School of Chemical Engineering, University of Birmingham, Birmingham, UK.
School of Physics and Astronomy, University of Birmingham, Birmingham, UK.
Sci Rep. 2022 Nov 14;12(1):19535. doi: 10.1038/s41598-022-24022-x.
Simulating the response of a radiation detector is a modelling challenge due to the stochastic nature of radiation, often complex geometries, and multi-stage signal processing. While sophisticated tools for Monte Carlo simulation have been developed for radiation transport, emulating signal processing and data loss must be accomplished using a simplified model of the electronics called the digitizer. Due to a large number of free parameters, calibrating a digitizer quickly becomes an optimisation problem. To address this, we propose a novel technique by which evolutionary algorithms calibrate a digitizer autonomously. We demonstrate this by calibrating six free parameters in a digitizer model for the ADAC Forte. The accuracy of solutions is quantified via a cost function measuring the absolute percent difference between simulated and experimental coincidence count rates across a robust characterisation data set, including three detector configurations and a range of source activities. Ultimately, this calibration produces a count rate response with 5.8% mean difference to the experiment, improving from 18.3% difference when manually calibrated. Using evolutionary algorithms for model calibration is a notable advancement because this method is novel, autonomous, fault-tolerant, and achieved through a direct comparison of simulation to reality. The software used in this work has been made freely available through a GitHub repository.
模拟辐射探测器的响应是一个建模挑战，因为辐射具有随机性，通常具有复杂的几何形状和多阶段的信号处理。虽然已经开发出用于辐射传输的蒙特卡罗模拟的复杂工具，但必须使用称为数字化仪的电子设备的简化模型来模拟信号处理和数据丢失。由于有大量的自由参数，因此校准数字化仪很快就成为一个优化问题。为了解决这个问题，我们提出了一种新的技术，通过进化算法自主校准数字化仪。我们通过校准 ADAC Forte 数字化仪模型中的六个自由参数来证明这一点。通过使用测量模拟和实验符合计数率之间绝对百分比差异的成本函数来量化解决方案的准确性，该函数跨越了一个稳健的特征数据集，包括三种探测器配置和一系列源活动。最终，这种校准产生的计数率响应与实验相比平均差异为 5.8%，而手动校准的差异为 18.3%。使用进化算法进行模型校准是一个显著的进步，因为这种方法是新颖的、自主的、容错的，并且通过模拟与现实的直接比较来实现。这项工作中使用的软件已经通过 GitHub 存储库免费提供。
