Reprinted from: The improving global economic outlook for polyolefins offers a strong advantage to those companies that are preparing for the increased demand, tighter schedules and more frequent type changes brought on by the recovering petrochemical business cycle. Inventory reduction and improved customer service initiatives mandate that petrochemical plants perform with greater reliability and cost efficiency. Equally important, these expectations will only increase further as this recovery continues. Preparing for the upturn. The SABIC Polyolefin polyethylene plant at Gelsenkirchen, Germany, prepared to take full advantage of the next polyolefins business upturn. As part of their business plan, SABIC sought methods to raise production rates at existing facilities, reduce transition time between product changes and minimize off-spec generation. Such efforts would cut operating cost while raising available product inventories. But what is the best way to achieve these goals? SABIC management elected to change the behavior of the facility and implemented a commercial nonlinear advanced control technology to achieve its manufacturing goals. This technology controls the nonlinear steady-state and dynamic process behaviors to ensure steady-state control and variability reduction for critical control parameters. Sustaining optimal control of product parameters during grade-change transitions reduces transition material generation. Likewise, best operating practices were devised that enable all process operators to run the unit optimally. This case history details the positive returns possible with advanced process control methods for the polymer industry. Extended downturn. For the past five years, most polyolefin businesses have experienced very tight, and even negative, margins resulting from several macroeconomic factors. Sharp excursions in natural gas prices and the general trend of increasing feedstock prices created significant margin pressure; these forces further amplified the lingering downside of the cyclical supplydemand balance of the petrochemical industry. A perfect storm occurred, in which major increases in feedstock prices joined forces with unstable and decreasing end-product selling prices PETROCHEMICAL DEVELOPMENTS April 2004 issue, pgs 45 49 Used with permission. www.hydrocarbonprocessing.com SPECIALREPORT Advanced control methods improve polymers business cycle Using nonlinear technologies, major polyolefins manufacturer increases plant capacity without equipment retrofits O. KARAGOZ and J. VERSTEEG, SABIC Polyolefine GmbH, Gelsenkirchen, Germany, and M. MERCER, Aspen Technology, Inc., Houston, Texas, and P. TURNER, Aspen Technology, Inc., Warrington, UK Capacity, million metric tons 60 50 40 30 20 10 FIG. 1 Capacity lags behind demand Demand Capacity Operating rate 0 97 98 99 00 01 02 Year Source: CMAI. 03 04 05 06 07 Major trends in the polypropylene market demand, capacity and utilization rate from 1997 to 2007. stemming from overcapacity. National events and world-wide recessions in the largest national economies exacerbated this very severe downturn in margins and financial performance. Better news for tomorrow. The good news is that industry analysts such as Chemical Market Associates, Inc., are anticipating the end of this extended downturn in the supply-demand balance of the polyolefin industry. Strong improvements are expected to occur through 2007 (Fig. 1). The approaching upturn in the business, according to industry leaders, could be the strongest peak in the last 40 years. As shown in Fig. 1, the falloff in operating rates at the end of the cycle is not steep; consequently, the next wave of capacity builds will be more measured than in previous cycles. For SABIC, market conditions are characterized by these drivers: Adapting to an increasingly complex product mix. The SABIC LD5 plant makes a wide range of products, including grades with different types of comonomers and differing comonomer contents. This unit frequently switches among these significantly different grades. The desire to achieve improved execution efficiency and results for these frequent and complex grade-change transitions was a driver to apply a technology solution for automating the grade-change execution to duplicate the execution and results of the best DCS operators. A polymer 100 95 90 85 80 75 Operating rate,%
SPECIALREPORT PETROCHEMICAL DEVELOPMENTS New EU safety guidelines impact polymer APC solutions An Eunite task force, assigned by the European Union (EU) to advise on applying smart technology in safetycritical operations, has recommended against using neural networks in closed-loop control of manufacturing processes. The report, Final Report of the Task Force on Safety Critical Systems, was issued in September 2003. What does the report say? An appendix to the report contains a paper, which describes the dangers in detail, Product Grade Transitions Exposing the Inherent and Latent Dangers of Neural Networks in Manufacturing Process Control. The task force concludes: Difficulties with vanishing and inverting gains are a characteristic of the feedforward neural networks and are better handled by other technologies (both of these problems have catastrophic implications for a closed-loop controller on a manufacturing facility) Closed-loop control of safety-related plant should use transparent models (i.e., not black boxes such as neural networks). These stark conclusions should be a wake-up call to all process manufacturers currently using or considering using neural networks for closed-loop control of their capital assets. Anticipating that neural networks were not a suitable algorithm for influencing valve movements on manufacturing processes, substantial investment was made in creating a suitable alternative technology. The resulting technology is specifically endorsed by the EU task force on safety critical systems. The full EU report and appendices can be downloaded from the following Website: http://www.eunite.org/eunite/task_forces/running/final ReportSafetyCriticalEunite2003.pdf http://www.eunite.org/eunite/task_forces/running/paper sateunite2003.pdf advanced control solution can significantly improve consistency in transition performance, and reduce off-spec product generation up to 40%. This improved performance capability supports and reinforces the plant s competitiveness to economically deliver a more complex product mix. Optimizing plant operation at very high capacities. SABIC recognized that additional capacity was available. Applying advanced control solutions could continuously operate plant production rates more closely to the actual physical constraints than would be possible with manual control. By reducing production rate and quality variability, and recognizing the physical constraints, new control technologies could enable maximizing production capacity. Typical figures for this industry are about 5%. Meeting innovation and improvements directives. SABIC continues to innovate and introduce new products and process technologies to remain competitive. Due to this environment, the chosen advanced control technology had to be robust and flexible to handle continuous changes in product property targets and reaction conditions, as well as changes in processing technologies related to the total reaction heat-removal capacity of the plant. These capabilities have been demonstrated and delivered with the polymer advanced control solution. Minimal additional maintenance is required to keep the control solution operating at high reliability. Resolving conflicts between total inventory and customer service goals. The improved process control delivered by the advanced control solution lowered product-property variation by 30% and completely eliminated steady-state off-spec product generation. The large decrease in steady-state off-spec production, combined with the improved reliability and reduced off-spec during transitions, has allowed SABIC to reduce off-spec inventory. More prime product is available in inventory to support ongoing customer needs. SABIC implemented this process control automation technology on the LD5 PE unit at the Gelsenkirchen site. Specific program goals included: Increased production rate Reduced product variability Lower transition times Reduced number of lab samples taken. Benefits and economic payback. To provide the plant with the capability to meet its operational excellence challenges, the polymer production control solution was implemented at Gelsenkirchen during 2002 to deliver these performance objectives: Predict polymer properties: melt index, density and comonomer content Reduce process and property variability Improve transition performance (reduced times and offspec quantities) Implement improved capabilities with rapid payback of investment of less than one year Ensure sustainability. The selected polymer production control solution was installed and proved very effective, at more than 20 polyolefin plant implementations over the last four years. This system has delivered demonstrated project results including: Up to 40% reduction in transition off-spec 50% to 100% reduction in steady-state off-spec. Because of external conditions (extended downturn for polymers), the longer-term benefit of capacity increase has not yet been tested. Yet, other benefits and economic paybacks have been experienced in several different ways. These benefits include up to 40% reduction in off-spec, 100% reduction in steady-state offspec, optimization of reactor conditions and feeds with resultant reduction in catalyst consumption, 30 40% reduction in variability of product, and steadier operation with fewer upsets (with resultant reduced risk of plant outages). To take full advantage of the polymer production controller, the total solution delivered to SABIC also included access to extensive polymer process experience, which was incorporated into the controller design and, as such, contributed significantly to capability and sustainability. Real-time prediction and control of properties. In the SABIC polyethylene plant, the instantaneous and bed-average melt index and density are predicted on a continuous basis with updates every minute. These predictions are particularly valuable because they can be used across the entire operating range of the plant: steady-state operation, transitions, different catalysts and co-catalysts, to name a few. These predictions and a reliable calculation of production rate are input to the controller. In turn,
PETROCHEMICAL DEVELOPMENTS SPECIALREPORT A breakthrough in polymer production control technology Beginning with the deployment of linear-model predictive control in the downstream oil industry in the 1980s, polymer companies have experimented with advanced control. But polymer plants are not oil refineries. Polyolefin plants do not run to a single grade; the relationships are highly nonlinear and interactive. Process dynamics can last many hours, and throughput-dependent, controlled variables are often measured in a laboratory. Product-grade transitions represent some of the most safety, reliability and cost-critical procedures done in a manufacturing environment. Unfortunately for polymer manufacturers, this holy grail of advanced control hasn t existed because mathematical transformations simply haven t coped with the demands of optimizing complex transition strategies. In the early 1990s, a technology known as neural networks was viewed as a possible empirical solution to the nonlinear control problem. However, the inability of neural networks to safely predict sensible process gains in a manufacturing environment has exposed this technology as unsuitable for real-time control of a process plant. Although linear-model predictive control had limited success on gas-phase reactors as applied to gas-concentration control, the many manifestations of nonlinear control (for product qualities) have been fraught with difficulty. Initial attempts to take mathematical transformations of key process variables in order to linearize the control problem were an immediate failure due to the incorrect assumption that the nonlinear relationships between critical manipulated variables and controlled variables were independent of other conditions in the reactor. Polymer production control is a challenge. Nonlinear relationships of the process are multivariable. For example, the sensitivity between gas concentration and melt index is dependent on reactor temperature. An alternative approach to dealing with nonlinearities has involved using vast resources in building first-principle models of the process, creating complex mathematical models of reactor kinetics, heat and mass balances, and reactor geometry. This approach, however, not only requires an unnecessarily (and possibly unsupportable) high degree of expertise; it is a mathematical overkill for the requirements of an advanced controller. Only about 0.5% of advanced controllers in industry are based on first-principle models. Good rigorous models require a significant investment of both money and resources. (A good rigorous model can take many months to build; anything less is simply a correlation model and unsuitable for nonlinear polymer predictive control.) Long-term support also demands maintenance with a high level of expertise. To control a plant optimally, all that is actually required are accurate predictions of the controlled variables, process gains, process dynamics and process delays, and a sensible infrastructure for dealing with unmeasured disturbances. Anything more than this is superfluous. This is why empirical models are the method of choice for the majority of process manufacturers they re straightforward to build, maintain and support. A recent report from the EU task force mandated to advise industry on the application of so-called smart technology to safety-critical systems has recently advised manufacturers against using neural networks in real-time manufacturing control applications due to safety and integrity of operation issues. In this challenging environment, a new nonlinear polymer production control technology represents a true breakthrough. This system uses empirical models with the power to predict any multivariable nonlinearity, but with the unique capability of incorporating first-principle process knowledge into the model-building process. Result: A model with the speed and efficiency of an empirical model coupled with the constraints and reality focus normally associated only with first-principle models. The technology is transparent and uses the same terminology understood by all process control engineers. Hence, models are easy to build, support and understand. When building a polymer model, the engineer specifies process knowledge that is taken from process recipes, operational experience and known plant behavior. For example, such information may include: minimum and maximum sensitivity (process gain) between hydrogen and melt index. Other information may include the minimum bed turnover time at high throughputs and maximum turnover time at lower throughputs. A small amount of historical data is used to calibrate the model. (The same amount of data is required as would be required to calibrate a first-principle model, typically a fullproduct wheel.) Once this is complete, the engineer has a fully nonlinear, dynamic model. The model can represent steady-state nonlinearities, throughput (or operating point)- dependent process time constants, variable dead times and directional dependent nonlinear dynamics. (Sometimes it is faster to go down than it is to go up.) The models can predict sensible process gains (analytically constrained via first-principles knowledge) and can therefore be safely used within the kernel of a nonlinear controller. HP the controller controls these parameters at target and minimizes the variability around the setpoint by controlling multiple process conditions simultaneously (Figs. 2 4). The improvement results of this approach are very significant. The Gelsenkirchen plant experienced property variability reductions in the range of 30 40%. Likewise, the facility completely eliminated off-spec product that had previously occurred within production runs due to the inability to recognize and factor variations and drifts in different process conditions. The ability to precisely control production rate is another direct
SPECIALREPORT PETROCHEMICAL DEVELOPMENTS FIG. 2 Polymer production control transition reliability via nonlinear control solutions. benefit. At the SABIC plant, the capability to tightly control production rate will deliver a significant increase in average capacity. The plant can take full advantage of the reduced variability to operate safely and reliably near the process constraints. The total economics of this debottlenecking approach are more favorable and beneficial for ROCE than traditional equipment-based methods in which millions of dollars are spent on pipe, pumps and other equipment to achieve higher capacity. Accordingly, production control should be the first step in any debottlenecking strategy ultimately, the investment in pipe, pumps, etc., may not be required. Full automation and optimization for product transitions. The SABIC Gelsenkirchen polyolefins plant significantly reduced average transition off-spec quantities (40% reduction) and lowered the range of transition off-spec quantity outcomes by the different shifts (Table 1). Transitions are now executed automatically with the advanced control solution. This procedure applies the methodology used by the best operators. This precise duplication of best practices across all shifts decreased transition performance results the ultimate objective of plant management and the business. FIG. 3 CV and MV Constraints/Ideal Resting Values MVs and DVs FIG. 4 Comparison of predicted polymer properties during product change (break point on transition material). Economic drivers Full nonlinear steady-state optimization Full nonlinear path optimization Nonlinear model capable of making accurate predictions of process gains and full extrapolation capability CV and MV constraints Model update System architecture of the nonlinear control strategy for polymers. MV setpoints Maintain control during new product trials. As described earlier, the fundamental technology used is an accurate depiction of the real process interactions and relationships in the plant. The relationships are mathematically identified with a bounding methodology that delivers the ability to accurately interpolate across regions of sparse process data (transitions), and also to extrapolate into regions outside the identified operating range (new product trials). The extrapolation is based on first-principles process knowledge. With this fundamentally sound extrapolation capability in place, the models used to predict process properties can be extrapolated within reasonable limits to successfully predict property relationships and process gains and dynamics for products and operating regions outside the normal operating window of the plant. This extrapolation capability was used to strong advantage by the SABIC LD5 plant in which significant process improvements to the reaction system heat removal methodology were completed. The advanced control system was able to well function in
PETROCHEMICAL DEVELOPMENTS SPECIALREPORT TABLE 1. Polymer production control demonstrated project results Factor Demonstrated project results, % Capacity increase 2 10 Transition off-spec reduction 25 50 Steady-state off-spec reduction 50 100 the new operating region; thereby, keeping the plant at the most cost-effective and highest quality operating point without waiting on data to retune the controller. Lower cost of ownership. The straightforward singlemodel approach provides the capability for efficient and costeffective transportability and migration of inferential predictions and controller across multiple lines with similar process technology. For example, the migration of the polymer production control solution to a SABIC LDPE plant in Saudi Arabia will be able to fully leverage what has been implemented at the LDPE plant in Germany. This can be a major deployment for efficiency and cost savings in polyolefin companies that have similar process technology installed at different locations. Following an implementation at a primary site, the developed solution can then be rolled out with high speed and efficiency across multiple sites to achieve a strategic advantage at the business unit level. By installing the new software at multiple similar sites across the enterprise, advancements and new learnings of better ways to control and transition the process are developed at one location and can be e-mailed to other sites for direct incorporation into the software. Result: The sharing of best practices can happen swiftly. Testing. Reducing analytical testing is a significant driver in polyolefin plants for several reasons. The most significant factor is that polyolefin plants can no longer afford to wait on lab results as the decision factor for switching from off-spec to prime, or from prime to off-spec during transitions. Operators need realtime information to make the switch at precisely the right time, so that no prime product is wasted, and no off-spec is allowed to contaminate the prime material. The capability for precise switching is provided via the real-time accurate property predictions. Result: No more waiting on lab results. There is also the opportunity to reduce the frequency of lab testing by replacement of lab tests with the information provided by the reliable real-time predictions. For example, the SABIC plant has made a significant reduction in lab test frequency with the solution in place (from one analysis per two hours to currently just once per shift), and further reductions in test frequency are planned. Outlook. The SABIC plant in Germany has prepared to take full advantage of the approaching financial opportunities in the next polyolefins business cycle. Using new nonlinear advance control solutions, the polyolefins unit can run at higher rates with repeatable, reliable transitions and cut expenses for catalyst, monomer and lab analysis. With operational excellence performance foundation in place, this facility can build upon best practice capabilities to further improve quality, reliability, productivity and cost performance. HP ABOUT SABIC POLYOLEFINE GMBH SABIC Polyolefine GmbH is part of SABIC EuroPetrochemicals. SABIC EuroPetrochemicals operates production sites in The Netherlands and in Gelsenkirchen, Germany. SABIC Polyolefine GmbH produces polyethylenes and polypropylenes. Oktay Karagoz has worked with Foxboro as application engineer for over nine years. He joined SABIC Polyolefine GmbH, Gelsenkirchen, Germany, in 1998 (at that time the facility was called DSM Polyolefine) as process control engineer on the LD5-Project. Mr. Karagoz participated in the complete APC-implementation project of this plant and as such, has in-depth experience with operational and maintenance aspects. He holds a degree in electrical engineering, (measurement and control technology). Jan Versteeg joined DSM in 1986 and has held many positions in process control, engineering, production and IT management. He was the project leader of the LD5-APC-project. Mr. Versteeg is now with the Competence Center of SABIC- EuroPetrochemicals BV in Geleen, The Netherlands. Mr. Versteeg holds degrees in mechanical engineering, (measurement and control technology) and process technology. Michael Mercer is the advisor to the AspenTech Polymer Team. He joined AspenTech in 2000 with 25 years of experience in the polyolefins industry with major chemical companies. His industrial experience includes leadership roles in the design and implementation of new polymer process technologies. Because of his extensive experience in polymer plant operations, he has a solid and in-depth understanding of both the day-to-day issues and challenges in a polymer plant, and the various polymer process technologies currently in industrial operation. Paul Turner is a senior technologist at Aspen Technology Inc. He has a PhD in advanced nonlinear control theory and has been involved in industrial nonlinear production control projects for the past 12 years. He is one of the principal developers of AspenTech s nonlinear production control solution. ARTICLE COPYRIGHT 2004 BY GULF PUBLISHING COMPANY. ALL RIGHTS RESERVED. PRINTED IN U.S.A.