed within step 6.
But actually the practitioner is especially interested in a seed model that is sufficient for final deployment, rather than just for constraint control during closed loop control model identification. This is where process models offer further potential. And it is also worth noting that process models are not limited to process simulators. The following example describes use of a "process model" developed within a spreadsheet.
The actual application is simplified for this discussion and is shown in Figure 2. It basically consists of a surge drum with three flows into and one flow from a drum. The control application has two MVs, two CVs and two DVs. The flow out of the drum is MV1. One of the flows into the drum is MV2. The two other flows into the drum are DV1 and DV2. The drum level is CV1 and the ratio of "DV1/(DV1+MV2+DV2)" is CV2. Vessel dimensions and level tap information from the unit Process and Instrumentation Diagrams (P&IDs) enable the process gains between CV1 and each of the MVs and DVs to be calculated by spreadsheet formula. Spreadsheet formula can also be used to calculate the process gains between ratio CV2 and DV1, MV2 and DV2. Here the gains depend on operating conditions but it is relatively trivial to calculate the variability of the gains across the operating region planned for the controller and to then select reasonable gains. (Alternately, the controller design could be revisited by looking to replace or transform CV2 or by deciding to dynamically update controller gains for CV2 online i.e. whilst the control application is in service.) The control models identified from this spreadsheet analysis were found to be consistent with control models obtained by conventional identification that used a few days of data mined from the DCS historian as input data. Based on this validation, the control models identified from spreadsheet analysis were used for commissioning of the multivariable controller without step testing. Here, the process model (spreadsheet formulae in this case) provided insight into gain nonlinearity across the process operating conditions that is not typically available from conventional identification.
More significant examples have also been discussed in the literature. One such example involved the design, test and installation of multivariable controllers during the engineering and construction phase of a 1300 mmscfd gas plant (Reference 1). Multivariable controllers were applied to the units for (a) feed gas separation and condensate stabilization; (b) dehydration and feed liquid separation and (c) NGL recovery. Dynamic process models developed during engineering design were enhanced for the multivariable control application development. In particular, multivariable process controllers were incorporated directly into the process models as unit operations. This has the advantage that the controller performs consistent with design intent at high model real time factors.
In the NGL unit, one enhancement arising from evaluation of regulatory control loop interactions within the process model was the addition of ratio controllers that were also used as MVs for the multivariable controller. The NGL unit process model also demonstrated that deployment of the multivariable controller application increased ethane recovery by 1.37% in ethane recovery mode, which has potential benefit of $ millions per annum. The controller commissioned on the virtual plant was designed consistent with the KISS (Keep It Simple, Sir) principle. However the NGL unit process model could also be used to demonstrate additional potential benefits through expanding the controller scope (MVs and CVs).
In this example the applicability of the control models identified from the process engineering model to the real plant was judged in part by comparing to control models identified by more traditional approach for a different application of the same gas plant process technology.
Conclusion
As has been discussed, the maturing of closed loop step testing and identification technology for advanced control reduces the effort required for testing and identification for an advanced control project. However, it requires a seed model and, as such, the interim milestone of providing this seed model will receive more focus. And, naturally, there will also be desire for the seed model to be more than just a seed model - it should basically be suitable for control deployment without step testing. This will lead to adaption of control project engineering work processes, on an "opportunity basis", to move towards meeting these desires. As the future evolves, work processes are likely to more widely incorporate use of:
-€¢ Latest features of data-centric modeling and efficiency tools
-€¢ Steady state process engineering model analysis
-€¢ Dynamic process models
-€¢ Organizationally supported knowledge management deployment
-€¢ Virtual platforms standardized for the organization
Of course, tools with proven historic value, such as process model engineering toolkits for control, will continue to be deployed where appropriate.
Reference
1. "Advanced Process Control in the Plant Engineering and Construction Phase"; V. Sakizlis, A. Coward, K. Vakamudi & I Mermans, Hydrocarbon Processing, Oct 2010.
Ken Allsford has 25 years of control and simulation experience and is currently a simulation discipline lead based in Houston with the Advanced Solutions group of Honeywell International. He has a PhD degree in Chemical Engineering from the University of Birmingham, United Kingdom. Email: Ken.Allsford@Honeywell.com
Bhaskar Iyer has 14 years of experience in implementing advanced control and is currently a control discipline lead based in Houston with the Advanced Solutions group of Honeywell International. He has a BTech in Chemical Technology from Nagpur University, India. Email: Bhaskar.Iyer@Honeywell.com
Aric Tomlins has 30 years of experience in implementing advanced control and is currently a control discipline lead based in Houston with the Advanced Solutions group of Honeywell International. He has a BSc in Chemical Engineering from Rensselaer Polytechnic Institute, USA. Email: Aric.Tomlins@Honeywell.com
For more information, please visit: www.honeywellprocess.com
Print this articleShare article:
Process and Control Today are not responsible for the content of submitted or externally produced articles and images. Click here to email us about any errors or omissions contained within this article.