A cigarette quality stability evaluation method and system based on a multivariate statistical process control-sensory evaluation fusion

Abstract

The application belongs to the field of cigarette quality stability evaluation, and discloses a cigarette quality stability evaluation method and system based on multivariate process statistics-sensory evaluation fusion. 2 T 2 The method takes three cigarette models, ML, XP and YY, as research objects, fuses multivariate process statistics and sensory evaluation, measures 25 chemical indexes such as conventional chemical components and alkaloids, constructs four derived indexes, and realizes multivariate collaborative fluctuation monitoring by means of Hotelling T The result shows that the conventional derived indexes and polyphenol indexes are key quality factors affecting the sensory quality stability of cigarettes, and the conventional components, alkaloids, multivariate acids and high-grade fatty acids are other quality factors; based on the difference in sample size, the application scenarios are divided, the statistical stability, early warning and risk interval of the key quality factors and other quality factors are determined, and quantitative basis and technical support are provided for the fine quality control of the whole process of cigarette production.

Description

Technical Field

[0001] The present invention belongs to, but is not limited to, the technical field of cigarette quality stability evaluation, and particularly relates to a cigarette quality stability evaluation method and system based on multivariate statistical process control-sensory evaluation. Background Technique

[0002] Cigarette quality stability is the core index for the production control of the tobacco industry, which directly determines the consistency of product sensory quality, brand reputation and market competitiveness. It is also the key prerequisite for the compliance of cigarette products to leave the factory and occupy the market for a long time. The tobacco industry in China is huge, with the total industry output value exceeding 1.5 trillion yuan for many consecutive years, and the tobacco tax accounting for more than 6% of the national fiscal revenue. The huge industrial scale puts extremely high requirements on the control of cigarette quality stability. In actual production, multiple factors such as the differences in the origin and year of tobacco leaf raw materials, the fine-tuning of silk-making process parameters, and the fluctuations in the working conditions of processing equipment will all cause fluctuations in cigarette chemical components, and then affect the sensory quality stability. Even for qualified finished cigarettes, the slight quality differences between different batches will reduce the consumer experience. Therefore, constructing a scientific, accurate and implementable cigarette quality stability evaluation method has become a technical issue that亟待解决的行业亟待解决的技术课题 in the industry.

[0003] At present, the evaluation of cigarette quality stability mainly relies on two technical paths: the detection of conventional chemical components and flue gas signature components of cigarettes and sensory quality evaluation. The industry has formed a relatively complete standard system, and special detection standards have been formulated for core chemical components such as total water-soluble sugar, total plant alkaloids, polyphenols, flue gas tar, flue gas nicotine, CO, etc. and key flue gas indicators. Sensory evaluation is based on YC / T138-1998 "Methods for Sensory Evaluation of Tobacco and Tobacco Products" as the core basis. However, the existing research and control means still have obvious shortcomings: First, traditional evaluations mostly focus on the fluctuations of single chemical component indicators, and use univariate control charts such as mean-standard deviation charts and mean-range charts for monitoring. They do not consider the synergistic coupling relationship between multiple chemical components, and cannot identify the problem of abnormal multi-index synergy. It is easy to have control loopholes where single indicators are qualified but the overall quality is unbalanced, and it is difficult to comprehensively characterize the internal quality stability of cigarettes; Second, although sensory evaluation is the core terminal means for judging cigarette quality, most existing research directly conducts sensory scoring evaluations on finished products, and does not quantify the response relationship between the fluctuation range of chemical components and the variation of sensory quality. It is difficult to accurately screen and clarify the key chemical indicators that cause sensory quality fluctuations, resulting in a lack of targeting in quality control; Third, most existing research focuses on the independent analysis of chemical components or sensory quality, lacking a key index screening and stability warning model constructed from the perspective of the association between chemical index fluctuations-sensory quality variation, and cannot provide technical support for the precise control of cigarette quality; Fourth, most research only conducts experiments on a single cigarette brand, and the research results are easily interfered by the characteristics of the brand, with insufficient universality and reference value.

[0004] To address the aforementioned industry pain points, multivariate statistical process control (MSPC) technology has been gradually applied to the field of cigarette quality control, among which Hotelling T... 2 Control charts, as a core tool for multivariate monitoring, can be used to monitor variables. 2 The statistical measure comprehensively reflects the degree of coordinated deviation among multiple indicators, effectively compensating for the shortcomings of univariate control charts. It has been widely applied and validated in multivariate quality control in fields such as food and chemicals. This invention innovatively incorporates Hotelling T... 2 This study integrates multivariate statistical process control with a three-point sensory evaluation method, selecting three different brands of cigarettes from Tobacco Company A as parallel research subjects. First, multi-dimensional chemical components were systematically measured to construct derived characteristic indicators for deeper analysis of component chemical information. Second, a controlled variable experimental design was used to screen samples with typical differences for testing. Furthermore, the results of the three-point sensory evaluation were used to determine the core key quality influencing factors of the cigarettes, and finally, the reasonable fluctuation control range of various quality factors was quantitatively defined. The aim is to construct a quality stability evaluation method that is adaptable to actual cigarette production, highly accurate, and universally applicable, thus overcoming the shortcomings of traditional evaluation methods and providing theoretical support and technical solutions for refined, digital, and quantitative control of the entire cigarette production process.

[0005] Based on the above analysis, the urgent technical problems that need to be solved in the existing technology are: Traditional univariate control charts fail to capture the intrinsic relationships between multiple components, making it impossible to identify synergistic anomalies among multiple indicators and thus difficult to comprehensively characterize the overall quality fluctuations of cigarettes. Furthermore, the lack of clarity regarding the correlation between chemical indicators and sensory quality presents an industry-wide challenge of quantitative correspondence. Summary of the Invention

[0006] To address the problems existing in the prior art, this invention provides a method and system for evaluating the stability of cigarette quality based on multivariate process statistics and sensory evaluation fusion.

[0007] This invention is implemented as follows: a method for evaluating the stability of cigarette quality based on the synergy of multivariate statistics and sensory evaluation, comprising the following steps: Step 1: Preparation of experimental samples and testing resources. Select multiple batches of cigarette samples under normal production conditions, confirm that the purity level of the reagents meets the requirements of chromatographic purity or analytical purity, and ensure that the testing instruments are calibrated and meet the accuracy requirements for detecting tobacco chemical components. Step two, chemical composition detection: conduct multi-category chemical composition detection on cigarette samples to obtain data on conventional chemical components, alkaloid compounds, polyphenolic compounds and organic acid compounds, and form a multi-index detection dataset; Step 3: Construction of derived indicators. Based on the results of conventional chemical component detection, derived indicators are calculated. The derived indicators include the ratio of total sugar to total alkaloids, the ratio of reducing sugar to total alkaloids, the ratio of reducing sugar to total sugar, and the ratio of total nitrogen to total alkaloids. Step 4: Multivariate fluctuation monitoring. Based on the multi-index detection dataset, a multivariate statistical monitoring model is established to perform synergistic fluctuation analysis on the quality indicators of cigarette samples. Step 5: Sensory evaluation and verification. The three-point test method is used to conduct sensory evaluation of cigarettes, and the results of sample difference identification are recorded through blind expert evaluation. Step 6: Screening of key quality factors. The results of multivariate statistical monitoring and sensory evaluation are correlated to identify key quality factors and other quality factors that affect the stability of cigarette quality and to establish their fluctuation control range.

[0008] Furthermore, the conventional chemical component detection includes the following indicators: Contents of water-soluble total sugars, reducing sugars, total alkaloids, total nitrogen, potassium, and chlorine.

[0009] This invention also provides a method for evaluating the stability of cigarette quality based on multivariate process statistics-sensory evaluation fusion, the method specifically including: S1: Selection of experimental materials. Three commercially available qualified finished cigarettes, namely ML, XP and YY, were selected. All reagents used in the experiment were of chromatographic grade or analytical grade. The water used in the experiment was ultrapure water. The detection instruments were calibrated and qualified, meeting the accuracy requirements for chemical component detection in the tobacco industry. S2: Chemical composition testing, which focuses on comprehensive testing of the core chemical components of cigarettes, covering four major categories: conventional chemical components, alkaloid compounds, polyphenolic compounds, and organic acid compounds, with a total of 25 indicators. All tests are implemented in accordance with the current industry standards. S3: Construction of derived indicators. Based on the results of conventional chemical component detection, four core derived indicators were constructed, namely total sugar / total alkaloids, reducing sugar / total alkaloids, reducing sugar / total sugar, and total nitrogen / total alkaloids. S4: Key Quality Factor Screening, using Hotelling T 2 Control charts enable multivariate coordinated fluctuation monitoring, and sensory evaluation verification is carried out in conjunction with the three-point test method. The system screens key quality factors and defines their fluctuation control range.

[0010] Furthermore, the routine chemical component testing strictly followed the tobacco industry's specific standards: water-soluble total sugar and reducing sugar according to YC / T 159-2019, total alkaloids according to YC / T 468-2021, total nitrogen according to YC / T 33-1996, potassium according to YC / T217-2007, and chlorine according to YC / T 162-2011, for a total of 6 routine chemical components.

[0011] Furthermore, the detection of alkaloid compounds is based on the YC / T 383-2010 standard method, with optimized experimental procedures to accurately determine four indicators: nicotine, nornicotine, pseudoequicein, and neonicotine.

[0012] Furthermore, the pretreatment and detection parameters for the polyphenolic compounds were optimized according to the YC / T 202-2006 standard, and five core polyphenol indicators, namely chlorogenic acid, neochlorogenic acid, cryptochlorogenic acid, rutin, and hyoscyamine, were determined.

[0013] Furthermore, the detection method for organic acid compounds is an improvement on the YC / T 288-2009 standard method, enabling the simultaneous detection of 10 organic acids: oxalic acid, malonic acid, succinic acid, malic acid, palmitic acid, myristic acid, citric acid, oleic acid, linoleic acid, and linolenic acid.

[0014] Furthermore, the determination of the key quality factors is carried out in accordance with the industry standard YC / T 138-1998. For each candidate quality factor, 27 professional sensory evaluations are conducted. The critical value of the number of experts who can accurately identify the differences in the samples is consulted in Appendix B1 of the standard (≥14 accurate identifications). If the number of experts who can identify the differences in the samples exceeds the critical value, the index is determined to be a key quality factor; otherwise, it is another quality factor that has no significant impact on the sensory quality stability of cigarettes. The sensory quality evaluation of the cigarettes involved nine professionals with extensive experience in tobacco product development and process research. All of them met the expert qualification requirements for sensory evaluation in section 4.1 of YC / T 138-1998, possessing solid smoking experience and accurate sensory discrimination capabilities. The evaluation experiments were strictly conducted according to the three-point test method in "Sensory Evaluation Methods for Tobacco and Tobacco Products" (YC / T 138-1998). Each group of samples included two normal samples (with small deviations in indicators) and one abnormal sample (with large deviations in indicators). Experts judged the differences between the samples through blind evaluation, and the number of times experts identified differences was accurately recorded. The evaluation process employed hypothesis testing for statistical determination: the null hypothesis was that there was no sensory difference between the two samples, with a correct identification probability of 1 / 3 (random guess level); the alternative hypothesis was that there was a sensory difference between the two samples, with a correct identification probability > 1 / 3, and a significance level of α = 0.05. The actual number of correct identifications was compared with the standard critical value. If the actual number of correct identifications did not reach the critical value, the null hypothesis could not be rejected, indicating that the difference in indicators did not affect sensory quality; if it exceeded the critical value, the null hypothesis was rejected, indicating that the difference in indicators had a significant impact on sensory quality.

[0015] Furthermore, the Hotelling T 2 The steps for constructing a control chart are as follows: (1) Construct a training set, collect multiple batches of cigarette samples under normal production conditions, measure p indicators, form an n×p data matrix, calculate the mean vector μ0 and covariance matrix S of the training set, and establish the production baseline state. (2) Calculate T 2 The statistic, with the formula T 2 =(x-μ0)ᵀS -1 (x-μ0), where x is the sample vector to be tested, which comprehensively reflects the degree of coordinated deviation of multiple indicators; (3) Calculate the upper control limit (UCL). When the sample size n < 50, the control limit is derived using the F distribution. The formula is UCL = [(n-1)p / (np)]F a When the sample size n > 50, the control limits are derived using the chi-square distribution, with the formula UCL = χ². 2 a p, α is 0.05.

[0016] Another objective of this invention is to provide a cigarette quality stability evaluation system based on multivariate process statistical-sensory evaluation fusion, the system specifically comprising: The experimental resource management module is used to manage cigarette sample information, verify the compliance of reagents and consumables, and confirm the calibration status of testing instruments. The chemical composition detection module is used to preset the detection standards for 25 indicators in four major categories, perform standardized testing, and store data for quality control. The indicator data processing module is used to retrieve conventional chemical composition data, calculate four derived indicators, and generate a complete dataset. Multivariate fluctuation monitoring module for integration with Hotelling T 2 Control chart models are used to visualize and monitor the coordinated fluctuations of multiple indicators. The sensory evaluation correlation module is used to manage the three-point test process, input evaluation results, and correlate and analyze statistical data. The key quality factor management module is used to screen key quality factors and automatically define their reasonable fluctuation control range; The evaluation report generation module is used to summarize key information and generate a standardized control report on the quality stability of cigarettes.

[0017] Based on the above technical solutions and the technical problems solved, the advantages and positive effects of the technical solution to be protected by this invention are as follows: Compared with the prior art, the technical essence of the present invention lies in the following: On the one hand, in addition to the detection of basic chemical components, a system of derived indicators with mechanistic significance, such as sugar-alkali ratio and sugar-nitrogen ratio, is further constructed. Although the prior art reveals that these indicators are related to sensory perception, they have not been systematically introduced as core parameters for stability evaluation. On the other hand, by establishing a multivariate statistical monitoring model, multi-indicator synergistic fluctuation analysis is achieved, and the sensory evaluation results of the three-point test method are further introduced for cross-validation and feedback, forming a closed-loop evaluation mechanism of data-driven + sensory verification.

[0018] Therefore, this method not only overcomes the problem of the independence between chemical analysis and sensory evaluation in existing technologies, but also realizes the fusion judgment of multi-source information and the accurate screening of key quality factors, thus upgrading the evaluation of cigarette quality stability from a single statistical analysis to a multi-dimensional collaborative decision-making model, which has significant technological progress and remarkable creativity.

[0019] This invention takes qualified finished cigarettes of three brands, ML, YY, and XP, as the research object and uses Hotelling T... 2 A fusion analysis method combining control charts and the three-point test was used to systematically screen key quality factors affecting the stability of cigarette quality.

[0020] The results showed that Hotelling T contained conventional components, alkaloids, polyphenols, polybasic acids, and higher fatty acids. 2 Control charts showed that all samples remained within the F-distribution control limits, with only a few slightly exceeding the chi-square control limits, and the overall content of various components remained stable. To further investigate the impact of key quality factors on sensory quality, the project team designed an experiment using the controlled variable method, focusing on the T-value among the above-mentioned categories of indicators. 2 For samples with significant deviations in statistical values, sensory evaluation was conducted according to the three-point test requirements in the "Sensory Evaluation Methods for Tobacco and Tobacco Products" (YC / T138-1998). Each variable had 27 participants. Referring to Clause 5.3.5 of the standard and Appendix Table B1, the evaluation critical value was ≥14 correct identifications. For the two key quality factors of conventional derived polyphenols, the number of correct identifications was >14, indicating that samples with significant differences in these two factors also showed significant differences in sensory quality. For samples with significant deviations selected based on other variables, the number of correct identifications in the three-point test sensory quality evaluation was less than 14, and the T-values ​​for all samples were... 2 Since the deviations of the statistics did not exceed the control limits of the F distribution, the conventional components, alkaloids, polybasic acids, and higher fatty acid components were identified as other quality factors.

[0021] Finally, based on Hotelling T 2Based on the principle of control charts, two application scenarios were set up according to the training set sample size, and the key quality factors and non-key quality factors T were calculated and determined respectively. 2 The range of statistical fluctuation control provides quantitative support for the production management and control of cigarette products. Attached Figure Description

[0022] Figure 1 This is a flowchart of the cigarette quality stability evaluation method based on multivariate process statistics-sensory evaluation fusion provided in the embodiments of the present invention; Figure 2 This is a block diagram of a cigarette quality stability evaluation system based on multivariate process statistics-sensory evaluation fusion provided in an embodiment of the present invention; Figure 3 The conventional component T provided in the embodiments of the present invention 2 Correlation diagram (ML) between control charts and three-point test evaluation results; Figure 4 The conventional component T provided in the embodiments of the present invention 2 Correlation diagram between control charts and three-point test evaluation results (YY); Figure 5 The conventional component T provided in the embodiments of the present invention 2 Correlation diagram between control charts and three-point test evaluation results (XP); Figure 6 The conventional derived index T provided in this embodiment of the invention. 2 Correlation diagram (ML) between control charts and three-point test evaluation results; Figure 7 The conventional derived index T provided in this embodiment of the invention. 2 Correlation diagram between control charts and three-point test evaluation results (YY); Figure 8 This is a conventional derived T provided in the embodiments of the present invention. 2 Correlation diagram between control charts and three-point test evaluation results (XP); Figure 9 The alkaloid T provided in the embodiments of the present invention 2 Correlation diagram (ML) between control charts and three-point test evaluation results; Figure 10 The alkaloid T provided in the embodiments of the present invention 2 Correlation diagram between control charts and three-point test evaluation results (YY); Figure 11 The alkaloid T provided in the embodiments of the present invention 2 Correlation diagram between control charts and three-point test evaluation results (XP); Figure 12 The polyphenol T provided in the embodiments of the present invention 2Correlation diagram (ML) between control charts and three-point test evaluation results; Figure 13 The polyphenol T provided in the embodiments of the present invention 2 Correlation diagram between control charts and three-point test evaluation results (YY); Figure 14 The polyphenol T provided in the embodiments of the present invention 2 Correlation diagram between control charts and three-point test evaluation results (XP); Figure 15 The T polyacids provided in the embodiments of the present invention are... 2 Correlation diagram (ML) between control charts and three-point test evaluation results; Figure 16 The T polyacids provided in the embodiments of the present invention are... 2 Correlation diagram between control charts and three-point test evaluation results (YY); Figure 17 The T polyacids provided in the embodiments of the present invention are... 2 Correlation diagram between control charts and three-point test evaluation results (XP); Figure 18 The higher fatty acid class T provided in the embodiments of the present invention 2 Correlation diagram (ML) between control charts and three-point test evaluation results; Figure 19 The higher fatty acid class T provided in the embodiments of the present invention 2 Correlation diagram between control charts and three-point test evaluation results (YY); Figure 20 The higher fatty acid class T provided in the embodiments of the present invention 2 Correlation diagram between control charts and three-point test evaluation results (XP). Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0024] like Figure 1 As shown in the embodiment of the present invention, the method for evaluating the stability of cigarette quality based on multivariate process statistics-sensory evaluation fusion specifically includes: S1: Selection of Experimental Materials. Three commercially available qualified finished cigarette brands—ML, XP, and YY—were selected to cover different styles, positioning, and formulation characteristics, avoiding experimental bias caused by a single brand and ensuring the representativeness and reliability of the research results. All reagents used in the experiment were of chromatographic or analytical grade, and the experimental water was ultrapure water. The testing instruments were calibrated and met the accuracy requirements for chemical component detection in the tobacco industry.

[0025] S2: Chemical composition testing, which involves comprehensive testing of the core chemical components of cigarettes, covering four major categories: conventional chemical components, alkaloid compounds, polyphenolic compounds, and organic acid compounds, with a total of 25 indicators. All tests are implemented in accordance with current industry standards to ensure the standardization and accuracy of the test results.

[0026] S3: Construction of derived indicators. A single conventional chemical component can only reflect local chemical characteristics and cannot reflect the synergistic effect between indicators. In order to deeply explore the quality information contained in chemical components, four core derived indicators were constructed based on the detection results of conventional chemical components, namely total sugar / total alkaloids, reducing sugar / total alkaloids, reducing sugar / total sugar, and total nitrogen / total alkaloids.

[0027] S4: Key Quality Factor Screening, using Hotelling T 2 Control charts enable multivariate coordinated fluctuation monitoring, and sensory evaluation verification is carried out in conjunction with the three-point test method. The system screens key quality factors and defines their fluctuation control range.

[0028] The routine chemical component testing strictly followed the tobacco industry's specific standards: water-soluble total sugar and reducing sugar according to YC / T 159-2019, total alkaloids according to YC / T 468-2021, total nitrogen according to YC / T 33-1996, potassium according to YC / T 217-2007, and chlorine according to YC / T 162-2011, for a total of 6 routine chemical components.

[0029] The alkaloid compound detection method is based on the YC / T 383-2010 standard method, with optimized experimental procedures to accurately determine four indicators: nicotine, nornicotine, pseudoesequine, and neonicotine. The optimized method improves the detection sensitivity of trace alkaloids and is suitable for the detection needs of complex matrices in cigarette samples.

[0030] The detection of polyphenolic compounds was carried out in accordance with the YC / T 202-2006 standard, which optimized the pretreatment and detection parameters. Five core polyphenol indicators, namely chlorogenic acid, neochlorogenic acid, cryptochlorogenic acid, rutin, and hyoscyamine, were determined, which effectively reduced matrix interference and improved the detection accuracy.

[0031] The method for detecting organic acid compounds is an improvement on the standard method of YC / T 288-2009, enabling the simultaneous detection of 10 organic acids, including oxalic acid, malonic acid, succinic acid, malic acid, palmitic acid, myristic acid, citric acid, oleic acid, linoleic acid, and linolenic acid. This method covers polybasic acids and higher fatty acids, thus improving the characterization system of cigarette chemical components.

[0032] The derived indicators focus on the core chemical ratios of cigarettes, which can more intuitively reflect the coordination of chemical components. They are key parameters for characterizing the intrinsic quality of cigarettes, making up for the limitations of single indicator characterization and providing richer data support for subsequent multivariate statistical analysis.

[0033] The key quality factor screening method uses a combination of the control variable method and Hotelling T. 2 The control chart method was used to screen key quality factors. The core logic was to explore the impact of deviations in a single category of chemical component on the sensory quality stability of cigarettes, while controlling for fluctuations in other categories of chemical components, thus identifying the core influencing factors. The study defined six candidate categories of key quality factors, including conventional chemical components, conventional derived indicators, alkaloids, polybasic acids, higher fatty acids, and polyphenols, covering all detection indicators and derived information.

[0034] Sample screening strictly follows the matching principle: relying on Hotelling T 2 For the control chart, paired samples were selected from those where a certain category of chemical component indicators deviated significantly from the training set mean, while the deviations from other categories of chemical component indicators were not significant, thus meeting the experimental requirements of the three-point test. To improve experimental reliability, for each candidate quality factor, three parallel samples corresponding to three cigarette brands were set up to conduct parallel experiments with multiple brands, eliminating the specific interference of a single brand.

[0035] The determination of key quality factors is carried out in accordance with the industry standard YC / T 138-1998. For each candidate quality factor, 27 professional sensory evaluations are conducted. The critical value of the corresponding number of test participants is consulted in Appendix B1 of the standard. If the number of experts who can identify the differences in the samples exceeds the critical value, the index is determined to be a key quality factor; otherwise, it is another quality factor that has no significant impact on the sensory quality stability of cigarettes.

[0036] The sensory quality evaluation of the cigarettes involved nine professionals with extensive experience in tobacco product development and process research. All nine met the expert qualification requirements for sensory evaluation as outlined in Section 4.1 of YC / T 138-1998, possessing solid smoking experience and accurate sensory judgment capabilities. The evaluation experiments were strictly conducted according to the three-point testing method in the "Sensory Evaluation Methods for Tobacco and Tobacco Products" (YC / T 138-1998). Each sample group consisted of two normal samples and one abnormal sample. Experts assessed the differences between the samples through blind evaluation, accurately recording the number of times each expert identified a difference.

[0037] The evaluation process employed hypothesis testing for statistical determination: the null hypothesis was that there was no sensory difference between the two samples, with a correct identification probability of 1 / 3; the alternative hypothesis was that there was a sensory difference between the two samples, with a correct identification probability greater than 1 / 3, and a significance level of α = 0.05. By comparing the actual number of correct identifications with the standard critical value, if the actual number of correct identifications did not reach the critical value, the null hypothesis could not be rejected, indicating that the difference in the indicator did not affect sensory quality; if it exceeded the critical value, the null hypothesis was rejected, indicating that the difference in the indicator had a significant impact on sensory quality.

[0038] The Hotelling T 2 Control charts are a core tool in multivariate statistical process control. Their key advantage lies in taking into account the correlations among multiple variables, using T-charts to... 2 The statistical measures comprehensively assess the overall deviation of the sample from the baseline state, solving the problem that univariate control charts cannot identify the coordinated anomalies of multiple indicators, and adapting to the needs of coordinated management of multiple chemical components in cigarettes.

[0039] The control chart calculation steps are as follows: (1) Data foundation: Construction of multivariate training set First, multiple batches of samples under normal production conditions need to be collected, forming the training set. Key indicators of cigarette quality are measured, such as p variables including total alkaloids and reducing sugars. Assuming n batches are collected, each batch contains 3 samples, and each sample measures p indicators, an n×p matrix X can be formed, representing the baseline state of the production process. Two key baseline parameters are calculated using the above data: the mean vector μ0 and the covariance matrix S of the training set. The mean vector μ0 represents the average level, and the covariance matrix S reflects the correlation and fluctuations among the variables.

[0040] (2) T 2 The calculation process of statistics The core of Hotelling control charts is through T 2 The statistic measures the overall deviation of the current batch from the baseline state. Its mathematical logic can be simplified as follows: Step 1: Calculate the mean vector of the training set μ0 = (μ1, μ2, ..., μ...). p ) T μ i Let p be the target mean of the i-th variable, p be the variable, and T be the transpose sign.

[0041] Step 2: Calculate the covariance matrix S of the training set. T ,in,( ) is the vector of deviations between the sample mean and the overall mean. ) T It is its transpose (row vector).

[0042] Step 3: Calculate the squared Mahalanobis distance between the validation set sample vector x or the training set sample and μ0, i.e., T. 2 The mean vector is x = (x1, x2, ..., xn). p (x) i (where T is the sample mean of the i-th variable) 2 =(x-μ0) T S -1 (x-μ0) The uniqueness of Mahalanobis distance lies in the fact that it considers both the absolute deviation of the indicators, such as how much sugar content deviates from the mean, and eliminates the influence of the indicator's dimensions and correlation through the covariance matrix. Mahalanobis distance can provide a unified measurement, and at the same time, the co-variance fluctuations of strongly correlated indicators will be reasonably weighted.

[0043] (3) Calculate the control limit UCL (Upper Control Limit) Hotelling T 2 The statistic measures the degree of deviation of multivariate data from the population center, and its UCL is based on the F distribution or chi-square (χ²) distribution. 2 The derivation of the distribution depends on the sample size and the number of variables, and mainly includes the following two methods: The control limit formula, based on the F distribution, is suitable for small sample sizes, n < 50.

[0044] It is significant at level α and has ( ) degrees of freedom The critical value of the F distribution. This refers to the size of the training set, which must satisfy > Otherwise, the covariance matrix is ​​not invertible. It is the number of variables.

[0045] The control limit formula, based on the chi-square distribution, is suitable for samples with a large number of samples, n>50.

[0046] It is the critical value of the chi-square distribution with a significance level of α and degrees of freedom of p.

[0047] The selection of the control limits, from the perspective of sample size, involves fewer than 50 samples for each grade in this invention, according to Hotelling T. 2The conventional application of control charts should prioritize using critical values ​​derived from the F-distribution as control limits, as this method is more suitable for the statistical characteristics of small sample data and reduces the risk of misjudgment in small sample scenarios. However, considering the actual sample characteristics of this invention, all sampled products were pre-judged as qualified, the overall deviation between samples was small, and the data fluctuation range was relatively narrow. To more comprehensively verify data stability, the team also calculated control limits derived from the chi-square distribution as a reference. Through comparative analysis of the two control limits, it is possible to further verify whether the fluctuation of the sample data is within a reasonable range, avoid the limitations that may exist with a single control limit, and provide a more rigorous statistical basis for subsequent multi-indicator collaborative anomaly identification.

[0048] T based on the training set 2 The distribution, combined with the significance level α, is used to calculate the UCL. The significance level α is typically set to 0.01, 0.05, etc., representing the probability of misclassifying data as abnormal. A value of 0.05 usually corresponds to a 95% confidence interval, indicating a 5% probability of normal data being misclassified as abnormal. To strictly control misclassification, α can be reduced, such as to 0.01. If the T of the new batch... 2 If the statistical value exceeds the UCL, it indicates that there may be an anomaly in the production process, and the combination of multiple variables has deviated from the normal state. Further investigation is needed to find the cause and take corresponding measures to restore the stability of the production process.

[0049] like Figure 2 As shown in the figure, the present invention provides a cigarette quality stability evaluation system based on multivariate process statistics-sensory evaluation fusion, which specifically includes: The experimental resource management module is used to manage cigarette sample information, verify the compliance of reagents and consumables, and confirm the calibration status of testing instruments. The chemical composition detection module is used to preset the detection standards for 25 indicators in four major categories, perform standardized testing, and store data for quality control. The indicator data processing module is used to retrieve conventional chemical composition data, calculate four derived indicators, and generate a complete dataset. Multivariate fluctuation monitoring module for integration with Hotelling T 2 Control chart models are used to visualize and monitor the coordinated fluctuations of multiple indicators. The sensory evaluation correlation module is used to manage the three-point test process, input evaluation results, and correlate and analyze statistical data. The key quality factor management module is used to screen key quality factors and automatically define their reasonable fluctuation control range; The evaluation report generation module is used to summarize key information and generate a standardized control report on the quality stability of cigarettes.

[0050] Evidence related to the technical effects obtained by the embodiments of the present invention.

[0051] 1. Based on T 2 Key quality factor screening using a combination of control charts and three-point testing Using eight batches of samples from the first phase as the training set (marked with blue crosses) and three batches of samples from the second phase as the validation set (marked with orange crosses), multivariate Hotelling T values ​​were plotted for various indicators such as common chemical components, alkaloids, and polyphenols. 2 Control charts, using T 2 Statistics are used to reflect the overall fluctuation of multiple variables. In the figure, the blue circles represent the batch means of the training set, and the orange circles represent the batch means of the validation set, thus reflecting the overall level of each batch; the red dashed line is the UCL based on the F distribution (α=0.05), and the green dashed line is the UCL based on the chi-square distribution (α=0.05). If these lines are exceeded, it indicates that there may be anomalies in the process. In the figure, paired samples evaluated using the three-point test are circled and connected by lines. The numbers next to the lines represent the number of evaluations that accurately identified sample differences. This figure also integrates Hotelling T 2 Multivariate analysis information from control charts and sensory evaluation results from the three-point test.

[0052] 1.1 Conventional Ingredients Depend on Figure 3 As can be seen, all samples in ML are within the control limits of the F distribution, and the T values ​​of most training and validation set samples are within the control limits. 2 The statistical values ​​did not exceed the chi-square distribution control limits, with only sample ML-7-2 falling outside the control limits. This indicates that the content of conventional chemical components in the ML samples is generally relatively stable, with only slight overall fluctuations in a few samples. To investigate whether deviations in conventional chemical components affect the sensory quality of cigarettes, T was selected. 2 For the ML-7-2 sample, which deviated significantly from the training set mean, and the ML-4-3 sample, which deviated less significantly, the sensory quality was evaluated using the three-point test. Three people identified the difference between the two samples.

[0053] Depend on Figure 4 It can be seen that the T values ​​of all samples in the YY sample training set and validation set are... 2 None of the statistical measures exceeded the F-distribution control limits, except for samples YY-1-2 and YY-2-1, which were above the chi-square distribution control limits. This result indicates that the content of conventional chemical components in sample YY generally fluctuated little and showed good stability, without significant deviations from the normal range. T was selected. 2 The YY-2-1 sample, with a relatively large statistical value, is related to T. 2 For the YY-4-2 sample, which has a relatively small statistical size, a three-point test was used to conduct a sensory quality comparison evaluation. The results showed that only two people identified the sensory differences between the two samples.

[0054] Depend on Figure 5 It can be seen that the T values ​​of all samples in the XP training and validation sets are... 2 None of the statistical measures exceeded the F-distribution control limit, except for XP-8, which approached the chi-square distribution control limit. This result indicates that the content of common chemical components in the XP samples generally fluctuated little, exhibiting good stability and without significant deviations from the normal range. T was selected. 2 XP-8 samples with relatively large statistical values, compared with T 2 For the XP-2-2 sample, which has a smaller statistical size, a three-point test was used to conduct a sensory quality comparison evaluation. The results showed that three people identified the sensory differences between the two samples.

[0055] For the conventional chemical composition category variables, the project team conducted sensory evaluations by 27 participants (covering three groups of samples from three cigarette brands). The results showed that the total number of correct identifications of a single sample across the three groups was 8. According to Clause 5.3.5 of the "Sensory Evaluation Methods for Tobacco and Tobacco Products" (YC / T 138-1998) and the requirements of Appendix Table B1, this type of indicator evaluation requires meeting the critical standard of ≥14 accurate identifications. The identification results of 8 participants for the conventional chemical composition category variables did not meet this requirement. (Combined with Hotelling T...) 2 Stability analysis of the control chart indicators and sensory impact verification using the three-point test show that slight deviations in the conventional chemical composition variables of cigarettes do not significantly affect their sensory quality stability; even professional smoke testers find it difficult to clearly distinguish the sensory differences caused by such fluctuations. This result indicates that, within a certain range, simple differences in the content of conventional chemical components are not the core key factor affecting the quality stability of cigarettes.

[0056] 1.2 Conventional Derived Indicators Depend on Figure 6 It can be seen that the T values ​​of the samples in the ML training and validation sets... 2 None of the statistical measures exceeded the F-distribution control limit, except for samples ML-1-1 and ML-2-2, which exceeded the chi-square distribution control limit. This indicates that the general derived indices of the ML samples are generally within a stable range, without significant deviations from normal levels, with only a few samples exhibiting slight overall fluctuations. To investigate whether slight deviations in the general derived indices affect the sensory quality of ML cigarettes, samples with a deviation slightly larger than the overall mean of the training set (T0) were selected. 2 ML-1-1 samples with higher statistical values ​​(T) and smaller deviations (T) 2 For the ML-4-1 sample (with a low statistical value), a three-point test was used to conduct a sensory quality comparison evaluation. The results showed that five people identified the sensory differences between the two samples.

[0057] Depend on Figure 7It can be seen that among the YY samples, only sample YY-1-2 exceeded the chi-square distribution control limit and was close to the F distribution control limit, while the other samples were all below both control limits. This indicates that the conventional derived indices of the YY samples are generally within a stable range, with only a few samples showing fluctuations. Selecting T... 2 For sample YY-1-2, whose statistical value deviated significantly from the training set mean, and sample YY-6-2, whose value deviated less, a three-point test was used to compare their sensory quality. The results showed that five people identified the sensory difference between the two samples.

[0058] Depend on Figure 8 It can be seen that the T values ​​of the samples in the XP training and validation sets... 2 None of the statistics exceeded the control limits of the F-distribution and chi-square distribution, except for sample XP-5-2, which approached the control limit of the chi-square distribution. This indicates that the content of conventional derived indices in XP is relatively stable and fluctuates little. T was selected. 2 Sensory quality was evaluated using a three-point test for samples XP-5-2, which had a slightly larger deviation from the training set mean, and samples XP-2-1, which had a smaller deviation. The results showed that five individuals successfully identified the sensory differences between the two samples.

[0059] For the corresponding conventional derived indicator variables, a total of 27 sensory evaluations were conducted (covering three groups of samples from three cigarette brands). The results showed that the total number of correct identifications of a single sample across the three groups was 15. Based on the critical standard of ≥14 accurate identifications in Appendix Table B1 of YC / T 138-1998, the 15 accurate identifications for the conventional derived indicator variables meet this requirement. (Combined with Hotelling T...) 2 The stability analysis of the control chart indicators and the sensory impact verification by the three-point test show that deviations from conventional derived indicator variables of cigarettes have a significant impact on the sensory quality stability of cigarettes. Professional smokers can clearly distinguish the sensory differences caused by such deviations. Therefore, conventional derived indicators can be regarded as key quality factors affecting the quality stability of cigarettes, providing targeted indicators for subsequent quality control and optimization.

[0060] 1.3 Alkaloids Depend on Figure 9 It can be seen that there are 2 samples T in ML. 2 The statistics (ML-4-2, ML11-1) deviated significantly from the training set mean, exceeding the control limits for both distributions. Samples M-6-2 and M-8-2 were close to the chi-square distribution control limits, showing greater deviation than the previously mentioned conventional components and derived indices. To investigate whether deviations in alkaloid components affect the sensory quality of cigarettes, T was selected. 2 Sensory quality was evaluated using a three-point test for the ML-11-1 sample, which had a larger deviation from the training set mean, and the ML-7-3 sample, which had a smaller deviation. The results showed that two people identified the difference between the two samples.

[0061] Depend on Figure 10 It can be seen that the T values ​​of all samples in the YY sample training set and validation set are... 2 All measurements did not exceed the F-distribution control limit, except for sample YY-4-3, which exceeded the chi-square distribution control limit. T was selected. 2 The YY-4-3 sample has a relatively large statistical value, compared to T. 2 For the YY-7-1 sample, which has a smaller statistical value, a three-point test was used to conduct a sensory quality comparison evaluation. The results showed that three people were able to identify the sensory differences between the two samples.

[0062] Depend on Figure 11 It can be seen that, except for the XP-9-1 sample, the T in XP... 2 The statistic exceeds the chi-square distribution control limit and is close to the F distribution control limit; other samples T 2 None of the statistics exceeded the control limits of the F-distribution and chi-square distribution, indicating that the alkaloid content in XP was relatively stable with minimal fluctuations, except for a few samples. T was selected. 2 Sensory quality was evaluated using a three-point test for XP-9-1 samples, which had a slightly larger deviation from the training set mean, and XP-5-1 samples, which had a smaller deviation. The results showed that one person was able to identify the sensory difference between the two samples.

[0063] For the alkaloid variable, a total of 27 sensory evaluations were conducted (covering three groups of samples from three cigarette brands). The results showed that the total number of correct identifications of a single sample across the three groups was 6. Based on the critical standard of ≥14 accurate identifications in Appendix Table B1 of YC / T 138-1998, the 6 identifications for the alkaloid variable did not meet this requirement. (This is in conjunction with Hotelling T...) 2 Stability analysis of the control chart indicators and sensory impact verification using the three-point test show that slight fluctuations in the alkaloid content of cigarettes do not significantly affect their sensory quality stability; even professional smoke tasters find it difficult to clearly distinguish the sensory differences caused by such fluctuations. This result indicates that, within a certain range, simple differences in alkaloid content are not a key factor affecting the quality stability of cigarettes.

[0064] 1.4 Polyphenols Depend on Figure 12 It can be seen that the samples T in the ML training set and validation set 2 None of the statistical values ​​exceeded the control limits of the F-distribution and chi-square distribution, indicating that the polyphenol content of the ML samples was generally within a stable range and did not deviate significantly from the normal level. Samples ML-6-3, ML-7-3, and ML-3-3 showed slight deviations, close to the chi-square distribution control limits. To investigate whether these slight deviations in polyphenol content would affect the sensory quality of ML cigarettes, samples with a deviation slightly larger than the overall mean of the training set (T0) were selected.2 ML-6-3 samples with higher statistical values ​​(T) and smaller deviations (T) 2 For the ML-4-3 sample (with a low statistical value), a three-point test was used to conduct a sensory quality comparison evaluation. The results showed that five people identified the sensory differences between the two samples.

[0065] Depend on Figure 13 It can be seen that the T values ​​of all samples in the YY sample training set and validation set are... 2 All measurements were within the F-distribution control limit, except for samples YY10-1 and YY-5-3, which exceeded the chi-square distribution control limit. This indicates that the polyphenol content of the YY samples was generally within a stable range, with only a few samples showing fluctuations. T was selected. 2 For sample YY-5-3, whose statistical value deviated significantly from the training set mean, and sample YY-2-3, whose value deviated less, a three-point test was used to compare their sensory quality. The results showed that five people were able to identify the sensory difference between the two samples.

[0066] Depend on Figure 14 It can be seen that the T values ​​of the samples in the XP training and validation sets... 2 None of the statistical values ​​exceeded the F-distribution control limit. Samples XP-7 and XP-8 exceeded the chi-square distribution control limit, while other batches were far below the chi-square distribution control limit. This indicates that, except for samples XP-7 and XP-8, the polyphenol content in XP fluctuated relatively little. (T was selected.) 2 For samples XP-8, which deviated significantly from the training set mean, and XP-3-2, which deviated less significantly, a three-point test was used for sensory quality evaluation. The results showed that five people identified the sensory difference between the two samples.

[0067] For the polyphenol variable, a total of 27 sensory evaluations were conducted (covering three groups of samples from three cigarette brands). The results showed that a total of 15 individuals correctly identified a single sample across the three groups. Based on the critical standard of ≥14 accurate identifications in Appendix Table B1 of YC / T 138-1998, 15 individuals achieved this requirement for the polyphenol variable. (This is in conjunction with Hotelling T...) 2 The stability analysis of the control chart indicators and the sensory impact verification by the three-point test show that deviations in the polyphenols variable of cigarettes have a significant impact on the sensory quality stability of cigarettes. Professional smokers can clearly distinguish the sensory differences caused by such deviations. Therefore, polyphenols can be regarded as a key quality factor affecting the quality stability of cigarettes, providing a targeted indicator for subsequent quality control and optimization.

[0068] 1.5 polyacids Depend on Figure 15 As can be seen, all samples in ML are within the control limits of the F distribution, and the T values ​​of most training and validation set samples are within the control limits. 2The statistical values ​​did not exceed the chi-square distribution control limits, with only the ML-2-2 sample's value falling outside the control limits. This indicates that the content of conventional chemical components in the ML samples is generally relatively stable, with only slight overall fluctuations in a few samples. To investigate whether deviations in polyacid components affect the sensory quality of cigarettes, T was selected. 2 Sensory quality was evaluated using a three-point test for samples ML-2-2, which showed a large deviation from the training set mean, and samples ML-5-3, which showed a smaller deviation. The results showed that the difference was detected by three individuals.

[0069] Depend on Figure 16 It can be seen that the T values ​​of all samples in the YY sample training set and validation set are... 2 All measurements were within the F-distribution control limit, except for sample YY-10-1, which exceeded the chi-square distribution control limit. T was selected. 2 The YY-10-1 sample has a relatively large statistical value, compared to T. 2 For the YY-9-1 sample, which has a smaller statistical value, a three-point test was used to conduct a sensory quality comparison evaluation. The results showed that three people were able to identify the sensory differences between the two samples.

[0070] Depend on Figure 17 It can be seen that the T values ​​of the samples in the XP training and validation sets... 2 None of the statistical measures exceeded the F-distribution control limit, except for the XP-7 sample, which exceeded the chi-square distribution control limit, while all other batches were below the chi-square distribution control limit. T was selected. 2 Sensory quality was evaluated using a three-point test for samples XP-7, which showed a large deviation from the training set mean, and XP-4-1, which showed a smaller deviation. The results showed that five individuals identified a sensory difference between the two samples. However, it should be noted that in XP-7, in addition to the polyacid components exceeding the chi-square control limit, the polyphenol components also exceeded the chi-square control limit. Therefore, the possibility that the sensory difference between the two samples was due to deviations in the polyphenol components cannot be ruled out.

[0071] Corresponding to the variable of polyacids, the project team conducted sensory evaluations by 27 individuals (covering three groups of samples from three cigarette brands). The results showed that the total number of correct identifications of a single sample across the three groups was 11. Based on the critical standard of ≥14 accurate identifications in Appendix Table B1 of YC / T 138-1998, the 11 identifications of the polyacid index did not meet this requirement. (This is in conjunction with Hotelling T...) 2 Stability analysis of the control chart indicators and sensory impact verification using the three-point test show that deviations in the content of polyacids in cigarettes did not significantly affect their sensory quality stability; even professional smoke tasters found it difficult to clearly distinguish the sensory differences caused by such fluctuations. This result indicates that, within a certain range, simple differences in the content of polyacids are not a key factor affecting the quality stability of cigarettes.

[0072] 1.6 Higher fatty acids Depend on Figure 18 It can be seen that the samples T in the ML training set and validation set 2 The statistics did not exceed the control limits of the F distribution for either sample ML-8-3 or ML-5-3. 2 The statistical value exceeded the chi-square distribution control limit. To investigate whether deviations in higher fatty acid indicators affect the sensory quality of jasmine-scented cigarettes, samples ML-8-3 (slightly larger deviation from the overall mean of the training set) and ML-4-1 (smaller deviation) were selected, and a three-point test was used to compare and evaluate their sensory quality. The results showed that four people identified the sensory differences between the two samples.

[0073] Depend on Figure 19 It can be seen that the T values ​​of the samples in the YY training and validation sets... 2 None of the statistical measures exceeded the control limits of the F-distribution and chi-square distribution, indicating that the higher fatty acid indicators of sample YY were generally within a stable range and did not deviate significantly from normal levels. T was selected. 2 The YY-6-3 sample has a relatively large statistical value, compared to T. 2 For the YY-8-3 sample, which has a relatively small statistical value, a three-point test was used to conduct a sensory quality comparison evaluation. The results showed that only one person was able to identify the sensory difference between the two samples, indicating a low recognition rate.

[0074] Depend on Figure 20 It can be seen that the T values ​​of the samples in the XP training and validation sets... 2 None of the statistical measures exceeded the F-distribution control limit, except for sample XP-9-1, which exceeded the chi-square distribution control limit, while other batches were below the chi-square distribution control limit. T was selected. 2 For XP-9-1 samples, which deviated significantly from the training set mean, and XP-3-1 samples, which deviated less significantly, a three-point test was used for sensory quality evaluation. The results showed that three people identified the sensory difference between the two samples.

[0075] For the variable of higher fatty acids, a total of 27 sensory evaluations were conducted (covering three groups of samples from three cigarette brands). The results showed that the total number of correct identifications of a single sample across the three groups was 8. Based on the critical standard of ≥14 accurate identifications in Appendix Table B1 of YC / T 138-1998, the 8 identification results for the higher fatty acid variable did not meet this requirement. (This is in conjunction with Hotelling T...) 2Stability analysis of the control chart indicators and sensory impact verification using the three-point test show that slight deviations in the higher fatty acid content of cigarettes do not significantly affect their sensory quality stability; even professional smoke tasters find it difficult to clearly distinguish the sensory differences caused by such fluctuations. This result indicates that, within a certain range, simple differences in the content of higher fatty acids are not a key factor affecting the quality stability of cigarettes.

[0076] 2. Cigarette quality factor T 2 Determining the range of fluctuation of statistics Based on the previous research findings, conventional derived indicators and polyphenol indicators have been identified as key quality factors affecting the stability of cigarette quality. Conventional components, alkaloids, polybasic acids, and higher fatty acids have relatively minor impacts on cigarette quality stability and are classified as other quality factors; their fluctuation control range can be appropriately relaxed. (Based on Hotelling T) 2 The core principle of control charts is to divide application scenarios into two categories based on the training set sample size n: n≦50 (small to medium scale) and n>50 (large scale). This is achieved by calculating and determining the T-values ​​of key quality factors and other quality factors separately. 2 The range of statistical fluctuations provides quantitative support for the production control of cigarette products.

[0077] 2.1 Key Quality Factor T 2 Conditions for adapting the control range of statistical quantities Key quality factor T 2 The analysis of the fluctuation range of statistics takes the baseline state of the training set as the core. The control limit formulas and applicable conditions for the two types of scenarios must follow the statistical distribution characteristics in Table 1.

[0078] Table 1 Key quality factor T in different scenarios 2 Effective control range of statistics

[0079] When the training set sample size n ≤ 50, the key quality factor T 2 Based on the statistics in Table 2, we divide the data into three intervals: stable, warning range, and risk. The specific parameters are as follows.

[0080] Table 2 Key Quality Factors T 2 Range of fluctuation of the statistic (n≦50)

[0081] When the number of training set samples n>50, the key quality factor T 2 The statistics do not have a warning range. Based on Table 3, only two intervals are divided: stable and risky. The specific indicator parameters are as follows.

[0082] Table 3 Key Quality Factors T 2Range of fluctuation of the statistic (n>50)

[0083] 2.2 Fluctuation range of other quality factors Other quality factors T 2 The analysis of the fluctuation range of statistics takes the baseline state of the training set as the core. The control limit formulas and applicable conditions for the two types of scenarios must follow the statistical distribution characteristics in Table 4.

[0084] Table 4 Other quality factors T under different scenarios 2 Effective control range of statistics

[0085] When the training set sample size n ≤ 50, other quality factors T 2 Based on the statistics in Table 5, we divide the data into two intervals: stable and risky. The specific parameters are as follows.

[0086] Table 5 Other quality factors T 2 Range of fluctuation of the statistic (n≦50)

[0087] When the number of training set samples n>50, other quality factors are divided into stable and risk intervals, and the specific index parameters in Table 6 are as follows.

[0088] Table 6 Other quality factors T 2 Range of fluctuation of the statistic (n>50)

[0089] 3. Conclusion This invention takes qualified finished cigarettes of three brands, ML, YY, and XP, as the research object and uses Hotelling T... 2 A fusion analysis method combining control charts and the three-point test was used to systematically screen key quality factors affecting the stability of cigarette quality.

[0090] The results showed that Hotelling T contained conventional components, alkaloids, polyphenols, polybasic acids, and higher fatty acids. 2 Control charts showed that, except for two ML cigarette samples whose alkaloid content exceeded the F-distribution control limit, all other samples were within the F-distribution control limit; only a few samples exceeded the chi-square distribution control limit, and overall, the content of various components was stable. To further explore the impact of key quality factors on sensory quality, the project team designed an experiment using the controlled variable method, targeting the T-distribution control limit among the above-mentioned categories of indicators. 2For samples with significant deviations in statistical values, sensory evaluation was conducted according to the three-point test requirements in the "Sensory Evaluation Methods for Tobacco and Tobacco Products" (YC / T138-1998). Each variable had 27 participants. Referring to Clause 5.3.5 of the standard and Appendix Table B1, the evaluation critical value was ≥14 correct identifications. For the two key quality factors, conventional derived polyphenols, the correct identifications were both >14, indicating that samples with significant differences in these two factors also showed significant differences in sensory quality. For samples with significant deviations selected based on other variables, the correct identifications in the three-point test were all less than 14. Therefore, conventional components, alkaloids, polybasic acids, and higher fatty acids were identified as other quality factors.

[0091] Finally, based on Hotelling T 2 Based on the principle of control charts, two application scenarios were set up according to the training set sample size, and the key quality factors and non-key quality factors T were calculated and determined respectively. 2 The range of statistical fluctuation control provides quantitative support for the production management and control of cigarette products.

[0092] 4. Composition of the monitoring system 4.1 Monitoring Objects The cigarettes produced under the three brands of Jasmine Fragrance, Taishan New Product, and Taishan Yan Yue are monitored according to production batches.

[0093] 4.2 Monitoring Indicators and Measurement Methods Monitoring indicators should be set up according to key quality indicators and other quality indicators. The measurement methods must comply with industry standards or laboratory optimization methods, as detailed in the table below: Table 7. Indicator Categories, Specific Indicators, and Measurement Methods

[0094] 4.2 Monitoring frequency in the monitoring process 4.2.1 Sample collection frequency: Regular short-cycle batches (≤7 days). When the leaf group formula or packaging materials are changed, the frequency needs to be increased, the reason for the adjustment should be recorded and the batch should be marked, and the regular frequency should be restored after stabilization.

[0095] 4.2.2 Frequency of chemical composition determination: The determination is carried out simultaneously with sample collection, and one determination is performed for each group of samples collected.

[0096] 4.2.3 Hotelling T2 control chart analysis frequency: Analyze once after each chemical composition determination; calibrate the control limits every quarter or when 30 new batches of data are added.

[0097] 4.2.4 Frequency of sensory evaluation: only when sample T 2The statistics are conducted when the data falls within the warning or risk range, and the results are used to determine product quality consistency.

[0098] 4.3 Equipment and Personnel Requirements 4.3.1 Testing Equipment and Environment Chemical composition determination: High performance liquid chromatography (polyphenols), gas chromatography-mass spectrometry (alkaloids), gas chromatography (polyacids and higher fatty acids), continuous flow analyzer (routine components); Data processing: Install statistical analysis software (such as Python, SPSS), and have Hotelling T... 2 Control chart plotting; Sensory evaluation: The evaluation room meets the environmental requirements of YC / T38-1998.

[0099] 4.3.2 Personnel Requirements Testing personnel: must have passed tobacco chemical composition testing training and have more than 2 years of relevant testing experience; Sensory evaluation experts: must meet the qualification requirements of YC / T38-1998 and be familiar with the sensory characteristics of the monitored cigarette brands; Data Analyst: Possesses multivariate statistical analysis skills and can independently complete HotellingT (a type of statistical analysis). 2 Control chart construction.

[0100] 4.4 Data Management System Establish a cigarette quality consistency monitoring database to record the following information: Basic sample information: brand, production batch, sampling date, sampler, sample condition; Test data: original values ​​of each indicator, calculated values ​​of derived indicators, and parallel sample deviations; Control chart results: control limits for each indicator category, sample deviation (whether the limit was exceeded); Sensory evaluation record: list of experts, sample numbers for three-point testing, number of correctly identified samples, and judgment conclusion; Anomaly handling record: results of anomaly batch tracing, corrective measures, and verification results.

[0101] 5. Monitoring Process 5.1 Sample Collection and Preparation 5.1.1 Basic principles of sampling: (1) Randomness: Ensure that each sampling unit has an equal probability of being selected, and avoid sample bias caused by human selection; (2) Representativeness: The sample should be able to reflect the overall quality status of the sampled batch of products, covering different stages in the production process, equipment operating status and other factors; (3) Timeliness: Sample in a timely manner at the specified key nodes to avoid the loss of timeliness of the sample due to delay, which would affect the quality judgment.

[0102] 5.1.2 Sampling timing: Sampling begins after the equipment has been running continuously for 1 hour in the cigarette manufacturing process, and then every 1.5 hours thereafter, with 3 parallel samples collected for each batch.

[0103] 5.1.3 Sampling method: Continuous sampling method was adopted. At the cigarette output channel of the cigarette rolling equipment, 200 cigarettes were continuously collected using a sampling box. The first and last 10 cigarettes were removed, and the remaining cigarettes were used as samples for various index tests.

[0104] 5.1.4 Sample preparation: Disassemble the cigarette sample to be tested, take out the tobacco, and bake it in an oven at 40℃ for 6 hours (adjust the baking time appropriately according to the ambient temperature, humidity and sample quantity). Then grind the tobacco into powder, sieve it through a 60-mesh sieve, and seal it in a self-sealing bag for storage.

[0105] 5.2 Determination of Chemical Composition 5.2.1 Determine the content of each index according to the method specified in Table 4.2. Perform two parallel experiments for each index. The relative deviation of the parallel samples should be ≤5%. Take the average value as the final result. 5.2.2 Calculate the conventional derived index and retain two decimal places.

[0106] 5.3 Hotelling T 2 Control chart analysis 5.3.1 Control Limits: The indicator data of the training set samples for each brand (grouped by indicator category) are analyzed using Hotelling T. 2 For control charts, depending on the number of training set samples, calculate the F-distribution control limits or chi-square control limits for α=0.05 (95% confidence interval) according to the requirements of Appendix 1.

[0107] 5.3.2 Sample Judgment: Substitute the index data of the batch of samples to be monitored into the corresponding control chart to determine whether they exceed the corresponding control limits.

[0108] 5.4 Sensory evaluation (three-point test method) 5.4.1 Sample Submission: In accordance with the requirements of YC / T38-1998, the samples to be evaluated (risk range or warning range) and normal samples (stable range) shall be randomly combined into 3 sample groups (2 normal and 1 abnormal), and submitted to the evaluation experts after being labeled with random numbers. 5.4.2 Results Statistics: Record the number of abnormal samples correctly identified by each expert and count the number of correctly identified samples (n). 5.4.3 Significance Determination: Referring to YC / T138-1998 (binomial distribution significance table, α=0.05), find the critical value (C) based on the total number of experts (N): If n < C: the null hypothesis cannot be rejected, and it is determined that there is no significant sensory difference between the samples; If n≥C: reject the null hypothesis and determine that there is a significant sensory difference between the samples.

[0109] 6. Quality Consistency Judgment Rules 6.1 Quality Consistency Judgment The following conditions must be met simultaneously: 6.1.1 Key Quality Indicators (2 in total): T values ​​of 2 indicators in the sample 2 All statistics are within a stable range; or the T values ​​of 1-2 indicators in the sample are within a stable range. 2 The statistic is within the warning or risk range, and the three-point test shows no significant sensory difference; 6.1.2 Other quality indicators (4 in total): T in the sample that are within the risk range 2 The number of statistical indicators should not exceed two; or the T values ​​in the sample should be within the risk range. 2 The number of statistical indicators exceeds two, and the three-point test determines that there is no significant sensory difference.

[0110] 6.2 Determination of Quality Inconsistency Meet any of the following conditions: 6.2.1 Key Quality Indicators (2 in total): T values ​​of 1-2 indicators in the sample 2 The statistical value is within the warning or risk range, and a significant sensory difference is determined by the three-point test; 6.2.2 Other quality indicators (4 in total): T in the sample that are within the risk range 2 The number of statistical indicators exceeds two, and a significant sensory difference is determined by the three-point test; 6.3 Judgment Prerequisites All judgments involving three-point inspection must be based on complete and valid three-point inspection results; if the three-point inspection is incomplete or the inspection results are unclear, the quality consistency will not be judged at this time.

[0111] 7. Handling of Abnormal Samples When a discrepancy in quality is identified, a quality stability early warning is immediately activated, and the information of the batch under warning (brand, production batch, reason for triggering the warning) is recorded and simultaneously pushed to the relevant departments responsible for quality control, providing early warning support for subsequent cigarette quality consistency control.

[0112] Example 1 ML cigarettes were selected as the research object. Fifteen batches of samples were continuously collected from the same production line, with 200 cigarettes randomly selected from each batch for chemical analysis. The contents of water-soluble total sugar, reducing sugar, total alkaloids, total nitrogen, potassium, and chlorine were determined according to the established testing procedure. Derivative indices such as total sugar / total alkaloids, reducing sugar / total alkaloids, reducing sugar / total sugar, and total nitrogen / total alkaloids were calculated. Simultaneously, four alkaloids (nicotine, nornicotine, pseudoequisetine, and neonicotinoids) and five polyphenols (chlorogenic acid, neochlorogenic acid, cryptochlorogenic acid, rutin, and hyoscyamine) were detected. Data for ten organic acid indicators (oxalic acid, malonic acid, succinic acid, malic acid, palmitic acid, myristic acid, citric acid, oleic acid, linoleic acid, and linolenic acid) were obtained through the organic acid detection procedure. The above indicators were divided into six groups: conventional components, conventional derived indicators, alkaloids, polyphenols, polybasic acids, and higher fatty acids. Data matrices were constructed for each group, and the mean vector and covariance matrix were calculated using data from each batch. Hotelling T was then calculated. 2 Statistics were measured and control charts were plotted to achieve joint monitoring of the coordinated fluctuations of multiple indicators. Statistical results show that the T-values ​​for all samples... 2 The statistics are all below the control limits calculated by the F-distribution, indicating that the overall chemical composition of the product remains stable across all dimensions, and no further sensory quality evaluation verification is required.

[0113] Example 2 XP cigarettes were selected as the evaluation object, and 20 batches of production samples were collected and tested for the same 25 chemical indicators. Four derived indicators were calculated based on the conventional chemical composition data: the ratio of total sugar to total alkaloids, the ratio of reducing sugar to total alkaloids, the ratio of reducing sugar to total sugar, and the ratio of total nitrogen to total alkaloids. Multivariate evaluation datasets were constructed for both the original and derived indicators. Statistical analysis revealed that the conventional derived statistic of one batch of samples exceeded the F-distribution control limit, placing it within the risk range; all other indicators did not exceed the chi-square distribution control limit. Further sensory evaluation was conducted, with two normal samples combined with the abnormal sample for blind evaluation. In 27 evaluations, the abnormal sample was correctly identified 18 times, exceeding the critical value by 14 times, indicating that sensory differences did indeed exist between the samples. Analysis of the statistical monitoring results revealed that the fluctuation mainly stemmed from the abnormally high ratio of reducing sugar to total alkaloids. This example demonstrates that combining multivariate statistics with sensory evaluation can accurately identify key indicators that have a real impact on cigarette quality.

[0114] Example 3 Quality stability analysis was conducted on YY cigarettes. Fifty-two batches of samples were collected consecutively, and 25 chemical indicators were tested. A multivariate statistical model was established based on the test data, and statistics were calculated. The results showed that the statistics for polyacids and polyphenols in the 7th batch of samples exceeded the chi-square control limits. Further analysis of the contribution of each indicator revealed significant deviations in citric acid and chlorogenic acid content in this batch. Sensory evaluation experiments were then conducted. The tasters identified 15 different samples in 27 tests, exceeding the critical value. Further review of the tobacco formula for this batch revealed high levels of citric acid and polyphenols in its raw materials. This example illustrates that the evaluation method can identify potential quality fluctuations in advance through statistical analysis and confirm their actual impact on cigarette quality through sensory verification.

[0115] Example 4 To verify the applicability of this method across different products, three types of cigarette products were selected and independent evaluation models were constructed for each. First, 10 batches of samples were collected for each product, and 25 chemical indicators were measured. Then, four derived indicators were calculated, and a statistical baseline model was established. The results showed that the mean vectors of the indicators differed significantly among the different products, but stable quality control ranges were formed within their respective models. When a batch of samples deviated from normal production levels, the statistics of several quality factors increased significantly. Sensory evaluation experiments verified that samples with higher statistics were identified 16 times out of 27 smoking tests, demonstrating the consistency between statistical anomalies and sensory differences. This example proves that the evaluation method is applicable to different types of cigarette products and can establish product-specific stability control models.

[0116] Example 5 In a real-world production environment, this method was applied to a cigarette quality monitoring system. The system automatically receives test data and calculates derived indicators in real time, while simultaneously using a statistical model to calculate statistical quantities. When the statistical quantity approaches the control limit, the system automatically issues an early warning and marks the relevant indicators. Production personnel adjust the tobacco leaf formula and process parameters according to the warning. Subsequent testing of the adjusted samples showed that the statistical quantity had returned to the normal range. After sensory evaluation verification, the tasters were unable to consistently identify samples with discrepancies, indicating that quality fluctuations were effectively controlled. This embodiment demonstrates that this method can achieve real-time quality monitoring in industrial production processes.

[0117] It should be noted that embodiments of the present invention can be implemented in hardware, software, or a combination of both. The hardware portion can be implemented using dedicated logic; the software portion can be stored in memory and executed by a suitable instruction execution system, such as a microprocessor or dedicated-design hardware. Those skilled in the art will understand that the above-described devices and methods can be implemented using computer-executable instructions and / or included in processor control code, for example, such code provided on a carrier medium such as a disk, CD, or DVD-ROM, a programmable memory such as read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. The devices and modules of the present invention can be implemented by hardware circuitry such as very large-scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, or programmable hardware devices such as field-programmable gate arrays, programmable logic devices, etc., or by software executed by various types of processors, or by a combination of the above-described hardware circuitry and software, such as firmware.

[0118] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications, equivalent substitutions, and improvements made by those skilled in the art within the scope of the technology disclosed in the present invention, and within the spirit and principles of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for evaluating the stability of cigarette quality based on the synergy of multivariate statistics and sensory evaluation, characterized in that, Includes the following steps: Step 1: Preparation of experimental samples and testing resources. Select multiple batches of cigarette samples under normal production conditions, confirm that the purity level of the reagents meets the requirements of chromatographic purity or analytical purity, and ensure that the testing instruments are calibrated and meet the accuracy requirements for detecting tobacco chemical components. Step two, chemical composition detection: conduct multi-category chemical composition detection on cigarette samples to obtain data on conventional chemical components, alkaloid compounds, polyphenolic compounds and organic acid compounds, and form a multi-index detection dataset; Step 3: Construction of derived indicators. Based on the results of conventional chemical component detection, derived indicators are calculated. The derived indicators include the ratio of total sugar to total alkaloids, the ratio of reducing sugar to total alkaloids, the ratio of reducing sugar to total sugar, and the ratio of total nitrogen to total alkaloids. Step 4: Multivariate fluctuation monitoring. Based on the multi-index detection dataset, a multivariate statistical monitoring model is established to perform synergistic fluctuation analysis on the quality indicators of cigarette samples. Step 5: Sensory evaluation and verification. The three-point test method is used to conduct sensory evaluation of cigarettes, and the results of sample difference identification are recorded through blind expert evaluation. Step 6: Screening of key quality factors. The results of multivariate statistical monitoring and sensory evaluation are correlated to identify key quality factors and other quality factors that affect the stability of cigarette quality and to establish their fluctuation control range.

2. The method for evaluating the stability of cigarette quality according to claim 1, characterized in that, The routine chemical composition detection includes the following indicators: Contents of water-soluble total sugars, reducing sugars, total alkaloids, total nitrogen, potassium, and chlorine.

3. The method for evaluating the quality stability of cigarettes according to claim 1, characterized in that, The alkaloid compound detection includes four indicators: nicotine, nornicotine, pseudoesequine, and neonicotine.

4. The method for evaluating the stability of cigarette quality according to claim 1, characterized in that, The detection of polyphenolic compounds includes five indicators: chlorogenic acid, neochlorogenic acid, cryptochlorogenic acid, rutin, and hyoscyamine.

5. The method for evaluating the stability of cigarette quality according to claim 1, characterized in that, The detection of organic acid compounds includes 10 indicators: oxalic acid, malonic acid, succinic acid, malic acid, palmitic acid, myristic acid, citric acid, oleic acid, linoleic acid, and linolenic acid.

6. The method for evaluating the stability of cigarette quality according to claim 1, characterized in that, Multivariate fluctuation monitoring uses Hotelling T 2 The control chart method for constructing a statistical monitoring model includes the following steps: Step 1: Establish a training dataset by collecting test data from multiple batches of cigarette samples under normal production conditions, forming a data matrix with n samples and p indicators, and calculating the mean vector and covariance matrix according to the indicator categories. Step 2, calculate the statistic T 2 The statistic T 2 The result is the product of the transpose matrix, the inverse matrix of the covariance matrix, and the difference vector between the sample index vector and the mean vector; Step 3: Determine the control limits. When the sample size is less than 50, calculate the upper control limit based on the F distribution. When the sample size is greater than 50, calculate the upper control limit based on the chi-square distribution. Step four: Compare the statistical measure of the sample to be tested with the upper control limit to determine whether there is a coordinated abnormal fluctuation in the cigarette quality indicators.

7. The method for evaluating the stability of cigarette quality according to claim 6, characterized in that, The Hotelling T 2 The steps for constructing a control chart are as follows: (1) Construct a training set, collect multiple batches of cigarette samples under normal production conditions, measure p indicators, form an n×p data matrix, calculate the mean vector μ0 and covariance matrix S of the training set, and establish the production baseline state. (2) Calculate T 2 The statistic, with the formula T 2 =(x-μ0)ᵀS -1 (x-μ0), where x is the sample vector to be tested, which comprehensively reflects the degree of coordinated deviation of multiple indicators; (3) Calculate the upper control limit (UCL). When the sample size n < 50, the control limit is derived using the F distribution. The formula is UCL = [(n-1)p / (np)]F a When the sample size n > 50, the control limits are derived using the chi-square distribution, with the formula UCL = χ². 2 a p, α is 0.

05.

8. A cigarette quality stability evaluation system based on the method of any one of claims 1 to 7, characterized in that, include: The experimental resource management module is used to manage cigarette sample information and confirm reagent purity levels and the calibration status of testing instruments. The chemical composition detection module is used to perform multi-category chemical composition detection and record the detection data. The indicator data processing module is used to calculate derived indicators and construct multi-indicator datasets; Multivariate fluctuation monitoring module, used for Hotelling T 2 Statistical models monitor the coordinated fluctuations of multiple indicators; The sensory evaluation correlation module is used to record the evaluation results of the three-point test and establish the correlation between statistical data and sensory data; The quality factor management module is used to screen key quality factors and set the fluctuation control range for key quality factors and other quality factors. The evaluation report generation module is used to generate cigarette quality stability evaluation reports.

9. The cigarette quality stability evaluation system according to claim 8, characterized in that, The multivariate fluctuation monitoring module includes: A data matrix construction unit is used to construct the multi-indicator dataset into six two-dimensional data matrices; The mean vector calculation unit is used to calculate the mean vector of the two-dimensional data matrix; The covariance matrix calculation unit is used to calculate the covariance matrix of the two-dimensional data matrix based on the mean vector. The statistical calculation unit is used to calculate Hotelling T based on the mean vector, the covariance matrix, and the preset sample data. 2 Statistic; The control limit determination unit is used to determine the control limit based on the significance level and to set the Hotelling T 2 The statistics are compared with the control limits to monitor the coordinated fluctuation of the multiple indicators.

10. The cigarette quality stability evaluation system according to claim 8, characterized in that, The sensory evaluation association module includes: The evaluation process management unit is used to manage the evaluation process of the three-point test method, including the preparation of evaluation samples, the arrangement of evaluation personnel, and the evaluation schedule. The evaluation data entry unit is used to enter the evaluation results of the three-point test method, which include the evaluators' judgment results and evaluation opinions on the sample differences. The statistical significance determination unit is used to perform statistical significance analysis on the evaluation results to determine whether there is a significant correlation between the statistical data and the sensory data; The quality factor identification unit is used to identify key quality factors and other quality factors that affect the sensory quality of cigarettes based on the statistical significance analysis results.

Images