All About Curve Sawing
By Rod Nelson. 2001
Long, long ago, before optimization, there was only one kind of tree in sawmills: Big Trees. The Big Trees were cut up by big saws with big kerfs into boards with big target sizes. But, trees were cheap, so nobody worried too much about efficiency. Until the Big Trees all went away...
Then came Small Trees and Optimization. Which came first is a philosophical question, similar to the Chicken or the Egg, but I tend to believe that Small Trees were always there, but no one could see them.
The earliest known small trees were Cylindrical Trees. Computer programs were developed that cleverly fit rectangular board cross sections into the circle and, viola, Optimization was born.
The Cylindrical Trees were ultimately replaced with Truncated Circular Cone Trees, then Elliptical Cone Trees. Luckily computers and optimization programs were evolving as fast as the trees, so the optimization programs could handle the more complicated trees. It was during this time the BOF Tree (best opening face) was discovered, but I believe they are now extinct.
Next came the Real Shape Trees. I believe the Real Shape Trees have always existed, but computers were not fast enough nor were the optimization programs clever enough to recognize them. Therefore they were treated the same as the more primitive small trees.
Optimization had finally peaked out. It would get the most lumber out of any tree. The programs and equipment were mature, so now the startups for these systems only took a day or two.
Mechanical equipment designers, who have always been secretly envious of optimization programmers, were not pleased with this situation. They started rumors that the Real Shape Trees were not straight and proceeded to design machines that would follow the curve of these trees. This in turn required new optimization programs and more sophisticated positioning systems. Startups again began taking two to three weeks. Mechanical equipment designers were again happy.
CURVE-SAWING GANG
The curve-sawing gang is what equipment designers developed both to handle curved trees and to keep optimization and controls startup people off of the golf course.
There are two basic breeds. One breed leaves the saws stationary and drives the cant through the saws on a curved path. This is most prevalent in single arbor vertical gangs integrated into or closely coupled to canters or a twin band. The other breed drives the cant straight through the saw box, and articulates the saw box to carve the desired path. This is most prevalent on decoupled horizontal gangs and most double arbor gangs.
Stationary saw gangs tend to use a centering infeed mechanism that forces the center of the log to a consistent position as it feeds into the saw cluster. If the centering mechanism is in a fixed position, it would tend to make a split taper solution. Some machines allow the centering mechanism to slew during the cut, thus allowing solutions at other tapers. I use the phrase "tend to" because the irregularities on the log (limb stumps, sharp bends, etc.) can make the path of the log center uncertain. Optimizers hate uncertainty!
The uncertainty of the log position results in lower recovery, but the log behavior is generally predictable. When it misbehaves, the cant is usually so misshapen that no one cares what kind of solutions you get as long as you don't destroy the saws.
The problems with this type of machine are:
1. Positioning
uncertainty. Lower recovery than articulating saw gang.
2. Sweep must be oriented in one direction.
3. Less predictable results on S shaped cants.
4. Pressure adjustments to make centering mechanism work are a mystery.
5. Difficult to troubleshoot when the cuts are mispositioned.
6. If cant is too bent, the boards will be too bent.
Articulating (Wiggle Box) saw gangs are similar to conventional gangs except the section with the saw arbors is cut free from the infeed and outfeed pressroll sections. The saw section is mounted on a base that can be rotated +/- 5° and shifted left and right. This approach tends to position the cuts more predictably. These machines usually deal with two sided cants, so most incorporate chip heads to clean up the outside boards.
The problems with this type of machine are:
1. Horrific
crashes.
2. Difficult to troubleshoot when the cuts are mispositioned.
3. More complicated servo controls.
The main problem with this type of machine is the horrific crashes. The award winner in the horrific crash category was a multiple cluster, saw box that shifted several feet during a cut. Many saws and guides were destroyed and one of the arbors was bent. All gang edgers, conventional or curve-following have crashes, but the crashes are more frequent and more damaging with the articulating saw box.
THE GOOD NEWS
Rest assured, it is not all bad news. Curve-following gangs will produce more lumber and/or lumber of a higher grade. The percentage improvement depends on the log length, the log diameter and the shape characteristics, but mainly it depends on the current machine's efficiency. If your current system is manually fed, improvements can range from 5% to 10%. If your current system is an optimized system with good scanning and full taper control, the improvement can range from 0.5% to 3.5%. Overall mill uplift depends on the percentage of the volume coming from the gang. If the gang produces 70% and the other 30% is in sideboards that go directly to board edgers, the expected mill uplift would be 7/10 of the recovery improvement in the gang
Let's look at one example that compares tapered straight sawing to tapered curve-sawing. The solution image above is from the Nelson Brothers Engineering cant optimizer.
The cant shown was generated using a cant simulation program with the following parameters:
* Length 20 ft.
* Small diameter 10 in.
* Taper 1.0 in. per 10 ft.
* X sweep 0.0 in. per 10 ft.
* Y sweep 0.5 in. per 10 ft.
* Ovality: X diameter 0.0 in. greater than Y diameter
* Diameter deviation 0.1 in.
* Cant thickness 6.0 in.
The optimizer cleverly selects the best taper and fills the cant image with boards, fitting a shorter 2x6 and 1x4-6 in the belly, then putting a little saddle wane on the full length 2x6 on the other side. With the following lumber prices, the value of this solution is $24.53.
|
Product List |
Prices in $ / 1000 bdft |
|||||||||||
|
Type |
Spec |
Thk |
Wth |
Grade |
6' |
8' |
10' |
12' |
14' |
16' |
18' |
20' |
With a curve-following gang the same cant is effectively straightened as shown below. The 2x6-16 becomes a 2x6-20, the 1x4-6 becomes a 1x4-10 and the saddle wane on the other side is replaced with full length wane. And most importantly, the value of this solution is $25.66. That is a 4.6% increase!
If your mill had a steady diet of 20 ft., 10 in. diameter cants with 0.5 in. in 10 ft. sweep, 1 in. in 10 ft. taper and you were only cutting 6 in. cants, the 4.8% uplift looks pretty good. A better assessment would be to run a large variety of lengths, diameters, sweeps, etc., then compare the difference in projected lumber value.
100 CANT SIMULATIONS
With the cant simulator in the NBE optimizer, we can specify ranges for all the parameters defined above. The optimizer randomly (uniform distribution) picks values between the ranges specified. When the 100 cant test is selected, the tally gets cleared and 100 cants are run.
TEST 1: 100 cant test, small log mill with the following log parameters:
|
16 |
|
20 |
|
Length range in
feet |
Resulting values: Straight = $2071.49; Curved = $2136.96; Improvement 3.2%
Test 2: 100 cant test, stud mill:
|
8 |
|
8 |
|
Length range in
feet |
Resulting values: Straight = $489.46; Curved = $494.30; Improvement 1.0%
Test 3: 100 cant test, larger log mill:
|
16 |
|
20 |
|
Length range in
feet |
Resulting values: Straight = $6271.33; Curved = $6430.21; Improvement 2.5%
WHAT'S IT ALL MEAN?
Any time you run a recovery test, either with real cants, real cant data or simulated cant data, you have to question all aspects of the test.
* Does the cant
sample represent the expected cant mix?
* How many samples are needed?
* Is the optimizer really getting the best solution?
* Are the wane rules appropriate for the product?
* Can the machine always make the optimum solution?
* Are the lumber prices and grade assessments valid?
When analyzing the test results, you invariably discover other measurements that should have been made or unexplainable results that need further investigation.
With simulated log tests, you also have to question how well the simulated logs represent real logs. In the sample tests, one should question if the log sweep is truly a uniform distribution or if the selected cant thicknesses were appropriate.
With so many uncertainties, what are you to do? Well... "A field fertile with uncertainty is good for raising options." So, here are my opinions on simulated cant tests.
An improvement in production demonstrated by a simulated cant test is usually valid. Cant optimization systems normally realize 97% to 99% of the lumber that they forecast. A straight sawing cant optimizer that historically realized 98% of its forecast would still realize 98% of its forecast if it was switched to curve-sawing. Therefore, if the lumber forecast increased by 3% for curve-sawing, the realized lumber would also increase by 3%.
Simulated cant data can be just as good and in many cases better than real data. With simulated data you can try thousands of cants. If the test results raise additional questions, they can be resolved by running another thousand cants. For example, if the distribution of diameters or sweep was uncertain, you could run tests with different ranges to see if recovery was sensitive to the differences. If recovery is insensitive, then the uncertainty is not important. In another example, if you believe the distribution was gaussian (not uniform) you could run multiple tests with different uniform ranges then combine the results.
Mill personnel benefit from running simulated cant data. They learn the effects of changes to wane rules, prices, log mix and machine capabilities. If the simulation program is the same one you intend to use in production, the simulation exercise becomes a great training tool. Some vendors make their optimizer programs available for this purpose, knowing that if you invest time learning their program, you will tend to buy their system. NBE hopes this is true! So, if you are interested in running simulated cant tests with the NBE cant optimizer, e-mail me at rodn@millsmart.com
Now that I've gotten my opinions out of the way, you still have to justify curve-sawing to yourself and more than likely to your banker.
Which one of the following justifications would you trust?
* Simulation
results
* An equipment vendor's recovery analysis
* Another mill's recovery results
* Recovery analysis by an independent company
Obviously, a study by an independent company is best. Several companies specialize in recovery analysis for sawmills. They have the expertise to collect cant data at your mill and forecast the production of various equipment configurations and optimization methods. They will also do performance studies to determine if the installed system meets the goals. These independent companies are less biased than a vendor, trained to ask the right questions and aware of the latest trends.
Assuming the test results justify the curve-sawing gang, what else should you consider?
CUT PATH
First, how to measure sweep? The NBE optimizer uses the maximum distance between the curved path and a 10 ft. straight path. We use a fixed length of 10 ft. since that defines a curved path independent of the log length.
So what does a 1 in. sweep in 10 ft. look like? To make the math reasonable, let's assume a parabolic curved path. The equation for a U shaped parabolic curve is:
y = S * (( x - L / 2 ) / 60) ^ 2
Where:
* L is the cant length in inches
* x is the distance down the tree length in inches
* S is the sweep parameter in inches per 10 ft. of length.
For various lengths and sweeps, the maximum distance from the straight path and the angle of the saws upon entering or exiting a parabolic path are as follows:
|
Length |
10' Sweep |
Max. Distance |
Max. Angle |
If you were designing a sawbox for a 20 ft. lineally scanned gang and you anticipated a 1 in. in 10 ft. maximum sweep limit, the saws would need to rotate +/- 4° for curves and an additional +/- 2° to accommodate skewed cants and various tapers.
Another concern is how does the saw fit in the curved path. If the distance from the saw entry point to the saw exit point is 20 in., a 1 in. in 10 ft. sweep would extend the cant 0.028 in. toward the saw plate at the middle of the saw. The side clearance on most gang saws is less than 0.028 in., so one might wonder how the saws handle the situation.
The obvious answer is that the saw will flex and back-cut will carve out some wood plus the saw will heat up. Luckily cants requiring the maximum sweep do not occur often, so the problem is not continuous. It would be a good test for a mill with a curved gang to disable curve-sawing and just run straight tapered sawing. It would be an expensive test since the production loss could exceed $500 per shift and it would take a number of shifts to be convinced of any change in maintenance costs.
The cut path described so far pertains to the articulating saw box where the curve path is directly controlled. For the stationary saw box with the centering infeed mechanism, the curved path is determined by the log's shape and the geometry of the infeed mechanism. If these systems are fed cants with severe sweeps or crooks, the resulting boards will be severely bowed and the saws will be abused.
Investigation into the mathematics of curve-following has helped to identify new species of trees that will soon be added to the history of trees. The Parabolic Tree just discussed is probably one of the Nth order polynomial trees but some believe it to be a trivial example of the more exotic Spline Fit Tree.
CURVED LUMBER HANDLING
What do you do with bowed boards? People selling curve-sawing systems will tell you that many of the boards from a curved sawn cant will lie flat. They will also tell you that the cuts follow the wood grain thus producing stronger lumber of a higher grade. That is true.
They may forget to say that many boards won't lie flat and you will have to modify downstream unscramblers, conveyors and stackers to get the lumber out of the mill. Once in the kiln, you may have to top the units with a couple tons of concrete to convince the boards to straighten out. During this experience you will be become familiar with the phrase "boards from hell."
Generally the required modifications are not significant, but you should have some idea of what to expect.
PICKING/PROFILING
The best problem yet to be solved for straight or curve-sawing gangs is the separation of the following products:
1. Boards that do
not need to be trimmed
2. Boards that need to be trimmed
3. Boards that need to be edged
4. Boards that need to be resawn
5. Trash edgings
If trash could be successfully separated, all the chip head problems and most of the horrific crashes would go away. If the boards needing edged or resawn could be separated from the trim boards, you would not need manual sorting. If boards that do not need to be trimmed could bypass the trimmer, the trimmer piece rates would be greatly reduced.
"You might say I'm a dreamer, but I'm not the only one." This problem is being addressed a number of ways. Soon, a winner will emerge.
Some of the newer small log processors try solving the problem by adding more chip heads. The additional heads will profile the boards that need to be edged. For example, the side 2x4s in a 4-6-6-4 cant solution would get made by the profiling heads in conjunction with the heads that are making the 6 in. cant. Time will tell if these machines can control the log well enough to reliably produce the profiled boards as predicted.
SCANNING METHODS
If you read my article on lineal versus transverse scanning for board edgers in the July/August 2001 edition of Timber Processing, you might guess that I would favor lineal scanning for gang edgers. Yes, lineal is better for all the same reasons: simplicity, accuracy, lower initial cost and a lower maintenance cost.
The lineal scanner collects most of its data from the rounded sides of the cant, while a transverse scanner collects most of it data from the flat tops and bottoms. Since the goal of optimization is to fit as much lumber as possible into the sides, it is important to have good side data.
The scan accuracy and cut position accuracy are of even greater importance on a gang since a 1/8 in. error on a sideboard can dramatically alter the board's face, while the same error on a board edger will only shift the wane pattern.
Lineal scanning does have two drawbacks when used on cants:
One, cants that have little or no face tend to roll as the cant goes through a belt scanner. This is bad, since the scanner does not know the cant is rolling, thus it assumes the cant is skewed on the belt. If chip heads are involved, they will be set in the wrong place and a disaster follows. Cants with little face also cause problems on positioning tables. These problems can usually be overcome mechanically. Resist the temptation of forcing a bigger face to assure stability, since it will result in lower recovery.
Two, there is no gap control after the scan. By the time the cant is scanned and optimized, there is little opportunity to increase the gap when additional time is needed to shift saw clusters. If the operator can make the cluster selections prior to loading the cant, the gap can be created by PLC controls. If the optimizer needs to make the cluster selection, the normal high production gap would not allow enough time for a cluster shift.
These may not be problems in most mills, and someone may already have solutions. If you are planning on-the-fly cluster shifts, you better know the solution before the startup.
OPTIMIZATION
Optimization for curved-following gang is amazingly simple. It estimates a path that follows the cant's shape and stays within the maximum curve limits. Then it straightens the cant to that path and uses the normal cant optimizer. That is why the cant images on an NBE optimizer look straight. You may have noticed a curved, dashed, blue line in the image. That line represents how much the cant was bent to remove the natural curve. Early optimization programs only tried a single curved path, since less than one-half second was available for optimization. This resulted in all cants being curve sawn.
In many cases, curve-sawing has no monetary advantage over straight tapered sawing. In some cases, curve-sawing loses money. Since the amount of machine motion for straight tapered sawing is significantly less than that for curve-sawing, it would be reasonable to favor a straight sawn over a curve sawn solution of equal value. This could be done with a value deduction for machine motion, similar to that for additional processing (i.e. re-edge, resaw, extra boards) and positioning risk (i.e. sideboards with little freedom). With the additional speed of newer computers, optimizers can assess multiple paths and reduce the value of solution that requires more machine motion.
Curve-following optimization requires additional parameters. On articulating gangs, the maximum curve allowance limits how severe a curve can be attempted. On the NBE systems this allowance is in inches of bow per 10 ft. of length. The allowance is normally set below 1.0"/10'. Some optimizers allow different curve allowances for different clusters, different species and different cant thicknesses.
The optimizer needs to provide a good set of analysis tools. The minimum is the ability to save and replay cant images with different optimization parameters. Also needed is the ability to graphically step through the decision process, displaying the best solution for each curve or straight path. The analysis should show all tapers attempted and hopefully the individual steps that assemble boards into the cant. Other nice features are simulated cant generation, manual cant generation and automated replay of saved or simulated cant sets.
CONTROLS
Controlling a curve-following gang can get violent. One pioneering controls programmer tuned the positioning servos by sitting on the toilet. If he got splashed with water when the curve-following gang shook the building, he would reduce the acceleration parameters.
Positioning the saw box rotation and translation is done two ways. For slower, constant speed gangs, positioning is usually possible with the point-to-point PLC modules that have been used for the last 10 years. When the feed speed exceeds 400 FPM, or the speed varies during the cut, positioning with a motion-controller module is becoming more prevalent.
Because of horrific crashes, the PLC controls should maintain a history of the previous five cants. The history data should include the optimizer solution data and servo position plots. Ideally, the PLC should have a mode that will reenact any of the last five cants, without having to feed a real cant.
The single most important controls feature is the controls startup person. Good ones will identify the mechanical problems, straighten out the optimizer guru, argue with saw filers and get the machine running. Talk with other mills to find out who is best, then demand the best.
THE FINAL DECISION
If you are planning a new gang edger system for a random length mill, the decision is a "no brainer." The additional cost for curve-sawing should be less than $100K, which would pay back in months with the 2% recovery increase.
For stud and fencing mills, the recovery gain of 1% or less may not have a fast enough payback to offset costs.
If you already have an optimized straight gang, a 2% recovery increase at the gang may not be enough to justify the $1000K price of a new curved gang, optimizer and control system. In these cases, a more exacting assessment of the recovery increase would be in order.