Visualisasi Prediksi Remaining Useful Life Bearing Menggunakan Regresi Bayesian

Abstract

Abstract. Bearing element are prone to failure, which can cause economic losses and even fatalities. Prediction of the remaining age is utilized to see conditions that may occur in order to avoid dissatisfaction. The Bayesian method is a method for estimating parameter distributions that have high accuracy. This thesis aims to apply the estimated parameters of the least squares method and Bayesian regression model to predict Remaining Useful Life (RUL) bearings. The bearing index degradation was obtained using principal components through a dimension reduction process. Time domain features are reduced from the corresponding vibration signals to construct Health Indicators (HI). Bayesian regression index degradation was used to predict RUL. The data used is secondary data on accelerated degradation related to China's XJTU-SY. RUL prediction results were acquired at tp of 60 minutes. For the horizontal direction on the standard deviation feature, RUL prediction values were obtained with KT of 54 minutes and Bayesian of 11 minutes, while for the kurtosis factor feature, RUL prediction values were earned with KT of 46 minutes and Bayesian of 40 minutes. For the vertical direction, the peak value feature with KT is 57 minutes, and Bayesian is 28 minutes. The RUL graph shows that the prediction line has an up or down trend, indicating that predictions using KT bearing degradation are slower than those utilizing Bayesian. It can be concluded that Bayesian predictions are more accurate than KT because, using Bayesian RUL value predictions, the bearing degradation is smaller, meaning that bearing degradation can be predicted more quickly. Maintenance can be carried out immediately to reduce maintenance costs.

Abstrak. Elemen bearing rentan terhadap kegagalan yang dapat menyebabkan kerugian secara ekonomi bahkan korban jiwa. Prediksi sisa usia digunakan untuk melihat kondisi kelayakan bearing guna menghindari terjadinya kegagalan. Metode Bayesian merupakan metode untuk mengestimasi parameter distribusi yang memiliki akurasi yang tinggi. Skripsi ini bertujuan untuk menerapkan estimasi parameter model regresi metode kuadrat terkecil dan Bayesian pada prediksi Remaining Useful Life (RUL) bearing. Indeks degradasi bearing diperoleh melalui proses reduksi dimensi menggunakan komponen utama. Fitur domain waktu di reduksi dari sinyal vibrasi bearing untuk membangun Health Indicator (HI). Regresi Bayesian indeks degradasi digunakan untuk memprediksi RUL. Data yang digunakan merupakan data sekunder akselerasi degradasi bearing XJTU-SY China. Didapatkan hasil prediksi RUL pada tp sebesar 60 menit, untuk arah horizontal pada fitur standar deviasi didapatkan nilai prediksi RUL dengan KT sebesar 54 menit dan Bayesian sebesar 11 menit sedangkan pada fitur faktor kurtosis didapatkan nilai prediksi RUL dengan KT sebesar 46 menit dan Bayesian sebesar 40 menit. Untuk arah vertikal pada fitur nilai puncak dengan KT sebesar 57 menit dan Bayesian sebesar 28 menit. Dilihat dari grafik RUL garis prediksi memiliki trend naik atau turun yang menunjukkan prediksi menggunakan KT degradasi bearing lebih lambat daripada menggunakan Bayesian. Dapat disimpulkan bahwa prediksi menggunakan Bayesian lebih akurat daripada KT karena menggunakan Bayesian nilai prediksi RUL degradasi bearing lebih kecil, artinya degradasi bearing dapat diprediksi lebih cepat dan dapat segera dilakukan perawatan untuk mereduksi biaya perawatan.