Prediksi Perambatan Retak Pada Plat Aluminum 1100-H16 Menggunakan Artificial Neural Networks Dan Metode Elemen Hingga

Airlangga, Prima Putra (2022) Prediksi Perambatan Retak Pada Plat Aluminum 1100-H16 Menggunakan Artificial Neural Networks Dan Metode Elemen Hingga. Undergraduate thesis, Institut Teknologi Sepuluh Nopember.

This is the latest version of this item.

[img] Text
02511740000100-Undergraduate_Thesis.pdf - Accepted Version
Restricted to Repository staff only

Download (3MB) | Request a copy
[img] Text
02511740000100-Undergraduate_Thesis.pdf
Restricted to Repository staff only

Download (3MB) | Request a copy

Abstract

Kegagalan ( failure ) material akibat pada sebuah struktur atau komponen menjadi permasalahan rumit dikarenakan banyaknya dan kompleksnya variabel yang mempengaruhi. Oleh karena itu maka dibutuhkan teknik komputasi untuk menganalisis perambatan retak. Penelitian ini bertujuan untuk menganalisis keakurasian dan efisiensi model neural networks Generalized Regression Neural Networks (GRNN) sebagai metode untuk memprediksi arah perambatan retak dibandingkan dengan tiga model neural networks yang berbeda, pengaruh learning rate pada model neural networks, dan pengaruh parameter input terhadap prediksi perambaan retak berupa panjang retak, sudut inklinasi retak, jarak ujung retak, dan jarak retak offset. Pada penelitian dilakukan dua metode yaitu simulasi metode elemen hingga (MEH) yang divalidasi secara kualitatif dan modelling neural networks yang divalidasi secara kuantitaif dan kualitatif. Dari proses simulasi menunjukkan bahwa hasil simulasi perambatan retak dengan MEH dapat divalidasi dengan hasil eksperimen. Model GRNN menghasilkan tingkat akurasi dan efisiensi paling tinggi dimana nilai MSE, MAE, dan R2 score sebesar 1,79; 0,50; dan 0,985 dengan epoch kurang dari 100 serta learning parameter optimum 0.001 dimana parameter input digunakan seluruhnya. ================================================================================================ Failure in a structure or component becomes intricate problems due to the complexity and the number of variables influencing on it. In consequences, a computational technique is necessary to analyze and predict crack propagation. This study aims to analyze the accuracy and efficiency of Generalized Regression Neural Networks (GRNN) model as a method for predicting the crack propagation direction by comparing three different neural network models, the effect of learning rate on neural network models, and the influence of input parameters on crack propagation prediction using crack length, crack inclination angle, the distance of two crack tips, and crack offset distance. In this study, two methods were imposed using Finite Element Method (FEM) which was validated subjectively and neural networks modelling which was validated quantitatively and qualitatively. Based on simulation process, it shows that the results of crack propagation by FEM can be validated properly by experimental results. In addition, The GRNN model produces the highest accuracy and efficiency values where the MSE, MAE, and R2 scores respectively are 1.79; 0.50; and 0.985 with epochs less than 100 and the optimum learning parameter is 0.001 using all input parameters.

Item Type: Thesis (Undergraduate)
Uncontrolled Keywords: Retak, GRNN, MEH, Crack, FEA
Subjects: T Technology > T Technology (General) > T57.62 Simulation
T Technology > TA Engineering (General). Civil engineering (General) > TA169.5 Failure analysis
T Technology > TA Engineering (General). Civil engineering (General) > TA347 Finite Element Method
Divisions: Faculty of Industrial Technology and Systems Engineering (INDSYS) > Material & Metallurgical Engineering > 28201-(S1) Undergraduate Thesis
Depositing User: Prima Putra Airlangga
Date Deposited: 11 Feb 2022 04:09
Last Modified: 11 Feb 2022 04:09
URI: https://repository.its.ac.id/id/eprint/93530

Available Versions of this Item

Actions (login required)

View Item View Item