Sistem Pengendalian Tekanan Pompa Air Berbasis Inverter dengan Neural Network

Sianipar, Yoel Yhokhanan (2026) Sistem Pengendalian Tekanan Pompa Air Berbasis Inverter dengan Neural Network. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5022221081-Undergraduate_Thesis.pdf Restricted to Repository staff only Download (31MB) | Request a copy

Abstract

Tekanan air pada pompa air rumah tangga cenderung tidak stabil, terutama ketika beberapa keran dibuka secara bersamaan, sehingga diperlukan sistem yang mampu menjaga tekanan air tetap stabil. Solusi yang umum digunakan berupa perangkat tambahan seperti *pressure switch* yang bekerja pada sisi keluaran pompa. Penelitian ini menawarkan pendekatan berbeda, yaitu mengendalikan tekanan dari sisi suplai daya melalui rangkaian PWM AC chopper bertopologi diode bridge dengan pengendali berbasis *neural network*. Pengendali yang digunakan adalah Multi-Layer Perceptron (MLP) 3-4-1 dengan masukan berupa galat tekanan, perubahan galat, dan *duty cycle*, serta keluaran berupa perubahan *duty cycle*, yang dilatih menggunakan data operasi *closed-loop* (*behavior cloning*). Hasil pengujian menunjukkan bahwa rangkaian PWM AC chopper mampu mengatur tegangan RMS keluaran secara linear terhadap *duty cycle* dengan selisih kurang dari 1% terhadap nilai teoritis, serta *dead time* terukur sebesar 1,48–1,49 µs sesuai dengan rancangan. Model *neural network* mencapai koefisien determinasi R² = 0,915 pada data uji, yang menunjukkan akurasi prediksi aksi kendali yang baik. Pada pengujian sistem, pengendali mampu mempertahankan tekanan pada *setpoint* dengan galat tunak ≤ 0,009 bar untuk variasi satu hingga empat keran terbuka. Pada pengujian perubahan *setpoint* (0,25–0,35 bar), *settling time* berkisar antara 4–7 detik, sedangkan pada pengujian penolakan gangguan akibat perubahan beban keran, tekanan kembali stabil dalam waktu 9–15 detik. Keterbatasan terjadi ketika seluruh keran tertutup atau saat empat keran dibuka secara bersamaan, sehingga tekanan minimum pompa melampaui *setpoint* dan menyebabkan aktuator mengalami saturasi fisik. Dengan demikian, sistem pengendalian tekanan pompa air berbasis *neural network* dari sisi suplai terbukti mampu bekerja dengan baik terhadap variasi beban.
===================================================================================================================================
Water pressure in household pumps tends to be unstable, particularly when multiple faucets are opened simultaneously, creating the need for a system capable of maintaining stable water pressure. Conventional solutions typically employ additional devices such as pressure switches installed on the pump outlet side. This study proposes an alternative approach by regulating pressure from the power supply side through a diode bridge-based PWM AC chopper controlled by a neural network. The controller employs a 3-4-1 Multi-Layer Perceptron (MLP) with pressure error, error change, and duty cycle as inputs, and duty cycle increment as the output. The model is trained using closed-loop operational data through a behavior cloning approach. Experimental results demonstrate that the PWM AC chopper regulates the output RMS voltage linearly with respect to the duty cycle, exhibiting less than 1% deviation from the theoretical value, while the measured dead time of 1.48–1.49 µs closely matches the design specification. The neural network model achieves a coefficient of determination (R²) of 0.915 on the test dataset, indicating high accuracy in predicting control actions. System-level evaluation shows that the controller maintains the desired pressure setpoint with a steady-state error of no more than 0.009 bar for operating conditions ranging from one to four open faucets. During setpoint tracking tests (0.25–0.35 bar), the settling time ranges from 4 to 7 seconds, while load disturbance rejection tests recover the desired pressure within 9–15 seconds. Performance limitations occur when all faucets are closed or when four faucets are opened simultaneously, causing the pump's minimum output pressure to exceed the setpoint and forcing the actuator into physical saturation. Overall, the proposed supply-side neural network-based water pump pressure control system is demonstrated to operate effectively under varying load conditions.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	PWM AC Chopper, Neural Network, Tekanan Pompa Air, Multi Layer Perceptron PWM AC Chopper, Neural Network, Water Pump Pressure, Multi Layer Perceptron
Subjects:	Q Science > QA Mathematics > QA76.87 Neural networks (Computer Science) T Technology > TJ Mechanical engineering and machinery > TJ910 Electric pumping machinery T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK2785 Electric motors, Induction. T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK2851 Voltage regulators.
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Electrical Engineering > 20201-(S1) Undergraduate Thesis
Depositing User:	Yoel Yhokhanan Sianipar
Date Deposited:	23 Jul 2026 01:20
Last Modified:	23 Jul 2026 01:20
URI:	http://repository.its.ac.id/id/eprint/135995

Actions (login required)

View Item