High-Dimensional Waveform Optimization Dataset for Piezoelectric Inkjet Printing

High-Dimensional Waveform Optimization Dataset for Piezoelectric Inkjet Printing 1. Overview This repository contains the experimental dataset Experimental_Data_Piezoelectric_Inkjet.csv generated for the study of multi-pulse driving waveform optimization in Piezoelectric Inkjet (PIJ) printing. The dataset was collected to address the constrained high-dimensional physical black-box optimization challenge, where the feasible region for fluid jetting within a 12-dimensional parameter space is highly sparse, and traditional unconstrained optimization can easily cause ink overflow or printhead damage. Using the proposed Classifier-Assisted Multi-Trust-Region Bayesian Optimization (CA-TuRBO-m) framework, the physical printing data was efficiently collected by combining a Random Forest active learning classifier to dynamically intercept high-risk parameter combinations and a multi-trust-region mechanism to balance exploration and exploitation. Instead of relying on complex computational fluid dynamics (CFD) models, this dataset utilizes static printed dot features on paper as a composite visual feedback source. This dataset is the foundation of the research presented in the following paper: J. Zhang, H. Zhan, Y. Zhang, Y. Zhang, "Classifier-Assisted Multi-Trust-Region Bayesian Optimization for High-Dimensional Waveform Design in Piezoelectric Inkjet Printing," Please cite this paper if you use this dataset in your research. 2. Data Description Experimental_Data_Piezoelectric_Inkjet.csv is a CSV file containing 5758 rows and 18 columns. Among them, 12 columns are input features and 6 columns are output targets. 3. Data Dictionary The dataset consists of 12 input features (composite waveform decision variables) and 6 output features (machine vision feedback and print quality evaluation metrics). Input Features (12-D Waveform Space) V_amp1: Voltage amplitude of Pulse I/II (V), range [3.0, 18.0] t_trans1: Voltage transition time of Pulse I/II (μs), range [0.3, 2.0] t_hold1: Voltage hold time of Pulse I/II (μs), range [0.3, 5.0] t_gap1: Gap time between Pulse I and II (μs), range [0.3, 3.0] t_gap2: Gap time between Pulse II and III (μs), range [0.3, 10.0] V_p: Positive amplitude of Pulse III (V), range [3.0, 18.0] t_trans_p: Positive transition time of Pulse III (μs), range [0.3, 2.0] t_hold_p: Positive hold time of Pulse III (μs), range [0.3, 5.0] V_n: Negative amplitude of Pulse III (V), range [18.0, 24.0] t_trans_n: Negative transition time of Pulse III (μs), range [0.3, 2.0] t_hold_n: Negative hold time of Pulse III (μs), range [0.3, 5.0] t_gap3: Associated gap time of Pulse III (μs) Output Features (Print Quality & Vision Feedback) classification_dot: Binary classification label (1.0 or 0.0), indicating whether the printed dot is successfully printed. avg_circ: Average circularity of the droplets. sigma_total_um: Standard deviation of the dot morphology in micrometers. sat_ratio: Satellite ratio, reflecting the proportion of undesirable satellite droplets. avg_area_um2: Average area of the main printed droplets in square micrometers. objective_score: Composite Print Quality Index (CPQI), reflecting the morphological score of the droplet; a lower score indicates a more ideal droplet morphology. 4. Experimental Conditions Printhead Model: Ricoh GEN6 (MH5320) Experimental Temperature: Maintained at 25 ± 1 °C. Physical Properties of the Testing Ink: Ink Density (kg/m³) Surface Tension (mN/m) Dynamic Viscosity (mPa·s) 1030.23 43.573 13.729 5. Contact If you have any questions about this dataset, please contact the corresponding author: Hongwu Zhan ([e-mail]).