Abstract Despite the winemaker’s efforts, deviations such as bacterial spoilage can occur during wine alcoholic fermentation resulting in economic losses and low quality wines. When a deviation is suspected, samples are usually sent to an oenological laboratory for the off-line analysis of specific quality control parameters. The use of ATR-MIR as a fast analytical tool to monitor the fermentation process could be very useful, as getting real-time information of the process allows making readjustments before the process ends. In this study, we aimed at detecting white wine spoilage during alcoholic fermentation due to the action of lactic bacteria using a portable ATR-MIR instrument and MSPC charts. A total of 33 small-scale alcoholic fermentations were conducted (25 in normal operation conditions (NOC) and 8 simulating a bacterial spoilage with the addition of lactic bacteria (MLF)) to evaluate the capability of the MSPC charts to detect deviations from NOC. MSPC control charts were developed based on Q residuals and Hotelling’s T2 statistics. Time-wise unfolding was applied to the original three-way data to build different PCA models, obtaining very satisfactory results: MLF samples were detected before the end of alcoholic fermentation in the Q residuals charts after 80 hours and Hotelling T2 chart could also differentiate the samples after 100 hours.

中文翻译：

---

使用 ATR-MIR 光谱仪和 MSPC 图表监测葡萄酒发酵偏差

摘要 尽管酿酒师付出了努力，但葡萄酒酒精发酵过程中仍会出现细菌腐败等偏差，造成经济损失和劣质葡萄酒。当怀疑存在偏差时，样品通常会被送到酿酒实验室进行特定质量控制参数的离线分析。使用 ATR-MIR 作为监测发酵过程的快速分析工具可能非常有用，因为获取过程的实时信息允许在过程结束前进行重新调整。在本研究中，我们旨在使用便携式 ATR-MIR 仪器和 MSPC 图表检测酒精发酵过程中由于乳酸菌的作用而导致的白葡萄酒腐败。总共进行了 33 次小规模酒精发酵（25 次在正常操作条件 (NOC) 下进行，8 次模拟添加乳酸菌 (MLF) 的细菌腐败），以评估 MSPC 图表检测与 NOC 偏差的能力。MSPC 控制图是基于 Q 残差和 Hotelling 的 T2 统计数据开发的。对原始三向数据进行时间展开来构建不同的PCA模型，得到了非常满意的结果：80小时后Q残差图中酒精发酵结束前检测到MLF样品，Hotelling T2图也可以区分100 小时后的样品。MSPC 控制图是基于 Q 残差和 Hotelling 的 T2 统计数据开发的。对原始三向数据进行时间展开来构建不同的PCA模型，得到了非常满意的结果：80小时后Q残差图中酒精发酵结束前检测到MLF样品，Hotelling T2图也可以区分100 小时后的样品。MSPC 控制图是基于 Q 残差和 Hotelling 的 T2 统计数据开发的。对原始三向数据进行时间展开来构建不同的PCA模型，得到了非常满意的结果：80小时后Q残差图中酒精发酵结束前检测到MLF样品，Hotelling T2图也可以区分100 小时后的样品。