Radioactive particle tracking (RPT) is one of the most widely used non‐intrusive velocimetry technique for multiphase reactors. The large volume of interrogation and the presence of internals limit the application of RPT in large‐scale real‐world systems. The main challenge lies in having fast reconstruction algorithms applicable to conventional (i.e., bubble columns, fluidized beds, etc.) as well as new vessels. In this contribution, a reconstruction methodology is proposed based on machine learning. Three machine‐learning algorithms, namely, artificial neural network (ANN), support vector regression (SVR), and relevance vector regression (RVR), have been employed for RPT reconstruction. The results show that the position reconstruction accuracy of SVR was best for all cases and that the accuracy of RVR was comparable to SVR for large training datasets. Whereas, in terms of reconstruction speed, RVR outperforms SVR significantly, owing to sparser RVR model. SVR and RVR based reconstruction algorithms expedite the position reconstruction.

中文翻译：

---

基于机器学习的放射性粒子跟踪实验的位置绘制算法

放射性粒子跟踪（RPT）是用于多相反应堆的最广泛使用的非侵入式测速技术之一。大量的询问和内部构件的存在限制了RPT在大规模实际系统中的应用。主要挑战在于拥有适用于常规（即气泡塔，流化床等）以及新容器的快速重建算法。在此贡献中，提出了一种基于机器学习的重构方法。RPT重建采用了三种机器学习算法，即人工神经网络（ANN），支持向量回归（SVR）和相关向量回归（RVR）。结果表明，在所有情况下，SVR的位置重建精度均最佳，对于大型训练数据集，RVR的精度可与SVR媲美。而在重建速度方面，由于RVR模型较稀疏，RVR的性能明显优于SVR。基于SVR和RVR的重建算法可加快位置重建。