Investigation of kerf width in CO2 laser fabrication of microchannels on aluminium-coated polymethyl methacrylate (PMMA) using adaptive neuro-fuzzy inference system (ANFIS)

Abstract

Kerf width affects the quality of microfluidic devices. For highly accurate results, kerf width should be minimum. It is challenging to attain minimum kerf width with laser micromachining. Relating the inputs to the outputs is also a critical challenge. Accordingly, an intelligent system with adaptive neuro-fuzzy inference system (ANFIS) has been built to attain the desired kerf widths of microchannels in CO₂ laser fabrication of polymethyl methacrylate (PMMA). The work samples were coated with a 1 μm layer of aluminium of purity 99.5 %. The inputs were speed (10, 15, and 20 mm/s), pulse rate (800, 900, and 1000 pulses/in.), and power (1.5, 3.0, and 4.5 W). The experiments were designed using response surface methodology. The total number of experiments was 54. The ANFIS toolbox was used for generating the ANFIS model in MATLAB R2022a. The model was trained and examined using the experimental results. The chosen membership function was Gaussian (gaussmf). The significance of all the inputs was examined using analysis of variance (ANOVA). Power was found to be the most significant. Speed was second in significance and pulse rate was the last. Mean relative error (MRE), correlation coefficient (R), and mean absolute error (MAE) were used to investigate how accurate the model was. MRE, R, and MAE were 0.003799, 0.99997116 (coefficient of determination, R² = 0.99998558) and 0.582 respectively. The training data had the root mean square error (RMSE) of 0.623405 while RMSE was 1.08345 for the checking data. The accuracy of the ANFIS model was high.