Design of a visible near infrared spectrometer and an application example

Recently spectroscopic analysis methods have very common usage in industry and agricultural studies to analyze and classify some kind of materials. Many institutions have laboratory to analyze inner structure of materials using spectrometer or spectrophotometer devices. These devices usually are very expensive and their sizes are usually large. Spectrometric devices commonly have optical and electronics components arranged in an appropriate way. Most of spectrometric analyze methods have sample preparation phases which are consist some chemical process. So that it is important to have knowledge on material science and chemical bonds when trying to analyze inner structure of any material or identifying what it is in. Electromagnetic spectrum is a whole of a spectral region, which covers all wavelengths of light. The spectrum has very wide range with wavelengths from fm's to thousand kilometers. The range that can be seen with human eyes called visible region is very narrow band of whole spectrum with 400 nm width. Electromagnetic wave is generated by moving charge. A steady charge has an electric field around itself. A moving charge such as negative charged electron creates moving electric field and that moving electric field creates changing electric flux. Maxwell show that changing electric flux creates a changing magnetic flux and vice versa. So that a moving charge generates an electromagnetic wave. All electromagnetic waves in the spectrum related with a charge movement. Spectrometric analysis is based on an interaction between matter and electromagnetic waves. Atoms have chemical bonds and some of these bonds have polarity. Electromagnetic waves interact with the polarity of molecules or if their wavelengths are small enough to penetrate into electron orbits they also interact with electrons. Some chemical bonds with appropriate polarity and geometry can absorb electromagnetic waves and with this energy, their vibrational movements increase. Therefore, the relation between chemical bonds and electromagnetic wave wavelength gives an opportunity to identify chemical bond and chemical characteristics. Spectrometers are designed to identify these differences to get the information that we need to know. Infrared wavelengths are very long for electronic transitions. Therefore, those electromagnetic waves interact with molecular polarity instead of electrons in this region. All chemical bonds have different polarity and different molecular geometry. They absorb different wavelengths in electromagnetic spectrum. When a molecule absorbs electromagnetic wave, because of the electromagnetic wave's enegry, molecular vibrations increase. So that infrared spectroscopy is sometimes called vibrational spectroscopy. Spectrometers are used for different purposes with different wavelengths. Although their main components are same. They have different light sources and sensors because of working on different wavelengths. Spectrometers mainly consist light source, sample holder or cuvette, grating or prism and sensor array. Spectrometers also have different constructions to different measurement types. In absorption spectroscopy, the light beam that comes from light source is directly sent to sample holder. Sample in the sample holder absorbs some of the electromagnetic waves, which have same vibrational frequency (or wavelength) with molecules of the sample. Some of electromagnetic waves pass through the sample to the grating or prism (or any other optic structure) and then they all go to sensor array. These structures are located in the same axes. When light (electromagnetic wave) has detected by sensor array with the information of its wavelength device can get the information of absorbance bands of sample. In reflectance spectroscopy, light beam goes to sample and then re-emitted by sample in different wavelengths and different directions. Grating or prism is generally located orthogonal with light source. The most important phase of spectrometric analysis is getting the information about the molecular structure of the sample looking at spectral data. Many methods have been used to analyze and identify the relation between infrared absorbency and chemical properties. Statistical methods are mostly used when there is a need to find a relation between two different properties. Some of the methods about regression and classification are Partial Least Squares Method (PLSR), Linear Discriminant Analysis (LDA) and Support Vector Machine (SVM) etc. All methods have some advantages and disadvantages related to process capability, spectral region, the chemical properties of sample and sample count. The choice of method, which is the best for any kind of application is a subject of another study and it has not been studied. Classification algorithms have two different methods named supervised and unsupervised. In both case, models are generated from measurements of different samples which are needed to be classified. After creating a model, the next sample is labeled with the class that the sample is belonged to. In supervised learning it is known that which sample is belonged to which material (class), it is not in unsupervised learning. In sections one, two and three, fundamentals of infrared spectroscopy, structure of a spectrometer and analysis methods were studied respectively and literature search was made. For the analysis of spectral data, LDA method was selected due to limited process power and easy to use with Accord.Net machine learning library in .NET. A machine-learning algorithm was run on MATLAB with simulated IR data of some of the soil compounds, which are collected from National Institute of Standards Technology (NIST). IR spectral data of ammonium nitrate, sodium nitrate and potassium nitrate compound were randomized and then put as an input into discriminant analysis function in MATLAB. Three different compounds were easily used. All spectral data was modeled with LDA algorithm using MATLAB and randomly generated data was successfully classified. In section four, structural, optical and electronic hardware/software design of visible near infrared spectrometer was considered. The sensor was selected within the spectral range of 640-1100nm. Hamamatsu C11708MA micro spectrometer sensor was used as a sensor array. It includes all optical components as micro designed in single package. A 4.8V 0.5 tungsten light bulb and an IR led were used as light source for 640-1100 nm range of spectrum. A black room, which has three holes to light source section, sensor section, and sample holder section, was constructed. Light from light source passes through a hole to the black room then it goes directly to the sample that is located in front of hole of the black room. Then it is reflected to the sensor hole and detected by the sensor. This structural design implements an optical isolation between environmental changes and spectrometer light source. An electronic board was designed to control C11708MA sensor with Atmega16A 8 bit microcontroller. ATMEGA16A sensor board controls generation of clock, reading analog input and sending through UART to any other digital media. One of the digital media is PC and other is STM3210E-Eval board. Sensor driver board has changeable resistors to adjust intensity of light from light source. It also has status LED to show what the internal status of the board is. STM3210E-Eval board has COM ports. A protocol between digital media and sensor board was implemented. Using this protocol, any other media can control the Hamamatsu spectrometer sensor and get spectral data. A Windows application was designed to collect NIR spectral data and create LDA model using Accord.Net machine-learning libraries. To collect data from sensor board, designed protocol on the top of the UART was used. Collection of the spectral data stored in data grid view component can be transferred to other platforms like excel. Hamamatsu C11708MA sensor has 256 pixels. 20 pixels were selected to reduce complexity of classification algorithm, which is running on ARM board. After computing LDA model eigenvectors and transformed class means are generated by machine -learning library. These vectors and transformed means of the classes can be transferred to any platform for further classification process. In section five an application of the spectrometer and machine-learning algorithm was considered. Two different potting soil were selected for the spectral measurements. 30 borosilicate tubes were used as sample holder for per potting soil type. Totally 60 tubes were used for collecting spectral data of soils. 20 tubes were selected as learning samples and 10 tubes were as test samples for per soil type. Spectral data was collected for four different angles and 10 times for per angle. Totally 800 spectral data for per soil type were collected by designed spectrometer and PC application. 1600 spectral data were used as an input to the LDA supervised machine-learning algorithm. A reference block was used for calibration to minimize light source changes. Calibration vector, eigenvectors and transformed means of two different soil types are transferred to STM3210E-Eval board. Classification process simply calculates the distance between the sample and transformed means. The minimum distance means that the sample belongs to that class. Analysis was performed in PC platform and classification process is performed in STM3210E-Eval board, which is located on the top of the spectrometer itself. 10 sample tubes of per soil types were used to test the classification success. 5 measurements for per tube were collected and totally 100 spectral data were collected to test the model. The model was successful using calibration data, which is collected when training data was collecting. In this study, a NIR spectrometer is designed with Hamamatsu C11708MA sensor and hardware/software are implemented to collect and analyze spectral data of two different potting soil type. In many other studies show that total organic carbon is main factor to classify two different soil in the region of visible short wave near infrared.