Electronic Theses and Dissertation

Minyak nilam (Pogostemon cablin Benth) merupakan salah satu komoditas unggulan Indonesia di sektor minyak atsiri. Kualitas ini dinilai melalui beberapa parameter, salah satunya adalah indeks bias, yang mencerminkan tingkat kemurnian dan konsistensi senyawa kimia di dalam minyak. Pengukuran indeks bias umumnya dilakukan dengan metode konvensional yang handal dan telah banyak digunakan. Namun dalam perkembangannya teknologi inframerah mulai diterapkan karena bersifat cepat, non-destruktif, dan ramah lingkungan. Penelitian ini bertujuan untuk menganalisis kemampuan Fourier Transform Inframerah Dekat (FT-NIR) dalam menduga nilai indeks bias pada minyak nilam serta menganalisis pengaruh pretreatment untuk meningkatkan performa model, dan menentukan rentang panjang gelombang optimum.
Penelitian ini menggunakan minyak nilam mentah Aceh yang didapat dari petani nilam melalui Atsiri research Center (ARC). Minyak nilam difraksinasi dan dicampurkan untuk menghasilkan variasi kadar patchouli alcohol (PA). Pengujian kadar PA dilakukan dengan metode Gas Cromatography and Mass Spectrometry (GCMS). Pengukuran indeks bias aktual dilakukan menggunakan hand refractometer. Selanjutnya dilakukan akuisisi spektrum untuk setiap sampel minyak menggunakan alat FT-NIR Thermo Nicolet Antaris II TM. Model pendugaan angka indeks bias dianalisis dengan metode Partial Least Square Regression (PLSR) dan diberikan pretreatment de-trending (DT) dan derivative-1 (D1).
Sebanyak 30 variasi sampel diperoleh dengan kadar PA dari 21,73% sampai 41,10 %. Nilai indeks bias aktual yang diperoleh adalah 1.5040—1.5085. Hasil penelitian ini menunjukkan bahwa FT-NIR dengan metode PLSR dapat digunakan untuk menduga indeks bias minyak nilam secara cepat, non-destruktif, dan efisien. Model tanpa pretreatment menghasilkan nilai koefisien korelasi (r) sebesar 0,85, koefisien determinasi (R²) sebesar 0,72, error (RMSEC) sebesar 0,00078 dan RPD sebesar 1,9. Hasil ini menunjukkan bahwa performa model awal masih rendah. Namun setelah dilakukan pretreatment DT dan D1 terbukti mampu meningkatkan kinerja model kalibrasi. Model terbaik adalah PLSR-D1 dengan LV-9 yang memiliki kriteria koefisien korelasi sebesar 0,97, koefisien determinasi sebesar 0,95, error sebesar 0,00034, serta nilai RPD sebesar 4,5. Hasil ini menunjukkan bahwa model yang dibangun tergolong akurasi prediksi yang sangat baik (excellent prediction accuracy). Hasil analisis loading plot dengan menggunakan Latent Variable (LV)-4 menunjukkan bahwa kontribusi paling signifikan terhadap pembentukan model kalibrasi terdapat pada rentang panjang gelombang optimum 1910-1980 nm.

Patchouli oil (Pogostemon cablin Benth) is one of Indonesia’s leading commodities in the essential oil sector. Its quality is assessed through several parameters, one of which is the refractive index, reflecting the purity and consistency of the chemical compounds within the oil. Measurement of the refractive index is generally carried out using conventional methods that are reliable and widely applied. However, with technological advancements, infrared spectroscopy has begun to be used due to its rapid, non-destructive, and environmentally friendly characteristics. This study aims to analyze the capability of Fourier Transform Near-Infrared (FT-NIR) spectroscopy in predicting the refractive index of patchouli oil, to evaluate the effect of spectral pretreatments in improving model performance, and to determine the optimum wavelength range. The research utilized raw Aceh patchouli oil obtained from local farmers through the Atsiri Research Center (ARC). The oil was fractionated and blended to produce variations in patchouli alcohol (PA) content. The PA content was determined using the Gas Chromatography–Mass Spectrometry (GC-MS) method, while the actual refractive index was measured using a hand refractometer. Subsequently, spectral acquisition was conducted for each oil sample using the FT-NIR Thermo Nicolet Antaris II™ instrument. The refractive index prediction model was developed using the Partial Least Squares Regression (PLSR) method with de-trending (DT) and first derivative (D1) pretreatments applied. A total of 30 samples were obtained with PA contents ranging from 21.73% to 41.10%, and actual refractive index values between 1.5040 and 1.5085. The results indicate that FT-NIR combined with PLSR can be effectively used to predict the refractive index of patchouli oil rapidly, non-destructively, and efficiently. The model without pretreatment yielded a correlation coefficient (r) of 0.85, coefficient of determination (R²) of 0.72, root mean square error of calibration (RMSEC) of 0.00078, and a residual predictive deviation (RPD) of 1.9, indicating low model performance. However, after applying DT and D1 pretreatments, model performance significantly improved. The best model was PLSR-D1 with 9 latent variables (LV-9), which achieved a correlation coefficient of 0.97, coefficient of determination of 0.95, RMSEC of 0.00034, and an RPD value of 4.5, indicating excellent prediction accuracy. The loading plot analysis using LV-4 revealed that the most significant contribution to the calibration model was found within the optimal wavelength range of 1910–1980 nm.