|
Becoming a low-cost, globally competitive mill requires more than just having the best available equipment. It also requires having the most thorough understanding possible of the endless array of process variables that affect sheet quality. Without such an understanding, changes to a variable that improves one aspect of sheet quality may cost dearly in terms of other sheet properties and overall production expense.
To make its processes as cost and quality efficient as possible, Boise Cascade's Jackson, Ala., mill has used an optimization method known as Multivariable Testing (MVT) since 1996. After a difficult recycle plant startup, the mill began using the eight-step MVT method devised by one of its consultants to improve plant performance.
Predominantly accomplished by mill operators and total quality managers, this method has now been used throughout the mill with great success. One recent MVT project was an optimization of the Jackson mill's state-of-the-art J3 uncoated free-sheet machine that was installed in 1997. This project improved CD stiffness and formation while lowering costs. In the pulping area, a fiber yield optimization has the potential to improve pulp brightness while also achieving costs savings.
|
FIGURE 1. Multivariable Testing (MVT) drives the process outside control limits set for the test to determine which factors affect the measures targeted for improvement, such as CD stiffness.
|
MULTIVARIABLE TESTING: AN EIGHT-STEP METHOD. Multivariable Testing is a statistically-based testing methodology stemming from a branch of mathematics known as experimental design. The specific eight-step method used at Boise Cascade was developed by a consulting company called QualPro, and it has been used for the past 20 years for improvements in a variety of industries, ranging from defense to process industries such as chemical and paper.
Using the traditional scientific method of changing one variable at a time and holding all others constant can result in an unrealistic number of test cases. For example, 10 variables designated as relative to improvement result in more than one million unique combinations, any one of which could bring the desired outcome. With MVT, statistically-designed "recipe" combinations limit the number of test cases to 12 in this example, selecting the best combination of variables for improvement.
J3 PAPER MACHINE OPTIMIZATION.Installed in 1997, the J3 uncoated free-sheet paper machine at Boise Cascade's mill is state of the art. Compared with machines of its 352-in. trim size, it had the fastest paper machine startup on record at 3,333 fpm. Since then, it has continued to set records, including a North American one of 4,101 fpm for printing papers in 1998.
|
TABLE 1. The J3 efficiency team created a list of 10 factors affecting CD stiffness, formation, fiber orientation, and image deletion. The minus (-) column represents the value typically used, while the plus (+) column shows the value that could improve key measures.
|
Despite the paper machine's success, Boise Cascade felt more improvements in the area of a CD stiffness and formation were attainable, even though the paper was meeting the required quality specifications. Total quality manager Joni Mazur reports that, due to the success of the QualPro MVT method in the recycle plant, the "mentality just exists that we should try MVT as a structured way to get accurate, data-proven results instead of trial and error."
In July 2000, Boise Cascade began applying the eight-step MVT method on the J3 paper machine, as the following sections describe.
Step 1: Create the environment. Creating the right environment for the J3 optimization project involved a combination of management support, identification and training of the efficiency team members, and the incorporation of team member experience into the project. According to J3 operator and efficiency team member Mike Whitehead, Jr., management support is "critical" due to the possibility of losing paper during the trials; but, with the MVT success in the recycle plant, support was solid, all the way to Richard Merson, vice president of Alabama operations. However, the plan presented to management stated that any one trial would be aborted if two reels (around 100 tons) of paper were lost.
The J3 efficiency team consisted of 10 members, including two wet end operators, four dry end operators, two winder operators, an information services representative, and a member of management functioning as team sponsor. However, Whitehead is quick to point out that the group mainly functions as a steering committee, since "everybody on the mill floor has input."
Training for MVT trials has evolved into a one-day orientation course in the eight-step method. Although the mill initially used a two-week intensive course on both the eight steps and the statistics supporting MVT, total quality facilitator Jerry Yarbrough says this approach resulted in team members "either forgetting it if the information was not used right away or becoming bogged down in statistics."
With the one-day orientation, team members work with a consultant on the eight-step procedure for their project in a hands-on fashion using their own data. The one-day training approach works, since the Jackson mill now has four years of experience using MVT. Also, Mazur and Yarbrough have gone through extensive training with the consulting company to help provide internal facilitation. In addition, QualPro consultant Bill Thompson visits the site every four to six weeks to check progress and help with issues.
Step 2: Define key measures and operational definitions. What did the efficiency team want to improve in the J3 process? The primary key measures were CD stiffness and formation. In addition, secondary key measures were defined as fiber orientation and image deletion (lack of ink absorbency resulting in a non-printed area).These secondary measures were targeted for observation, not necessarily improvement, to ensure that changes benefiting the primary key measures would not compromise other sheet parameters. Standard definitions of key measures were provided to all efficiency team members and operators to ensure consistency.
Step 3: Evaluate the measurement system. Were the systems used to measure the primary and secondary key measures at the Jackson mill reliable? The efficiency team determined they were. To evaluate formation and fiber orientation, an AutoLab system was used, working from a machine-width strip of paper off each reel produced on the J3 machine. Calibrated weekly by lab personnel, the measurement variation for this system was negligible.
Variations were higher for measuring CD stiffness and image deletions. Image deletion measurements, taken at Boise-Cascade's associated converting facility in Jackson, Ala., varied by 25%. CD stiffness measurement variation ran a bit higher at 33%. According to J3 operator and efficiency team member Danny Smith, these variations were operator induced, because it is often difficult to precisely cut the 1-in. by 1.5-in. strips that are fed into the Gurley stiffness meter. During the twenty years that the mill's consulting company has applied MVT methodology, it has been found that measurement variation in the range of 30% or less is acceptable for MVT success.
Step 4: Examine process stability. To determine if the J3 process was stable before proceeding with the MVT, the efficiency team gathered data relating to the primary and secondary key measures and calculated control limits for them. With the exception of one point related to CD stiffness (Figure 1), all data fell within the control limits for the key measures, showing a stable process before the MVT runs began. Yarbrough notes that these control limits, calculated with actual process data, were well below actual specification reject limits.
Step 5: Identify special causes and correct them. To identify the cause for the CD stiffness data point outside the control limit, the efficiency team examined trending data with the mill's PI System and Bailey DCS for the J3 machine, looking for furnish changes or other factors impacting the limit violation. No apparent cause was found, so corrective action was not possible, nor deemed detrimental to the MVT, since this was only one special cause identified in a large number of data points. However, Smith designates the inability to trace the source of the stiffness variation as "very uncommon."
Step 6: Look for easy fixes. Were there any simple or obvious changes that could bring about the required improvements in stiffness and formation? The MVT method used by the mill defines such easy fixes as "gimmes." Although none were found for the J3 optimization project, Smith describes such a "gimme" on another MVT project associated with the J3 machine.
"Our first project as an efficiency team for the J3 was fairly simple-to reduce our lineal footage variation," says Smith. "We found that we were getting high/low because we had a caliper variation, so we lowered the set point and reduced our lineal footage standard deviation almost 500 ft."
Step 7: Determine if process is capable of meeting requirements. The efficiency team had to determine if the J3 process was capable of meeting actual specifications for the primary and secondary key measures. To do so, the mill had to calculate capabilty, or the Cpk index.
|
FIGURE 3. An interaction between two test factors-caustic strength at 15% and using Eop filtrate for the D top shower supply-improved DN brightness by 1.6 points.
|
The Cpk index looks at the spread in the process data and the distance of the process average from the specification. When evaluating this index, data should be normally distributed, and the process should also be deemed stable, which the team had already determined. Cpk takes into consideration where the process is centered by evaluating the upper half of the process distribution against the upper specification (Cpu) and the lower half against the lower specification (Cpl). The Cpk value is the lower of the two values (Cpu, Cpl).
Generally, a process is not considered fully capable until a Cpk of 1.33 is achieved. This means that at least a safety margin of one sigma exists between the tail of the distribution and the closest specification limit. The efficiency team discovered that the system was capable, but only marginally so, since Cpk for CD stiffness was only 1.0. But, since the system was capable, why experiment with it as described in Step 8?
"We chose to proceed because we wanted our process and quality the best they possibly can be, not just marginal," describes Smith. "When situations such as the closing of the International Paper mill in Mobile occur, it really hits close to home. We want to focus on continually improving."
Step 8: Experiment with the process. The J3 efficiency team brainstormed and gathered information from machine operators to create a list of factors affecting the primary and secondary key measures of CD stiffness, formation, fiber orientation, and image deletion, coming up with about 20. To further narrow the list to the 10 factors shown in Table 1, the team looked at whether or not the factor was cost effective, feasible, and practical.
|
TABLE 2. For the fiber yield project, the team listed 12 factors that would potentially improve DN and D brightness and cost per ton. The minus (-) column represents the value typically used, while the plus (+) column shows the value that might improve key measures.
|
As Table 1 shows, the ten factors were: ash content, percentage of recycle, jet-to-wire, slice flow, basis weight profile, reel moisture, headbox tilt, basis weight, hardwood freeness, and starch pick-up. In the table, the minus (-) column represents the value typically used to run the machine, while the plus (+) column shows the value that could improve key measures, as determined by team members and operators.
Because of its statistically-based method of eliminating combinations, the MVT allowed Boise Cascade to run just 24 combinations of the 10 factors affecting key measures, eliminating those combinations that would yield insignificant results. The various combinations were run randomly, as if chosen from a hat, and data from each was gathered from two test reels. On average, another two transition reels were run in between combinations. All 24 runs were completed within just three and a half days during August 2000.
"To have completed this in such a short timeframe is a tribute to the training we have had and to the operators, who know what to look for and what we need when they see us on the floor," says Smith. "They don't wait until the next day to ask somebody what to do; they react immediately."
Figure 1 shows CD stiffness data before and during the MVT runs. As Smith describes, "you want to drive your process outside the control limits to see what factors are affecting you positively and negatively." Despite this, all paper produced during the trials was within specification and salable, he says. A Pareto chart from the MVT run data showed that four factors, called main effects, improved the primary key measure of CD stiffness. Figures 2a-d show the effects of the following factors on CD stiffness:
• Ash 14%
• Percent recycle 10%
• Jet-to-wire 1.04
• Basis weight 50.8 lb
Pareto analysis showed only one main effect for the other primary key measure of formation-an ash content of 14%. By lowering ash content from 16%, formation improved almost 4.5 points.
In addition to the main effects impacting CD stiffness and formation, there were four interactions between two of the test factors that produced formation improvements:
• Jet-to-wire = 1.04 with slice flow = 33,300 gpm
• Starch pickup = 65 lb/ton with basis weight = 50.3 lb
• Jet-to-wire = 1.06 with hardwood freeness = 475 to 525
• Reel moisture = 4.3% with percent recycle = 25%
According to Yarbrough, discovery of such multi-factor interactions is a bonus that you would not find testing one factor at a time. Operator Smith notes that these interactions are "pretty amazing from a practical papermaking standpoint-you just don't think that way when you are working hands-on with the paper machine." However, years of experience with applying MVT in process industries such as pulp and paper have shown that higher-order interactions involving three or more variables usually produce statistically insignificant results, according to Thompson.
As hoped, the secondary key measures of fiber orientation and image deletions were not negatively impacted during the trials, functioning as intended to ensure that quality was not compromised during the 24 runs.
So, with a variety of factors identified for potentially improving CD stiffness and formation, which ones would actually be implemented as a permanent part of the J3 papermaking process? The team next assessed the cost impacts. Of the four main effects improving CD stiffness, only a jet-to-wire setting of 1.04 would not impact cost, and it was this discovery that surprised the team.
"As papermakers, we knew that lowering ash and recycle content would improve stiffness, just as raising basis weight would," explains Smith. "But lowering the jet-to-wire setting was the one that we did not expect to help us, and the trials showed it has as much impact on stiffness as reducing ash content. It turned out to be the one with no associated cost."
With all four main effects for CD stiffness implemented, the Cpk index would rise from 1.0 to 4.6. However, with the cost-free jet-to-wire change, the index would still significantly rise to 3.0.
At the same time, the J3 efficiency team was examining the potential changes to improve formation, including the main effects and the multi-factor interactions. Again, lowering ash content was cost prohibitive. Also, a jet-to-wire of 1.06 in one interaction contradicted with the 1.04 setting that would improve stiffness. Lowering reel moisture and starch pickup, factors in two other interactions, also impacted cost. However, a jet-to-wire of 1.04 with slice flow at 33,300 gpm could provide the needed formation improvements without financial impact. By implementing the jet-to-wire/slice flow combination, Cpk for formation jumped from 2.2 to 3.9.
In addition to improving quality, the jet-to-wire and slice flow changess have saved considerable dollars.
"We knew we were going to have to improve CD stiffness, and we thought lowering the ash content was the way to go," explains Smith. "That would have cost us approximately $2 million/year for the extra fiber, which is what the MVT saved us by discovering the jet-to-wire adjustment."
APPLYING MVT TO IMPROVE FIBER YIELD. When Boise Cascade began applying the eight-step MVT method in its pulp mill, the goal was to provide 800 a.d. tpd of bleached stock consistently and cost effectively while maintaining quality standards and complying with environmental regulations. The single-line pulp mill supplies virgin pulp to both the J1 and J3 paper machines. The single line pulp mill swings between hardwood and softwood depending on storage tower inventories. The pulping process at Jackson includes six batch digesters, five brown stock washers, and a four stage bleaching sequence, D-Eop-Dn-D. The furnish recycle content comes from an on-site recycle plant.
Step 1: Create the environment. As with the J3 efficiency project, the fiber yield project enjoyed full management support at the Jackson mill, including vice president of Alabama operations, Richard Merson. Help was also available from the total quality department at Jackson, the mill's MVT consultant, and representatives from information services, accounting, process control, and the technical department.
The core fiber yield team was made up of 12 members from various functions throughout the pulp mill, including operators from the bleach plant, brown stock washing, digesters, causticizing, recovery, and wood scaler. Also, a woodyard crew leader, pulp mill crew leader, pulp mill engineer, and assistant pulp mill superintendent were part of the team. From the beginning, the team realized that this was a very large project. Yield can be affected all along the fiber line from the woodyard through the bleach plant, in some areas more than others. Since the mill was in the process of replacing their woodyard equipment, the team decided to divide the project into pieces and experiment their way from the bleach plant to the woodyard. The project in this article is the first of at least three screening experiments.
|
FIGURE 3. An interaction between two test factors-caustic strength at 15% and using Eop filtrate for the D top shower supply-improved DN brightness by 1.6 points.
|
Step 2: Define key measures and operational definitions. Key measures were defined as DN brightness (next to last stage), D brightness (last stage before HD storage towers), and cost per ton (cost/ton).
Step 3: Evaluate the measurement system. With brightness measurements showing just a 10% variation, Yarbrough reports that the first reaction was to think that something was "either great or very wrong." The former proved to be true, and the low measurement variation was attributed to well-trained technicians and ISO-documented calibration procedures.
As for evaluating costs, the fiber yield team relied on help from the mill's accounting department, where costs were compared to actual accounting figures. However, the end-process tonnage measurement was not reliable, since the mill had experienced problems with this consistency meter in the past. After conducting a measurement study that showed a 149% measurement variation with the old meter, mill management replaced it with a new one, which proved to have just 35% variation.
Step 4: Examine process stability. The team gathered data relating to the key measures and set control limits for them. After monitoring brightness data from April 2000 to June 2000, the DN and D brightness stayed well within the control limits calculated for the fiber yield MVT key measures.
Step 5: Identify special causes and correct them. Although there were no limit violations observed when monitoring brightness, there were nine points in a row above the average. However, after reviewing process data from the PI system, the team discovered that high-bright pulp was run that day, so the data anomaly was deemed a normal occurrence, and no corrective action was taken.
Step 6: Look for easy fixes. As with the J3 efficiency project, there were no easy fixes related to improving brightness and cost/ton. As Yarbrough notes, "you won't find many of these Œgimmes' if you know how to manage your systems."
Step 7: Determine if process is capable of meeting requirements. Cpk for DN brightness and D brightness was approximately 0.70, which was well below the 1.33 designating a capable process. In addition, Cpk for cost/ton was even worse at 0.25, so the fiber yield team had an excellent chance to improve the key measures with MVT.
Step 8: Experiment with the process. During its initial brainstorming session, the fiber yield team came up with 73 factors that might improve its key measures. Additional factors were also gathered from other pulp mill operators. However, after the team analyzed the factors for practicality, cost efficiency, and feasibility, it came up with the list of 12 shown in Table 2.
The factors included species, consistency of brown stock going to the bleach plant, caustic strength, D top shower supply and caustic application, Eop temperature, peroxide, oxygen, Eop pH control location, DN temperature, and D temperature. The team also identified the values typically run (shown in the minus column) as opposed to those values that might bring improvement (shown in the plus column).
By applying MVT, Boise Cascade was able to run just 24 combinations of the 12 factors affecting the key measures of DN and D brightness and cost/ton. The various combinations were run randomly, and pulp for each trial ran through the pulp mill for at least 10 hours. Combined with eight hours of transition time, each trial took a minimum of 18 hours. Overall, this portion of the fiber yield MVT took one month, which Yarbrough describes as "pretty good for a process like this."
So, what were the results? Pareto charting of the MVT data showed only one factor, or main effect, for improving the key measure of DN brightness. Boosting brightness by 2.5 points, this main effect was the running of 100% hardwood, which is understandable, though not feasible, since hardwood bleaches more easily than softwood. Also, an interaction between two test factors-caustic strength at 15% and using Eop filtrate for the D top shower supply-improved DN brightness by 1.6 points, as Figure 3 shows.
For the key measure of D brightness, two main effects were found. Maintaining peroxide at 10 lb/ton improved this key measure, but only by 1.1 points. Also, maintaining caustic strength at 15% improved D brightness by 0.9 points. One interaction-changing Eop pH control from before mixer to after and keeping D temperature at 170 degrees F-improved the key measure by 1.2 points.
Pareto analysis showed two main effects for improving cost: hardwood species 100% and using no peroxide.
Based on the results, the team recommended four factors for further investigation and refining:
• No peroxide
• 15% caustic strength
• 15% caustic strength and D top shower supply with Eop filtrate
• Eop pH control after mixer and D temperature at 170 degrees F
According to Yarbrough, the potential savings for the entire fiber yield project could amount to $5.1 million/year. This estimate was derived from the mill's Cost of Quality model comparing bleached pulp production in 1999 with an optimum benchmark of 800 a.d. tpd. For example, in the bleach plant, using no peroxide saves over $500,000/year. Although peroxide use does improve brightness, Yarbrough notes that "for just one point of brightness, you have to wonder if it's worth the bang for the buck." Also, using the Eop filtrate as D top shower supply instead of fresh water along with the 15% caustic strength would save another $100,000/year, as well as boost brightness back almost to the point of using peroxide. The mill does, however, plan to keep peroxide at the bleach plant for its high-bright grade.
|
