Model Interpretability, Explainability and Trust for Manufacturing 4.0

Manufacturing is currently characterized by a widespread availability of multiple streams of big data (e.g., signals, images, video-images, 3-dimensional voxel and mesh-based reconstructions of volumes and surfaces). Manufacturing 4.0 refers to the paradigm shift involving appropriate use of all this rich data environment for decision making in prognostic, monitoring, optimization and control of the manufacturing processes. The paper discusses how the new advent of Artificial Intelligence for manufacturing data mining poses new challenges on model interpretability, explainability and trust. Starting from this general overview, the paper then focuses on examples of big data mining in Additive Manufacturing. A real case study focusing on spatter modeling for process optimization is discussed, where a solution based on robust functional analysis of variance is proposed.