ﺟﺪاﺳﺎزي ﺧﻮدﮐﺎر ﭘﻠﯿﻤﺮPVC ﺑﺎ اﺳﺘﻔﺎده از ﺑﯿﻨﺎب ﻧﻤﺎﯾﯽ ﻓﺮوﺷﮑﺴﺖ اﻟﻘﺎﯾﯽ ﻟﯿﺰري و ﺷﺒﮑﻪ ﻋﺼﺒﯽ LQV

Automated Separation Of PVC Polymer By Laser Induced Breakdown Spectroscopy an‎d LVQ Neural Network

ه: در اﯾﻦ ﭘﮋوﻫﺶ ﺑﺎ ﺑﮑﺎرﮔﯿﺮي روش ﺑﺮدار ﭼﻨﺪي ﺳﺎز ﯾﺎدﮔﯿﺮ LVQ در ﺣﻮزه ي ﺷﺒﮑﻪﻫﺎي ﻋﺼﺒﯽ ﻣﺼﻨﻮﻋﯽ و روش ﺗﺤﻠﯿﻞ ﻣﻮﻟﻔﻪ اﺻﻠﯽ PCA، ﺗﻮاﻧﺎﯾﯽ روش ﺑﯿﻨﺎب ﻧﻤﺎﯾﯽ ﻓﺮوﺷﮑﺴﺖ اﻟﻘﺎﯾﯽ ﻟﯿﺰري در ﺟﺪاﺳﺎزي ﭘﻠﯿﻤﺮ ﭘﻠﯽ وﯾﻨﯿﻞ ﮐﻠﺮاﯾﺪ )PVC( از ﺳﺎﯾﺮ ﭘﻠﯿﻤﺮﻫﺎ در ﻓﺮآﯾﻨﺪ ﺑﺎزﯾﺎﻓﺖ ﻣﻮرد ارزﯾﺎﺑﯽ ﻗﺮار ﻣﯽﮔﯿﺮد. در اﺑﺘﺪا 5 ﻧﻮع ﭘﻠﯿﻤﺮ آﻣﺎده ﺷﺪ و ﺷﺪتﻫﺎي ﻣﺮﺗﺒﻂ ﺑﺎ ﺧﻄﻮط اﺗﻤﯽ، ﯾﻮﻧﯽ و ﻫﻤﭽﻨﯿﻦ ﮔﺴﯿﻞﻫﺎي ﻣﻮﻟﮑﻮﻟﯽ ﻣﺮﺗﺒﻂ ﺑﺎ ﻫﺮ ﻧﻤﻮﻧﻪ ﺛﺒﺖ ﺷﺪ، ﺳﭙﺲ روش PCA ﺑﻪ ﻋﻨﻮان ﯾﮑﯽ از روشﻫﺎي آﻣﺎري ﭼﻨﺪ ﻣﺘﻐﯿﺮه ﺑﻪ ﻣﻨﻈﻮر ﮐﺎﻫﺶ اﺑﻌﺎد دادهﻫﺎ و ﻫﻤﭽﻨﯿﻦ ﯾﺎﻓﺘﻦ ﯾﮏ اﻟﮕﻮ در ﺑﯿﻨﺎب LIBS ﻧﻤﻮﻧﻪﻫﺎ، ﻣﻮرد اﺳﺘﻔﺎده ﻗﺮار ﻣﯽﮔﯿﺮد. ﭘﺲ از آن روش ﺑﺮدار ﭼﻨﺪي ﺳﺎز ﯾﺎدﮔﯿﺮ ﮐﻪ اﺧﯿﺮا در ﺣﻮزهي روشﻫﺎي ﯾﺎدﮔﯿﺮي ﻣﺎﺷﯿﻦ ﻣﻄﺮح ﺷﺪه اﺳﺖ را ﺑﺮاي ﺟﺪاﺳﺎزي ﭘﻠﯿﻤﺮ PVC از ﺳﺎﯾﺮ ﭘﻠﯿﻤﺮﻫﺎ ﺑﮑﺎر ﻣﯽﮔﯿﺮﯾﻢ. ﻧﺘﺎﯾﺞ ﻧﺸﺎن ﻣﯽدﻫﺪ ﮐﻪ ﺗﻨﻬﺎ ﺑﺎ دو ﻣﺘﻐﯿﺮ ﺧﺮوﺟﯽ از روش PCA ﮐﻪ ﺑﻪ ﻋﻨﻮان ﻣﺘﻐﯿﺮﻫﺎي ورودي ﺑﺮاي روش LVQ از آﻧﻬﺎ اﺳﺘﻔﺎده ﮐﺮدﯾﻢ، ﻣﯽﺗﻮان ﭘﻠﯿﻤﺮﻫﺎي PVC را از ﺳﺎﯾﺮ ﭘﻠﯿﻤﺮﻫﺎي ﻣﻮرد آﻧﺎﻟﯿﺰ در اﯾﻦ ﭘﮋوﻫﺶ ﺑﺎ دﻗﺖ 93/3 % ﺗﻔﮑﯿﮏ ﮐﺮد.

in this study by applying learning vector quantization (LVQ) in artificial neural networks (ANN) and principal component analysis (PCA), capability of laser-induced breakdown spectroscopy (LIBS) is evaluated in order to separate Polyvinyl chloride (PVC) from other polymers in the recycling process. At first, five kinds of polymer samples were prepared and intensities related to atomic and ionic lines as well as molecular emissions for each sample were recorded, then PCA is applied as one of the statistical multivariate analysis methods with aim of reducing dimensions of data and also finding a pattern in LIB spectra. Then learning vector quantization (LVQ) as a method that has recently been reported in the machine learning literature, is applied for discriminating PVC from other ones. The results show that with only two output variables from PCA method which are used as input variables for LVQ ANN, we can discriminate PVC from other analyzed polymers with precision of 93.3%.