Ultrasound Pallet Part Evaluator/grader and Cant Scanner

Sorting and grading of wooden pallet parts are key factors for manufacturing quality and durable pallets. The feasibility of ultrasonic scanning for defect detection and classification has been examined in this report. Defects, such as sound and unsound knots, decay, bark pockets, wane, and holes were scanned on both red oak (Quercus rubra, L.) and yellow-poplar (Liriodendron tulipifera, L.) pallet materials. Scanning was conducted by two pressure-contact rolling transducers in a pitch-catch arrangement. Pallet parts, such as deckboards, stringers, and cants were fed through the transducers, and data were collected, stored, and processed with software written in the LabViewTM environment. Defects were characterized on the basis of time of flight, pulse energy, and pulse duration of the received ultrasonic signals. Significant losses of energies were observed through these defects. Time of flight is less sensitive to defects compared to other parameters. This relative change of parameter values, with respect to values for clear wood, can be used to locate, identify, and quantify various pallet part degrades. Two-dimensional images were constructed using multi-line scanning data. The reconstructed images are able to show the position and surface area of the defects. Defects were classified using a multi-layer perceptron (MLP), a probabilistic neural network (PNN), and a K-nearest neighbor (KNN) classifier. Defective wood was classified quite clearly and accurately by all of these networks with high recognition rates. Decay has a higher recognition rate than the other defects. Wane and holes were readily confused owing to their common loss of transducer contact. The MLP were found to be more efficient for classifying these defects. Results demonstrate that real-time, on-line inspection and classification of defects in wooden pallet parts are possible by ultrasonic scanning. INTRODUCTION Approximately forty percent of the hardwoods produced in the United States are used for manufacturing pallets. Wooden pallets are the largest single use of sawn hardwoods, consuming 4.5 billion board feet of lumber for manufacturing 400 million pallets annually. Pallets are integral to the US transportation infrastructure, as almost all products spend a part of their life on a pallet, either as component parts or following final assembly. Typically, a wooden pallet consists of two parts-stingers, the structural center members that support the load, and deckboards-the top and bottom members that provide dimensional stability and product placement. There are many types of pallet designs based on the size, number, and position of the stringers and deckboards. Usually, pallet parts are produced from the low quality lumber or from the center cant materials of logs. These cants have a high percentage of defects, and therefore have less market value for other solid wood products. High quality pallet parts produce higher grade and longer lasting pallets with increased material handling safety, and permit multiple trips before repair, recycling or remanufacturing. The presorting and optimized sawing of cants can reduce processing cost and will produce higher grade pallet parts. Manual grading and sorting of pallet parts is a slow and inaccurate process, which depends on the individual skill of the grader. Moreover, the presence, location, and extent of defects in pallet parts are often difficult to determine accurately, making 30th Hardwood Symposium Proceedings 95 May 30 June 1, 2002 manual grading complicated. An automated, nondestructive machine would be very useful for pallet industry for the sorting and grading of these pallet parts. Detection and classification of defects in a pallet part is a challenging and complex task. Several nondestructive methods have been employed for detecting defects in wood, including X-ray, microwave, dielectric, optical, acoustic/ultrasonic, and laser (Szymani and McDonald 1981). Each of these methods has distinct advantages and limitations. Ultrasonic scanning has many advantages-through transmission (able to scan inside of the board/wood), faster, nondestructive and non-hazardous. The most common defects found in wood or pallet parts are sound and unsound knots, cross grain, decay, bark pockets, splits, holes, and wane. Many researchers have investigated the feasibility of using ultrasound for detecting these defects in wood (McDonald 1980, Schmoldt et al. 1994, 1997, Ross et al. 1992, Fuller et al. 1996, Niemz et al. 1999, Raczkowski et al. 1999, Karsulovic et al. 2000). The most common way of detecting defects using ultrasound is the measurement of transmission time through clear and defective wood. The measurement of transmission time is useful when there is only a single type of defect in a board. Some defect types may not respond well to the transmission time, but may respond to the other ultrasonic parameters, e.g., peak amplitude, centroid time, frequency domain energy, etc. Recent studies showed that the frequency domain analysis (Halabe et al. 1994, 1996) and energy loss (Brashaw et al. 2000, Kabir et al. 2002) parameters are able to detect defects efficiently in wood. The classification and recognition of defect types in a single sample is very difficult and complicated using any automated scanning system. The changes in the measurement parameters for many of these defects are similar, making classification systems difficult. Once we are able to detect, locate, and classify defects, grading of pallet parts is possible using established grading rules. MATERIALS AND METHODS The scanning equipment was provided by the former Ultrasonic Group, Forest Products Division of Perceptron Inc. The system consists of in-feed and out-feed roll beds, two pinch rollers for part movement, and two rolling transducers which are mounted in an ultrasonic ring. The bottom transducer transmits ultrasonic pulses and the top transducer receives those pulses. Pallet parts move through the system, lying on a face, and ultrasonic signals propagate through the part’s thickness. The necessary electronics and software to control material movement, signal generation, data collection, and analysis were supplied by Perceptron. The desired scanning resolution can be achieved by controlling roller speed and the number of pulses generated per second. All measurements were carried out at 120 kHz transmitting frequency and received signals were sampled at 500 kHz. Deckboards, stringers, and cants were collected from local sawmills for both yellow-poplar (Liriodendron tulipifera, L.) and red oak (Quercus rubra, L.). They were fresh cut and unplaned, and were kept immediately in cold storage to reduce their drying rate and to keep their moisture content above fiber saturation point. Twenty-five deckboards, eighteen stringers, and eight cants were scanned for each species. Each sample contains both clear and defective wood and the scanning was conducted in two ways. First, a line was marked on the board through a defect of interest and scanning was done along this line. Second, six scan lines were marked longitudinally along the face of each sample 1.27 cm apart across the width of the board, and scanning was performed along these six lines. These multi-line scans were used to characterize the entire deckboard and to construct a 2-D image. ULTRASONIC VARIABLES The ultrasonic scanning involves measurement of many parameters-three for time-of flight, two for ultrasound pulse energy, one using ultrasound pulse duration, and peak frequency. The wave energy of the received signal can be expressed as the time integral of the voltage: