Please use this identifier to cite or link to this item:
https://bura.brunel.ac.uk/handle/2438/34206| Title: | Retrospective Evaluation of the In-Process Intelligent Compensation Method for Mass-Customized Manufacturing of Freeform Lenses Based on Ensemble Learning and Dynamic EWMA Fusion |
| Authors: | Chen, Xin [0000-0003-3171-0809] Cheng, Kai Zhao, Liang Fu, Yuanzheng |
| Keywords: | freeform lens manufacturing;in-process intelligent compensation;ensemble learning;dynamic EWMA tracking;high-precision manufacturing |
| Issue Date: | 16-Nov-2026 |
| Publisher: | MDPI |
| Citation: | Chen, X. et al. (2026) 'Retrospective Evaluation of the In-Process Intelligent Compensation Method for Mass-Customized Manufacturing of Freeform Lenses Based on Ensemble Learning and Dynamic EWMA Fusion', Journal of Manufacturing and Materials Processing, 10(11), 514, pp. 1–30. doi: 10.3390/jmmp10110514. |
| Abstract: | Freeform ophthalmic lens manufacturing requires compensation for prescription-dependent deviations and process drift. This study retrospectively evaluates an order-to-order compensation method combining ensemble prediction, exponentially weighted moving average (EWMA), residual tracking, and a proposed material-specific three-level gate. The 2245 manufacturing records are partitioned by whole chronological days into training (1356), validation (484), and final-test (405) sets. Thirteen models are compared, with controller settings selected using validation data only. Under unit-gain replay, joint pass rates for spherical and cylindrical power deviations within the study-specific ±0.10 D band are 57.04% without added compensation, 68.40% with historical fixed-offset correction, 75.06% with static AI, 65.43% with pure EWMA, 74.81% with global AI + EWMA, and 76.54% with material-specific tracking and the gate. The material strategy gains 19.51 percentage points over the baseline (95% date-bootstrap interval: 14.57–25.70). Paired intervals comparing material tracking with static AI and global fusion include zero in the robustness analysis. Global fusion achieves higher joint pass rates than static AI in all five expanding-window evaluations under the robustness protocol. The gate remains FULL in the primary replay, producing identical commands to ungated tracking; one material performs below the baseline. These findings motivate prospective assessment of machine response, manufacturing yield, and gate protection. |
| Description: | Data Availability Statement: The data presented in this study are available on request from the authors. |
| URI: | https://bura.brunel.ac.uk/handle/2438/34206 |
| DOI: | https://doi.org/10.3390/jmmp10110514 |
| Appears in Collections: | Department of Mechanical and Aerospace Engineering Research Papers |
Files in This Item:
| File | Description | Size | Format | |
|---|---|---|---|---|
| FullText.pdf | Copyright © 2026 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license. | 8.61 MB | Adobe PDF | View/Open |
This item is licensed under a Creative Commons License