AdaBoost ensemble learning algorithm for precision manufacturing of aerosol jet printing

Aerosol jet printing (AJP) is a hotspot technic applied in customized microelectronic component manufacturing due to its high design flexibility and capability for fine feature deposition. However, its complex processing parameters and no-mask nature make it challenging to produce patterns with a specific line width. Machine learning (ML) is an effective approach for predicting the line width of AJP, but the traditional Gaussian process regression (GPR) method exhibits significant prediction errors. To enhance prediction accuracy, this study proposed an ML approach based on AdaBoost ensemble learning, employing multilayer perceptron (MLP) as base learners. Key hyperparameters were optimized using grid search to construct a high-precision prediction model. Furthermore, the Genetic algorithm (GA) was utilized to optimize the processing parameter space for a desired line width. Experimental results demonstrated that the relative error of predicted linewidth was within 1%, and the model fitting accuracy was significantly enhanced, with the R² of 0.96 and the MSE of 0.024. This approach offers novel theoretical insights and practical guidance for high-precision manufacturing utilizing AJP in the fields of microelectronics and flexible electronics.