Progress 10/01/03 to 09/30/09
Outputs
OUTPUTS: The project investigated predictive modeling of the physical properties of wood composites using advanced computational algorithms. Research rationale and justification for the project comes from the importance of the forest products industry and the engineered panel industry to the U.S. economy. In 2002 this industry contributed more than $240 billion to the U.S. economy and employed more than 1,000,000 Americans in 22,231 primary wood products manufacturing facilities. Engineered wood panel manufacturing processes have a large number of differing, but interdependent process variables that have complex functional forms which influence properties. Key process parameters may include mat-forming consistency, line speed, press temperature, press closing rates, wood chip dimensions, fiber dimension, fiber-resin formation, etc. At the time of production, the quality of engineered wood is unknown, i.e., samples are analyzed at a later time in the lab using destructive testing. The time span between destructive tests may vary from two to six hours where unacceptable product (wood waste) may go undetected. The engineered wood panel sector produced 64.3 billion square feet of panels in 2003 with wood waste ranging from three percent to nine percent. Reducing wood waste by only one percent can translate into annual savings of almost $700,000 for a typical producer. The five research objectives were: 1) develop multisensor data fusion structures for wood composites; 2) investigate the feasibility of using advanced computational algorithms to predict the physical properties of wood composites in industrial settings; 3) validate the advanced computational algorithms solutions at United States wood composite manufacturers; 4) develop patentable software for real-time prediction using advanced statistical and computational algorithms prediction system; and 5) investigate the use of a fully automated prediction system for wood composite manufacturers that will be a first generation artificial intelligence system. All five study objectives were completed. The study resulted in a real-time genetic algorithm and neural network (MIGANN) predictive modeling system of the physical properties of wood composites. MIGANN was based on the fusion of real-time univariate process data with event-based destructive test data. The system was initially validated at one southeastern U.S. medium density fiberboard (MDF) and one southeastern oriented strand board (OSB) plants. Root mean square of prediction (RMSEP) for the internal bond (IB) at the MDF plant was approximately 12% and RMSEP for the concentrated static load (CSL) at the OSB plant was approximately 15%. Final validation was conducted at a southeastern U.S. particleboard plant after improvements were made to the data fusion portion of the system (e.g., time-lagging of sensor data). The improved modeling system also allowed for simultaneous prediction of two product attributes (e.g., particleboard IB and Modulus of Rupture or MOR). The time-lagging improvement to the real-time data fusion resulted in a RMSEP of 10% for IB and a RMSEP of 8% for MOR. Residuals were approximately normal for all test sites. PARTICIPANTS: Individuals: Timothy M. Young, PhD Associate Professor University of Tennessee Nicolas Andre, PhD Post Doctoral Research Associate University of Tennessee Partner Organizations: QMS, Inc. Knoxville, TN University of Tennessee Research Foundation (UTRF) Knoxville, TN USDA Small Business and Innovative Research Program Washington, D.C, Collaborators: Georgia-Pacific Chemicals, LLC Atlanta, GA Georgia-Pacific Corp. Atlanta, GA. Louisiana-Pacific Corp.,Nashville, TN, Flakeboard, LLC Bennettsville, SC, Langboard MDF, LLC, Valdosta, GA. Training and Professional Development provided for: Georgia-Pacific Chemicals, LLC Atlanta, GA, Georgia-Pacific Corp., Inc., Atlanta, GA, Louisiana-Pacific Corp., Nashville, TN, Carolina Flakeboard, LLC, Bennettsville, SC, Langboard MDF, LLC, Valdosta, GA. TARGET AUDIENCES: Forest products industry, bioenergy/biofuels industries, researchers, paper industry PROJECT MODIFICATIONS: Nothing significant to report during this reporting period.
Impacts
Key outcomes of the research were a real-time data fusion system and modeling system for the wood composites and engineered panel industry. The project generated successful Phase I and Phase II proposals for the USDA Small Business Innovation Research (SBIR) grant program partnering with QMS, Inc. in Knoxville, TN. Commercialization and licensing of MIGANN is being pursued between the University of Tennessee Research Foundation (UTRF) and QMS, Inc. QMS, Inc. hired two new employees as result of this project. The medium density fiberboard and oriented strand board plants used for validation were able to reduce resin usage from use of the genetic algorithm system. Cost savings from reduced resin and wood use during the validation studies varied from $700,000 to $1.2 million at the test mills. The modeling system has led to detection of unknown sources of variation and reduced wood waste at the test sites. The MIGANN system is envisioned to be available to the forest products industry in 2010.
Publications
- Kim, H., F.M. Guess, and T.M. Young. 2009. An extension of regression trees to generate better predictive models. IIE Transactions. Accepted.
- Chastain, J., T.M. Young, F.M. Guess, H. Bensmail, and R.V. Leon. 2009. Using reliability analyses to characterize wood strand thickness for oriented strand board panels. International Journal of Reliability and Application. 10(4):in press.
- Crookston, K.A. 2009. Reliability of wood plastic composites and improving lower percentile estimation via induced percentile censoring. M.S. Thesis. The University of Tennessee, Department of Statistics. Knoxville. (advisors: T.M. Young, F.M. Guess and R.L. Zaretski) 126p.
- Chastain, J.S. 2009. A statistical reliability analysis of the variability and upper percentiles of the wood strand thickness of oriented strand board. M.S. Thesis. The University of Tennessee, Department of Statistics. Knoxville. (advisors: T.M. Young, F.M. Guess and R.V. Leon) 138p.
- Steele, J.C. 2006. Function domain sets confidence intervals for the mean residual life function with applications in production in medium density fiberboard. M.S. Thesis. The University of Tennessee, Department of Statistics. Knoxville. (advisors: T.M. Young, F.M. Guess and R.V. Leon) 108p.
- Perhac, D.G. 2007. An applied statistical reliability analysis of the modulus of elasticity and modulus of rupture for wood-plastic composites. M.S. Thesis. The University of Tennessee, Department of Statistics. Knoxville. (advisors: T.M. Young, F.M. Guess and R.V. Leon) 102p.
- Shaffer, L.B. 2007. Examining regression analysis beyond the mean of the distribution using quantile regression. M.S. Thesis. The University of Tennessee, Department of Statistics.Knoxville. (advisors: T.M. Young, F.M. Guess and R.V. Leon) 94p.
- Young, T.M., J. Chastain, F.M. Guess, and R.V. Leon. 2009. Estimating upper percentiles of strand thickness for oriented strand board. Forest Products Journal. Accepted.
- Wang, Yang. 2007. Reliability analysis of oriented strand board's strength with a simulation study of the median censored method for estimating of lower percentile strength. M.S. Thesis. The University of Tennessee, Department of Statistics. Knoxville. (advisors: T.M. Young, F.M. Guess and R.V. Leon) 85p.
|
Progress 01/01/08 to 12/31/08
Outputs
OUTPUTS: The project investigates predictive modeling of the physical properties of wood composites using advanced computational algorithms. There were five research objectives: 1) develop multisensor data fusion structures for wood composites; 2) investigate the feasibility of using advanced computational algorithms to predict the physical properties of wood composites in industrial settings; 3) validate the advanced computational algorithms solutions at United States wood composite manufacturers; 4) develop patentable software for real-time prediction using advanced statistical and computational algorithms prediction system; and 5) investigate the use of a fully automated prediction system for wood composite manufacturers that will be a first generation artificial intelligence system. All five study objectives are complete. The study has resulted in a real-time genetic algorithm and neural network (MIGANN) predictive modeling system of the physical properties of wood composites. The system was validated at one U.S. medium density fiberboard and one oriented strand board manufacturing facilities. Continued validation is on-going at two mill sites. MIGANN was written in C++ and has a Visual Studio.Net human machine interface. MIGANN was based on the fusion of real-time univariate process data with event-based destructive test data. Residuals for all products were less than three percent of the median. Residuals were approximately normal. The real-time data fusion system was an important outcome of the research. Commercialization and licensing of MIGANN is being pursued. PARTICIPANTS: Individuals: Timothy M. Young, PhD Research Associate Professor The University of Tennessee. Nicolas Andre, PhD Post Doctoral Research Associate The University of Tennessee. Partner Organizations: QMS, Inc. Knoxville, TN. USDA Small Business and Innovative Research Program Washington, D.C, Collaborators: Georgia-Pacific Chemicals, LLC Atlanta, GA. Georgia-Pacific Corp. Atlanta, GA. Louisiana-Pacific Corp. Nashville, TN. Carolina Flakeboard, LLC Bennettsville, SC. Langboard MDF, LLC Valdosta, GA. Training and Professional Development provided for: Georgia-Pacific Chemicals, LLC Atlanta, GA. Georgia-Pacific Corp., Inc. Atlanta, GA. Louisiana-Pacific Corp. Nashville, TN. Carolina Flakeboard, LLC Bennettsville, SC. Langboard MDF, LLC Valdosta, GA. TARGET AUDIENCES: The target audience is the North American and possibly global forest products industry. Wood costs are the largest component of total manufacturing costs for forest products manufacturers. U.S. manufacturers have wood costs as high as 40% of total manufacturing costs. In 2003, the engineered wood panel sector produced 64.3 billion square feet of panels of which wood waste ranged from 3% to 9%. Reducing wood waste by 1% can translate into annual savings of $500,000 to $700,000 per producer and save 1.9 to 5.9 billion square feet of wood. Two of the largest contributors to wood waste in engineered wood manufacture are rejected panels and high density targets. This research addressed the problems of wood waste and poor wood yield in engineered wood manufacture by developing a real-time prediction system for physical properties using a hybrid Genetic Algorithm/Neural Network (MIGANN) with distributed data fusion. The MIGANN system will lower the rate of rejected panels, optimize throughput, identify key process parameters, optimize wood usage, promote lower resin use, lower energy use and improve wood yield. PROJECT MODIFICATIONS: First, statistical (e.g., ridge regression, partial least squares, etc.) and heuristic (e.g., genetic algorithm/neural network) real-time modeling systems were combined as an "ensemble" process modeling system for the bio-based industries. The systems are unique with completely automated, real-time, data-fusion of multi-sensor data. The systems are operational in two oriented strand board mills and one particleboard mill. Research has also been initiated in the development of automated modeling systems for biofuels manufacturing systems.
Impacts
Engineered wood manufacturing has a large number of differing, but interdependent process variables that have complex functional forms which influence properties. Wood passes through many processing stages that may influence the final properties. Key process parameters may include mat-forming consistency, line speed, press temperature, press closing rates, wood chip dimensions, fiber dimension, fiber-resin formation, etc. At the time of production, the quality of engineered wood is unknown, i.e., samples are analyzed at a later time in the lab using destructive testing. The time span between destructive tests may vary from two to six hours. Hours of unacceptable engineered wood production may go undetected between these tests. Many engineered wood manufacturers run higher than needed density targets to make up for this gap in product quality knowledge. The medium density fiberboard and oriented strand board plants used for validation were able to reduce resin usage from use of the genetic algorithm system. Cost savings from reduced resin and wood use during the validation studies varied from $700,000 to $1.2 million at the test mills. The modeling system has led to detection of unknown sources of variation and ultimately resulted in reduced variation at both test sites. The MIGANN systems is envisioned to result in lower wood waste, faster throughput, lower chemical usage, lower energy use and improved wood yield.
Publications
- Crookston, K.A., F.M. Guess, T.M. Young and D. Harper. 2009. A comparison of two wood plastic composite extrusion processes using statistical reliability analysis. Holzforschung. In Press.
- Young, T.M. 2008. Reducing variation, the role of statistical process control in advancing product quality. Engineered Wood Journal. 11(2):41-42.
- Young, T.M., L.B. Shaffer, F.M. Guess, H. Bensmail and R.V. Leon. 2008. A comparison of multiple linear regression and quantile regression for modeling the internal bond of medium density fiberboard. Forest Products Journal. 58(4):39-48.
- Young, T.M., D.G. Perhac, F.M. Guess and R.V. Leon. 2008. Bootstrap confidence intervals for percentiles of reliability data for wood plastic composites. Forest Products Journal. 58(11):106-114.
- Clapp, N.E., Jr., T.M. Young and F.M. Guess. 2008. Predictive modeling the internal bond of medium density fiberboard using principal component analysis. Forest Products Journal. 58(4):49-55.
- Andre, N., H.W. Cho, S.H. Baek, M.K. Jeong and T.M. Young. 2008. Enhanced prediction of internal bond strength in a medium density fiberboard process using supervised probabilistic regression and feature selection. Wood Science and Technology. Wood Science and Technology. 42:521-534. DOI 10.1007/s00226-008-0204-7.
- Labbe, N., I.S. Swamidoss, N. Andre, M.Z. Martin, T.M. Young and T.G. Rials. 2008. Extraction of information from laser-induced breakdown spectroscopy spectral data by multivariate analysis. Applied Optics. 47: G158-G165.
- Shaffer, L.B., T.M. Young, F.M. Guess, H. Bensmail and R.V. Leon. 2008. Using R software for reliability data analysis. International Journal of Reliability and Application. 9(1):53-70.
- Young, T.M., B.H. Bond and J. Wiedenbeck. 2007. Implementation of a real-time statistical process control system in hardwood sawmills. Forest Products Journal. 57(9):54-62.
- Perhac, D.G., T.M. Young, F.M. Guess and R.V. Leon. 2007. Exploring reliability of wood-plastic composites: stiffness and flexural strengths. International Journal of Reliability and Application. 8(2):153-173.
- Young, T.M., D.G. Hodges and T.G. Rials. 2007. The Forest Products Economy of Tennessee. Forest Products Journal. 57(4):12-19.
- Wang, Yingjin. 2008. Comparing linear discriminant analysis with classification trees using forest landowner survey data as a case study with considerations for optimal biorefinery siting. M.S. Thesis. The University of Tennessee. Knoxville. (advisors: T.M. Young, F.M. Guess and R.L. Zaretski) 114p.
|
Progress 01/01/07 to 12/31/07
Outputs
OUTPUTS: The project investigates predictive modeling of the physical properties of wood composites using advanced computational algorithms. There were five research objectives: 1) develop multisensor data fusion structures for wood composites; 2) investigate the feasibility of using advanced computational algorithms to predict the physical properties of wood composites in industrial settings; 3) validate the advanced computational algorithms solutions at United States wood composite manufacturers; 4) develop patentable software for real-time prediction using advanced statistical and computational algorithms prediction system; and 5) investigate the use of a fully automated prediction system for wood composite manufacturers that will be a first generation artificial intelligence system. All five study objectives are complete. The study has resulted in a real-time genetic algorithm and neural network (MIGANN) predictive modeling system of the physical properties of wood campsites. The system was validated at one U.S. medium density fiberboard and one oriented strand board manufacturing facilities. Continued validation is on-going at two mill sites. MIGANN was written in C++ and has a Visual Studio.Net human machine interface. MIGANN was based on the fusion of real-time univariate process data with event-based destructive test data. Residuals for all products were less than three percent of the median. Residuals were approximately normal. The real-time data fusion system was an important outcome of the research. Commercialization and licensing of MIGANN is being pursued. PARTICIPANTS: Individuals: Timothy M. Young Research Associate Professor The University of Tennessee Nicolas Andre Post Doctoral Research Associate The University of Tennessee Jeroen van Houts Consultant New Zealand Partner Organizations: QMS, Inc. Knoxville, TN USDA Small Business and Innovative Research Program Washington, D.C, Collaborators: Georgia-Pacific Chemicals, LLC Atlanta, GA Georgia-Pacific Corp., Inc. Atlanta, GA Langboard MDF, LLC Valdosta, GA Training and Professional Development provided for: Georgia-Pacific Chemicals, LLC Atlanta, GA Georgia-Pacific Corp., Inc. Atlanta, GA Langboard MDF, LLC Valdosta, GA TARGET AUDIENCES: The target audience is the North American and possibly global forest products industry. Wood costs are the largest component of total manufacturing costs for forest products manufacturers. U.S. manufacturers have wood costs as high as 40% of total manufacturing costs. In 2003, the engineered wood panel sector produced 64.3 billion square feet of panels of which wood waste ranged from 3% to 9%. Reducing wood waste by 1% can translate into annual savings of $500,000 to $700,000 per producer and save 1.9 to 5.9 billion square feet of wood. Two of the largest contributors to wood waste in engineered wood manufacture are rejected panels and high density targets. This research addressed the problems of wood waste and poor wood yield in engineered wood manufacture by developing a real-time prediction system for physical properties using a hybrid Genetic Algorithm/Neural Network (MIGANN) with distributed data fusion. The MIGANN system will lower the rate of rejected panels, optimize throughput, identify key process parameters, optimize wood usage, promote lower resin use, lower energy use and improve wood yield. PROJECT MODIFICATIONS: Describe major changes in approach and reason(s) for these major changes. If applicable, provide special and/or additional reporting requirements specified in the award Terms and Conditions.
Impacts
Engineered wood manufacturing have a large number of differing, but interdependent process variables that have complex functional forms which influence properties. Wood passes through many processing stages that may influence the final properties. Key process parameters may include mat-forming consistency, line speed, press temperature, press closing rates, wood chip dimensions, fiber dimension, fiber-resin formation, etc. At the time of production, the quality of engineered wood is unknown, i.e., samples are analyzed at a later time in the lab using destructive testing. The time span between destructive tests may vary from two to six hours. Hours of unacceptable engineered wood production may go undetected between these tests. Many engineered wood manufacturers run higher than needed density targets to make up for this gap in product quality knowledge. The medium density fiberboard and oriented strand board plants used for validation were able to reduce resin usage from use of the genetic algorithm system. Cost savings from reduced resin and wood use during the validation studies varied from $700,000 to $1.2 million at the test mills. The modeling system has led to detection of unknown sources of variation and ultimately resulted in reduced variation at both test sites. The MIGANN systems is envisioned to result in lower wood waste, faster throughput, lower chemical usage, lower energy use and improve wood yield.
Publications
- Leslie B. Shaffer. 2007. University of Tennessee, Department of Forestry, Wildlife and Fisheries. Graduated 08/07. M.S. Thesis. Examining Regression Analysis Beyond the Mean of the Distribution using Quantile Regression. 94p.
- Yang Wang. 2007. University of Tennessee, Department of Forestry, Wildlife and Fisheries. Graduated 08/07. M.S. Thesis. Reliability Analysis of Oriented Strand Board's Strength with a Simulation Study of the Median Censored Method for Estimating of Lower Percentile Strength.
- Wang, Y., T.M. Young, F.M. Guess and R.V. Leon. 2007. Exploring reliability of oriented strand board's tensile and stiffness strengths. International Journal of Reliability and Application. 8(1): 113-126.
- Guess, F.M., J.C. Steele, T.M. Young and R.V. Leon. 2006. Applying novel mean residual life confidence intervals. International Journal of Reliability and Application. 7(2):177-186.
- Diane G. Perhac. 2007. University of Tennessee, Department of Forestry, Wildlife and Fisheries. Graduated 08/07. M.S. Thesis. An Applied Statistical Reliability Analysis of the Modulus of Elasticity and Modulus of Rupture for Wood-Plastic Composites. 102p.
|
Progress 01/01/06 to 12/31/06
Outputs
The project investigates predictive modeling of the physical properties of wood composites using advanced computational algorithms. There were five research objectives: 1) develop multisensor data fusion structures for wood composites; 2) investigate the feasibility of using advanced computational algorithms to predict the physical properties of wood composites in industrial settings; 3) validate the advanced computational algorithms solutions at United States wood composite manufacturers; 4) develop patentable software for real-time prediction using advanced statistical and computational algorithms prediction system; and 5) investigate the use of a fully automated prediction system for wood composite manufacturers that will be a first generation artificial intelligence system. Four of the five study objectives were completed. The final objective is close to completion. The study has resulted in a real-time genetic algorithm and neural network (MIGANN) predictive
modeling system of the physical properties of wood campsites. The system was validated at one U.S. medium density fiberboard and one oriented strand board manufacturing facilities. MIGANN was written in C++ and has a Visual Studio.Net human machine interface. MIGANN was based on the fusion of real-time univariate process data with event-based destructive test data. Residuals for all products were less than three percent of the median. Residuals were approximately normal. The real-time data fusion system was an important outcome of the research. Commercialization and licensing of MIGANN is being pursued. METHOD FOR CONTROLLING A PRODUCT PRODUCTION PROCESS Patent Application No. 11/088,651 Filing Date: 03/24/2005 LNG File No. 59176.US 7800.0
Impacts
Engineered wood manufacturing have a large number of differing, but interdependent process variables that have complex functional forms which influence properties. Wood passes through many processing stages that may influence the final properties. Key process parameters may include mat-forming consistency, line speed, press temperature, press closing rates, wood chip dimensions, fiber dimension, fiber-resin formation, etc. At the time of production, the quality of engineered wood is unknown, i.e., samples are analyzed at a later time in the lab using destructive testing. The time span between destructive tests may vary from two to six hours. Hours of unacceptable engineered wood production may go undetected between these tests. Many engineered wood manufacturers run higher than needed density targets to make up for this gap in product quality knowledge. The medium density fiberboard and oriented strand board plants used for validation were able to reduce resin usage
from use of the genetic algorithm system. Cost savings from reduced resin use during the six-month validation study were as large as $700,000 at one test mill. One test site was able to reduce final target density by one-half of one percent by using the system resulting in excess of one million dollar per year in savings. The modeling system may lead to a lower wood waste, faster throughput, lower chemical usage, lower energy use and improve wood yield.
Publications
- Andre, N., T.M. Young and T.G. Rials. 2006. On-line monitoring of the buffer capacity of particleboard furnish by near-infrared pectroscopy. Applied Spectroscopy. 60(10): 1204-1209.
- Chen, W., R.V. Leon, T.M. Young and F.M. Guess. 2006. Applying a forced censoring technique with accelerated modeling for improving estimation of extremely small percentiles of strengths. International Journal of Reliability and Application. 7(1):27-39.
- Jonathan Cody Steele. 2006. University of Tennessee, Department of Forestry, Wildlife and Fisheries. Graduated 08/06. M.S. Thesis. "Function domain sets" confidence intervals for the mean residual life function with applications in production of medium density fiberboard. 108p.
- Weiwei Chen. 2005. University of Tennessee, Department of Forestry, Wildlife and Fisheries. Graduated 12/05. M.S. Thesis. A Reliability Case Study on Estimating Extremely Small Percentiles of Strength Data for the Continuous Improvement of Medium Density Fiberboard Product Quality. 69p.
|
Progress 01/01/05 to 12/31/05
Outputs
The project investigates predictive modeling of the physical properties of wood composites using advanced computational algorithms. There were five research objectives: 1) develop multisensor data fusion structures for wood composites; 2) investigate the feasibility of using advanced computational algorithms to predict the physical properties of wood composites in industrial settings; 3) validate the advanced computational algorithms solutions at United States wood composite manufacturers; 4) develop patentable software for real-time prediction using advanced statistical and computational algorithms prediction system; and 5) investigate the use of a fully automated prediction system for wood composite manufacturers that will be a first generation artificial intelligence system. Four of the five study objectives were completed. The final objective is in progress. A heuristic algorithmic method using genetic algorithms with real-time distributed data fusion was
developed to predict the internal bond of medium density fiberboard and the parallel elasticity index for oriented strand board. The system incorporated real-time lags and statistical estimates of 285 critical process parameters with data quality verification algorithms. The real-time relational data fusion system was completely automated and represented the infrastructure of the genetic algorithm prediction system. The real-time data fusion system was an important outcome of the research. Additional validation and enhancement of the system is ongoing at one medium density fiberboard plant and one oriented strand board plant in North America. Results of the genetic algorithm modeling system indicate predictions within 5% of the median physical properties. Time-ordered residuals accurately detect trends and are normally distributed.
Impacts
Engineered wood manufacturing have a large number of differing, but interdependent process variables that have complex functional forms which influence properties. Wood passes through many processing stages that may influence the final properties. Key process parameters may include mat-forming consistency, line speed, press temperature, press closing rates, wood chip dimensions, fiber dimension, fiber-resin formation, etc. At the time of production, the quality of engineered wood is unknown, i.e., samples are analyzed at a later time in the lab using destructive testing. The time span between destructive tests may vary from two to six hours. Hours of unacceptable engineered wood production may go undetected between these tests. Many engineered wood manufacturers run higher than needed density targets to make up for this gap in product quality knowledge. The medium density fiberboard and oriented strand board plants used for validation are able to reduce resin usage
from use of the genetic algorithm system. Cost savings from reduced resin use during the six-month validation study are as large as $700,000 at one test mill. The modeling system may lead to a lower wood waste, faster throughput, lower chemical usage, lower energy use and improve wood yield.
Publications
- Chen, W., R.V. Leon, T.M. Young and F.M. Guess. 2005. Applying a forced censoring technique with accelerated modeling for improving estimation of extremely small percentiles of strengths. International Journal of Reliability and Application. 6(4).c
- Guess, F.M., X. Zhang, T.M. Young and R.V. Leon. 2005. Using mean residual life functions for unique insights into strengths of materials data. International Journal of Reliability and Application. 6(4).
- Guess, F.M., R.V. Leon, W. Chen and T.M. Young. 2004. Forcing a closer fit in the lower tails of a distribution for better estimating extremely small percentiles of strengths. International Journal of Reliability and Application. 5(4):129-145.
- Wang, S., P.M. Winistorfer and T.M. Young. 2004. Fundamentals of vertical density profile formation in wood composites Part 3. MDF density formation during hot-pressing. Wood and Fiber Science. 36(1):17-25.
|
Progress 01/01/04 to 12/31/04
Outputs
The project investigated predictive modeling of the physical properties of wood composites using advanced computational algorithms. There were five research objectives: 1) develop multisensor data fusion structures for wood composites; 2) investigate the feasibility of using advanced computational algorithms to predict the physical properties of wood composites in industrial settings; 3) validate the advanced computational algorithms solutions at United States wood composite manufacturers; 4) develop patentable software for real-time prediction using advanced statistical and computational algorithms prediction system; and 5) investigate the use of a fully automated prediction system for wood composite manufacturers that will be a first generation artificial intelligence system. Three of the five study objectives were completed. The final two objectives are in progress. A heuristic algorithmic method using genetic algorithms with real-time distributed data fusion was
developed to predict the internal bond of medium density fiberboard. The system incorporated real-time lags and statistical estimates of 285 critical process parameters with data quality verification algorithms. The real-time relational data fusion system was completely automated and represented the infrastructure of the genetic algorithm prediction system. The real-time data fusion system was written in Microsoft Transact SQL and was an important outcome of the research. Validation of the system was performed at a modern medium density fiberboard plant in North America which uses a continuous press. Results of the genetic algorithm prediction system were promising. Validation results of the genetic algorithm prediction system had mean and median residuals for all product types of 1.19 and -0.13 pounds per square inch, respectively. Mean and median validation residuals had an error of less than 5% error of the average or target internal bond. First-order derivatives of time-ordered
residuals accurately detected slope changes in physical properties. Validation residuals were normally distributed.
Impacts
Engineered wood manufacturing may have a large number of differing, but interdependent process variables that may have complex functional forms, which influence properties. Wood passes through many processing stages that may influence the final properties. Key process parameters may include mat-forming consistency, line speed, press temperature, press closing rates, wood chip dimensions, fiber dimension, fiber-resin formation, etc. At the time of production, the quality of engineered wood is unknown, i.e., samples are analyzed at a later time in the lab using destructive testing. The time span between destructive tests may vary from two to six hours. Hours of unacceptable engineered wood production may go undetected between these tests. Many engineered wood manufacturers run higher than needed density targets to make up for this gap in product quality knowledge. The medium density fiberboard plant used for the validation study was able to reduce resin usage from use
of the genetic algorithm system. Cost savings from reduced resin use during the six-month validation study were approximately $700,000 at the test mill. The proposed system may also lead to a lower rate of rejected panels, faster throughput, identify key sources of variability, lower wood usage, lower energy use and improve wood yield.
Publications
- Young, T.M., N. Andre and C.W. Huber. 2004. Predictive modeling of the internal bond of MDF using genetic algorithms with distributed data fusion. Proc. 8th European Panel Products Symposium. Llandudno, United Kingdom. p. 45-59.
- Young, T.M. and C.W. Huber. 2004. Predictive modeling of the physical properties of wood composites using genetic algorithms with considerations for distributed data fusion. Proc. of the 38th International Particleboard/Composite Materials Symposium. Washington State University. Pullman, WA. p. 71-86.
- Guess, F.M., D.J. Edwards, T.M. Pickrell and T.M. Young. 2003. Exploring graphically and statistically the reliability of medium density fiberboard. International Journal of Reliability and Application. 4(4):157-170.
- Wang, S., P.M. Winistorfer and T.M. Young. 2004. Fundamentals of vertical density profile formation in wood composites - Part 3. MDF density formation during hot-pressing. Wood and Fiber Science. 36(1):17-25.
|
|