Ratri, Widhi Sulistyaning (2017) Perbandingan Rentang Frekuensi Pengukuran Sifat Biolistrik Secara Longitudinal Untuk Pendugaan Nilai Pol Dan Brix Tebu (Saccharum Officinarum) Berbasis Dielektrik Dan Jaringan Saraf Tiruan. Sarjana thesis, Universitas Brawijaya.
Abstract
Indonesia mengalami peningkatan kebutuhan gula dari tahun ke tahun. Direktorat Jendral Perkebunan mencatat konsumsi gula Indonesia tahun 2014 mencapai 5,700 juta ton. Peningkatan kebutuhan gula belum diimbangi produksi gula setiap tahun. Rendemen tebu merupakan salah satu indikator keberhasilan produksi gula. Penentuan rendemen tebu mengacu pada perhitungan Rendemen Sementara (RS). Perhitungan RS dipengaruhi nilai pol dan brix tebu. Brix adalah jumlah zat kering (gula dan non gula) terlarut (dalam gram) setiap 100 g larutan. Pol adalah jumlah (gram) gula terlarut dalam 100 gram larutan yang diukur menggunakan polarimeter. Pengukuran pol dan brix tebu yang diterapkan di pabrik gula selama ini tergolong konvensional sehingga kurang efektif. Upaya mempercepat pengukuran RS dapat dilakukan dengan merancang alat yang praktis dan cepat menggunakan metode non-destruktif, salah satunya berbasis dielectric properties dan Jaringan Syaraf Tiruan (JST). Pengukuran sifat biolistrik menghasilkan nilai yang khas karena adanya pengaruh eksternal dan internal bahan. Salah satu pengaruh eksternal bahan yaitu frekuensi pengukuran. Frekuensi pengukuran sifat biolistrik mempengaruhi nilai yang didapat. Perbedaan nilai biolistrik yang diperoleh membuktikan adanya aktivitas perpindahan ion bahan yang berbeda. Untuk mengetahui tingkat pergerakan ion melalui nilai biolistrik dilakukan pembagian rentang frekuensi pengukuran sifat biolistrik menjadi frekuensi rendah (50-1000 Hz), tengah (1100-10,000 Hz) dan tinggi (11,000-200,000 Hz). Data sifat biolistrik memerlukan preprocessing sebelum diolah menggunakan JST. Alternatif preprocessing data digunakan dalam penelitian ini adalah data cleaning sedangkan preprocessing pada JST menggunakan normalisasi dengan metode minimum maksimum (minmax) dan Standar Deviasi (std). Hasil pendugaan dengan kombinsi arsitektur jaringan meliputi fungsi aktivasi logsig purelin, fungsi pembelajaran trianlm, learning rate 0.1, momentum 0.9 dan iterasi 5000 menunjukkan rentang frekuensi pengukuran tengah menghasilkan prediksi pol dan brix terbaik. Model jaringan terbaik yaitu 4-25-25-2 (4 neuron input, 25 neuron hidden layer 1, 25 neuron hidden layer 2, 2 neuron output). Nilai MSE testing topologi terbaik sebesar 0.013 dengan R2 testing sebesar 0.99. Preprocessing data terbaik pada penelitian ini yaitu normalisasi metode minimum maksimum. Hasil pendugaan nilai pol dan brix digunakan untuk menghitung RS prediksi sehingga diketahui selisih RS prediksi dan aktual.
English Abstract
Indonesia experienced an increase in sugar demand from year to year. Directorate General of Plantation noted that Indonesia's sugar consumption in 2014 reached 5.700 million tons. The increase of sugar demand has not been balanced with the production of sugar in every year. Sugarcane yield is one indicator of success of sugar production. Determination of sugar cane yield refers to the calculation of temporary yield. The calculation is influenced by sugarcane pol and brix. Brix is the amount of dry matter (sugar and non-sugar) dissolved (in grams) in every 100 g of solution. Pol is the amount (grams) of dissolved sugar in 100 grams of a solution measured by using a polarimeter. The measurement of sugarcane pol and brix applied in the sugar factory so far is conventional. Therefore, it is less effective. Efforts to speed up the yield measurements can be done by designing practical and fast tools with non-destructive methods, one of which is based on dielectric properties and Artificial Neural Network (ANN). The measurement of bioelectrical properties results in a distinctive value due to the external and internal effects of the material. One of the external influences of the material is the measurement of the frequency. Frequency measurement of biolelectrical properties affects the value which is obtained. Differences in the biolectric values obtained prove to the activity of ionic movement of different materials. To know the level of ion movement through biolelectric value, dividing the frequency range measurement of biolistrik to low frequency (50-1000 Hz), middle (1100-10,000 Hz) and high (11,000-200.000 Hz) is needed. Bioelectrical properties data which is required the preprocessing before being processed using Artificial Neural Network (ANN). Alternative preprocessing data used in this research is normalization with minimum maximum method (minmax) and Standard Deviation (std). The result of the estimation with architectural combination of activation function of logsig purelin, trianlm learning function, 0.1 learning rate, 0.9 momentum, and 5000 iteration showed the result of the middle measurement of frequency range pol and brix prediction. The best network model is 4-25-25-2 (4 input neurons, 25 hidden layers 1 neuron, 25 hidden layer 2 neurons, 2 output neurons). The best MSE topology testing score is 0.013 with R testing of 0.99. In this study, the best preprocessing data is normalization of minimum maximum methods. The result of prediction of pol and brix value is used to calculate the RS prediction so that the predicted and actual RS differences are known
Other obstract
-
| Item Type: | Thesis (Sarjana) |
|---|---|
| Identification Number: | SKR/FTP/2017/309/051706453 |
| Uncontrolled Keywords: | Brix, Dielektrik, JST, Pol, Preprocessinng data, Tebu |
| Subjects: | 300 Social sciences > 338 Production > 338.1 Agriculture |
| Divisions: | Fakultas Teknologi Pertanian > Teknologi Industri Pertanian |
| Depositing User: | Kustati |
| Date Deposited: | 09 Aug 2017 01:46 |
| Last Modified: | 29 Nov 2021 02:51 |
| URI: | http://repository.ub.ac.id/id/eprint/1126 |
Preview |
Text
Widhi Sulistyaning Ratri.pdf Download (2MB) | Preview |
Actions (login required)
 |
View Item |