Abstract:Glass with good processability, chemical inertness, and optical transparency has been widely used in optical and aesthetic products, many of which require curve pro-files with high precision. To meet the increasingly tightened geometrical tolerances and fast product updating rates, the traditional approach of developing a thermoform-ing process through trials and errors can cause a large waste of time and resources and often end up with failure. Hence, there is a need to develop an efficient predictive model, replacing the costly simulations or experiments, to assist the design of preci-sion glass thermoforming. In this work, we report a dimensionless back-propagation neural network (BPNN) that can adequately predict the form errors and thus compen-sate for these errors in mold design to achieve precision glass molding. Based on the precision molds, also discussed is the issue of error magnification considering that cover glass for AR/VR glasses or smartphones, with extremely large scale of produc-tion, may require a lower level of mold machining accuracy. It is expected that this BPNN will also be implementable in the glass-manufacturing industry, i.e., trained using industrial data for precision mold designs.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is: **How to design precision glass thermo - forming molds efficiently and accurately to meet the requirements of high - precision geometric tolerances and rapid product updates?** Specifically, the traditional method of developing glass thermo - forming processes through trial - and - error wastes a great deal of time and resources and often fails. Therefore, there is an urgent need to develop an efficient prediction model to replace expensive simulations or experiments and assist in the design of precision glass thermo - forming molds. ### Specific Background of the Problem 1. **Application Requirements of Glass**: - Glass is widely used in optical and aesthetic products because of its good processability, chemical inertness, and optical transparency. - Many applications (such as cover glasses for AR/VR glasses and smartphones) require products with high - precision curved surfaces. 2. **Limitations of Traditional Methods**: - Traditional grinding, polishing, and lapping methods are very time - consuming and expensive. - Methods such as single - point diamond turning can produce complex glass profiles, but it may take hours or even tens of hours to produce one product. - Although the precision glass molding (PGM) technology can be rapidly formed at high temperatures, it requires error compensation to ensure high precision. 3. **Importance of Error Compensation**: - Adding an error - compensation profile in the mold design can offset the shape error of the final glass product. - The "Displacement Adjustment" (DA) method and the "Force Descriptor" method in metal forming have been successfully applied in other fields, but glass thermo - forming has additional error sources, such as viscous flow at high temperatures and shrinkage during the cooling process. ### Solution To overcome the above problems, the author proposes a model based on a dimensionless back - propagation neural network (BPNN) to predict the error - compensation profile in mold design. The main features of this model include: - **Input Variables**: Geometric features (such as coordinates, thickness, Gaussian curvature, and inclination angle) and forming parameters (such as annealing rate and forming temperature). - **Output Variables**: Error - compensation profiles (FECs), that is, the amount of error compensation in mold design. - **Data Generation**: Generate a training data set through finite - element simulation, covering glass products with different geometric shapes and forming conditions. - **Model Structure**: Adopt a BPNN with 8 hidden layers, the activation function is ReLU, the loss function is mean - square error, and the optimizer is Adam. - **Verification and Application**: Verify the prediction ability of the model by comparing simulation results and actual data, and test its generalization performance for glass products of different sizes. ### Conclusion This BPNN model can predict the error - compensation profiles required for rotationally symmetric glass products with different geometric shapes and sizes with an error of less than 0.01%, significantly improving the efficiency and accuracy of precision glass thermo - forming mold design. In addition, this model can also be applied to glass products in large - scale production, such as cover glasses for smartphones and tablets, thereby reducing costs and improving production efficiency.

Precision Glass Thermoforming Assisted by Neural Networks

Finite Element Analysis on Non-Isothermal Glass Molding

Numerical Simulation of Glass Bottle Forming Process

Application of Orthogonal Test in Numerical Simulation of Glass Lens Molding

Influences of Model’s Shape on Molding Time in Glass Molding Press

A Framework of Cooperative Design for Aspheric Glass-Lens Mold

Evaluation of GlassNet for physics-informed machine learning of glass stability and glass-forming ability

Real-Time Quality Control in Thin Glass Forming Using Infrared Thermography and Deep Learning

Efficient mold manufacturing for precision glass molding

Evaluation of GlassNet for physics‐informed machine learning of glass stability and glass‐forming ability

Process modeling and optimization in laser drilling of bulk metallic glasses based on GABPNN and machine vision

Glass Optimization Using Neural Network

Toward Interpreting the Thermally Activated β Dynamics in Metallic Glass Using the Structural Constraint Neural Network

Simulation and Experimental Study on the Precision Molding of Irregular Vehicle Glass Components

Artificial Neural Network Modeling for Predicting Melt Blowing Processing

Deformation Analysis of the Glass Preform in the Progress of Precision Glass Molding for Fabricating Chalcogenide Glass Diffractive Optics with the Finite Element Method

Pre-Compensation of Mold in Precision Glass Molding Based on Mathematical Analysis

GlassNet: a multitask deep neural network for predicting many glass properties

Precision glass molding: Toward an optimal fabrication of optical lenses

Designing optical glasses by machine learning coupled with a genetic algorithm

A simulation-based optimization approach for the heating process of precision glass molding