Bidirectional Process Prediction in the Laser‐Induced‐Graphene Production Using Blackbox Deep Learning

ABSTRACT Machine learning offers significant potential for automating manufacturing processes through prediction of parameters and properties. However, its adoption is limited by the needs for process‐specific expertise, complex analytical models, and large, high‐quality training datasets. This study introduces a streamlined, non‐analytical blackbox deep‐learning approach tailored to resource‐ and knowledge‐limited settings. Applied to laser‐induced graphene, an emerging technique for fabricating flexible electronics, this method uses a simple neural network for bidirectional predictions: forward estimation of graphene properties from laser parameters, and inverse prediction of laser settings from target properties. Using only a small dataset of 500 moderate‐quality, high‐variance samples, the model attained joint prediction accuracies of up to 95% on validation samples, and 96% in a functional flexible circuit. These results match or surpass more elaborate physics‐informed or multi‐component models in the literature, with minimal resources and no preprocessing. The method is widely applicable to various functional materials and fabrication techniques, minimizing dependence on prior knowledge, manual optimization, and specialized equipment. This demonstrates that high predictive performance can emerge from minimal overhead, making machine learning a practical, accessible instrument that democratizes intelligent manufacturing in graphene‐based and related systems, enabling efficient and scalable real‐world innovation.