Application of CANFIS for modelling and predicting multiple output performances for different materials in μEDM
Micro Electrical Discharge Machining (μEDM) is one of the most demanding manufacturing processes available today. The selection of EDM parameters remains a challenge since it is frequently based on machinist intuition and heuristic approaches. Artificial intelligence algorithms have been used to mod...
Saved in:
Main Authors: | , , |
---|---|
Format: | Article |
Language: | English |
Published: |
Elsevier Ltd
2022
|
Subjects: | |
Online Access: | http://irep.iium.edu.my/97437/1/1-s2.0-S1755581722000499-main.pdf http://irep.iium.edu.my/97437/ https://www.sciencedirect.com/science/article/abs/pii/S1755581722000499 https://doi.org/10.1016/j.cirpj.2022.02.021 |
Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
Summary: | Micro Electrical Discharge Machining (μEDM) is one of the most demanding manufacturing processes available today. The selection of EDM parameters remains a challenge since it is frequently based on machinist intuition and heuristic approaches. Artificial intelligence algorithms have been used to model and predict the μEDM machining process in recent years. However, artificial intelligence has not been established for predicting μEDM performances based on material properties. Therefore, this paper has proposed a model that considers the material properties, such as thermal conductivity, melting point, and electrical resistivity. Since μEDM is a non-linear and stochastic process, Coactive Neuro-Fuzzy Inference Systems (CANFIS) was proposed to model and predict the multiple μEDM performances on various materials. The material properties, feed rate, capacitance, and gap voltage are input parameters in a three-level design based on a full factorial experiment. The CANFIS model can accurately predict the material removal rate (MRR), total discharge pulse, overcut, and taperness in a single model. The mean average percentage error (MAPE) of various outputs (predicted by the model) for test dataset such as MRR, total discharge pulse, overcut, and taper angle were found to be 4.5% (95.4% accuracy), 6.8% (93.2% accuracy), 15.4% (84.6% accuracy) and 15.2% (84.8% accuracy) respectively. |
---|