Perancangan Sistem Deteksi Pararel Misalignment Pada Motor Induksi Mengunakan Sensor Arus Dengan Metode Convolution Neural Network

Pada penelitian ini dirancang sistem deteksi kesalahan pararel misalignment pada motor induksi menggunakan pengukuran arus dengan convolution neural network. Perancangan sistem ini dibuat dengan menggunakan sensor arus jenis current transformator dan Arduino mikrokontroller untuk pengolahan sinyal arus. Sinyal kemudian diolah dengan Fast Fourier Transform untuk mengubah output Arduino dari domain waktu menjadi domain frekuensi. Selanjutnya sinyal diolah untuk mendeteksi adanya peak di daerah f+nfr. Hasil dari fast fourier transform dijadikan input sistem convolution neural network untuk mengetahui tingkat akurasinya. Dalam eksperimen deteksi motor induksi dengan malakukan variasi pergantian kopling normal dan pararel misalignment didapatkan hasil pembacaan sinyal ketika diubah kedalam domain frekuensi yang terjadi adalah peak dalam kondisi kopling normal hanya muncul di daerah f = 13,5 Hz sedangkan kondisi pararel 1 cm ditemukan peak didaerah f+fr = 20 hz, Pararel 2 cm ada di f+fr (23 hz) dan f+2fr (30 hz). Dengan Convolution Neural Network tingkat akurasi sebesar 87,5 %. Dengan iterasi/epoch 100 kali, untuk data training, kondisi mulai stabil pada epoch 60 dengan akurasi 0,87. Sedangkan pada data testing kondisi mulai stabil di epoch 20 dengan akurasi 0,87. ====================================================================================================== In this study, a parallel misalignment error detection system was designed for an induction motor using current measurement with a convoluted neural network. The design of this system is made using a transformer type current sensor and an Arduino microcontroller for current signal processing. The signal is then processed with Fast Fourier Transform to convert Arduino output from time domain to frequency domain. Furthermore, the signal is used to detect the presence of a peak in the f+nfr region. The results of the fast Fourier transform are used as input for the convolution neural network system to determine the level of accuracy. In the induction motor detection experiment by varying the normal coupling change and parallel misalignment, it was found that the reading occurred when it was converted into a frequency domain which was a peak under normal coupling conditions only appeared in the area f = 13.5 Hz in parallel conditions 1 cm found a peak in the f + fr area. = 20 hz, 2 cm parallel is at f+fr (23 hz) and f+2fr (30 hz). With Convolution Neural Network the accuracy rate is 87.5%. With 100 iterations/epochs, for training data, the condition starts to stabilize at 60 epochs with an accuracy of 0.87. Meanwhile, in data testing, the condition began to stabilize at epoch 20 with an accuracy of 0.87