Martin Caron Comparaison de l`optimisation selon le prix et selon le
Transcription
Martin Caron Comparaison de l`optimisation selon le prix et selon le
Martin Caron Comparaison de l’optimisation selon le prix et selon le rendement matière dans les usines de débitage de composants de bois franc Mémoire présenté à la Faculté des études supérieures pour l’obtention du grade de maître ès sciences (M.Sc.) Département des sciences du bois et de la forêt FACULTÉ DE FORESTERIE ET GÉOMATIQUE UNIVERSITÉ LAVAL AVRIL 2003 © Martin Caron, 2003 Résumé La recherche sur les procédés et les indicateurs de performance de l’industrie des composants de bois franc est souvent réalisée sur la base du rendement matière. Plusieurs études ont été effectuées à l’aide de logiciels informatiques afin de simuler les procédés de débitage et de trouver des stratégies et techniques pour améliorer le rendement matière. La présente étude a été réalisée afin de comparer l’optimisation selon la valeur des pièces produites versus l’optimisation selon le rendement matière, à l’aide d’une structure formelle de prix inspirée du système de classement du bois franc de la National Hardwood Lumber Association. Les résultats de cette étude démontrent clairement qu’une optimisation selon le rendement matière est sous-optimale en termes de la valeur des pièces produites par rapport à une optimisation selon la valeur et n’apporte pas le retour monétaire maximum par rapport au bois utilisé. Pour de l’érable à sucre de grade No. 1 Commun avec un procédé de délignage en tête, changer d’une optimisation selon la surface à une optimisation selon la valeur produite des pièces a résulté en une diminution relative de rendement de 10.6% et en une augmentation relative du retour monétaire apporté entre la valeur produite et le coût de bois, de 107.6%. © Martin Caron ii Remerciements Merci au Centre de Recherche Industrielle du Québec (CRIQ) pour m’avoir ouvert les horizons dans le domaine du bois autant par le développement que par l’utilisation du logiciel « Optimiseur deux axes ». Merci à Michel R. Bouchard pour son support éclairé lors de la rédaction de ce mémoire et sa précieuse aide dans le développement des outils d’analyse sur « Matlab ». Merci à Robert Beauregard pour ses contacts, son support au cours de la rédaction de ce mémoire et sa connaissance éclairée du domaine du bois. Je souhaite au lecteur autant de plaisir et d’émerveillement à lire et comprendre ce mémoire que j’en ai eu à l’écrire. © Martin Caron iii Avant-propos Le mémoire est constitué d’une introduction et d’une conclusion générales en français ainsi que d’un article intercalé en anglais entre les deux autres sections. L’article intitulé “Comparing a value based and a surface based part prioritization in the hardwood rough mill” a été soumis pour publication dans le Forest Products Journal. Les auteurs de l’article sont Martin Caron, Michel R. Bouchard et Robert Beauregard. Le premier auteur a réalisé les simulations, les analyses statistiques et il a rédigé le manuscrit sous la direction des deux co-auteurs, respectivement du CRIQ et de l’Université Laval. © Martin Caron iv Table des matières Résumé ................................................................................................................ ii Remerciements .................................................................................................. iii Avant-propos ..................................................................................................... iv Table des matières ..............................................................................................v Liste des tableaux.............................................................................................. vi Liste des figures................................................................................................ vii 1. Introduction..................................................................................................1 2. Comparing a value based and a surface based part prioritization in the hardwood rough mill. .........................................................................................6 2.1. Introduction .............................................................................................................6 2.2. Materials and methods..........................................................................................11 2.2.1. Definitions .......................................................................................................11 2.2.2. Board database description.............................................................................12 2.2.3. Part grade definition .......................................................................................15 2.2.4. Defects definition.............................................................................................15 2.2.5. Part grades in relation with defects ................................................................16 2.2.6. Cutting bill.......................................................................................................16 2.2.7. The pricing structure .......................................................................................17 2.2.8. NHLA dimension definitions............................................................................17 2.2.9. Value in the pricing structure depend on parts dimensions ............................18 2.2.10. Simulation process...........................................................................................20 2.2.11. Statistical analysis ...........................................................................................20 2.3. Results and discussion ...........................................................................................22 2.3.1. Effect of the processes on yield........................................................................24 2.3.2. Effect of the process on RPBF.........................................................................26 2.3.3. Effect of the prioritization mechanism on yield...............................................29 2.3.4. Effect of the prioritization mechanism on RPBF.............................................30 2.4. Conclusions.............................................................................................................33 2.5. Litterature cited .....................................................................................................36 3. Conclusion...................................................................................................51 4. Liste des références ....................................................................................54 Annexe 1 : Prix calculés....................................................................................57 © Martin Caron v Liste des tableaux TABLE 1. — DESCRIPTION OF THE HARD MAPLE AND BLACK CHERRY BOARD DATABASE ........40 TABLE 2. — PART GRADES DESCRIPTION ................................................................................40 TABLE 3. — DESCRIPTIONS OF DEFECTS IN THE BOARD DATABASE ..........................................41 TABLE 4 . — DEFINITION OF PART GRADES IN RELATIONS WITH DEFECTS ...............................41 TABLE 5. —CUTTING BILL USED IN THE SIMULATIONS ............................................................42 TABLE 6. — CUTTING DIMENSIONS IN THE NHLA GRADING SYSTEM ......................................43 TABLE 7. — PART DIMENSION LIMITS USED FOR THE PRICING STRUCTURE OF THE CUTTING BILL .........................................................................................................................................43 TABLE 8. — HARD MAPLE PRICING, AND WIDTH/LENGTH FACTORS .........................................43 TABLE 9. — PRODUCT / GRADES PRICING AND PRICE FACTOR ..................................................44 TABLE 10. —OPTIMISATION PARAMETERS ...............................................................................45 TABLE 11. — OPTIMIZATION RESULTS FOR THE HARD MAPLE DATABASE ................................46 TABLE 12. — OPTIMIZATION RESULTS FOR THE BLACK CHERRY DATABASE ............................47 © Martin Caron vi Liste des figures FIGURE 1. —BOREALSCAN IMAGE ACQUISITION SYSTEM ..........................................................48 FIGURE 2. ― PRODUCTS DISTRIBUTION FOR HARD MAPLE .......................................................49 FIGURE 3. ― PRODUCTS DISTRIBUTION FOR BLACK CHERRY ...................................................50 © Martin Caron vii 1. Introduction La base d’analyse de la rentabilité pour l’industrie du bois franc est souvent le rendement matière. Face aux pressions du marché, les entreprises spécialisées dans le débitage du bois franc ont mis l’accent sur l’augmentation de leur rendement matière (Buehlmann 1998). Déjà au début des années 1980, la loi de l’offre et de la demande combinée avec le déclin de la ressource naturelle avaient provoqué une hausse drastique des prix (Hall et al. 1980). Cette tendance s’est poursuivie au cours des années 1990 et 2000. Au Québec, une partie grandissante des billes utilisées au sciage provient maintenant d’importations à cause de la piètre qualité et de la rareté des billes locales (Industrie Canada 2002). Forcée par l’augmentation des coûts de main d’oeuvre en Amérique du nord, l’augmentation du prix de la matière première et la mondialisation des marchés, l’industrie doit trouver de nouvelles façons de faire pour maintenir sa position concurrentielle (El Radi et al. 1994, Hoff et al. 1997). Pour les industriels du domaine du bois, l’optimisation des procédés et le développement de nouveaux produits sont devenus des moyens nécessaires pour assurer leur survie. L’industrie du débitage du bois franc demeure loin derrière les autres industries en ce qui a trait à l’innovation technologique (Buehlmann 1998). Les sources potentielles d’amélioration de la compétitivité doivent être analysées. L’industrie doit se différencier, trouver des moyens de diminuer les coûts, d’augmenter sa productivité et mettre au point de nouvelles façons de faire. Il est nécessaire de voir plus loin, de viser un accroissement de la rentabilité par la recherche sur le procédé (Hoff et al. 1997). La recherche d’aujourd’hui se base sur des logiciels de simulations de débitage (Thomas 1996b, 1997), des systèmes de vision (Conners et al. 1997), des traitements de la matière première en procédant à une présélection ou une amélioration des planches afin d’obtenir un gain de rendement matière (Gatchell 1990, Gatchell et Thomas 1997, 1998, Wiedenbeck et Scheerer 1996, Wiedenbeck 2001, Steele et Wiedenbeck 1999). L’optimisation du procédé de débitage, considérant la valeur des pièces produites ainsi que leur contribution en rendement a été suggérée dans des études antérieures (Thomas 1996b, Mitchell 2002). Une méthode formelle permettant de générer une structure de prix et les effets sur le procédé que provoque une optimisation selon le prix, n’ont jamais été étudiés directement par simulation. Le procédé conventionnel de débitage du bois franc a longtemps été considéré comme celui procédant par le tronçonnage en tête (Thomas 1997, Buehlmann 1998). Le tronçonnage en tête consiste à faire des coupes sur les planches selon la largeur de celles-ci. La rareté des billes de fortes dimensions, les prix de la matière première en hausse, les besoins changeants de l’industrie en fonction des besoins des clients et de la complexité des paniers de produits, ainsi que des cadences de production plus élevées ont partiellement et graduellement poussés les producteurs de composants vers un procédé de délignage en tête (Mullin 1990, Wiedenbeck et Scheerer 1996, Wiedenbeck 2001). Le délignage en tête consiste à faire des coupes sur les planches selon la longueur de celles-ci. Il faut noter cependant que chacun de ces deux procédés de base possède ses avantages et offre un potentiel supérieur dans certaines conditions. En usine, les planches sont transformées en bandes (délignés ou tronçons) afin d’optimiser le rendement en surface sur le premier axe traité (Thomas 1996a). Ces bandes sont ensuite traitées pour en purger les défauts et respecter les dimensions spécifiées par le carnet de commande, tout en tentant de maximiser le rendement matière. Il a été mentionné que le délignage en tête tend à produire des pièces plus longues que le tronçonnage en tête et offre généralement un meilleur rendement dans les grades intermédiaires (No. 1 commun et No. 2A commun). Le tronçonnage en tête permet de purger des zones de défauts concentrés sur une courte longueur et offre généralement un meilleur rendement dans les grades inférieurs (No. 3 commun) (Wiedenbeck 2001). Les procédés de « délignage toutes longueurs » (plancher, moulures) et « tronçonnage toutes largeurs » (panneaux) sont des extensions des procédés de délignage en tête et de tronçonnage en tête offrant des avantages du fait qu’ils possèdent un axe de liberté sur une dimension pour le placement des pièces. © Martin Caron 2 La recherche sur les procédés de débitage est effectuée avec différentes techniques, l’une d’elles étant la simulation par logiciels informatiques. Ces outils de simulation utilisent des bases de données de planches numérisées, un carnet de commande contenant les informations sur les pièces à produire et des paramètres représentant le procédé. Une base de données consiste en un ensemble de planches classifiées selon certaines règles ou critères, dont les dimensions sont connues et dont les défauts sont identifiés et localisés. Un carnet de commande consiste en une liste de pièces à produire dans une usine de débitage . Les dimensions, grades, quantités et prix y sont contenus pour chaque pièce (Buehlmann 1998). Un travail de comparaison entre l’Optimiseur deux axes (Caron 2001) et les logiciels du USDA ROMI-RIP et ROMI-CROSS (Thomas 1996a, 1997) pour du délignage en tête et du tronçonnage en tête a été effectué et permet d’établir un niveau de performance relatif entre ces outils (Bouffard et al. 2002). L’étude se base sur les résultats de simulations produits par l’Optimiseur deux axes du système BorealScan développé par le CRIQ. La base de données est constituée de bois classifié selon les règles de la National Hardwood Lumber Association (NHLA 1998) et contient plus de 2700 planches numérisées d’érable à sucre (Acer saccharum Marsh.) et de cerisier noir (Prunus serotina Ehrh.) de tous les grades NHLA. Cette base de donnée a été crée et est maintenue par le CRIQ. Un carnet de commande standardisé a été défini à partir d’usines typiques de producteurs de composants du Québec. La plupart des simulateurs industriels se basent sur une optimisation « un axe » et produisent des délignures ou des tronçons qui sont re-traités par une deuxième opération. Contrairement à ROMI-RIP et à ROMI-CROSS (Thomas 1996a, 1997), l’Optimiseur deux axes (Caron 2001) permet de produire une solution de débitage optimale (pour la période de temps d’optimisation allouée) en considérant simultanément les deux axes d’optimisation et génère une solution contenant une série de pièces produites, une série de positions de coupe et des informations sur les produits en sortie. © Martin Caron 3 L’Optimiseur deux axes du CRIQ supporte les différents procédés de base de l’industrie : 1) Tronçonnage en tête 2) Délignage en tête 3) Tronçonnage toutes largeurs (panneaux) 4) Délignage toutes longueurs (bois de plancher) L’Optimiseur deux axes permet de trouver le placement optimal de pièces à l’intérieur de planches comportant un certain nombre de défauts. La recherche de la solution optimale est faite en utilisant un parcours de graphe exhaustif pour chaque possibilité de placement pour un grade donné. Les placements sont faits premièrement selon l’axe primaire de coupe et deuxièmement en remplissant les espaces libres dans la planche. Une limite de temps maximale est allouée à la recherche de la solution. Les débits à produire proviennent d’un carnet de commande de l’usine. L’optimiseur réalise les quantités demandées dans le carnet de commande et cherche à optimiser le placement dans le but de maximiser l’utilisation du bois ou de maximiser la valeur monétaire totale produite (Caron 2001). L’optimiseur est conçu pour être intégré à un équipement spécialisé (mode production) ou pour fonctionner en mode autonome afin de réaliser des simulations. Dans cette étude, il est important de faire une distinction entre les différents procédés de débitage permis par l’Optimiseur deux axes. Il peut générer des solutions de débitage basées sur un procédé conventionnel, respectant un seul axe de coupe, soit le délignage en tête ou le tronçonnage en tête. Ces procédés sont appelés: « non-sélectifs ». Il peut aussi, faire une sélection entre deux solutions de débitage sur les deux axes de coupe, soit la meilleure solution de délignage en tête ou de tronçonnage en tête. Ces procédés sont appelés: « sélectifs ». De plus, il permet de composer des procédés de débitage hybrides qui combinent plusieurs types de produits dans la même solution, en optimisation selon le rendement matière et selon le prix Le but de cette recherche est de comparer les effets d’une optimisation selon le prix par rapport à une optimisation selon la surface. Cette comparaison est effectuée en bâtissant une structure de prix formelle provenant des standards dimensionnels de la norme NHLA et des prix du marché. L’analyse est faite en considérant des procédés conventionnels de débitage (délignage en tête, tronçonnage en tête), et une solution de débitage développée par le CRIQ © Martin Caron 4 résultant en un choix entre la meilleure solution du délignage en tête et du tronçonnage en tête. La difficulté et la nécessité du choix entre le tronçonnage en tête et le délignage en tête au niveau du rendement matière ont fait l’objet d’études et discussions antérieures (Mullin 1990, Wiedenbeck 2001) mais l’effet que ces procédés ont sur la valeur produite des pièces est relativement inconnu. Le but de cette étude est de scruter ces différents procédés sous l’angle de la valeur monétaire produite, en bâtissant une structure de prix permettant de comparer des optimisations selon la surface avec des optimisations selon le prix. Le rendement matière, la valeur produite (selon les pièces générées), le retour monétaire obtenu (valeur des pièces produites moins le coût des matières premières) et la distribution des produits générés seront utilisés comme éléments de comparaison. Des études récentes ont permis la validation du logiciel ROMI-RIP dans un contexte de production (Hamner et al. 2002, Thomas et Buehlman 2002). Ces études utilisent des facteurs statiques et dynamiques inclus dans ROMI-RIP pour contrôler la priorisation de la production des pièces du carnet de commande (i.e. : la longueur au carré multipliée par la largeur des pièces, un coefficient à exposant dynamique). L’utilisation d’un facteur modifiant la pondération d’une pièce par rapport à une autre biaise inévitablement le rendement matière obtenu (Thomas 1996b, Mitchell 2002). Utiliser un facteur de pondération afin de favoriser le placement de pièces longues et/ou larges dans une optimisation revient à accepter une diminution de rendement. Une étude comparant ROMI-RIP avec l’optimiseur de débitage du CRIQ (Bouffard et al. 2002) indique qu’en optimisation selon la surface, sans facteurs statiques ou dynamiques, ROMI-RIP favorise le placement de longues et larges pièces, ce qui semble indiquer l’existence d’un processus interne d’établissement de priorités sur les pièces du carnet de commande, et non une pure optimisation en rendement matière. L’hypothèse de la présente étude est que l’utilisation d’une structure formelle de prix basée sur des prix réels de composants, permettra de contrôler le placement des pièces dans un simulateur d’une manière rentable, tout en évitant l’utilisation de facteurs complexes ou dynamiques. De plus, une optimisation maximisant la valeur devrait permettre de générer une valeur produite par volume entré plus grande qu’une optimisation maximisant le rendement matière et ce, peu importe le procédé de débitage utilisé. © Martin Caron 5 2. Comparing a value based and a surface based part prioritization in the hardwood rough mill. 2.1. Introduction Traditionally, profitability in the hardwood industry has been evaluated through yield. Due to high market pressure, the hardwood industry has put a lot of focus on increasing yield in their operations (Buehlmann 1998). The increasing costs of labor in North America, the globalization of markets and the coupling of higher lumber cost with the decrease in lumber quality forced the industry to find new ways to do business and survive (El Radi et al. 1994, Hoff et al. 1997). For the hardwood component industry, process optimization and product development are now vital for survival. The hardwood component rough mill industry is still far behind the general industrial level in terms of technological innovation (Buehlmann 1998). The industry must go through phases of improvement, decrease costs, increase productivity and change the traditional way of doing things which has been a roadblock to innovation (Hoff et al. 1997). Today’s research in the rough mill goes through software simulation (Thomas 1996a, 1997), vision systems (Conners et al. 1997), the improvement of rough mill yield by a selection of boards using specific criteria or upgrading of boards before the cutting operation (Gatchell 1990, Gatchell and Thomas 1997, 1998, Wiedenbeck and Scheerer 1996, Wiedenbeck 2001, Steele and © Martin Caron 6 Wiedenbeck 1999). Optimization of the rough mill process using part prices versus yield has been suggested in prior works (Thomas 1996b, Mitchell 2002). A formal method to generate a pricing structure and the effects of prioritizing the rough mill operation using such a structure has not yet been studied thoroughly. The conventional way of processing hardwood is to cut boards in relation to a first axis in order to produce strips which maximize yield. (Thomas 1996a). These strips are then reworked to remove unacceptable defects and respect cutting bill requirements while keeping an acceptable yield. Traditionally, the component rough mill process has been crosscut-first (Thomas 1997, Buehlmann 1998). Increasing scarcity of large lumber, increasing lumber price and higher production needs, slowly pushed wood component producers to rip-first. (Mullin 1990, Wiedenbeck and Scheerer 1996, Wiedenbeck 2001). These processes (crosscut-first and rip-first) have their respective advantages. Ripping first can more easily remove defects along the side of the boards or concentrated along a longitudinal strip and usually allows for higher yields in the intermediate grades of lumber (No. 1 Common and No. 2A Common). Crosscutting-first can more easily remove defects concentrated on a relatively narrow zone by removing the zone and usually provides higher yields on boards with crook and wane and/or of lower grades (No. 3A Common) (Wiedenbeck 2001). The processes of “rip-first random lengths“ (flooring, mouldings) and “crosscut-first random width” (pannelling) are extensions of the rip-first and crosscut-first processes that present specific advantages due to their free dimensional axis. © Martin Caron 7 Process research on rough mills is pursued using various techniques; one of them is through computer simulation software. A comparative study between the “Two-Axis Optimizer“ (Caron 2001) and the USDA software ROMI-RIP 2.1 and ROMI-CROSS (Thomas 1996a, 1997) has been recently done (Bouffard et al. 2002). This study is based on the CRIQ “Two-Axis Optimiser” of the BorealScan system, using a database of lumber graded using the National Hardwood Lumber Association rules. Two different species were used in the study : hard maple (Acer saccharum Marsh.) and black cherry (Prunus serotina Ehrh.). The procedure to digitize the database is described in Bouffard et al. (2002). It used crayon mark reading, with a color camera, of manually identified defects. Most industrial rough mill simulators are based on a “one axis” optimization technique producing maximized strip yield reworked by a secondary processing operation. ROMI-RIP, ROMI-CROSS, and the CRIQ “Two-Axis Optimiser” consider the whole board before producing an optimal cutting solution. Cutting positions, yields and grade information is given as output (Thomas 1996a, 1997, Caron 2001). The CRIQ “Two-Axis Optimiser” can simulate the following industrial processes : 1) Crosscut-first. 2) Rip-first. 3) Crosscut first, random width (panel). 4) Rip-first, random length (flooring). © Martin Caron 8 This software brings innovative elements such as hybrid solutions containing various types of products and being able to yield the best solution between two or more rough mill processes. The are two types of simulation processes used in the simulator, 1) processes based on a single cutting axes and 2) processes based on a selection between two cutting axes simultaneously. Processes based on a single cutting axis, namingly rip-first and crosscut-first, will be called “non-selective” and processes based on a selection between the best of the two cutting axes will be called “selective”. The software supports different prioritization strategies such as value, pure yield and yield obtained with static factors applied to parts. The goal of this study is to compare the effects of an optimization strategy based on the value of the parts produced versus an optimization based on yield, using a formal pricing structure based on dimensional standards of the NHLA grading rules and market prices. This analysis is done considering conventional hardwood component rough mills processes (rip-first, crosscut-first) and a process selecting the best of rip-first or crosscut-first optimizations developed by the CRIQ. Yield differences between rip-first and crosscut-first have been the subject of prior works and discussions (Mullin 1990, Wiedenbeck 2001) but the effects of the process on the value of the produced parts is relatively unknown. The goal of this study is to open new research paths using the maximization of value produced instead of the maximization of yield using different rough mill processes. Results will be compared using yield, total value produced, return value (value of parts minus the cost of wood) and the distribution of products generated. © Martin Caron 9 Recent studies validated ROMI-RIP in rough mill operations (Hamner et al. 2002, Thomas and Buehlman 2002). These studies used static and dynamic prioritization factors included in ROMI-RIP to control the use of parts in the simulation process (i.e: square length of parts multiplied by the width, complex dynamic exponent factor applied to parts) because a simple surface optimization could not compare easily to a real rough mill. Using a factor to modify the part prioritization in relation to others will inevitably change the final yield obtained (Thomas 1996b, Mitchell 2002). Applying a surface modification factor to long and/or wide parts is accepting a reduction of yield according to the specifications of the factor used, this strategy is also called “force cutting”. A study comparing ROMI-RIP and the CRIQ “Two-Axis Optimiser” (Bouffard et al. 2002) has shown that in pure yield optimization, without any prioritization factors, ROMI-RIP was still prioritizing long and wide parts over smaller ones. This result was counter-intuitive and points to the existence within ROMI-RIP of an internal prioritization mechanism. This mechanism appears to apply even when the pure area prioritization strategy is chosen. It is our belief that the use of an explicit pricing structure will allow a better solution to the prioritization of parts while avoiding the use of prioritization factors that are not necessarily based on sound theoretical ground and may yield unexpected results. Furthermore, it is believed that the use of a part prioritization strategy based on value of parts should provide results with a higher return value than a pure area prioritization strategy, and that whatever the NHLA lumber grade used. © Martin Caron 10 2.2. Materials and methods 2.2.1. Definitions Volume of input The volume of input is the volume of lumber, accounted in board feet, entered in the simulation process (void defects are not included). Volume of output The volume of output is the volume of parts, accounted in board feet, produced in the simulation process. Yield Yield in this study accounts to the “ratio of the surface area of output to the surface area of input of common thickness expressed as a percent” (Buehlmann 1998). Value produced Value produced in this study accounts to the total value of part produced expressed in US dollars. © Martin Caron 11 Value produced per board feet entered Value produced per board feet (VPBF), accounts to the total value of parts produced divided by the input volume entered to produce those parts. By dividing the return by the B.F entered we can get a VPBF value independent of simulation B.F variations. Return The return is a monetary value that accounts to the cumulated value of parts produced in a simulation minus the cost of the wood entered to produce those parts expressed in US dollars. Return per board feet entered Return produced per board feet (RPBF) for the total value of parts produced in the simulation minus the cost of the wood used to produce those parts divided by the input volume. By dividing the return by the B.F entered we get a RPBF value independent of simulation B.F variations. This is an integrated value combining the value of parts produced, the number of parts produced, the volume of wood inputed and finally the cost of the wood used to produce those parts. This performance indicator is believed to be the most meaningful for comparing different rough mill processes with a similar pricing structure for parts in our study. 2.2.2. Board database description The board database used was built using the image acquisition system BorealScan built by the Centre de Recherche Industrielle du Québec. All boards are contained within three NHLA grades (No. 1 Common, No. 2A Common and No. 3A Common) and two species (hard maple and black cherry) representative of the supply of a specific component © Martin Caron 12 manufacturer from Quebec. Boards were graded by NHLA certified graders within the standard tolerances of the industry. All boards were dried rough wood. The boards were not ripped prior to scanning nor planed. Various color crayons were used to mark defects on both faces of the boards. Hand-marks placed by the operators were then scanned in the BorealScan system. As such, the precision of defect detection was limited to the precision of the hand-markers and the size of the crayon used (3mm wide). Three defect types were originally marked on the boards : 1) unacceptable defects (wane, large splits, large holes, unsound knots), 2) colorations defects (heartwood), and 3) sound defects, that are acceptable on the back face of certain products (sound knots, burl, etc...). Scanning of boards was done with two high speed, high resolution color cameras providing images with a resolution of 0.2 mm along the width of boards and 1 mm along the length. A digitized file representing the type and position of the defects of each board was obtained in output. The lower left corner was used as a reference point for all boards. The BorealScan image acquisition system is shown in figure 1. All marked defects were thoroughly re-worked using an image processing software. Images were manually modified to change the type of certain defects according to the specifications of Quebec component stock factories. Thus, coloration (heartwood) was split into two different types of defects 1) light coloration (heartwood where contrast with sapwood is © Martin Caron 13 minor), 2) dark coloration (heartwood where contrast with sapwood is major). Knots of less than 3mm in diameter were made a separate class as “pin knots”. The same images and database were used in a study comparing the CRIQ optimiser, ROMI-RIP and ROMI-CROSS (Bouffard et al. 2002). Information on the boards contained in the database is detailed in Table 1. From the observation of the database, the following comments can be made: • No 1 Common black cherry is of lower average length than the No 1 Common hard maple. • No 3 Common hard maple is of lower average length than the No 3 Common black cherry. • Standard deviation on length and width is higher for hard maple than for black cherry, especially for lower grades. Black cherry boards were precision trimmed and sorted by length. The black cherry database contained fixed lengths of four, five and six feet for No. 1 Common and lengths of seven and eight feet for No. 2A and No. 3A Common. Hard maple boards were also trimmed and sorted by length. Boards of six, seven and eight feet were used in No. 1 common, boards of four, five, seven and eight feet were used in No. 2A Common and boards of four, five, six, seven and eight feet were used in No. 3A Common. The average length of the No. 1 Common black cherry is sensibly shorter than No. 2A Common and No. 3A Common. There is more variability on lengths in the hard maple database because there are more lengths included. © Martin Caron 14 It is important to note that there is no information in the database concerning the crook of boards. All board dimensions were determined by the smallest rectangle encompassing the wooden section of the board. Void defects were included to represent the sections in the rectangle where no wood was scanned. Void sections were not considered in the input surface used for yield calculations in the simulator. 2.2.3. Part grade definition The part grades used in the simulator can be found in Table 2. They are sorted from the highest priority to the lowest. There are 7 different grades based on 3 general product types. The product specifications were based on actual products from a typical Quebec component factory. The three types of products used were fixed dimensions components, random width panels and random length flooring strips. 2.2.4. Defects definition There were three defect types determined and based on the actual rough mill process of a Quebec component factory (coloration, sound defects and unacceptable defects). Two defects were manually added with the image processing software. Coloration was split into “light heartwood” and “dark heartwood”. “Pin knots” were extracted from the sound defects class. A sixth defect type (void) was added by the image processing software. Table 3 describes the defect types used in the rough mill simulator. © Martin Caron 15 2.2.5. Part grades in relation with defects Table 4 describes the relationships between part grades and defects used in the rough mill simulator. The “―“ symbol means that this type of defect cannot be accepted on a part of this grade. The “1F” symbol means that this defect can be accepted on the back face of parts of this grade. Finally, the “2F” symbol means that this defect can be accepted on both faces of a part of this grade. These relationships were built according to the rough mill specifications of a Quebec component factory. 2.2.6. Cutting bill Various cutting bills were obtained from the same rough mill in Quebec. Based on these existing cutting bills, a standardised cutting bill was built for this study and it is an extension of the cutting bill used by Bouffard et al. (2002). Standardization was performed to reduce the number of actual lengths and widths into a limited number of dimensions that were easier to manage by the simulator but were still covering the actual range of dimension values. The cutting bill requests for three different types of products in infinite quantities. The cutting bill contains 8 different widths (from 45mm to 125mm) and 15 different lengths (from 290mm to 1800mm). Panel parts width were variable and ranged from 38mm to 76mm. Flooring parts length were variable and ranged from 350mm to 1850mm. Two grades are used for fixed dimension components (1F and 2F), two grades are used for panel parts (Pan 1F and Pan 2F) and three grades are used to define parts of flooring (Sup, Med, Inf). The complete cutting bill can be seen in Table 5. © Martin Caron 16 2.2.7. The pricing structure The use of a pricing structure is to provide a solid ground for the prioritization of parts while using a value based prioritization mechanism. With a pricing structure and a value prioritization mechanism, the usage of higher value parts is favored. Previous studies mentioned that a pricing structure applied to a cutting bill could emulate a dynamic part prioritization using dynamic weight factors (Thomas 1996b, Mitchel 2002). In order to build the pricing structure it was necessary to determine what attributes make a part more or less wanted. Three attributes were determined as being responsible for the value of parts : 1) the length, 2) the width, and 3) the grade. Rules defining the pricing structure in the present study were based on the establishment of relationships between parts attributes, NHLA hardwood grading rules and NHLA lumber value. 2.2.8. NHLA dimension definitions The pricing structure aims at establishing a formal base for defining a way to generate prices for every part in a cutting bill. Hence the price structure was not based on single part novelties or producer niche or market specialization. The pricing structure was rather based on dimensions from cuttings used in NHLA hardwood grading which is the core of the current hardwood trade. In NHLA grading, length, width and positioning of defects are used to determine the grade of each board. Three different length groups and three different width groups were built based on the NHLA hardwood grading cutting dimensions per grade and the standardized cutting bill. NHLA standard cutting dimensions are summarized in Table 6. Table 7 describes the dimension limits used to qualify pieces of the standardized cutting bill © Martin Caron 17 parts in the pricing structure established, based on similar relationships as seen in the NHLA grading system. 2.2.9. Value in the pricing structure depend on parts dimensions NHLA grading is the trade standard for the hardwood industry, thus every NHLA grade has an average price value attached to it in the market. Pricing differences between NHLA lumber grades are used in the pricing structure to create factor steps for the dimension groups. The pricing structure is based on formal steps aiming at being as close as possible to reality. The pricing structure is not a universal and an exact representation of what is happening in the market, but these factors could be modified in the simulation of a specific mill to map exactly a specific pricing structure. Table 8 shows the prices of green, unselected, hard maple. A dimension group will be linked to a specific NHLA grade to determine a basic level of comparison. Every other dimension group and grade will be compared to this basic level to determine its pricing factor. Thus, every part having dimensions or grade lower than this basic level will receive a penalty factor and every part having dimensions or grades higher than this basic level will receive a bonus factor. There is an important difference between lengths in the NHLA cuttings and a much smaller difference between widths. To determine pricing factors, “Short” will be linked to No. 2A Common because the “Short” group lengths are similar to the cutting lengths of No. 2A common. In the same way, “Medium” will be linked to No. 1 Common and “Long” will be linked to FAS / Select. No. 2A Common width dimensions are narrower compared to the © Martin Caron 18 other NHLA cutting width dimensions so 2A Common will be linked to “Narrow” in the width factors, No.1 Common will be linked to “Medium” and FAS / Select to “Wide”. A pricing factor needs to be given to grades of parts. Aside from a few special rules for positioning of splits and wane (NHLA 1998), all NHLA grades have a similar acceptation of defects. All NHLA cuttings are considered to have one clear face and one sound face. Those characteristics compare generally to the basic grade 1F used in the simulator. To determine the pricing factor scale between all seven grades, a survey was made from five Quebec rough mills to acquire average panel and fixed dimension components prices. A trade journal (Hardwood review weekly 2002) was used to determine prices for random length strip flooring. The Quebec Wood Export Bureau was consulted to examine and validate the pricing structure obtained (Labbé 2002). Pricing factors obtained for each grade are presented in Table 9. Product prices were obtained from standard furniture component and cabinet factories in Quebec, these produced part prices did not contain specifications for black cherry which is more of a marginal product. Thus prices for hard maple products were used in all grades specifications, including those for black cherry. Lumber prices were hence those of unselected hard maple based on a northern pricing and this for both species. The final price of a part is computed as the unit surface price multiplied by the part surface and by the total computed part factor (the length factor multiplied by the width factor multiplied by the grade factor). The base unit price (1.0) was priced at US$ 1700 per 1000 © Martin Caron 19 B.F of hard maple 1F components. The price of each part can be computed using this information. Using these guidelines and the pricing scale we can allocate a dollar value to all the parts in the standardised cutting bill. For example, a piece of 290 mm long and 65 mm wide of grade 2F would have a price of 0.548$. A 1600mm long piece by 65 mm wide of 2F grade would be allocated a price of 7.988$. 2.2.10. Simulation process Fifteen randomly generated board files each containing 65 boards (which amounts to approximately 200 B.F per file) were built for each wood species and NHLA grade. Optimization data was gathered in two prioritization modes (yield, and value) and three optimization processes (rip first, crosscut first, best of rip first / crosscut first). General optimization parameters are summarized in Table 10. All simulations were allowed 60 seconds per board to execute, after which time the simulator would return the best cutting solution found. Simulations using the process of “best of rip-first / crosscut-first” were allowed 120 seconds to keep the time per process constant. 2.2.11. Statistical analysis Simulation results were analysed using a two-way factorial ANOVA design. The significance of the prioritization strategy, the process and their interactions were examined in the model. Two different wood species and three NHLA grades were used which accounted © Martin Caron 20 to 6 ANOVA analyses. Fifteen observations were made per test (n=15). The statistical threshold used of significance was fixed at 99 %. All ANOVA tests presented here are based on the null hypothesis that “ the effects of the prioritization strategy and the rough mill process are equal for a given NHLA grade and a species of wood and have no effects on the simulation factor analysed (yield, value produced, return…)”. The factorial ANOVA model used is the following : y ijk = µ + α . j + β i. + γ ij + ε ijk , Where: 1. yijk is the matrix of observations. 2. µ is a matrix containing the averages of the observations (over-all means). 3. α . j is the prioritization effect. 4. β i. is the process effect. 5. γ ij is the interaction term. 6. ε ijk is the error term. Since our ANOVA analyses were comparing three processes with different values, a significant process difference noted in the ANOVA results did not necessarily mean that two optimization results from two different processes were significantly different when looked © Martin Caron 21 together. So paired t-tests (α<0,01) were done to compare results between each process values in our ANOVA table. The two way classification analysis is based on the Student t test. Thus, for this test to be valid, variances associated with different observations must be homogeneous. Cochran’s test for the homogeneity of variances will be used to determine if variances are equal. The computation with the variances S is the following (Bowker and Lieberman 1972) : G= highest S 2 , S12 + S22 +L+ Sk2 where k is the total number of observed variances. Homogeneity of variances is accepted if G ≤ gα (n) , where g α is given in a Cochran’s chart (Bowker and Lieberman 1972) and n is the number of repetitions (n=15). Cochran’s test proved the homogeneity of variances, hence the ANOVA could be performed validly. 2.3. Results and discussion The statistical analysis has shown that the prioritization strategy and the rough mill process both have significant effects on the output values from the simulator. The effect of the prioritization strategy on output values was the highest, followed by the optimization process. Moreover, it was seen that there is always an interaction between the prioritization and the rough mill process used, at a significance level of α ≤ 0.01. The interaction means that there © Martin Caron 22 is a direct relationship between the prioritization mechanism and the criterion evaluated, but this effect is different across processes. Three performance indicators have been used to compare results between the different simulations : 1) yield, 2) VPBF and 3) RPBF. Yield and RPBF will be discussed and examined in this section. VPBF is necessary as an intermediate result providing the total value of parts produced in a simulation for the calculation of RPBF. The optimization results are shown on Table 11 for hard maple and Table 12 for black cherry in all three NHLA grades (No.1 Common, No.2 Common and No.3 Common). The intuitive reasoning behind value prioritization suggested by Mitchell (2002) has been proven by the results. Simulations using value prioritization produced significantly less yield than simulations using surface prioritization, but produced significantly more VPBF, thus resulting in a much higher overall RPBF. General product distribution follows the products priority requirements. The most valuable product required in the product mix and the one that was most produced was dimension stock (Figures 2 and 3). Panels and flooring were used to fill gaps and improve overall value. Dimension stock production decreased while flooring production increased as lower NHLA grade lumber was processed in the simulations. Panel production also increased with a lower NHLA grade but to a lesser extent than flooring. The black cherry database produced a lower RPBF value in the intermediate and lower grades (No.2 and No.3 Common) than the hard maple which is confirmed by the product © Martin Caron 23 distribution. There appears to be a difference between black cherry and hard maple considering the proportions of dimension stocks and flooring components produced. Black cherry produced less dimension stock and more flooring in all NHLA grades than hard maple. These differences are explained by the higher proportion of dark heartwood and sound defects (sap stain mostly) in the black cherry lumber. 2.3.1. Effect of the processes on yield Using a surface prioritization mechanism for hard maple, the highest yield has been obtained from the selective process in all lumber grades. Between the two “non-selective” processes, rip-first has provided the best yield in No.1 Common. Yields were similar for both “nonselective” processes in No.2 Common. Crosscut-first provided the best yield for No.3 Common. Using a surface prioritization mechanism for black cherry, the highest yield has been provided by the selective process in all lumber grades. Between the two “non-selective” processes, rip-first has provided the best yield in No.1 and No.2 Common lumber. Crosscutfirst provided the best yield for No.3 Common lumber. Yield results using a surface prioritization mechanism are higher in both species using a ripfirst process in the higher grade of lumber used (No.1 Common). The difference between yields of the two "non-selective" processes diminished for No.2 Common and even disappeared for hard maple. For No.3 Common, crosscut-first provided higher yields for © Martin Caron 24 both species. These results confirm the relationships suggested by Wiedenbeck (2001) in her crosscut-first and rip-first comparison. The results also show the high yield conservation potential provided by a "selective" process. The yield is kept higher by the "selective" process which adapts more easily to board dimensional and defects characteristics. For example, using the rip-first process, yields varied from 73.4% to 61.7% between No.1 Common and No.3 Common. With the crosscut-first process, yields varied from 69.2% to 64.4%. By using the selective process, yields varied from 75.2% to 67.5%. The results tend to show that a selective process based solely on yield, used in the rough mill would provide a systematic higher yield and also give results that are more robust to lumber mix variations. Using a value prioritization mechanism for hard maple, the best yield giving process for No.1 Common lumber was the selective process, followed by the rip-first process, and finally the crosscut-first process. For No.2 and No.3 Common, the best yield giving process was crosscut-first, followed by the selective process and then finally the rip-first process. Using a value prioritization mechanism for black cherry, the best yield giving process for No.1 Common was the selective process, followed by the rip-first and crosscut-first processes which provided similar results. For No.2 Common, the crosscut-first process and the selective process were providing similar yields, followed by the rip-first process. For No.3 Common lumber, the best yield was provided by the crosscut-first process, followed by the selective process and finally the rip-first process. © Martin Caron 25 In the value prioritization mechanism, yield was not the criterion being maximized, thus yield results vary depending on the “value adding” potential brought by an addition of yield for each process. The "selective" process did not provide the best yields in all lumber grades, because the selection between rip-first solutions and crosscut-first solutions were done on the basis of value, not yield. Despite the fact that prioritization was not done on yield but value, yield results between the two "non-selective" processes showed a similar trend to the one observed when using a prioritization mechanism based on the maximization of surface. The rip-first process provided higher yield in the higher NHLA grade used, the gap between the two "non-selective" processes diminished in the intermediate grade (No.2 Common) while slightly being in favor of the crosscut-first process, and finally, the crosscut-first process provided a higher yield in the lower NHLA grade (No.3 Common). For example, yields obtained using No.1 Common hard maple were 65.6% in rip-first and 61.8% in crosscut-first. For No.2 Common, a yield of 64.7% was obtained in crosscut-first and 62.3% in rip-first. For No.3 Common, a yield of 59.5% was obtained in crosscut-first and 56.6% in rip-first. The best overall yield for both species and for both prioritization mechanisms were obtained using No.1 Common with a selective process. 2.3.2. Effect of the process on RPBF Using a surface prioritization mechanism for hard maple with No.1 Common lumber, the highest RPBF was provided by the selective process, followed by the rip-first process and finally the crosscut-first process. For No.2 Common, the highest RPBF were provided by the © Martin Caron 26 rip-first and selective processes whose values were similar, followed by the crosscut-first process. For No.3 Common, the highest RPBF was provided by the selective process, followed by the rip-first and crosscut-first processes which provided similar values. Using a surface prioritization mechanism for black cherry with No.1 Common lumber, the highest RPBF was provided by the rip-first process, followed by the best of rip-first and crosscut-first process and finally crosscut-first. For No.2 Common, RPBF obtained from the three different processes were not significantly different. For No.3 Common, highest RPBF were provided by the best of rip-first and crosscut-first and crosscut-first processes which provided similar values, followed by the rip-first process. Using a value prioritization mechanism for both species with No.1 Common and No.2 Common lumber the highest RPBF were obtained from the best of rip-first and crosscut-first and rip-first processes which provided similar values, followed by the crosscut-first process. For No.3 Common, the best RPBF giving process was the best of rip-first and crosscut-first process, followed by the rip-first process and finally the crosscut-first process. Value, and thus RPBF are not maximized using a surface prioritization mechanism. Results show only a small variation between the highest RPBF and the lowest RPBF obtained. This little variation is directly caused by the fact that value was not the criterion being maximized, thus parts produced have no prioritization on length and grade. Thus smaller pieces and lower grades are used to fill boards and obtain the highest possible yield. Comparing the two "non-selective" processes, rip first provided the best RPBF in the higher grades of lumber © Martin Caron 27 used (No.1 Common). For No.2 Common, the gap between the RPBF from the rip-first and crosscut-first processes diminished. For No.3 Common, crosscut-first provided the best RPBF. Since yield is being maximized these results follow the same trend and point directly to the results obtained from yield in a surface prioritization. For example, using No.1 Common hard maple, RPBF obtained in rip-first was 1.05 and in crosscut-first 0.95. For No.2 Common, a RPBF of 1.34 was obtained in rip-first and 1.29 in crosscut-first. For No.3 Common, a RPBF of 1.24 was obtained in rip-first and 1.23 in crosscut-first. RPBF results obtained using a value prioritization mechanism clearly show the "value adding" potential of the rip-first process in both species of lumber used. The variation of RPBF obtained between the two "non-selective" processes is very important and always in favor of the rip-first process except for No.3 Common black cherry. For No.1 Common and No.2 Common, rip-first provides a similar RPBF value to the selective process. For No.3 Common, the selective process provided a higher RPBF. than the rip-first process for both species. These results clearly point to a higher RPBF produced in a rip-first process for the higher and intermediate grades of lumber (No.1 and No.2 Common) . There are two possible sources for this significant increase of produced value :1) a longer average length of parts produced, 2) a higher average grade of parts produced. Results observed gave no insights or information pointing to a higher grade of parts being produced in rip-first versus crosscut first. This higher value is obtained from longer parts being produced from boards which are cut along their length axis in a rip-first process coupled with a value based part prioritization. For example, using No.1 Common hard maple, RPBF obtained in rip-first was 2.18 and in crosscut-first 1.60. For No.2 Common, a RPBF of 2.24 was obtained in rip-first and 1.79 in © Martin Caron 28 crosscut-first. For No.3 Common, a RPBF of 1.73 was obtained in rip-first and 1.55 in crosscut-first. The best overall RPBF for both species was obtained with the value prioritization mechanism with a rip-first or selective process. For hard maple, the best RPBF value was obtained for No.2 Common and for black cherry, the best RPBF value was obtained for No.1 Common. 2.3.3. Effect of the prioritization mechanism on yield Switching from a surface prioritization mechanism to a value prioritization mechanism for No.1 Common hard maple caused a significant yield diminution of 10.6% in rip-first, 10.7% in crosscut-first and 12.0% in the selective process. For No.2 Common hard maple switching prioritization mechanisms accounted for a significant yield diminution of 11.9% in rip-first, 8.4% in crosscut-first and 13.8% in the selective process. For No.3 Common hard maple the change of prioritization mechanism accounted for a significant yield diminution of 9.9% in rip-first, 7.6% in crosscut-first and 13.2% in the selective process. The change of prioritization mechanism from surface to value for No.1 Common black cherry accounted for a significant yield diminution of 14.8% in rip-first, 13.9% in crosscutfirst and 16.8% in the selective process. For No.2 Common black cherry, switching prioritization mechanism from surface to value produced accounted for a significant yield diminution of 13.7% in rip-first, 9.2% in crosscut-first and 15.9% in the selective process. For No.3 Common black cherry the change of prioritization mechanism accounted for a © Martin Caron 29 significant yield diminution of 13.2% in rip-first, 9.0% in crosscut-first and 16.9% in the selective process. For both species, the "selective" process suffered from the highest yield diminution by changing the prioritization mechanism from the maximization of surface to the maximization of value. The results have also shown that between the "non-selective" processes, rip-first suffers from a higher yield diminution than crosscut-first with a change of prioritization mechanism. These results support the basic hypothesis of this study and are in agreement with the comments by Mitchell (2002). Switching from a surface based part prioritization mechanism to a value based part prioritization mechanism has the overall effect of decreasing obtained yield. The results also proved that a value based part prioritization has an important effect on the selection of parts produced because of the important yield variations observed. 2.3.4. Effect of the prioritization mechanism on RPBF Switching from a surface prioritization mechanism to a value prioritization mechanism for No.1 Common hard maple caused a RPBF increase of 107.6% in rip-first, 68.4% in crosscut-first and 101.8% in the best of rip-first and crosscut-first. For No.2 Common hard maple, the change of prioritization strategy accounted for a RPBF increase of 67.2% in ripfirst, 38.8% in crosscut-first and 68.6% in the best of rip-first and crosscut-first. For No.3 Common hard maple, the change of prioritization strategy accounted for a RPBF increase of 39.5% in rip-first, 26.0% in crosscut-first and 43.8% in the best of rip-first and crosscut-first. © Martin Caron 30 The change of prioritization strategy from yield to value for No.1 Common black cherry accounted for a yield diminution of 115.2% in rip-first, 101.5% in crosscut-first and 133.7% in the best of rip-first and crosscut-first. For No.2 Common black cherry the change of prioritization strategy accounted for a RPBF increase of 65.6% in rip-first, 38.9% in crosscutfirst and 66.7% in the best of rip-first and crosscut-first. For No.3 Common black cherry, the change of prioritization strategy accounted for a yield diminution of 41.9% in rip-first, 29.0% in crosscut-first and 43.6% in the best of rip-first and crosscut-first. These results proved the second part of the study hypothesis. RPBF, is considerably higher, when prioritizing parts with a value prioritization mechanism rather than a surface prioritization mechanism. The results also confirmed that a value based part prioritization has an important effect on the selection of parts produced. Obtaining a much higher produced RPBF while observing a yield diminution proves that the selection of parts is based on a sound basis which is the pricing structure used for parts. For both species, the results point to higher benefits from switching from a surface to a value prioritization mechanism for the higher NHLA grades, the obtained benefit tends to decrease with grade. Results have shown for both species that surface prioritization gave the smallest RPBF from the higher grades of lumber. This means that while yield obtained was high, products issued from a surface prioritization mechanism were clearly composed of parts of low value (shorter and lower grade parts). This result is important because it proves that an increase of yield obtained from a complex cutting bill does not necessarily provide an increase of RPBF. The © Martin Caron 31 maximization of surface does not provide the best value obtained from the board in a rough mill, because hardwood pricing is based on length, width, and grade. As shown by this results on RPBF both concerning the effects of the process and the effects of prioritization, the maximization of surface tends to produce shorter or lower grade parts. Results also confirmed the strength of RPBF in comparing different rough mill processes, producing different product mixes and using different grades of lumber as input. They have shown that yield is a complex and potentially incomplete value when used as a performance index between different rough mills. Yield by definition does not contain information on the produced parts, the volume of wood required for production nor the cost of wood used for the production. © Martin Caron 32 2.4. Conclusions This study was made to compare the effects for hard maple and black cherry of prioritizing rough mill operations with the maximization of value using a part pricing structure inspired by the NHLA hardwood grading rules, versus the prioritization of rough mill operations based on pure yield. The comparison required the examination of the direct effects and interactions of the three rough mill processes (rip-first, crosscut-first and best of rip-first and crosscut-first) and the two prioritization strategies (maximization of the yield, maximization of the value) used in the CRIQ “Two Axis Optimiser”. Overall results confirmed the hypothesis in favor of a valued based part prioritization strategy. Simulations using the value prioritization mechanism produced less yield than simulations using the surface prioritization mechanism, but always produced significantly better RPBF. This result is important because it emphasizes an aspect that has not been thoroughly studied in past work. Higher yield is not directly related to a higher monetary return value. The current standard for rough mill performance (i.e. yield) is based on a result that is not always producing the optimum value of parts, thus not giving the component rough mill plant the best possible value for the wood it is using. The yield results obtained from a surface prioritization mechanism have been higher in both species using rip-first in the higher grade of lumber used (No.1 Common). The difference between yields of the two "non-selective" processes was smaller for No.2 Common and even © Martin Caron 33 non-significant for hard maple. Crosscut-first provided higher yields for both species in the lower grade of NHLA lumber used (No.3 Common). The results also show the high yield conservation potential provided by a "selective" process like the best of rip-first and crosscut-first process with a surface maximization mechanism. Even as lower NHLA grade lumber is processed, the yield is kept higher by the "selective" process which adapts more easily to board characteristics and dimensions. This tends to show that a selective process used in the rough mill would provide a constant higher yield and also give yield results more robust to lumber mix variations, when yield is used as a performance indicator. These results clearly point to a higher RPBF produced in a rip-first process for the higher and intermediate grades of lumber (No.1 and No.2 Common) . There are two possible sources for this significant increase of produced value :1) a longer average length of parts produced, 2) a higher average grade of parts produced. Results observed gave no insights or information regarding the production of parts of a higher grade in rip-first versus crosscut first. This higher value is obtained from longer parts being produced from boards which are cut along their length axes in a rip-first process coupled with a value based part prioritization. This study has confirmed the strength of RPBF in comparing different rough mill processes, producing different product mixes and using different grades of lumber as input. They have shown that yield is a complex and potentially incomplete index to evaluate performance between different rough mills. Yield by definition does not contain information on the © Martin Caron 34 produced parts, the volume of wood required for the production nor the cost of wood used for the production. It is important to note that this study was done requiring infinite part quantities because of the component producer context it was linked to. The results obtained with infinite quantities point to the use of a pricing structure for parts in a simulator with value-based part prioritization as a means to provide better results than a “pure yield” strategy. Future work could be done comparing the effects of part prioritization between value and surface using finite cutting bills to determine the potential industrial use of a pricing structure in other production contexts such as the furniture industry. © Martin Caron 35 2.5. Litterature cited Bouffard J.-F., Beauregard R. et T. Lihra. 2002. Hardwood rough mill optimization : comparison of various approaches and simulation software. Minutes du 4ième Groupe de travail de l’IUFRO WP S5-01-04 « Connection between Forest Ressources and Wood Quality : Modelling Approaches and Simulation Software », Harrison (Vancouver) BC, 9-13 septembre 2002. Bowker, A. H. et G.J. Lieberman.1972. Engineering statistics 2nd Edition. Prentice-Hall press. 1972. p264. Buehlmann, U. P. 1998. Understanding the relationship of lumber yield and cutting bill requirements : a statistical approach. Doctoral dissertation, unpublished, Virginia Polytechnic Institude and State University, Blacksburg, VO, 209p. Conners, R. W., D. E. Kline, P. A. Araman, an T. H. Drayer 1997. Machine vision technology for the forest product industry. IEEE computer innovative technology for computer professionals 30(7):43-48. Caron, M. 2001. Optimiseur 2 axes : Manuel de l’utilisateur [CRIQ rough mill simulator user's guide in french]. Centre de Recherche Industrielle du Québec. Sainte-Foy, Quebec. 10-12-2001. 55p. © Martin Caron 36 El-Radi, T. E., S. H. Bullard and P. H. Steele. 1994. A performance evaluation of edging and trimming operations in U.S. hardwood sawmills. Canadian Journal of Forestry Research 23:1450-1456. Gatchell, C. J. 1990. The effect of crook on yields when processing narrow lumber with a fixed arbor gang ripsaw. Forest Products Journal 40(5):9-17. Gatchell, C. J. and R. E. Thomas. 1997. Within-grade quality differences for 1 and 2A Common lumber affect processing and yields when gang-ripping red oak lumber. Forest Products Journal 47(10):85-90. Gatchell, C.J and R. E. Thomas. 1998. Data bank for kiln-dried red oak lumber. Research paper NE-245 USDA Forest service, Northeastern Forest Experiment Station, Radnor, PA. 60p. Hamner, P. C., Bond, B. Wiedenbeck, J. 2002. The effect of lumber length on part yields in gang-rip-first rough mills. Forest Products Journal 52(5):71-76. Hardwood review weekly. 2002. Hardwood pricing, hard maple green unselected, northern pricing. March 2002. Vol. 18. Issue 28. © Martin Caron 37 Hoff, K., N. Fisher, S. Miller, and A. Webb. 1997. Sources of competitiveness for secondary wood products firms : A review of literature and research issues. Forest Products Journal 47(2):31-37 Labbé S. 2002. CEO of the Québec Wood Export Bureau. Personnal communication. Mitchell, P. 2002. Setting priorities on rough mill cutoff saws. Wood Digest May 2002 : 41-45. Mullin, S. 1990. Why switch to rip-first ? Furniture design & manufacturing, 62(9):36-42. National Hardwood Lumber Association. 1998. NHLA 1998 Hardwood Grading rules and inspection handbook. Memphis, TN, USA. 90p. Steele, H. P. And Wiedenbeck, J. K. 1999. The influence of lumber grade on machine productivity in the rough mill. Forest products Journal 49(9) : 48-54. Thomas, R. E. 1996a. ROMI-RIP : An analysis tool for rip-first rough mill operations. Forest Products Journal 46(2):57-60. Thomas, R. E. 1996b. Prioritizing parts from cutting bills when gang ripping first. Forest Products Journal 46(10):61-66. © Martin Caron 38 Thomas, R. E. 1997. ROMI-CROSS : An analysis tool for crosscut-first rough mill operations. Forest Products Journal 48(2):68-72. Thomas, R. E. And Buehlmann. U. 2002.Validation of the ROMI-RIP rough mill simulator. Forest Products Journal 52(2):23-29. Wiedenbeck, J. K. 2001. Deciding between crosscut and Rip-first processing. Wood and Wood Products August 2001. 5p. Wiedenbeck, J. K. And C. Scheerer. 1996. A report on rough mill yield practices and performances – how well are we doing ? Twenty-fourth annual Hardwood Symposium of the Hardwood Research Council, 121-128. © Martin Caron 39 Table 1. — Description of the hard maple and black cherry board database Average length Name (mm) Std. Average Std. Average Total deviation width deviation lumber lumber length width volume volume (mm) (mm) (mm) (B.F) (B.F) Hard maple – 1 Common 2268.56 277.54 159.69 47.38 3.67 503.12 Hard maple – 2 Common 2095.05 449.08 136.63 37.34 2.89 512.19 Hard maple – 3 Common 1734.95 500.68 136.76 35.14 2.37 320.54 Black cherry – 1 Common 1746.43 197.68 157.35 41.00 2.77 529.62 Black cherry – 2 Common 2399.56 128.68 138.47 33.27 3.31 580.01 Black cherry – 3 Common 2412.43 118.90 118.32 24.27 2.78 387.02 Table 2. — Part grades description Grade name Product Example Pan 2F Clear panel 2 faces Kitchen cabinet doors 2F 2 clear face dimension part Bed rim part Pan 1F Clear panel 1 face Table tops 1F 1 clear face dimension part Front drawer part Sup Superior grade flooring component « Select & better » flooring part Med Medium grade flooring component « Traditionnal » flooring part Inf Inferior grade flooring component « Western » flooring part © Martin Caron 40 Table 3. — Descriptions of defects in the board database Defect Description Void Sections around the board that are included in the rectangular section but do not contain wood (crook, taper ). Unacceptable Defects that were considered high removal priority by the markers : deep wane, rot, large splits, holes, shake. Sound defect Defects that are keeping the structural integrity of the board (sound knots, sap stain, mineral stain...). Pin knots Knots that are smaller than 3mm in diameter. Light heartwood Heartwood where colour contrast with sapwood is light. Dark heartwood Heartwood where colour contrast with sapwood is heavy. Table 4 . — Definition of part grades in relation with defects Grade name Pan 2F Pan 1F 2F 1F Sup Med Inf © Martin Caron Void Unacceptable Sound defects Pin knots Light heartwood Dark heartwood ― ― ― ― ― ― ― ― ― ― ― ― ― ― ― ― ― ― 1F 1F 2F ― 1F 2F 2F 2F 2F 2F ― 1F ― 1F 1F 2F 2F ― 1F ― 1F 1F 2F 2F 41 Table 5. —Cutting bill used in the simulations Length (mm) Width (mm) Grade 290 290 290 350 350 350 400 450 450 450 450 450 500 500 500 500 550 600 600 650 650 650 650 750 750 750 850 1050 1050 1050 1050 1250 1250 1500 1500 1500 1600 1600 1600 1600 1800 1800 1800 1800 350 400 450 500 550 650 850 1050 1250 1500 1800 random random random 65 76 85 65 70 76 65 45 70 76 85 125 65 70 76 85 105 65 85 55 65 85 105 65 85 125 65 45 65 85 105 65 85 65 75 85 65 75 85 105 55 65 75 85 random random random random random random random random random random random 65 90 115 2F, 1F 2F, 1F 2F, 1F 2F, 1F 2F, 1F 2F, 1F 2F, 1F 2F, 1F 2F, 1F 2F, 1F 2F, 1F 2F, 1F 2F, 1F 2F, 1F 2F, 1F 2F, 1F 2F, 1F 2F, 1F 2F, 1F 2F, 1F 2F, 1F 2F, 1F 2F, 1F 2F, 1F 2F, 1F 2F, 1F 2F, 1F 2F, 1F 2F, 1F 2F, 1F 2F, 1F 2F, 1F 2F, 1F 2F, 1F 2F, 1F 2F, 1F 2F, 1F 2F, 1F 2F, 1F 2F, 1F 2F, 1F 2F, 1F 2F, 1F 2F, 1F Pan2F, Pan1F Pan2F, Pan1F Pan2F, Pan1F Pan2F, Pan1F Pan2F, Pan1F Pan2F, Pan1F Pan2F, Pan1F Pan2F, Pan1F Pan2F, Pan1F Pan2F, Pan1F Pan2F, Pan1F Sup, Med, Inf Sup, Med, Inf Sup, Med, Inf © Martin Caron 42 Table 6. — Cutting dimensions in the NHLA grading system Length (mm) Width (mm) 2 Common 610 51 1 Common 915 / 610 75 / 102 2133 / 1524 75 / 102 NHLA grade FAS Table 7. — Part dimension limits used for the pricing structure of the cutting bill Dimension Length Width Attribute Dimension stock Panels Flooring components Minimum (mm) Maximum (mm) Minimum (mm) Maximum (mm) Minimum (mm) Maximum (mm) Short 290 650 350 650 x x Medium 750 1250 850 1250 350 1850 Long 1500 1800 1500 1800 x x Narrow 45 55 x x 65 65 Medium 65 76 38 76 90 90 Wide 85 125 x x 115 115 Table 8. — Hard maple pricing, and width/length factors [Unselected, Green, net tally] (Hardwood Review Weekly March 2002) NHLA grade Price $ Length factor Width factor 2 Common 590 1.0 0.95 1 Common 1020 1.73 1.0 FAS 1560 2.64 1.05 © Martin Caron 43 Table 9. — Product / grades pricing and price factor Real Product Simulator grade $ / 1000 B.F Price factor Kitchen Cabinet 2F Pan 2F 3000 1.79 Dimension stock 2F 2F 2700 1.59 Kitchen cabinet 1F Pan 1F 2500 1.47 Dimension stock 1F 1F 1700 1.00 Strip flooring - Sel/Btr Sup 1420 0.84 Strip flooring - 1 Common Med 1135 0.67 Strip flooring - 2 Common Inf 935 0.55 © Martin Caron 44 Table 10. —Optimisation parameters Parameter Value Product priority Dimension stocks, pannels, flooring Optimisation process Rip first, crosscut first, best of rip first / crosscut first Maximum optimisation time 60s per board (per process) Rip saw kerf 4mm Crosscut saw kerf 4mm Saw blades All blades moveable Prioritisation technique Yield, Value © Martin Caron 45 238.6 238.6 Surface Value 188.9 188.9 Surface Value 154.2 154.2 Surface Value 61.8% (0.7) 65.6% (0.7) 59.5% (0.9) 55.6% (0.8) 58.6% (0.8) abc abc 64.4% (1.0) 61.7% (0.8) 67.5% (1.0) Yield Yield Paired (Cross) (Rip+Cross) tests¹ Yield (Rip) 0.000 abc aab 70.6% (0.6) 64.7% (0.8) 70.7% (1.1) 62.3% (1.2) 74.0% (0.6) 63.8% (0.9) Yield Yield Paired (Cross) (Rip+Cross) tests¹ abc Yield (Rip) 0.000 66.2% (0.7) abc 69.2% (1.0) 73.4% (0.5) 75.2% (0.5) Yield Yield Paired (Cross) (Rip+Cross) tests¹ Yield (Rip) 1.60 (0.05) 2.13 (0.05) VPBF³ (Rip) 1.63 (0.04) © Martin Caron ³ VPBF and RPBF are expressed in US $ per board feet ² results of the two-way ANOVA classification 2.24 (0.05) 0.000 1.95 (0.05) abc 1.73 (0.05) Paired RPBF³ VPBF³ VPBF³ (Cross) (Rip+Cross) tests¹ (Rip) 1.24 1.63 1.67 aab (0.04) (0.04) (0.05) 0.000 aba 1.55 (0.05) RPBF³ (Cross) 1.23 (0.05) 1.29 (0.06) 1.79 (0.09) aba 1.34 (0.07) 2.24 (0.11) 1.96 (0.08) 2.90 (0.12) 1.93 (0.08) 2.83 (0.12) 1.88 (0.06) 2.38 (0.10) RPBF³ (Cross) Paired RPBF³ VPBF³ VPBF³ (Cross) (Rip+Cross) tests¹ (Rip) VPBF³ (Rip) 0.000 aba 2.18 (0.06) 3.24 (0.06) 3.20 (0.06) 2.69 (0.05) RPBF³ (Cross) 0.95 (0.04) Paired RPBF³ VPBF³ VPBF³ (Cross) (Rip+Cross) tests¹ (Rip) 1.97 2.12 1.05 abc (0.04) (0.04) (0.04) VPBF³ (Rip) 2.07 (0.04) 0.000 p-values² ¹ results of the paired test comparison between the processes for each output variable B.F Priorization Hard maple – No.3 common p-values² B.F Priorization Hard maple – No.2 common p-values² B.F Priorization Hard maple – No.1 common Table 11. — Optimization results for the hard maple database aba aba aba 0.000 1.84 (0.05) abc Paired RPBF³ (Rip+Cross) tests¹ 1.28 aab (0.04) 0.000 1.37 (0.08) 2.31 (0.11) Paired RPBF³ (Rip+Cross) tests¹ 0.000 2.22 (0.06) Paired RPBF³ (Rip+Cross) tests¹ 1.10 abc (0.06) 46 180.6 180.6 Surface Value 215.9 215.9 Surface Value 180.4 180.4 Surface Value 62.7% (1.0) 63.5% (1.0) 61.4% (1.2) 53.7% (1.4) 58.9% (1.15) abc abc 67.5% (1.6) 61.9% (1.2) 70.9% (1.3) Yield Yield Paired (Cross) (Rip+Cross) tests¹ Yield (Rip) 0.000 abb abc 66.7% (0.8) 60.6% (0.9) 67.7% (0.9) 58.4% (1.3) 72.2% (0.7) 60.7% (1.4) Yield Yield Paired (Cross) (Rip+Cross) tests¹ aba Yield (Rip) 0.000 64.3% (1.1) abc 72.8% (1.1) 74.5% (0.7) 77.3% (0.8) Yield Yield Paired (Cross) (Rip+Cross) tests¹ Yield (Rip) 1.37 (0.08) 1.61 (0.05) VPBF³ (Rip) 1.25 (0.04) © Martin Caron ³ VPBF and RPBF are expressed in US $ per board feet ² results of the two-way ANOVA classification 1.74 (0.05) 0.000 1.59 (0.05) aab 1.22 (0.05) Paired RPBF³ VPBF³ VPBF³ (Cross) (Rip+Cross) tests¹ (Rip) 0.86 1.32 1.33 abb (0.04) (0.04) (0.03) 0.000 abc 1.20 (0.05) RPBF³ (Cross) 0.93 (0.04) 0.90 (0.06) 1.25 (0.06) aaa 0.90 (0.06) 1.49 (0.08) 1.52 (0.07) 2.14 (0.07) 1.49 (0.06) 2.08 (0.07) 1.49 (0.06) 1.84 (0.06) RPBF³ (Cross) Paired RPBF³ VPBF³ VPBF³ (Cross) (Rip+Cross) tests¹ (Rip) VPBF³ (Rip) 0.000 aba 1.98 (0.10) 3.03 (0.10) 3.00 (0.10) 2.39 (0.08) RPBF³ (Cross) 0.68 (0.05) Paired RPBF³ VPBF³ VPBF³ (Cross) (Rip+Cross) tests¹ (Rip) 1.70 1.88 0.92 abc (0.05) (0.07) (0.06) VPBF³ (Rip) 1.94 (0.06) 0.000 p-values² ¹ results of the paired test comparison between the processes for each output variable B.F Priorization Black cherry –No.3 common p-values² B.F Priorization Black cherry –No.2 common p-values² B.F Priorization Black cherry –No.1 common Table 12. — Optimization results for the black cherry database aba abc aaa 0.000 1.35 (0.05) aab Paired RPBF³ (Rip+Cross) tests¹ 0.94 abb (0.03) 0.000 0.93 (0.07) 1.55 (0.07) Paired RPBF³ (Rip+Cross) tests¹ 0.000 2.01 (0.10) Paired RPBF³ (Rip+Cross) tests¹ 0.86 abc (0.07) 47 Figure 1. —BorealScan image acquisition system © Martin Caron 48 Figure 2. ― Products distribution for hard maple Products distribution for hard maple 100% Percentage of production (%) 90% 80% 70% 60% Flooring 50% Pannels 40% Dimension stock 30% 20% 10% 0% 1 Common surface 1 Common value 2 Common surface 2 Common value 3 Common surface 3 Common value NHLA grade - prioritization © Martin Caron 49 Figure 3. ― Products distribution for black cherry Products distribution for black cherry 100% Percentage of production (%) 90% 80% 70% 60% Flooring 50% Pannels 40% Dimension stock 30% 20% 10% 0% 1 Common surface 1 Common value 2 Common - 2 Common surface value 3 Common surface 3 Common value NHLA grade - prioritization © Martin Caron 50 3. Conclusion Cette étude a été accomplie dans le but de comparer les effets de prioriser les opérations d’une usine de débitage de composants de bois francs selon la valeur produite des pièces, en utilisant une structure de prix inspirée des règles de classification des bois francs de la NHLA, en comparaison avec l’utilisation d’une stratégie de placement basée seulement sur le rendement matière. La comparaison nécessitait l’examen des effets et des interactions entre trois procédés de débitage (délignage en tête, tronçonnage en tête et meilleur entre délignage en tête et tronçonnage en tête) et deux mécanismes de priorisation (la maximisation de la valeur produite et la maximisation du rendement matière) utilisés dans « l’Optimiseur deux axes » développé par le CRIQ. Les résultats globaux ont confirmé l’hypothèse de la supériorité d’une stratégie de priorisation des pièces selon la maximisation de la valeur produite. Les simulations utilisant un mécanisme de priorisation selon la valeur des pièces ont obtenu un rendement inférieur aux simulations maximisant le rendement matière, mais dans tous les cas ont produit un RPBF nettement supérieur. Ce résultat est majeur car il met en lumière un fait qui n’a pas été profondément analysé dans les ouvrages précédents ; « Faire plus, ne veut pas dire, nécessairement faire mieux ». Un rendement matière supérieur n’apporte pas directement un retour monétaire plus élevé. Ceci démontre que les indicateurs courants de performance pour les usines de débitage de bois franc (i.e. le rendement matière) sont basés sur un résultat qui n’apporte pas nécessairement une valeur optimale de pièces, donc qui n’apporte pas à l’usine et l’entreprise la meilleure valeur possible en retour sur le bois qu’elle utilise. Les rendements obtenus dans une optimisation maximisant le surface produite ont été plus élevés pour les deux essences de bois en utilisant le délignage en tête pour les grades supérieurs utilisés (No.1 Commun). La différence entre les rendements entre les deux procédés « non-sélectifs » était beaucoup plus faible dans le grade No.2 Commun (voir © Martin Caron 51 même nulle dans le cas de l’érable). Le tronçonnage en tête a apporté de meilleurs rendements pour les deux essences de bois quand le grade inférieur était utilisé (No.3 Commun). Ces résultats confirment les relations suggérées par Wiedenbeck (2001) dans sa comparaison entre les procédés de délignage en tête et de tronçonnage en tête. Les résultats pointent aussi vers le haut potentiel apporté par un procédé « sélectif » comme le procédé de sélection entre la meilleure solution du délignage en tête et du tronçonnage en tête avec un mécanisme de priorisation basé sur la maximisation de la surface. Les rendements obtenus sont supérieurs et conservés à un niveau plus élevés malgré l’utilisation de grades NHLA inférieurs. Le procédé « sélectif » s’adapte aux dimensions et caractéristiques des planches. Ceci démontre qu’un procédé « sélectif » utilisé dans une usine de débitage apporterait un rendement matière plus élevé, plus constant et plus robuste aux variations de mélange de bois utilisé lors du débitage, quand le rendement est utilisé comme indicateur de performance. Les résultats démontrent aussi clairement un RPBF nettement supérieur produit pour les grades supérieur et intermédiaires utilisés (No.1 Commun et No.2 Commun) en délignage en tête avec une maximisation de la valeur produite des pièces. Il y a deux sources possibles pour cette hausse significative de valeur produite : 1) une longueur moyenne plus élevée pour les pièces produites, 2) un grade moyen plus élevé pour les pièces produites. Les résultats observés n’ont pas démontré d’augmentation du grade des pièces produites en délignage en tête versus en tronçonnage en tête. Les résultats indiquent donc qu’en priorisant le débitage selon la maximisation de la valeur des pièces produites, le délignage en tête produit des pièces plus longues que le tronçonnage en tête. Les résultats ont confirmé la force du RPBF pour comparer différents procédés de débitage utilisant différents mélanges de produits et différents grades de bois en entrée. Ils démontrent que le rendement est une indicateur complexe et potentiellement incomplet quand utilisé pour comparer des usines différentes. Par sa définition, le rendement matière ne contient pas d’information sur les pièces produites, le volume de bois nécessaire pour produire ces pièces, ni le coût du bois nécessaire pour la production. © Martin Caron 52 Il est important de noter que cette étude a été faite sur la base d’une demande infinie de pièces à produire en relation avec le contexte de producteur de composants. Les résultats de cette étude tendent à supposer que l’utilisation d’une structure de prix pour les pièces dans un simulateur utilisant une stratégie de priorisation de valeur produite va apporter de meilleurs résultats qu’une stratégie d’optimisation selon le rendement matière. Des recherches futures comparant les effets de la stratégie de priorisation entre la valeur produite et la surface produite devrait être faites en utilisant une demande finie pour les pièces du carnet de commande. Ceci correspondant au contexte de l’industrie du meuble par exemple. © Martin Caron 53 4. Liste des références Bouffard J.-F., Beauregard R. et T. Lihra. 2002. Hardwood rough mill optimization : comparison of various approaches and simulation software. Minutes du 4ième Groupe de travail de l’IUFRO WP S5-01-04 « Connection between Forest Resources and Wood Quality : Modelling Approaches and Simulation Software », Harrison (Vancouver) BC, 9-13 septembre 2002. Bowker, A. H. et G.J. Lieberman.1972. Engineering statistics 2nd Edition. Prentice-Hall press. 1972. p264. Buehlmann, U. P. 1998. Understanding the relationship of lumber yield and cutting bill requirements : a statistical approach. Thèse de doctorat, non publiée, Virginia Polytechnic Institute and State University, Blacksburg, VO, 209p. Conners, R. W., D. E. Kline, P. A. Araman, an T. H. Drayer 1997. Machine vision technology for the forest product industry. IEEE computer innovative technology for computer professionals 30(7):43-48. Caron, M. 2001. Optimiseur 2 axes : Manuel de l’utilisateur [CRIQ rough mill simulator user’s guide version française seulement]. Centre de Recherche Industrielle du Québec. Sainte-Foy, Québec, 10-12-2001. 55p. El-Radi, T. E., S. H. Bullard and P. H. Steele. 1994. A performance evaluation of edging and trimming operations in U.S. hardwood sawmills. Canadian Journal of Forestry Research 23:1450-1456. Gatchell, C. J. 1990 The effect of crook on yields when processing narrow lumber with a fixed arbor gang ripsaw. Forest Products Journal 40(5):9-17. © Martin Caron 54 Gatchell, C. J. and R. E. Thomas. 1997. Within-grade quality differences for 1 and 2A Common lumber affect processing and yields when gang-ripping red oak lumber. Forest Products Journal 47(10):85-90. Gatchell, C.J and R. E. Thomas. 1998. Data bank for kiln-dried red oak lumber. Research paper NE-245 USDA Forest service, Northeastern Forest Experiment Station, Radnor, PA. 60p. Hall, S. P. , R. A. Wysk, E. M. Wengert, and M. H. Agee. 1980. Yield distribution and cost comparison of a cross-cut first and a gang-rip-first rough mill producing hardwood dimension stair parts. Forest Products Journal 30(5):34-39. Hamner, P. C., Bond, B. Wiedenbeck, J. 2002. The effect of lumber length on part yields in gang-rip-first rough mills. Forest Products Journal 52(5):71-76. Hardwood review weekly. 2002. Hardwood pricing, hard maple green unselected, northern pricing. March 2002. Vol 18. Issue 28. Hoff, K., N. Fisher, S. Miller, and A. Webb. 1997. Sources of competitiveness for secondary wood products firms : A review of literature and research issues. Forest Products Journal 47(2):31-37. Industrie Canada. 2002. La série des cadres de compétivité sectorielle : Les produits forestiers. Site web Strategis : http://strategis.ic.gc.ca/SSGF/fb01003f.html Labbé S. 2002. CEO of the Québec Wood Export Bureau. Communication personnelle. Mitchell, P. 2002. Setting priorities on rough mill cutoff saws. Wood Digest May 2002 : 41-45. Mullin, S. 1990. Why switch to rip-first ? Furniture design & manufacturing, 62(9):36-42. © Martin Caron 55 National Hardwood Lumber Association. 1998. NHLA 1998 Hardwood Grading rules and inspection handbook. Memphis, TN, USA. 90p. Steele, H. P. And Wiedenbeck, J. K. 1999. The influence of lumber grade on machine productivity in the rough mill. Forest products Journal 49(9) : 48-54. Thomas, R. E. 1996a. ROMI-RIP : An analysis tool for rip-first rough mill operations. Forest Products Journal 46(2):57-60. Thomas, R. E. 1996b. Prioritizing parts from cutting bills when gang ripping first. Forest Products Journal 46(10):61-66. Thomas, R. E. 1997. ROMI-CROSS : An analysis tool for crosscut-first rough mill operations. Forest Products Journal. 48(2):68-72. Thomas, R. E. And Buehlmann. U. 2002.Validation of the ROMI-RIP rough mill simulator. Forest Products Journal 52(2) : 23-29. Wiedenbeck, J. K. 2001. Deciding between crosscut and Rip-first processing. Wood and Wood Products August 2001. 5p. Wiedenbeck, J. K. And C. Scheerer. 1996. A report on rough mill yield practices and performances – how well are we doing ? Twenty-fourth annual Hardwood Symposium of the Hardwood Research Council, p121-128. © Martin Caron 56 Annexe 1 : Prix calculés Longueur Largeur Surface (mm) (mm) (mm²) Grade Facteur Facteur Longueur Largeur Facteur Grade prix/mm² Prix calculé 290.00 65.00 18850 2F 1 1 1.59 1.8298E-05 0.54842 290.00 76.00 22040 2F 1 1 1.59 1.8298E-05 0.64123 290.00 85.00 24650 2F 1 1 1.59 1.8298E-05 0.71716 350.00 65.00 22750 2F 1 1 1.59 1.8298E-05 0.66188 350.00 70.00 24500 2F 1 1 1.59 1.8298E-05 0.71280 350.00 76.00 26600 2F 1 1 1.59 1.8298E-05 0.77390 400.00 65.00 26000 2F 1 1 1.59 1.8298E-05 0.75644 450.00 45.00 20250 2F 1 0.95 1.59 1.8298E-05 0.55969 450.00 70.00 31500 2F 1 1 1.59 1.8298E-05 0.91646 450.00 76.00 34200 2F 1 1 1.59 1.8298E-05 0.99501 450.00 85.00 38250 2F 1 1.05 1.59 1.8298E-05 1.16848 450.00 125.00 56250 2F 1 1.05 1.59 1.8298E-05 1.71835 500.00 65.00 32500 2F 1 1 1.59 1.8298E-05 0.94555 500.00 70.00 35000 2F 1 1 1.59 1.8298E-05 1.01828 500.00 76.00 38000 2F 1 1 1.59 1.8298E-05 1.10557 500.00 85.00 42500 2F 1 1.05 1.59 1.8298E-05 1.29831 550.00 105.00 57750 2F 1 1.05 1.59 1.8298E-05 1.76418 600.00 65.00 39000 2F 1 1 1.59 1.8298E-05 1.13466 600.00 85.00 51000 2F 1 1 1.59 1.8298E-05 1.48378 650.00 55.00 35750 2F 1 0.95 1.59 1.8298E-05 0.98810 650.00 65.00 42250 2F 1 1 1.59 1.8298E-05 1.22921 650.00 85.00 55250 2F 1 1.05 1.59 1.8298E-05 1.68781 650.00 105.00 68250 2F 1 1.05 1.59 1.8298E-05 2.08494 750.00 65.00 48750 2F 1.73 1 1.59 1.8298E-05 2.45370 750.00 85.00 63750 2F 1.73 1.05 1.59 1.8298E-05 3.36912 750.00 125.00 93750 2F 1.73 1.05 1.59 1.8298E-05 4.95459 850.00 65.00 55250 2F 1.73 1 1.59 1.8298E-05 2.78086 1050.00 45.00 47250 2F 1.73 0.95 1.59 1.8298E-05 2.25929 1050.00 65.00 68250 2F 1.73 1 1.59 1.8298E-05 3.43518 1050.00 85.00 89250 2F 1.73 1.05 1.59 1.8298E-05 4.71677 1050.00 105.00 110250 2F 1.73 1.05 1.59 1.8298E-05 5.82659 2F 1.73 1 1.59 1.8298E-05 4.08950 1250.00 65.00 81250 1250.00 85.00 106250 2F 1.73 1.05 1.59 1.8298E-05 5.61520 1500.00 65.00 97500 2F 2.64 1 1.59 1.8298E-05 7.48875 1500.00 75.00 112500 2F 2.64 1 1.59 1.8298E-05 8.64086 © Martin Caron 57 1500.00 85.00 127500 2F 2.64 1.05 1.59 1.8298E-05 10.28263 1600.00 65.00 104000 2F 2.64 1 1.59 1.8298E-05 7.98800 1600.00 75.00 120000 2F 2.64 1 1.59 1.8298E-05 9.21692 1600.00 85.00 136000 2F 2.64 1.05 1.59 1.8298E-05 10.96814 1600.00 105.00 168000 2F 2.64 1.05 1.59 1.8298E-05 13.54888 1800.00 55.00 99000 2F 2.64 0.95 1.59 1.8298E-05 7.22376 1800.00 65.00 117000 2F 2.64 1 1.59 1.8298E-05 8.98650 1800.00 75.00 135000 2F 2.64 1 1.59 1.8298E-05 10.36904 1800.00 85.00 153000 2F 2.64 1.05 1.59 1.8298E-05 12.33915 290.00 65.00 18850 1F 1 1 1 1.8298E-05 0.34492 290.00 76.00 22040 1F 1 1 1 1.8298E-05 0.40329 290.00 85.00 24650 1F 1 1.05 1 1.8298E-05 0.47360 350.00 65.00 22750 1F 1 1 1 1.8298E-05 0.41628 350.00 70.00 24500 1F 1 1 1 1.8298E-05 0.44830 350.00 76.00 26600 1F 1 1 1 1.8298E-05 0.48673 400.00 65.00 26000 1F 1 1 1 1.8298E-05 0.47575 450.00 45.00 20250 1F 1 0.95 1 1.8298E-05 0.35201 450.00 70.00 31500 1F 1 1 1 1.8298E-05 0.57639 450.00 76.00 34200 1F 1 1 1 1.8298E-05 0.62579 450.00 85.00 38250 1F 1 1.05 1 1.8298E-05 0.73489 450.00 125.00 56250 1F 1 1.05 1 1.8298E-05 1.08073 500.00 65.00 32500 1F 1 1 1 1.8298E-05 0.59469 500.00 70.00 35000 1F 1 1 1 1.8298E-05 0.64043 500.00 76.00 38000 1F 1 1 1 1.8298E-05 0.69532 500.00 85.00 42500 1F 1 1.05 1 1.8298E-05 0.81655 550.00 105.00 57750 1F 1 1.05 1 1.8298E-05 1.10954 600.00 65.00 39000 1F 1 1 1 1.8298E-05 0.71362 600.00 85.00 51000 1F 1 1.05 1 1.8298E-05 0.97986 650.00 55.00 35750 1F 1 0.95 1 1.8298E-05 0.62145 650.00 65.00 42250 1F 1 1 1 1.8298E-05 0.77309 650.00 85.00 55250 1F 1 1.05 1 1.8298E-05 1.06151 650.00 105.00 68250 1F 1 1.05 1 1.8298E-05 1.31128 750.00 65.00 48750 1F 1.73 1 1 1.8298E-05 1.54321 750.00 85.00 63750 1F 1.73 1.05 1 1.8298E-05 2.11894 750.00 125.00 93750 1F 1.73 1.05 1 1.8298E-05 3.11609 850.00 65.00 55250 1F 1.73 1 1 1.8298E-05 1.74897 1050.00 45.00 47250 1F 1.73 0.95 1 1.8298E-05 1.42094 1050.00 65.00 68250 1F 1.73 1 1 1.8298E-05 2.16049 1050.00 85.00 89250 1F 1.73 1.05 1 1.8298E-05 2.96652 1050.00 105.00 110250 1F 1.73 1.05 1 1.8298E-05 3.66452 1250.00 65.00 81250 1F 1.73 1 1 1.8298E-05 2.57201 1250.00 85.00 106250 1F 1.73 1.05 1 1.8298E-05 3.53157 1500.00 65.00 97500 1F 2.64 1 1 1.8298E-05 4.70991 1500.00 75.00 112500 1F 2.64 1 1 1.8298E-05 5.43451 © Martin Caron 58 1500.00 85.00 127500 1F 2.64 1.05 1 1.8298E-05 6.46706 1600.00 65.00 104000 1F 2.64 1 1 1.8298E-05 5.02390 1600.00 75.00 120000 1F 2.64 1 1 1.8298E-05 5.79681 1600.00 85.00 136000 1F 2.64 1.05 1 1.8298E-05 6.89820 1600.00 105.00 168000 1F 2.64 1.05 1 1.8298E-05 8.52131 1800.00 55.00 99000 1F 2.64 0.95 1 1.8298E-05 4.54325 1800.00 65.00 117000 1F 2.64 1 1 1.8298E-05 5.65189 1800.00 75.00 135000 1F 2.64 1 1 1.8298E-05 6.52141 1800.00 85.00 153000 1F 2.64 1.05 1 1.8298E-05 7.76047 350.00 1.00 350 Pan 2F 1 1 1.76 1.8298E-05 0.01127 400.00 1.00 400 Pan 2F 1 1 1.76 1.8298E-05 0.01288 450.00 1.00 450 Pan 2F 1 1 1.76 1.8298E-05 0.01449 500.00 1.00 500 Pan 2F 1 1 1.76 1.8298E-05 0.01610 550.00 1.00 550 Pan 2F 1 1 1.76 1.8298E-05 0.01771 650.00 1.00 650 Pan 2F 1 1 1.76 1.8298E-05 0.02093 850.00 1.00 850 Pan 2F 1.73 1 1.76 1.8298E-05 0.04736 1050.00 1.00 1050 Pan 2F 1.73 1 1.76 1.8298E-05 0.05850 1250.00 1.00 1250 Pan 2F 1.73 1 1.76 1.8298E-05 0.06964 1500.00 1.00 1500 Pan 2F 2.64 1 1.76 1.8298E-05 0.12753 1800.00 1.00 1800 Pan 2F 2.64 1 1.76 1.8298E-05 0.15304 350.00 1.00 350 Pan 1F 1 1 1.47 1.8298E-05 0.00941 400.00 1.00 400 Pan 1F 1 1 1.47 1.8298E-05 0.01076 450.00 1.00 450 Pan 1F 1 1 1.47 1.8298E-05 0.01210 500.00 1.00 500 Pan 1F 1 1 1.47 1.8298E-05 0.01345 550.00 1.00 550 Pan 1F 1 1 1.47 1.8298E-05 0.01479 650.00 1.00 650 Pan 1F 1 1 1.47 1.8298E-05 0.01748 850.00 1.00 850 Pan 1F 1.73 1 1.47 1.8298E-05 0.03955 1050.00 1.00 1050 Pan 1F 1.73 1 1.47 1.8298E-05 0.04886 1250.00 1.00 1250 Pan 1F 1.73 1 1.47 1.8298E-05 0.05817 1500.00 1.00 1500 Pan 1F 2.64 1 1.47 1.8298E-05 0.10652 1800.00 1.00 1800 Pan 1F 2.64 1 1.47 1.8298E-05 0.12782 1.00 65.00 1 0.95 0.84 1.8298E-05 0.00095 1.00 90.00 90 Sup 1 1 0.84 1.8298E-05 0.00138 1.00 115.00 115 Sup 1 1.05 0.84 1.8298E-05 0.00186 65 Med 1 0.95 0.67 1.8298E-05 0.00076 65 Sup 1.00 65.00 1.00 90.00 90 Med 1 1 0.67 1.8298E-05 0.00110 1.00 115.00 115 Med 1 1.05 0.67 1.8298E-05 0.00148 0.95 0.55 1.8298E-05 0.00062 1.00 65.00 65 Inf 1 1.00 90.00 90 Inf 1 1 0.55 1.8298E-05 0.00091 1.00 115.00 115 Inf 1 1.05 0.55 1.8298E-05 0.00122 © Martin Caron 59