Rancang Bangun Smart Fish Feeder Berbasis IoT dan Fuzzy Sugeno dengan Evaluasi Akurasi dan Repeatability

Abstract

Smart fish feeder diperlukan untuk membantu pemberian pakan secara otomatis, tetapi massa pakan yang dihasilkan belum selalu sesuai target dan dapat berbeda pada pengujian berulang. Penelitian ini bertujuan merancang dan membangun smart fish feeder berbasis IoT dan Fuzzy Sugeno serta mengevaluasi akurasi dan repeatability massa keluarannya. Sistem dikembangkan menggunakan metode prototyping dengan ESP32, sensor US-100, servo, RTC, dan antarmuka web. Pengujian dilakukan pada tiga target porsi, yaitu 60 g, 80 g, dan 105 g, serta tujuh level pakan dari 100% hingga 40%. Setiap kombinasi diuji lima kali sehingga diperoleh 21 skenario dan 105 data massa. Hasil menunjukkan sistem berhasil diimplementasikan dan fitur IoT berfungsi sesuai skenario pengujian. Evaluasi menghasilkan MAE gabungan sebesar 5,82 g, rata-rata signed error +3,34 g, dan rata-rata CV 3,21% dengan rentang 1,19–5,20%. Absolute error terendah diperoleh pada target 60 g dan level 60%, sedangkan tertinggi pada target 105 g dan level 40%. Kondisi paling akurat tidak selalu sama dengan kondisi paling repeatable.

A smart fish feeder is needed to support automatic feeding, but the resulting feed mass does not always match the target and may vary across repeated tests. This study aimed to design and develop an IoT and Fuzzy Sugeno-based smart fish feeder and to evaluate the accuracy and repeatability of its output mass. The system was developed using a prototyping method with an ESP32, a US-100 ultrasonic sensor, a servo, an RTC, and a web interface. Tests were conducted at three portion targets, namely 60 g, 80 g, and 105 g, and seven feed levels ranging from 100% to 40%. Each combination was tested five times, resulting in 21 scenarios and 105 mass data points. The results showed that the system was successfully implemented and that the IoT features functioned as expected across the tested scenarios. The evaluation yielded a combined MAE of 5.82 g, a mean signed error of +3.34 g, and a mean CV of 3.21%, ranging from 1.19% to 5.20%. The lowest absolute error was obtained at the 60 g target and 60% feed level, while the highest was obtained at the 105 g target and 40% feed level. The most accurate condition was not always the same as the most repeatable condition.