The paucity of readily available medical data poses a major challenge for the development of AI (artificial intelligence)-based healthcare applications and devices. To aid in overcoming this challenge, we propose a sensor-based medical time series data synthesis system especially designed for the training of diabetic foot diagnosis models. The proposed system utilizes statistical methods, augmentation techniques, and the NeuralProphet model to accomplish its purpose while still maintaining medical validity. Our results show that the generated synthetic time series data follow the trends and tendencies of real data. We also verify our work using machine learning-based clustering. By successfully clustering the synthetic data generated by our proposed system, we prove that our system is capable of meeting its objectives.