Application of neural network analysis in nondestr

  • Detail

Application of near IR determination neural network analysis in nondestructive identification of waste plastics

Application of near IR determination of Chinese plastics testing/neural network analysis in nondestructive identification of waste plastics Meng Pingrui and Li liangbo (Department of chemical engineering, Shandong Building Materials University, Jinan 250022) introduced the method of using near ir/neural network analysis to quickly identify a variety of plastics. At near IR 1 About 300 samples of 51 kinds of plastics in the range of 3~2.3, um were input into the neural network element of a three-layer visual sensor using standardized second-order differential spectrum data. The results showed that the average hit rate of 50 kinds of plastics was 0. The treatment of waste plastics mainly included direct burial, incineration and recycling. Direct burial is limited due to its large volume; Incineration is also inappropriate due to the generation of toxic substances. Therefore, it is imperative to promote the recycling of waste plastics from the perspective of the earth's environmental protection or the earth's resources where raw materials (crude oil) are exhausted. Recycling of plastics includes reprocessing of plastics, decomposition into raw material monomers, cracking into gas or liquid, and recycling as heat and energy. But neither method can be separated from the identification of plastic. About the identification methods of plastics, relative density, X-ray diffraction and near-infrared spectroscopy have been put into practice, especially suitable for the study of polymers such as plastics. 9 '10, the second derivative of near IR spectrum is carried out, and then the neural network analysis 1112 is carried out. A new method for rapid identification of a variety of plastics is discussed. 1 method 1.1 samples in order to develop a plastic type identification method for recycling waste plastics, all samples are collected from plastic wastes provided by factories and families. There are various shapes (different brands, grades, additives and uses of similar plastics are different samples) and colors range from transparent to very deep. A total of 287 samples of 51 kinds of plastics were collected (see Table 1 for details). Polystyrene and polyethylene samples account for the majority. 1. Measuring device the reflectance spectrum of the above sample is measured with Japan AOTF near infrared spectrometer. The wavelength is 1.0~2.5, and the resolution is 0.0005mm. Measure at 1.3~2.3mm, scan 2000 times, measure 5 times, and take the average value. 1.3 data processing input the above measured data into the neural network system for the following operations: take the average of the above measured spectral data for 2000 scans to calculate the quadratic differential spectrogram. (0.9~0.9 range) to standardize it (see the results). Collect the * big value or * small value in the spectrum Table 1 name and sample number symbol name number symbol name number polystyrene polysulfone polyethylene polymethylpentene polypropylene polysulfide acrylonitrile/butadiene/styrene copolymer polyurethane polyglycol hard rubber polyvinyl chloride epoxy resin acrylonitrile/styrene synthetic rubber polycarbonate maleamide polycarbonate acrylonitrile/butadiene/styrene copolymer polybutene polyethylene terephthalate Glycol ester polyester rubber polymethacrylate to improve battery performance ester polymer to achieve the spring Before the surface of metal structure is painted, clear the rust, paint, scale and strengthen the surface The most ideal equipment for eliminating internal stress is naphthalene glycol ester nylon 6 polyimide nylon 66 polyphenylene ether cellulose polyphenylene sulfide melamine polytetrafluoroethylene polybutylene terephthalate polyvinylidene chloride phenolic polyvinylidene fluoride urea formaldehyde resin styrene/butadiene copolymer ethylene vinyl acetate copolymer synthetic rubber ion bonded polymer silicone resin polyaryl ether styrene maleic acid copolymer polyimide styrene terpolymer Aryl compounds thermoplastic synthetic rubber peek thermoplastic rubber polyethyleneimine thermoplastic polyurethane 1.4 program system is a hierarchical neural network. The statistical analysis of main components and peak clusters was carried out using the einsight software made by infometix. 1.5 neural network training composition. Input the near IR quadratic differential data of various plastics in the input layer, and practice with the identification number of various plastics as the standard in the output layer. The number of units in the input layer is 198 (2 ⑴ quadratic differential processing lacks both ends). When identifying 51 kinds of plastics, the output layer is set as 6 units, and the plastic identification number is converted into binary as standard data for learning. The middle layer consists of 30 units. 2 Results 2.1 principal component analysis in order to investigate the possibility of distinguishing 50 kinds of plastics by this method, first calculate the average spectrum of 50 kinds of plastics. Using this data for principal component analysis, the first principal component (factor 1) and the second principal component (factor 2) are plotted as abscissa and ordinate respectively. It can be seen that all kinds of plastics are dispersed and distributed without overlap in space. Therefore, it is predicted that 50 kinds of plastics can be distinguished by near IR data. 2.2 discrimination experiment in order to investigate the feasibility of the discrimination method, the following experiments are carried out: take the plastic with more than 10 samples as the measurement, analyze the peak clusters of the spectra of 10 test objects, and divide the samples into 2-4 peak clusters. Calculate the average spectrum of each peak cluster. The average spectrum is also calculated for non test objects. These average spectra were used for neural network training. Six sets of accessories for various mechanical experiments, such as various insulation systems, adhesives, flexible water-resistant putty, are added. The spectral data of the test object is input into the neural network, and the identification number of the spectrum is calculated from the value of the output layer to calculate the hit rate. The hit rate of each sample calculated according to the above operation sequence is shown in Table 2 Among the 10 kinds of plastics tested, he estimated that about 1/3 of the company's business came from FRP Bridge materials. POM, as and PS all achieved a very high hit rate, while ABS hit rate * was low, and the overall average hit rate was 77%. 3 discussion the average hit rate of the plastic identification method described in this paper was 77%, which does not seem to meet the practical requirements. However, the data used in the neural network analysis of the discrimination test experiment is only 2-4 peak cluster data. For the number of plastic samples with a very low hit rate, the hit rate of the number of samples hit by the number of peak clusters/% in total, we expect to further increase the peak cluster data and improve the hit rate. When developing a practical identification device for recycling waste plastics, the hit rate of all plastic samples should be close to 100% as far as possible Therefore, this method is expected to become a very useful plastic. The hit rate of ABS and pc/abs is very low, mainly because the measured spectrum of these plastics is very different, and the composition of ABS copolymers is very different. However, using very few peak cluster data for 50 kinds of plastics, the average hit rate can be close to 80%, 7 rock yuan Mufu, river wild Chengfu, fish live pure. Introduction to near infrared spectroscopy. Lucky study, 8 Miyazaki yukiyo, Kawabata Cong. Near infrared spectrophotometry. The publication of July17 by the society also proves the possibility that a variety of methods for rapid identification of plastics can be developed. Since the types of plastics can be increased arbitrarily by using neural network analysis, even if the types of plastics are increased to more than 100, this method can also be used to identify. The near IR measuring device used in this method is small and cheap, and can quickly identify a variety of plastics, which is very suitable for on-site identification of waste plastic treatment

Copyright © 2011 JIN SHI