Machine vision analyzes lumber
In the United States, the forest-products industry is concentrated in the southeast and northeast regions and in the Pacific Northwest. In these areas, loaded log trucks are a familiar sight, as are stacks of logs waiting to be processed at regional mills.
Laser illumination, linescan cameras, and image processing are combined to automate lumber processing.
Joe Hallett, Contributing Editor
In the United States, the forest-products industry is concentrated in the southeast and northeast regions and in the Pacific Northwest. In these areas, loaded log trucks are a familiar sight, as are stacks of logs waiting to be processed at regional mills. Visitors to the various regions are struck by the differences in size of logs, and machinery in these regions has been developed to deal with properties of available logs. While a mill in one region is optimized to use relatively small logs, another may depend upon large, older trees.
USNR Log optimizers use Hermary scanners to throw a linear array of light on the wood. These can be seen in blue.
Two types of mills use logs as raw material to produce both boards and thin sheets of wood suitable for lamination into plywood. After initial sorting, logs are cut to the proper length and then analyzed to determine the most efficient use of the material. After a longitudinal cut is made as a reference surface for subsequent sawing operations, additional operations produce cut material of a designated size, defects are identified, and sorting and grading take place.
All of these mill operations involve large pieces of wood moving at relatively high speed. Because of this, laser illumination is frequently used along with linescan cameras and image-processing systems to make the precise measurements that are essential to automated cutting and inspection.
Outside the mill, bark from tree-length logs is removed in a rotating debarker machine. This machine also cuts the wood to saw logs that are typically between 8 and 24 ft. A variety of lengthwise saw cuts are possible, and a large mill will set goals to obtain either the greatest possible volume of cut wood or specific lengths. This process—known as log optimization—involves calculating the saw cuts that will produce the best possible yield from each log (see Fig. 1).
Band saws are often used for the primary breakdown, in which an opening cut is made, exposing a flat surface that is the base for further cuts. "Most often, the timber passes through a band saw with a number of bands to cut a number of boards at one time," says Sven Durland, software research-and-development manager for Coe Manufacturing (Tigard, OR), a manufacturer of large equipment for sawmills.
In operation, the machine grabs a log and keeps it from rolling as it moves through a scan zone. This is typically a 1- to 2-s operation. "Logs are positioned in 0.001-in. offsets relative to the center of the machine to avoid making a cut that wastes wood," adds Richard Lord, a senior systems engineer for Coe. "By feeding logs in a transverse (lateral) motion, we can take a cross section as fast as the electronics allow."
FIGURE 1. After data are processed, a three-dimensional model of the log is created and the cutting machine optimized to maximize the use of wood to be sliced.
US Natural Resources (Woodland, WA) manufactures an optimization system that sends logs on a linear path that must be about three times the length of the longest log to be processed. Logs are scanned by an array of laser reference beams that are arranged in the same plane as a linear imaging system using laser profile scanners. According to the scanner manufacturer, Hermary Opto Electronics (Coquitlam, BC, Canada), space requirements for coplaner scanners are minimal, typically a few inches, whereas sheet-of-light scanners require substantial space in the z direction to achieve reasonable triangulation. In addition, coplaner systems view a narrow strip directly across from itself, whereas sheet-of-light scanners view a large area either upstream or downstream, making it more difficult to shield from plant lighting. Most important, coplaner scanners can visualize an entire surface, allowing three-dimensional profile scanning and image reconstruction (see Fig. 2).
Scanning for defects
Scanning for defects involves more than just dimensional measurements. In the Coe defect-scanning system, rough boards or sheets of plywood pass through an imaging system in which illumination is provided both by multiple infrared lasers and by halogen lamps. Several linear CCD cameras from Lockheed Martin Fairchild Systems (Milpitas, CA) are positioned so that they can image the laser reference beams inside a predetermined scan zone and also provide an image of the wood's surface. An optical shaft encoder senses the position of the board or sheet as it passes through the scanner.
Six cameras view the top of the boards and another six view below, picking up both the laser light and the light from halogen lamps. Linescan cameras are used for board and veneer scanning. "There's a new scan every 0.050 in. of travel," says Durland, "producing a camera image that consists of 0.050 x 0.050-in. pixels when projected upon the wood."
"Illumination is key," says Durland. "Automatic calibration is provided. Automatic alarms in the software indicate lamp failure or reduced light output. Long-life bulbs last more than a year in production. Blowers keep the lower cameras clean." Image processing and identification of defects are performed by proprietary hardware and software, interfacing to Windows-based computers via a communication port.
"Filters remove wood grain from the images. Defects are classified according to knots, splits, decay, and stains. After defects are detected, software performs image analysis to identify defects. PC-based frame buffers and pipeline image processors provide real-time pixel processing," says Durland.
Sorting is performed after scanning. The system keeps track of grading results and tracks the location to which boards or sheets are sent. "Grading criteria such as the size and number of knots can be set by the customer. The display also provides signal monitoring in real time and a defect/grading display and history," he says. A computer analyzes and displays the image and suggests what to do next. An operator decides whether to use the recommendation or to override it.
The main use of Coe's defect-detection system has been for veneer and plywood. "Veneer grading by operators gives 75%-85% accuracy while the scanner is capable of 95%-100%," says Durland.
Another development is the lumber scanner, which also performs knot detection through a saw mill using machine vision," says Durland. "Two of the systems have been sold so far, in addition to more than 100 systems without the knot-detection feature."
Work is underway to produce a similar scanner for green (wet) veneer. "In a veneer mill, logs are cut to approximately 8-ft sections that are peeled into continuous ribbons. A continuous vision system to assess defects in dry veneer is already available, and the continuous green-veneer system will be available next year," says Durland.
To process small logs, higher throughput is required. A rate of 15 logs per minute through scan and saw—one every 4 s—has been fast enough. But future machines will take advantage of higher processing power that can gather more cross sectional data.
Portland, Oregon 97223
Hermary Opto Electronics
Coquitlam BC V3K 7A1 Canada
Lockheed Martin Fairchild Systems
Milpitas, CA 95035-7407
US Natural Resources
Woodland, WA 98674