"Dynamic trimming" could incorporate such factors as customer requirements, machine limits, and downstream equipment availability to optimize total production

Can Mills Use Computer Power to Trim Rolls More Efficiently?

By Paul W. Lail

The trimming process in a paper mill, where winder cutting patterns are generated from the mix of orders to be manufactured, has primarily been considered a one-time, user-initiated event. For the typical trim solution, a production planner will execute a computer-based linear programming (LP) or heuristic trim algorithm after all orders in a manufacturing block are known and as the actual production date approaches. Most planners trim between one and four days in advance of the manufacture date. Trimming too far in advance risks rework if customers change ordered quantities or sizes.

Trim algorithms, because of their computational nature, consume large amounts of computer processing time. In the past, running a trim algorithm was slow and could affect the performance of an entire system. As processor speeds have increased, and with the advent of the personal computer and distributed computing environments, the computer resources consumed in running trim algorithms are now largely inconsequential.

Many vendors of trim algorithms now offer their packages on personal computers or can install a separate server where trim problems run independent from the overall transaction system. And yet, most planners still run trim algorithms once per manufacturing run, with the functional designs of the supporting systems encouraging this "one-time" trimming approach.

CHANGING PROCESSES. Close examination of the entire planning and manufacturing process, however, reveals many changing factors that can influence the effectiveness and appropriateness of any given trim solution. These begin occurring when orders are first received into manufacturing blocks and continue until the manufacturing run is complete (Figure 1). Some mills have installed "retrim" capabilities where a winder operator can revise the remaining trim patterns in a run to account for off-quality rolls lost during manufacture.

Still, this is a process that must be user-initiated and is generally another one-time event during a manufacturing run, since operators normally wait until close to the end of the run before re-trimming. Several examples can serve to illustrate the point that different decisions might be made and higher efficiencies obtained if manufacturing runs were trimmed at other points in the process.

Many mills have particular grades and manufacturing runs that often do not achieve acceptable levels of trim efficiency. This is due to the typical roll size and quantity combinations ordered by customers not filling the deckle of the machine on which they are manufactured.

Where the problem occurs frequently, some mills have instituted policies that orders taken must provide a minimum trim before they are accepted. Where the problem occurs intermittently, planners may first discover the problem when trimming the manufacturing run-which occurs within a few days of manufacture. This leads to a scramble on the planner's part to either find a trim/side roll to manufacture or to find alternate orders that can be manufactured early or moved from another machine.

In these cases, could a better decision be made if the planner knew the existence of the problem when the order contributing to poor trim was first considered or taken? A trim model that executed each time an inquiry or orderwas received could also suggest many options to the planner, including:

• Optional roll sizes and quantities that would enhance the trim solution
• Orders in other runs or on other machines that might make sense incorporating in the current run
• Application of existing stock inventory to current orders
• Potential inclusion of stock or "help" roll sizes (based on sales history or lists of stocked items) that could change depending on trim efficiency and forecasted stocking levels.

In some trim models, the production planner can influence the number of slitter changes, but it is difficult to forecast in advance the situation at the winder when trimming in advance of actual production. In some cases, more slitter changes would not cause a big problem, while in others, fewer slitter changes are required to prevent downtime and reduce winder backlogs. Many times, planners have a fixed number of changes allowed.

In these situations, could a better solution result if the trim model was executed just prior to manufacture and the trim model took into account winder backlog (or scheduled winder downtime) when considering the relative cost of slitter changes? Similarly, the schedule of finishing or downstream converting equipment could also be considered by the trim model. These schedules may be difficult or impossible for the planner to predict when runs are currently trimmed, but they have tremendous impact on the product flow of rolls produced at the winder throughout subsequent operations.

One example of this is where mills use rewinders to complement operation of the winder. Large rolls are produced at the winder and then sent to the rewinder (planned rewinds), often because of slitter or size limitations at the winder. The rewinder backlog will also be influenced by the number of corrective rolls to be rewound for winding or quality defects. If rewinder backlogs are high, it may be better for the mill to produce fewer planned rewind rolls and to incur a higher trim loss in doing so. If rewinder backlogs are light, more rewind rolls can be produced.

Taking these concepts a step further, the winder schedule produced by the trim model is seen as a key influence on the flow of product through the finishing and shipping. Manufacturers have long utilized the concepts of materials requirements planning and capacity requirements planning to assist in managing this product flow, and systems incorporating these techniques are critical in the daily planning process.

Because of the uniqueness of the winder and trim models in the paper manufacturing process, it has been difficult or impractical to apply many of these same techniques. Even so, trim models have mostly concentrated on maximizing trim efficiency, for the most part ignoring the impact on downstream operations. In some cases, trim models have been built that allow planners to specify constraints on the trim solution, such as the number of open loads or the percentage of rewind rolls generated. In other cases, manufacturing runs are broken into segments and trimmed separately.
Two primary problems exist:

• Using fixed rules ignores information about current operating conditions.
• Without true integration/feedback between trim and other downstream schedules, the impact of the trim solution downstream is still unknown.

Another area of opportunity involves the maximum width, or deckle, that can be achieved in manufacturing a set at the winder. Planners must provide a single maximum width as input to the trim algorithm-is generally fixed as agreed with manufacturing and rarely changed (variations may exist based on grade or weight).

This width is generally a number that manufacturing agrees can always be produced. However, as mills have examined ways to improve machine efficiency, the realization has occurred that there is often 1 to 3 inches of variation between the actual usable deckle produced from reel to reel and run to run. While pondwidth provides an absolute maximum, good running conditions on the machine may provide usable paper in excess of the maximum width used by the production planner. In some cases, trim efficiency can appreciably increase with a small increase in usable width.

While this observation has been used to justify more than a few new headboxes, is there another way to look at this problem? What if the trim solution could take into account the usable machine deckle produced for each reel, at the time it is produced, rather than relying on a maximum deckle that more likely represents the minimum that is consistently achievable?

QUALITY DATA ISSUES. Another large area of potential surrounds the use of quality data. There are many conditions where variability in the production process causes a mismatch between the quality characteristics of the set and the needed quality characteristics of the customer sizes (and orders) shown in the trim pattern. This is especially pronounced when customers with varying quality demands are grouped together in the same manufacturing run.

In many cases, the off-spec roll is simply downgraded, trim positions within the set are swapped, or a different trim pattern is selected where roll sizes for less demanding customers can be cut. Some systems are sophisticated enough to assign appropriate customer orders to particular pattern positions or to highlight off-quality conditions based on customer quality specifications within the current pattern. But each quality test represents new information that could be incorporated into a new trim solution, rather than simply used to adjust a trim pattern that already exists.

What if the quality characteristics of uncut reels/sets could be considered and the trim solution resolved with this additional input? There are examples of trim packages today that trim around "bad spots" in one or more reels, essentially inserting a small roll, which will contain the defect, into the trim pattern at the appropriate width when resolving the trim problem. Could quality data across the web be considered and the trim problem continually resolved, whenever additional information is available?

DOWNGRADED ROLLS. A final area concerns makeup of downgraded rolls themselves. Often, as rolls are lost, a list is kept (either manually or on the system) until enough rolls have been accumulated to produce a complete set or the end of the run is imminent. A makeup set is then cut to an existing or created pattern, or if re-trim is available, the balance of the run is re-trimmed to create this pattern.

Instead of waiting until this point, what if the trim problem was resolved each time an additional roll was lost? Generally, the more options available as an input to trim, the better the solution. As a manufacturing run progresses and additional sets are cut, the available options become less and less.

This same concept regarding makeup might also be useful where customers order by weight or lineal footage and the number of rolls needed to fill an order may change depending on actual production. Some mills have large fluctuations in roll densities, attributed to furnish variations or winding tensions.

Whatever the reason, these variations may lead to cases where the planned number of rolls are cut, but the customer's order tolerance is not met. By considering actual production versus a customer's order requirements, a trim model could adjust the planned number of rolls in a matter similar to that regarding makeup rolls. Planning the "correct" number of rolls is a continual source of grief and frustration for many mills. Maybe it should be recognized that the estimated number of rolls is only that-an estimate-and that the number of rolls produced should be reevaluated as production occurs.

A BETTER WAY? All of these examples point to the need and value of a more iterative, dynamic trimming (Dtrim) process. Dtrim algorithms could work continually, re-solving trim problems each time additional information is received. It is conceivable that several different Dtrim models could coexist within the same system for addressing different needs (Figure 1).

Figure 1: Current trim applications are limited to initial creation of the winder schedule and winder retrim models. Future trim applications will cover much more of the order processing cycle.


For example, one Dtrim model could execute for trimming manufacturing runs when orders are still being received, while another would be specialized for manufacturing runs currently in production. Dtrim models would execute behind the scenes, without user initiation. To be most effective, it may be necessary to allow users to "freeze" particular manufacturing runs, trim solutions, or patterns from further Dtrim manipulation. Otherwise, system users could become confused.

The role of the production planner might also change. Instead of focusing on one manufacturing run at a time, planners could review Dtrim results across the whole system and adjust planned quantities and trim parameters in a continuous effort to improve efficiency. The review and release of a trim solution to the winder would probably still be necessary, however. Dtrim models could also be tightly coupled with Advanced Planning and Scheduling (APS) systems and could assist in determining the profitability impact of order-taking or schedule changes. With an APS, trim can be considered as one of many factors in schedule optimization.

Obviously, work is required, both to the trim models as well as to the supporting transaction systems, to make Dtrim models a reality. In particular, these models would not be accepted if they confused users by continually presenting changes or allowed changes to trim solutions or patterns which users wanted to designate as "fixed." But advances in technology have continually allowed operations which formerly happened infrequently to happen more continuously. The same can hold true for the application of trim models within the paper mill planning environment.

PAUL W. LAIL is business unit consultant, Champion International Corp., Hamilton, Ohio. He can be contacted at lailp@champint.com