New Mathematical Model Improves Lumber Drying: Wagner Electronic Products Implementing Process for Greater Accuracy in Kiln-Drying

- Advertisement -

Supplier Makes Strides in Improving Accuracy in Kiln-Drying

Expectations based on sets of assumptions drive industrial processes, including lumber drying. So, the expected values used to monitor processes must be derived from accurate assumptions.
Average height and average weight are terms we all understand. Plot a histogram (bar graph) of height or weight for a large enough sample of people (say, 30 or more), draw a line that connects the bar tops, and a familiar pattern emerges: a curve that looks like a bell.
Many measurements cluster around an average value in a way that fits a bell curve or what statisticians call a normal distribution. (In a perfectly symmetrical normal distribution, the mean (average value) and the median (midpoint of all the values) are equal.) Height and weight are two types of measurements that follow a normal distribution for most human populations.
When measurements follow a normal distribution, there is great power in tying decisions to mean value or the departure from it. For instance, many manufacturing processes define tolerances for deviations in a finished product by distance from the mean on either the plus or minus side. If the distance is too great, a product must be rejected or retooled.
Assuming an underlying normal distribution where there is none, however, can pose problems. For instance, even in nature, many phenomena do not fit a bell curve. Longevity is one of them. Plot the age of death for the U.S. human population, and the curve will be skewed to the right (i.e., it will have a long tail on one side). Although most people will die by age 70, which will be a distinct valley on the curve, the few that live longer may live past 100 — and those long-lived individuals will give the curve a long tail.
Yet even a skewed curve has predictive value. Longevity follows a lognormal curve. Plot the log (base 10) value of age at death, and the curve will take the shape of a bell.
What does this have to do with drying lumber? If kiln drying schedules are constructed with the assumption that wood dries according to a normal distribution pattern, when it does not, over drying and under drying may be problems. The better we understand a species of wood and the way it dries, the better the correct approximation of when a kiln charge of lumber of that species is done. And the more closely the performance of a kiln can be fine tuned.
Several years ago, Ed Wagner, president of Wagner Electronic Products, Inc., in Rogue River, Ore., started talking with Thomas Maness at University of British Columbia about kiln drying. Thomas had already been working on statistical process controls in other parts of the mill and he suggested applying them to kiln drying. He already had ideas about developing a mathematical model that would make use of information from inline dried wood (pre- or post-planer) to extract the best performance from kilns.
The conversations between Thomas and Ed kindled a process that ultimately led to a masters’ thesis project for Catalin Ristea, Thomas’ student. By 2006, Catalin, Thomas and a representative of the group of engineers at Wagner that worked with them, James Felsheim, obtained a patent for a method that uses moisture content data from kiln-dried lumber under set conditions to create kiln drying schedules based on a lognormal distribution. Wagner Electronic Products is the co-owner of the patent with the University of British Columbia.
The method covered by the patent takes data from moisture detectors downline to evaluate results of successive kiln charges. It uses analysis of the down-line data to keep kilns in optimal working form and to refine kiln drying schedules, making them more accurate the more the method is used.
Plenty of things can go wrong in a kiln that will cause subtle changes. An incorrectly set baffle or a fan that fails will reduce the effectiveness of a kiln. So will a faulty steam trap. The mathematical model Wagner is using allows kiln operators to take down-line information, assess it rapidly, and quickly determine if they should take a close look at the kiln.
The bottom line is that information that can be used to correct or reset elements of the kiln becomes available within days of wood leaving a kiln. Looking at the big picture, analysis of lumber drying must account for variations in the wood and the kilns. The better we understand the process of drying a certain species of wood, the better the drying schedule can be designed for the lumber.
The new mathematical model to control the evaluation of wood drying, which the inventors use, offers a way to make the most of all information in the most expedient way. Data is added to a model for drying lumber, and the model is refined and becomes more robust as the process of drying goes along, charge after charge, thanks to the down-line sensors that gather readings from individual boards. The method takes into account that even lumber of the same species and manufactured to the same specifications will experience variations in drying – even when dried according to the same kiln schedule for time and temperature.
Because the model uses information from an inline moisture detector that takes a reading from every board that comes through and compiles all that information, more confidence can be had in the model for determining the best drying schedule. In one test of the lognormal model as a predictor of optimal kiln drying, 5,000 boards of 1×4 Southern Yellow Pine (SYP) were dried in a single charge. The results demonstrated that assuming a normal distribution of drying would lead to over-drying and degrade. Lognormal was the best way to devise drying schedules to achieve the mean target of 12% moisture content.
Wagner Electronic Products is at work now with the first implementation of the process based on the patented lognormal model. Tim Duncan of Wagner summed it up this way in a recent presentation on the method: “Using the proper statistical method results in more accurate statistical informational inputs to monitoring processes,” he said. “Proper statistical process control models lead to much better managerial decisions. Properly monitoring your data will improve your bottom line.”
(The accompanying article looks at the progress that Wagner Electronic Products is making with in-line use of a moisture reading system designed to enhance grade recovery.)