Pilot production lines bridge the gap between increasingly rapid new material developments and mass production in battery cell manufacturing. They allow to develop advanced and specifically adapted process technologies and demonstrate the transferability of laboratory research results to the industrial level. With the increased use of machine learning methods in digital twins, particularly for process monitoring and quality control, the question arises to which extent machine learning tools can be developed, trained and tested on pilot production lines. Since the output of pilot facilities is typically limited by a small production capacity, dataset generation is particularly constrained. In our work we examine the implications of constrained resources for dataset generation on the applicability of machine learning tools. We point out the relevance of research findings on machine learning in the small data regime for pilot battery cell manufacturing and discuss the applicability of recently suggested frameworks for the development of machine learning in the small data regime for pilot battery cell manufacturing lines. Starting from computer experiments on small data analysis in exemplary models [2] we employ design of experiments (DOE) strategies for dataset generation in pilot production. The applicability of knowledge-driven feature engineering and other methods for dimensionality reduction of original data to align the machine learning models with the available training data volume are examined. Machine learning based approaches can be applied in the small data regime for the development of virtual quality gates (VQG), which we illustrate for the pilot production line housed at the Fraunhofer Institute for Silicate Research (ISC) in Germany.