CC BY 4.0Youssef, ShahendaShahendaYoussef2025-08-062025-08-062025https://publica.fraunhofer.de/handle/publica/490241https://doi.org/10.24406/publica-502110.24406/publica-5021Causality-driven AI focuses on enabling interpretable, robust, and actionable decision-making, with a specific focus on manufacturing systems. Existing methods, including Structural Causal Models and Propensity Score Matching, have demonstrated significant applications in process optimization and fault detection. Despite significant advancements, these methods often face limitations with seamlessly incorporating domain knowledge, efficiently handling high-dimensional and heterogeneous data, and leveraging causality for scalable deep learning architectures. This work highlights these gaps and proposes a unified direction to enhance causality in deep learning by integrating causal inference methods that embed causal priors into neural networks, optimizing causal discovery algorithms, and improving both the performance and interpretability of deep learning models through causal attributions and loss function optimization.enconference paper