Putting More Control Engineers in the Loop
By Pulp & Paper Canada
By Pulp & Paper Canada
The pulp and paper industry needs to follow the example of the information technology industry and invest in people. “One process engineer can cost you about $75 000 but you can save a million dollars…
The pulp and paper industry needs to follow the example of the information technology industry and invest in people. “One process engineer can cost you about $75 000 but you can save a million dollars,” said Frank Dottori, Tembec president and CEO. “The IT industry invests in people.” This industry wants to put in systems and get rid of people, he said.
“A good control engineer and some competent equipment,” said Dottori, and you can save a couple million in capital. “The technology — I know it’s there.”
Dottori was addressing the 138 delegates that attended the Control Systems 2000 conference held May 1 to 4, 2000, in Victoria, BC. Sixty-eight papers and posters were presented, as well as three plenary sessions.
At the seminar’s first plenary session he said that some of the same issues apply as when he started in the industry in the early 60’s.
However, one of the greatest pressures now facing the industry, he said, is financial performance. “Investors think it is a dead skunk.” According to PricewaterhouseCoopers’ 1999 Forest Industry in Canada report the rate of return on capital employed (ROCE) over the last four years has been below 4% for most companies. To encourage shareholders to maintain their stake in the pulp and paper industry the industry must improve its profitability. Process control is now seen as one of the success factors needed for the improvement of ROCE.
Environmental accountability is also important. “We got 98% of the blame for dioxins even though we only produce 2% of the problem.”
Other pressures: Excess capacity, lack of fibre, customer demands. The industry needs to improve its use of fibre resources, decrease waste, reduce costs and improve quality. There are “intangible benefits of these actions” said Dottori, such as an increased public relations image. “Every one wants to be associated with a winner.” Improved customer relations also follows. “The fact that you are a winner helps.”
The industry needs to take advantage of untapped resources. He discussed the benefits of chemometrics, broadly defined as statistics for chemistry. This area of expertise is being used today in the pharmaceutical, food, cosmetics, petrochemical and now in the pulp and paper industry.
“Chemometrics is a visual tool. It reduces troubleshooting effort, provides process understanding.” Chemometric tools allow us to see how our processes work together as a system instead of individual parts said Dottori.
“Process simulation and design is a quick optimization tool…. It’s low cost, avoids design errors…. When you sell your project, put it in economic terms.
“We have to really address this issue of analysis. I think that’s where the big bucks are. This industry has to move up the food chain — a lot of people in Europe are way ahead in this area.”
Dottori said that the Canadian industry needs to “show the rest of the world that we are not the backward bumpkins that we are perceived to be.”
Dilution of talent
The industry as a whole also has a problem of brain strain. In the second plenary session Karl strm, Lund Institute of Technology, questioned how the pulp and paper industry could compete with high tech and knowledge industries to attract the highest calibre candidates. How to recruit and stimulate skilled engineers to work at remotely located mills, he asked.
Mergers in the industry as well as among the system suppliers also result in the dilution of the talent. A team consisting of an experienced instrument engineer and a good graduate control engineer can have a major impact. It is difficult to maintain them in the turmoil of many mergers. strm said that consultants and system vendors offer the knowledge services, but it is still difficult to find the right combination of skills.
The third plenary session also talked about improving the knowledge in the pulp and paper industry, but this time through means of multivariate analytical techniques.
Svante Wold, Ume University, said that some of the benefits of multivariate analysis (MVA) are that
It gives an excellent overview of data and whole processes;
Troubleshooting is good;
You get good early fault detection.
He gave a history of multivariate modelling and its developments dating back to the work of Descartes in the 1600s up to recent industry developments. “Processes today have big problems,” according to Wold. “There are too many demands and too little resources.” The processes have more complicated requirements and because of the number of controllers there is an abundance of data. “We drown in the data” or else throw it away. MVA uses the data as points and describes it mathematically and statistically. The increasing masses of process data measured in this industry should be analyzed properly, and used for the purpose of process diagnostics and the improvement of the processes and of our understanding of the processes.
For the future Wold said that we would need something superior to MVA. We will need easier ways to interpret data and easier ways to present this information.
A study in control
The benefits of control were discussed in the first session. Fraser Forbes, University of Alberta, said that his study tried to systematize benefits analysis and find a way to evaluate and communicate the real worth of process improvements.
A six-step procedure for estimating the expected performance increase is proposed.
Similarly, Tom Marlin, McMaster University, presented his six-step plan to evaluating benefits. The study determined the potential benefits from improved automation in the process industries and the appropriate steps to achieve these benefits. To evaluate the process, he said, you first need to define the goals of the study. Once that is done there are six “rules”:
1. Interview plant personnel.
2. Establish economic benefit. The people at the control levels do not always know the economic value, he said. Obtain information at a higher level.
3. Evaluate base case performance. Base case process performance as evaluated from the data may be very different from impressions of the plant personnel.
4. Generate opportunity list and trim to the best.
5. Calculate the benefits of an individual performance.
6. Select controls for implementation. Do not assume a) that a fixed percentage can be saved, b) that variance can be reduced by a fixed percent, c) that optimum steady state can be achieved, or d) that the best historical operation can be achieved.
Benefits studies would profit from a better working relationship between engineers and management. Managers should not expect risk free investments, but should expect engineers to provide a thorough quantitative analysis.
Bill Strand, Pacific Simulation, presented a study on the Economic benefits from advanced quality control of TMP mills. The objective of advanced quality control (AQC) is improved economics of the papermaking. Strand said that TMP use in newsprint mill consumes more energy today than 20 years ago, but makes better pulp.
TMP advanced quality control was implemented at Ponderay Newsprint Company: Mainline refiner control of freeness and long fibre; reject refiner control of freeness and long fibre; screen room control of freeness and long fibre; and post refiner control of freeness and long fibre. Strand said that one economic quantifiable benefit was the decrease in specific energy consumption, resulting in about a $400 000 annual savings. Kraft use was decreased and paper machine speed was increased.
Continuing the theme, Seyan Nuyan, Neles Automation, also spoke about Harvesting the benefits of process control stressing the importance of good documentation and knowledgeable personnel. Projects are poorly documented, he said. Process and control understanding is insufficient. Improper and malfunctioning equipment are used. “These are problems that need to be solved.” Nuyan cited an example of a machine start-up optimization package to be installed on an off-machine coater.
“A program to monitor and document the start up performance was implemented, and limiting factors for each were d
ocumented,” said Nuyan. After the evaluation period the economic benefits were evaluated.
Nuyan said that benefits have to be properly documented using appropriate tools. Process and control understanding is key. Control experts with logging and analysis tools and skill are fundamental.
Part of the process
Antii Nissinen, Neles Automation spoke about removing the obstacles of process control, interpreting the information that you gather and using it profitably. Nissinen spoke of a process modifications and observations in an SC paper mill in Scandinavia. A new dilution headbox was added with a variety of sensors for the process.
He said that control alone cannot ensure stability. Suitable design and operation are needed. Multivariable control is required, incorporating wet-end consistency and ash measurements.
Hong Cui, Royal Institute of Technology, in Stockholm spoke about the Optimum design of buffers in plants with recycle. The sponsors of this project, said Cui, looked at how system closure affects the system dynamics and looked at adding buffers for control purposes, “to smooth the process access”.
By cascading a buffer with the plant, the buffer design can be applied to a plant with recycle.
She concluded that buffers could be designed to handle disturbances that cannot be effectively handled by feedback control system. Buffer location is a crucial decision with respect to required buffer size.
Mohamad Masudy, Tembec Industries, presented a paper on integrated approach for design of blow heat recovery system, using simulation and design of experiments. Design of Experiment (DOE) methodology “enables you to manipulate many variables simultaneously and still interpret the results. This methodology is used at Tembec to identify and model actual processes.”
Based on the results presented in Masudy’s report, the combined heat exchanger and pumping capacity was shown to be the most significant factor in meeting the process design goals. This requires increasing the hot condensate pump capacity, the heat exchanger capacity, and the fresh pump capacity.
Donald Brewster spoke about process control and the optimization of a stone groundwood process that was implemented over several years at Champion International mills.
Modelling of process dynamics is a critical step in project success, according to a presentation by Andrew Waite, EnTech Control Engineering, who developed a dynamic simulator for a project at Domtar Eddy to replace a headbox with a dilution headbox. Modifications began in 1996 and start-up in 1998.
To verify the simulator’s accuracy the project looked at control tuning, pumps, control valves, calibration, process dynamics and white water dynamics especially during transition.
The final simulation consisted of 70 000 VisSim blocks and 400 differential equations. The simulation design calculation took into account the supply side of the valve and the discharge side of the valve.
“The dynamic simulation got good results,” said Waite. It predicted problems that were later corrected. The model development took six to eight months, and was developed at the same time as the mill was gathering data.
Digester & bleach plant
Gerry Pageau, Howe Sound Pulp and Paper, spoke about two-stage oxygen delignification controls optimized for constant kappa output. A mathematical model for the delignification process taken from the literature has proven to be a useful tool for troubleshooting a mill-scale system, Pageau reported. It has also provided the basis for model reference control.
An inferential control strategy was used to determine the effect of chip level variations on the final pulp quality in a Kamyr digester. Effective alkali and chip level measurements are used as secondary measurements for the estimation of the Kappa number.
Jay Lee, Purdue University, said that there are many causes for chip level variation. Appropriate response strategy requires that the source be known. Controlling chip level alone will not ensure kappa number control. Three different strategies for chip level control were found. Whatever chip level control strategy one adopts disturbances will cause significant kappa number variations. There are limitations on relying on chip level control for kappa control.
Stephane Renou’s presentation looked at nonlinear control design for pulp bleaching. A chlorine dioxide reactor was modelled by mass balances, leading to a partial differential equations system. Using global differences, an approximate input/output model was obtained and this model was used to design a nonlinear control law. The non-linear controller, when compared to dynamic matrix control (DMC), shows better feedforward behaviour. It also showed better reduction of lignin input variability since its feedforward capacities are not affected by a penalty on control action as DMC.
A session on PID tuning presented three new methods. Lahoucine Ettalab, Paprican, said that poor loop performance is generally associated with one or more of the following defects: defective actuator or sensor; poor control tuning or; interaction with other poor performing loops. The processes considered are modelled by first-order-plus-time-delay approximations, and controlled by PI controls.
We have, he said, derived a simple second-order approximation model for feedback systems where the process is described by a first-order-plus-delay transfer function controlled by a PI control. The tuning method provides a quick fix for loops that are poorly tuned. The method uses operating data and can be combined with on-line performance monitoring tool.
S. Ogawa presented a new systematic approach to design and tuning of VPC assuming both valves are controlled by PI controllers. The new approach was applied to a saveall stock tower consistency control loop with good results.
PID-type controllers were used “for practical reasons.” The major and minor loops were analyzed as a unit using the process transfer functions and robust tuning rules. The analysis produced a new systematic methodology to the selection of controller structure and tuning.
This methodology was applied to a saveall storage consistency control loop, which sometimes oscillates due to a long time constant in the major dilution loop. The major and minor dilution processes were identified with experiments where each valve was manually adjusted. The two PI controllers were tuned based on robust tuning rules and process models. The major dilution loop was changed to a PI type from a dead-band I. The new configuration worked well, and the oscillations were eliminated for a wide range of operating conditions.
Hlne Panagopolous’ presentation looked at a new interactive tool for system identification and controller design. The tool has been applied to a starch boiler and to a steam pressure control in a drying section of a paper machine at the pulp and paper company Modo Paper in Husum, Sweden. Panagopolous said that there was a demand for an easy-to-use tool that matches the needs and knowledge of process engineers.
The tool is based on open loop step response, and it has pre-defined model structure. The user can add the step responses by dragging on the screen.
The existing controller for the boiler was a PI controller. The control problem was to regulate the temperature. PID control was proved to be better as variance was higher with PI.
During the questions and answers period following the session one delegate asked the presenters to check if these models actually work with backslash and stiction. Otherwise, he said, it is a pretty nice theoretical work but not very useful in the mill.
Methods of monitoring
A session on control performance and monitoring featured two papers on performance, two on CD control and one on the problem of diagnostics. Among them: Alexander Horsch, presented a simple method to diagnose and characterize oscillations in process control loops. This new method is based on the cross-correlation between control input and process output. It has been shown to correctly distinguish static friction (stic
tion) in control valves from other sources of oscillation. The most common sources are external disturbances and too tight tuning in connection with nonlinearities. The method should be used, according to the researchers, for self-regulating processes with non-compressible media
Ferhan Kayihan, IETek Integrated Engineering Technologies, spoke about an evaluation of performance index for CD control. He performed a practical assessment of CD control performance for a fine paper machine. His challenge was to estimate the best control feasible given scanner data and other process control information.
Stephen Duncan, Oxford University said that he wanted to develop a system that takes basic CD control feedback loop and builds a system that takes actual performance input and comes up with optimal response with the stated disturbances.
Model-based methods give a more realistic view, particularly where input constraints are taken into account.
Another study sought to improve paper machine productivity through increased insight into operating conditions. The study, presented by Michael Davies, University of British Columbia showed that the use of wavelets and wavelet packets in paper machine data analysis was effective in determining the controllable and uncontrollable variations in the cross machine direction. Process monitoring and control performance assessments were achieved by first separating controllable and uncontrollable variations in the cross machine direction profile using multi-resolution analysis. The performance and the potential of the system, according to the report, were then evaluated using both wavelet and wavelet packets and the results were compared. Some advantages are that wavelet packet analysis can achieve an accurate separation of CD controllable and uncontrollable variation and control performance assessment. A control performance index can be calculated on-line to provide the operator with a quick assessment of the process.
Jorge Castro Velez, University of Delaware presented a control study of the fibre line of a pulp mill. Cascade control using low-level feedforward control for disturbance rejection is coupled with PI control for setpoint tracking. This method is compared with model predictive control to take advantage of the remaining manipulated variables for better control and operation of the process and to obtain the best possible plantwide strategy for the process.
In another study PLS modelling alone and in combination with physical modelling are compared for successful grade change automation. The hybrid model is applied to the modelling of paper web moisture and the results are compared with actual process measurements of grade changes.
The hybrid method is a good way to model a nonlinear process if small data sample size is advantageous and a simple physical model can be constructed. The benefits from this approach are short commissioning times and easy portability to different paper machines.
Maryam Khanbaghi, Paprican, proposed a multivariable, nonlinear controller for pressure screens. Most mills, according to the study, use linear univariate control for the pressure screen. Increasing demands in both contaminant removal and fibre fractionation applications require that screens run as close to failure limit as is possible. This is not possible using the current approach because of the nonlinearities and coupling in the process. The proposed multivariate controller performs better than linear univariate controllers, enabling tighter control of the pressure screen. The performance of the controller was evaluated in simulation and in pilot plant experiments.
Ahmed Ismail, University of British Columbia, presented a dual adaptive controller to regulate coat weight on bent blade coaters. The work proposes a model of the coat weight /blade angle nonlinearity that encompasses the effect of blade wear.
The researchers believe this to be the first time such a model has been developed. A dual adaptive coater is applied to the coating process. The controller is coupled with an adapted Kalman filter, which has been used successfully to as an improved parameter estimator to track gain variations. Simulations have been successful, and mill trials of the strategy were underway at the time of the conference, to be presented at a later date.
An MPC management strategy to reduce the variation of the pulp properties by adjusting the broke ratio was proposed in a work presented by Michel Perrier, cole Polytechnique de Montral.
Implementation of the MPC management strategy should improve the paper machine runnability and stability and reduce the variability of the paper properties.
Preprints of the proceeding can be obtained by contacting PAPTAC at 514-392-0265.
The next Control Systems conference will be held in co-operation with SPCI. The dates are June 4 to 6, 2002 in Stockholm.