Desain Sensor Cerdas Berbasis Near Infrared (NIR) untuk Pemantauan Kadar Rhodinol dalam Fraksi Minyak Sereh Wangi

Abstract

Pada era modern ini, pengembangan teknologi sensor makin maju dan memberikan kontribusi besar dalam berbagai aplikasi di berbagai bidang. Sensor cerdas berbasis Near Infrared (NIR) telah menjadi area penelitian yang menarik karena kemampuannya dalam analisis non-destruktif, cepat, dan akurat terhadap berbagai komponen kimia dalam sampel. Salah satu potensi penggunaan sensor NIR adalah dalam pemantauan kadar senyawa aromatik seperti rhodinol dalam fraksi minyak sereh wangi. Fraksi minyak sereh wangi mengandung senyawa rhodinol yang berperan penting dalam industri parfum dan kosmetik. Namun, saat ini deteksi kadar rhodinol dalam aliran distilat masih mengandalkan metode konvensional seperti Gas Chromatography-Mass Spectrometry (GC-MS) di laboratorium, yang lambat dan mahal. Oleh karena itu, diperlukan solusi cepat dan real-time. Sensor cerdas berbasis NIR hadir sebagai alternatif, menggunakan teknik sensor optik di rentang gelombang NIR untuk mengukur kadar rhodinol dalam aliran distilat minyak sereh wangi. Dengan algoritma machine learning, sensor ini mampu memberikan estimasi kadar rhodinol secara real-time, mendukung pemantauan kualitas produk dalam proses fraksinasi dengan efisien dan biaya lebih rendah. Penelitian ini bertujuan untuk menghasilkan rancangan sistem sensor cerdas berbasis NIR untuk pemantauan kadar rhodinol dalam fraksi minyak sereh wangi. Beberapa tantangan penelitian yang telah diselesaikan yaitu investigasi dan identifikasi parameter sensor cerdas, formulasi sensor NIR dalam pendeteksian kadar rhodinol dalam fraksi minyak sereh wangi, integrasi blok Cyber-Physical System (CPS) melalui desain arsitektur fungsional, logikal, dan fisikal dalam sebuah prototipe sistem sensor cerdas untuk pemantauan kadar rhodinol dalam fraksi minyak sereh wangi, serta uji coba dan validasi hasil kinerja sensor cerdas yang dijalankan melalui program aplikasi sistem komputasi. Metode yang digunakan menggunakan pendekatan rekayasa sistem berbasis model (MBSE) dengan beberapa tahapan penelitian. Tahap investigasi dan identifikasi parameter sensor cerdas menggunakan pengujian spektra pada berbagai varian konsentrasi rhodinol dalam fraksi minyak sereh wangi. Pengujian spektrum transmitansi dilakukan menggunakan Near Infrared Spectroscopy NIRFLEX liquid N-500 buatan BUCHI. Sebanyak 16 sampel minyak sereh wangi @ 2 ml dipindai sebanyak 1 kali pada bentangan 10.000 cm-1 (1.000 nm) - 4.000 cm-1 (2.500 nm). Selanjutnya, rangkaian modul sensor optik berupa sensor Vis-NIR spektral AS7263 digunakan untuk mendeteksi reflektansi dari emisi energi elektromagnetik yang dilepaskan oleh larutan sampel minyak sereh wangi. Sampel minyak sereh wangi dideteksi pada bentangan panjang gelombang, λ = 600-870 nm dengan 6 near-IR channels: 610 nm, 680 nm, 730 nm, 760 nm, 810 nm dan 860 nm. Emisi yang diubah menjadi data spektral kemudian digunakan untuk menganalisis komposisi kadar rhodinol dalam minyak sereh wangi melalui kurva kalibrasi. Output data NIR berupa nilai tegangan yang tidak memiliki satuan pengukuran atau disebut nilai indeks. Kemudian, integrasi blok CPS melalui desain arsitektur fungsional, arsitektur logikal dan arsitektur fisikal sensor dirancang untuk menunjukkan keterkaitan antara hardware dan software dalam suatu sistem sensor cerdas. Perangkat keras seperti mikrokontroler, array diode, dan sensor terintegrasi dalam satu sistem dan fungsionalitas sistem digambarkan melalui komponen logis dan koneksinya. Komponen fisik sebenarnya dari sistem sensor cerdas yaitu sensor Vis-NIR spektral AS7263 dan mikrokontroler Arduino Uno R3 ATmega2560 digambarkan secara jelas. Dataset hasil pengukuran tegangan oleh sensor diolah menggunakan algoritma machine learning. Parameter proses fraksinasi (suhu, tekanan, waktu) diperoleh berdasarkan pengamatan dan data historis dari penelitian sebelumnya. Analisis pada data binary classification menggunakan beberapa algoritma yaitu Decision Tree, SVM, Random Forest dan Naïve Bayes. Perbandingan model klasifikasi berdasarkan nilai validasi model seperti AUC, CA, F1 dan Precision. Data mining diolah dengan bantuan software Weka dan Orange data mining. Prediksi kadar rhodinol dalam fraksi minyak sereh wangi dilakukan dengan menggunakan algoritma regresi polinomial pada orde 4. Pengolahan data mining secara komputasi ditampilkan pada GUI desktop yang dirancang menggunakan API Python PyQT. Hasil penelitian menunjukkan bahwa parameter sensor cerdas yang dipilih berdasarkan reflektansi spektrum dari kadar rhodinol dalam fraksi minyak sereh wangi di daerah NIR. Formulasi sensor NIR yang digunakan untuk mendeteksi kadar rhodinol dalam fraksi minyak sereh wangi adalah sebesar 810 nm dengan kemampuan ideal deteksi kadar rhodinol rentang 35-95%. Integrasi blok CPS melalui desain arsitektur fungsional, logikal dan fisikal yang terdiri atas rangkaian modul sensor Vis-NIR spektral AS7263, Arduino Uno R3 dan perangkat komputer yang ditanam algoritma machine learning menghasilkan sebuah prototipe sensor cerdas berbasis NIR. Sistem komputasi berbasis machine learning model klasifikasi menggunakan algoritma Decision Tree dengan max-depth= 7 diperoleh nilai F1 score = 0,989 dan nilai AUC = 0,994. Selanjutnya model prediksi menggunakan algoritma regresi polinomial dapat memprediksi kadar rhodinol dalam fraksi minyak sereh wangi dengan persamaan model prediksi: y = -188.48x4 + 481.75x3 - 439.57x2 + 219.46x + 21.525, R2 = 0.981 dengan nilai akurasi sebesar 98,14% dan presisi sebesar 99,97%. Penelitian lebih lanjut yang disarankan dalam penelitian ini adalah integrasi CPS antara hardware dan software sistem sensor cerdas dengan sistem kendali proses pada alat fraksinasi sehingga dapat mempermudah kontrol, pemantauan, dan efisiensi operasional dalam proses fraksinasi secara kontinu. Selain itu, spektrum NIRs minyak sereh wangi dapat digunakan untuk mengkaji dan memperdalam analisis vibrasi fonon dalam material untuk memperoleh nilai optik transversal (TO) dan nilai optik longitudinal (LO) dengan metode Kramer-Kronig sehingga dapat menghubungkan bagian real dan imajiner fungsi respons optik material.

In this modern era, the advancement of sensor technology has greatly contributed to various applications in different fields. Smart sensors based on Near Infrared (NIR) have emerged as an intriguing area of research due to their ability to provide non-destructive, fast, and accurate analysis of various chemical components in samples. One potential application of NIR sensors is monitoring aromatic compounds such as rhodinol levels in citronella oil fractions. Citronella oil fraction contains rhodinol compounds that play an essential role in perfume and cosmetics. However, detecting rhodinol levels in distillate streams still relies on conventional methods, such as Gas Chromatography-Mass Spectrometry (GC-MS) in laboratories, which are slow and expensive. Therefore, a fast and real-time solution is needed. NIR-based intelligent sensors are available as an alternative, using optical sensor techniques in the NIR wave range to measure rhodinol levels in citronella oil distillate streams. With machine learning algorithms, this sensor can estimate rhodinol levels in real-time, supporting product quality monitoring in fractionation processes efficiently and at lower costs. This research aims to develop a design for a NIR-based smart sensor system to monitor the levels of rhodinol in citronella oil fractions. Several research challenges were addressed, including investigating and identifying the parameters for designing smart sensors, formulating the NIR sensor's ability to detect rhodinol levels in citronella oil fractions, Integrating Cyber-Physical System (CPS) designing functional, logical, and physical architectures to a prototype smart sensor system for monitoring rhodinol levels, and testing and validating the performance of the smart sensor to build computation application program. The method used a Model Based System Engineering (MBSE) approach with several stages of research. Investigation stage and identification of smart sensor design parameters using spectra testing on various variants of rhodinol concentrations in citronella oil fraction. The transmittance spectra testing was performed using BUCHI's Near Infrared Spectroscopy NIRFLEX liquid N-500. Sixteen samples of citronella oil measuring 2 ml were scanned once in the 10.000 cm-1 (1.000 nm) - 4.000 cm-1 (2.500 nm) range. Subsequently, an array of optical sensor modules, specifically the AS7263 Vis-NIR spectral sensor, was used to detect the remaining light intensity after absorption by the citronella oil samples. The citronella oil samples were detected in the wavelength range λ = 600-870 nm with 6 near-IR channels: 610 nm, 680 nm, 730 nm, 760 nm, 810 nm, and 860 nm. The obtained information was used to analyze the composition of rhodinol levels in citronella oil through a calibration curve. The output NIR data consists of voltage values that do not have specific measurement units, often called index values. Subsequently, the sensor's functional, logical, and physical architecture design was developed to illustrate the smart sensor design's relationship between hardware and software. Hardware components such as microcontrollers, diode arrays, and integrated sensors were depicted as a single system, and the system's functionality was described through logical components and their connections. The actual physical components of the smart sensor system, namely the AS7263 Vis-NIR spectral sensor and Arduino Uno R3 microcontroller, were depicted. The voltage measurement dataset obtained from the sensor was processed using machine learning algorithms. The process parameters of the fractionation (temperature, pressure, time) were obtained based on observations and historical data from previous research. Binary classification analysis was performed using several algorithms, including Decision Tree, SVM, Random Forest, and Naïve Bayes. Model classification comparisons were based on validation metrics such as AUC, CA, F1, and Precision. Data mining was processed with the assistance of Weka and Orange data mining software. The prediction of rhodinol levels in citronella oil fractions was conducted using a polynomial regression algorithm of order 4. The computational data processing was displayed on a desktop GUI using API Python PyQT. The results showed that the parameters of the intelligent sensor were selected based on the reflectance spectrum of the rhodinol content in the citronella oil fraction in the NIR region. The NIR sensor formulation used to detect rhodinol levels in the citronella oil fraction is 810 nm, with an ideal ability to detect rhodinol levels in the 35-95% range. Integrating the CPS block through functional, logical and physical architectural design consisting of a series of AS7263 spectral Vis-NIR sensor modules, Arduino Uno R3, and computer devices embedded with machine learning algorithms produces a prototype of an intelligent sensor based on NIR. A computational system based on machine learning classification models using the Decision Tree algorithm with max-depth = 7 obtained F1 score = 0.989 and AUC value = 0.994. Furthermore, the prediction model using the polynomial regression algorithm can predict the rhodinol content in the citronella oil fraction with the prediction model equation: y = -335.12x4 + 800.45x3 – 658.29x2 + 264.72x + 22.524, R2 value of 0.975 with an accuracy value of 98.14% and a precision of 99.97%. Further research suggested in this study is the integration of CPS between intelligent sensor system hardware and software with process control systems in fractionation equipment to facilitate control, monitoring, and operational efficiency in the continuous fractionation process. In addition, the NIRs spectrum of citronella oil can be used to study and deepen the analysis of phonon vibrations in materials to obtain transverse optical values (TO) and longitudinal optical values (LO) using the Kramer-Kronig method so that they can relate the real and imaginary parts of the material's optical response function.