Rapid Manufacturing Solid Microneedle Array by Integrated Digital Light Processing with Machine Learning (DLP-ML)

Microneedle arrays (mu NA) are gaining popularity in the field of drug delivery due to their potential to minimize side effects compared to conventional hypodermic needles. Generally, mu NA is manufacturing through three or two stages which lack of flexibility in design. 3d printing is one of the powerful method that could be employed in three or two stages. However, it could be extended to single stage by directly manufacturing mu NA from designated material. Digital light processing (DLP) is a kind of 3d printing which is using in this study. A further understanding of DLP printing parameters would be investigated such as LED current (mA), curing time (sec), z-thickness (mu m), and grayscale level image (0 to 255). This study explores the integration of DLP technology and machine learning (ML), namely DLP-ML, to manufacture mu NA with customizable size, rapid production process, and reduced material wastage. The initial process in developing ML model is preparing a dataset obtained from experimental approach. An orthogonal table is simplified the number of experiment. The orthogonal table is contained five factors and five levels, L-25. The dataset is divided into input and output and substituted to ML system. In this approach, ML algorithms are employed to determine the optimal digital mask patterns and printing parameters for DLP-ML manufacturing. The study is employed various ML configurations in term of algorithm and number of hidden layer. Under different algorithm, the sort of larger to lower accuracy is arranged Bayesian regularization backpropagation (BR10), Levenberg-Marquardt backpropagation (LM10), and scaled conjugate gradient backpropagation (SCG(10)), respectively. It also notices that the deviation of BR10 is proving that mu NA is manufactured uniformly on a substrate. It is ultimately selecting Bayesian regularization backpropagation for its superior accuracy, exceeding 90%. For different number of hidden layer, the default value, 10, is also resulting highest accuracy among others. Then, BR10 could be concluded as the best ML configuration in this case. This algorithm empowers DLP-ML manufacturing to produce mu NA arrays with various aspect ratio (AR), representing the ratio between the height and base diameter of mu NA. The AR of 15 is highest that could be performed and the critical AR of mu NA bending is found at 7 (700/100). By harnessing the synergy of DLP-ML technologies, this research presents a promising avenue for the advancement of drug delivery systems, offering enhanced precision, scalability, and therapeutic efficacy in medical applications. Finally, DLP-ML is successfully developed a rapid mu NA manufacturing with different dimensions very accurately.