Scholarlink

<p>Total polar compounds (TPC) are important indicator for evaluating the quality of frying oils. Low-field nuclear magnetic resonance (LF-NMR) technology, known for its rapid and non-destructive characteristics, has emerged as an ideal method for TPC detection. However, the existing models based on this technique often suffer from poor generalizability. In this study, the variation patterns between LF-NMR signals and TPC during the frying process of soybean oil were systematically investigated, along with the signal similarity among different soybean oils, to evaluate the feasibility of constructing a universal TPC prediction model based on LF-NMR technology. The analysis revealed that with the prolonging of frying time, both TPC content and the T<sub>2</sub> signal attenuation rate increased significantly. A strong correlation was observed between LF-NMR signals and TPC content. The Jaccard similarity coefficient indicated high similarity in LF-NMR signals existed in oil grades, with coefficients exceeding 0.8297. Notably, the similarity of T<sub>2</sub> relaxation curve data surpassed that of conventional relaxation parameters. A generalized prediction model was ultimately developed using the 0–200&#xa0;ms relaxation interval data, combined with orthogonal signal correction (OSC) and support vector regression (SVR). This model presented the best performance, with RMSECV of 1.2860, 1.3150, and 1.3223 for first-grade soybean oil (FG-SFO), third-grade soybean oil (TG-SFO), and their mixture, respectively. The average relative error was below 6%, significantly outperforming traditional partial least squares (PLS) regression models. The proposed universal model enables the determination of polar compound content in different types of soybean oil, reducing the need for repeated model construction and thereby minimizing labor, material, and time costs. It offers an efficient and rapid monitoring method for frying oil quality, with substantial practical value for ensuring the safety of fried foods.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Construction of a Universal Model for Total Polar Compounds in Fried Soybean Oil Based on Low-Field NMR

  • 期刊论文

  • 发表日期：

    2026年2月5日

查看全文
  • Guangjie Zhang,
  • Shaoshuai Zheng,
  • Yunxia Xue,
  • Qianwen Zhang,
  • Jingnan Chen,
  • Dan Peng

摘要

Total polar compounds (TPC) are important indicator for evaluating the quality of frying oils. Low-field nuclear magnetic resonance (LF-NMR) technology, known for its rapid and non-destructive characteristics, has emerged as an ideal method for TPC detection. However, the existing models based on this technique often suffer from poor generalizability. In this study, the variation patterns between LF-NMR signals and TPC during the frying process of soybean oil were systematically investigated, along with the signal similarity among different soybean oils, to evaluate the feasibility of constructing a universal TPC prediction model based on LF-NMR technology. The analysis revealed that with the prolonging of frying time, both TPC content and the T2 signal attenuation rate increased significantly. A strong correlation was observed between LF-NMR signals and TPC content. The Jaccard similarity coefficient indicated high similarity in LF-NMR signals existed in oil grades, with coefficients exceeding 0.8297. Notably, the similarity of T2 relaxation curve data surpassed that of conventional relaxation parameters. A generalized prediction model was ultimately developed using the 0–200 ms relaxation interval data, combined with orthogonal signal correction (OSC) and support vector regression (SVR). This model presented the best performance, with RMSECV of 1.2860, 1.3150, and 1.3223 for first-grade soybean oil (FG-SFO), third-grade soybean oil (TG-SFO), and their mixture, respectively. The average relative error was below 6%, significantly outperforming traditional partial least squares (PLS) regression models. The proposed universal model enables the determination of polar compound content in different types of soybean oil, reducing the need for repeated model construction and thereby minimizing labor, material, and time costs. It offers an efficient and rapid monitoring method for frying oil quality, with substantial practical value for ensuring the safety of fried foods.