Font Size: a A A

Research On NIR Spectral Migration Modeling Method For Prediction Of Mechanical Strength Of Solid Wood Panels For Furniture

Posted on:2024-01-21Degree:DoctorType:Dissertation
Country:ChinaCandidate:J H ChenFull Text:PDF
GTID:1521306932980239Subject:Forestry engineering automation
Abstract/Summary:
The mechanical properties are important physical attributes of solid wood,which directly affect the grade and safe application of solid wood products.Near-infrared spectroscopy analysis,as an emerging detection technology,has the advantages of rapid,non-destructive,and real-time online analysis,and is currently widely used in the field of non-destructive testing for woodworking.However,in infrared spectroscopy analysis,there are problems such as weak signal intensity,wide spectral bandwidth,and overlapping spectral bands,making it difficult to directly extract qualitative and quantitative information of substances from the spectral signal.In addition,in practical applications,changes in temperature and humidity,aging of instrument components,and attachment replacements can affect the stability of spectral models.Improving the stability of the model is also an important research topic in near-infrared spectroscopy technology.Finally,compared with handheld near-infrared spectrometers,multispectral camera line scanning can collect information on a large scale and provide more macroscopic and accurate evaluation of product quality,making it more suitable for production line testing.Therefore,how to achieve spectral information sharing between the two instruments and experimental testing and precise analysis are also important considerations in practical applications.Regarding the above issue,this article takes the solid wood veneer mechanical testing as the research object and firstly studies the nonlinear modeling method of solid wood veneer.Then,the spectral data transfer technology is studied to effectively expand the near-infrared spectroscopy and improve model stability.Finally,the near-infrared fiber spectral data is transferred to the hyperspectral camera,realizing the accurate macro prediction of the mechanical properties of solid wood veneer.The specific research contents and innovative points are as follows:1.To address the issue of poor model stability caused by interference such as scattered light,baseline drift,and high-frequency noise in near-infrared spectroscopic quantitative analysis,an integrated solid wood mechanical property modeling method combining OPLS-PSA-MIX-PLS is proposed: This method can effectively suppress the interference such as baseline drift,noise,and brightness changes in the original spectral matrix through OPLS;extract spectral bands effectively through SPA;and accurately associate the feature spectrum with the mechanical properties of solid wood veneer through MIX-PLS.Experiments verify the stability of the method through spectral data acquisition and modeling analysis of the same sample at different times.2.Model transfer is a spectral calibration method used to eliminate spectral differences between different instruments,enabling a model established by one instrument(Device A)to be applicable to spectra measured by other instruments(Device B).To address the problem of reduced prediction accuracy of spectra measured by instruments caused by redundant spectral data when using existing methods for model transfer,the SWCSS-GFK-SVM near-infrared spectroscopy transfer method for mechanical properties of solid wood was proposed.The method used the NIRQuest512 near-infrared fiber spectrometer(Device A)and the NIR-NT-spectrometer-OEM-system spectrometer(Device B)to collect near-infrared spectra of the same batch of specimens.The spectra were preprocessed using SNV correction combined with the SG convolution smoothing algorithm,and the spectral features collected by both devices were restricted to the same spectral range.The SWCSS method was used to extract characteristic spectral bands,and finally,the GFK-SVM transfer model was used to construct a prediction model for mechanical properties of solid wood based on near-infrared spectra measured by Devices A and B.The validity of the method was verified through modeling and analysis of mechanical properties of larch wood specimens.3.A hyperspectral camera can be used to observe solid wood veneers over a large area,enabling more accurate evaluation of their mechanical properties.In order to correct for the response differences between the near-infrared spectrometer and the hyperspectral equipment,a spectral transfer model was designed to achieve the transfer of different types of spectrometers,to share data and improve the model’s accuracy.Two new model transfer methods based on structural equations were proposed,using the same batch of specimens’ spectra collected by both a near-infrared fiber spectrometer(NIRQuest512 near-infrared fiber spectrometer)and a hyperspectral camera(SPECIMFX17 hyperspectral camera);then,SNV+SG and wavelength range consistent preprocessing were performed;afterwards,PLS-SEM-FA and PLS-SEM-SST were designed for model transfer;finally,PLS was used to model the transfer spectra of the two different types of equipment,achieving a stable prediction of the mechanical properties of solid wood through hyperspectral analysis.4.A spectral fusion analysis system was established with the mechanical properties of solid wood veneer as the detection target.The system has functions such as high spectral information acquisition,transmission of solid wood veneer,mechanical property analysis,and veneer sorting.It has achieved the spectral information acquisition of the mechanical properties of solid wood,near-infrared spectral data analysis,and the transfer of different types of spectral data.Through mechanical property prediction experiments on solid wood veneer,the effectiveness of near-infrared spectral modeling,near-infrared spectral transfer,and spectral transfer between different equipment were verified.
Keywords/Search Tags:Solid wood panels for furniture, mechanical strengthprediction, transfer learning, near-infrared spectroscopy, hyperspectral imaging
Related items