Abstract Understanding yeast dynamics during fermentation is important for quality control, whether monitoring fermentation consistency or identifying aberrant events, such as premature yeast flocculation (PYF). Previous models of fermentation dynamics tend to be parameter rich and require large time series, which are rare in industry. This research investigates five simpler models to 1) describe fermentation dynamics, 2) refine quality control sampling regimes to improve model fit, and 3) identify PYF fermentations. The ability of these models to describe yeast dynamics was evaluated using model fitting with time series data and Akaike Information Criterion (AIC) model selection. Data simulated from large time series was used with this model fitting approach to improve sampling schedules without increasing sampling effort. Lastly, PYF was identified in fermentations of fungal-contaminated malt using linear discriminant analysis (LDA). For large data sets, a four-parameter extension of the normal curve performed best while smaller data sets were better described by the 2-parameter gamma model. Moving sampling effort nearer the population peak improved model fits. Lastly, all models detected PYF, however the two-parameter gamma model provided a simple metric for distinguishing PYF. This research provides guidelines on appropriate model use, improving sampling regimes, and identifying PYF.

中文翻译：

---

在实验室和商业发酵过程中对悬浮液中的酵母进行建模以检测异常发酵过程

摘要 了解发酵过程中的酵母动态对于质量控制很重要，无论是监测发酵一致性还是识别异常事件，如酵母过早絮凝 (PYF)。以前的发酵动力学模型往往参数丰富，需要大的时间序列，这在行业中很少见。本研究调查了五个更简单的模型，以 1) 描述发酵动力学，2) 改进质量控制采样制度以改进模型拟合，以及 3) 识别 PYF 发酵。使用时间序列数据的模型拟合和赤池信息准则 (AIC) 模型选择来评估这些模型描述酵母动力学的能力。从大型时间序列模拟的数据与这种模型拟合方法一起使用，以在不增加采样工作的情况下改进采样计划。最后，使用线性判别分析 (LDA) 在受真菌污染的麦芽发酵中鉴定 PYF。对于大型数据集，正态曲线的四参数扩展表现最佳，而较小的数据集则由 2 参数伽玛模型更好地描述。将抽样工作移近总体峰值改进模型拟合。最后，所有模型都检测到 PYF，但是双参数伽马模型提供了一个简单的度量来区分 PYF。这项研究提供了有关适当模型使用、改进抽样制度和识别 PYF 的指南。将抽样工作移近总体峰值改进模型拟合。最后，所有模型都检测到 PYF，但是双参数伽马模型提供了一个简单的度量来区分 PYF。这项研究提供了有关适当模型使用、改进抽样制度和识别 PYF 的指南。将抽样工作移近总体峰值改进模型拟合。最后，所有模型都检测到 PYF，但是双参数伽马模型提供了一个简单的度量来区分 PYF。本研究提供了有关适当模型使用、改进抽样制度和识别 PYF 的指南。