Quality control method for mixing process of intestine-strengthening and diarrhea-stopping pills based on NIRS spectrum combined with MSPC technology

Abstract

The invention belongs to the field of quality control of modern traditional Chinese medicines, and discloses a quality control method for a mixing process of intestine-strengthening and diarrhea-stopping pills based on combination of NIRS spectrum and MSPC technology. The method comprises the following steps: pre-treating medicinal materials according to a prescription proportion; an NIRS spectrometer is adopted to collect data; performing variable expansion and dimensionality reduction on the three-dimensional spectrum data, establishing an MSPC model in batches based on a training set, and setting a control limit according to a standard deviation + / -3 times of a mean value of a principal component score (PC1), Hotelling Tand DModX statistical magnitude; and a new batch of spectrum data is input in real time, and feeding abnormity and process abnormity are jointly monitored through three types of statistics. A PLS quantitative model is synchronously established, and dynamic content prediction of the six chemical components such as neochlorogenic acid is achieved. According to the method, real-time early warning of abnormity in the mixing process and dynamic analysis of chemical components are achieved, and the uniformity and stability of the quality of the intestine-strengthening and diarrhea-stopping pills are remarkably improved.

Description

Technical Field

[0001] The present invention belongs to the field of modern traditional Chinese medicine quality control, and specifically discloses a quality control method for the mixing process of Guchang Zhixie Pills based on NIRS spectroscopy combined with MSPC technology. Background Art

[0002] As the material basis of the traditional Chinese medicine industry chain, the quality of traditional Chinese medicine is the foundation for ensuring the effectiveness and safety of clinical medications, and is the key to promoting the internationalization, industrialization, and modernization of traditional Chinese medicine. Traditional Chinese medicine preparations are usually produced in batches. The differences in the properties of raw medicinal materials between batches and the fluctuations of intermediate materials during the production process will directly lead to quality differences in traditional Chinese medicine preparations. The quality of traditional Chinese medicine preparations depends not only on the inspection methods, but more on all aspects of product manufacturing such as design and development, production control, and standardized management. Due to the long-term reliance on operational experience in traditional Chinese medicine production, there are no effective monitoring means in the traditional Chinese medicine manufacturing process. The influence of various operating parameters on quality attributes is not yet clear enough, and the control of the quality of intermediates or products can only rely on post hoc inspection methods. Once the product quality is unqualified, this batch needs to be discarded or reprocessed, resulting in waste of resources. With the acceleration of the internationalization process of traditional Chinese medicine and the country's attention to the traditional Chinese medicine industry, how to effectively improve the quality of traditional Chinese medicine and ensure the "uniformity and stability" of traditional Chinese medicine quality has become an important problem faced by the modernization and standardization of traditional Chinese medicine. Systematic quality management and control have also become an urgent need for the development of the traditional Chinese medicine industry.

[0003] Guchang Zhixie Pills is one of the advantageous traditional Chinese medicine varieties of "Qin medicine". Based on the "Wumei Pills" in Zhang Zhongjing's "Treatise on Febrile and Miscellaneous Diseases", it was later modified according to the causes and locations of intermittent dysentery, chronic dysentery, intestinal mass, etc. It is a national protected traditional Chinese medicine variety with broad application value. The whole formula consists of six ingredients: smoked plum flesh, Coptis chinensis, dried ginger, Aucklandia lappa, Corydalis yanhusuo, and poppy shell, and has the effects of regulating the liver and spleen, and astringing the intestine to stop pain. It has good curative effects in treating diarrhea and abdominal pain, and chronic non-specific ulcerative colitis clinically. The monarch drug smoked plum flesh in the formula needs to be softened by water or steamed and then de-nucleated during production feeding. Since the processing technology of smoked plum flesh is still in the traditional empirical regulation stage and lacks clear technical parameters, it is difficult to ensure the stability and uniformity of the quality of smoked plum flesh, and then affect the quality consistency of the finished Guchang Zhixie Pills. Therefore, the research group previously used the entropy weight method combined with the analytic hierarchy process and the backpropagation neural network to optimize the best processing technology of smoked plum flesh as adding 22.5% water, moistening for 1 h, and steaming for 30 min.

[0004] The mixing process is a critical operational unit in the production of Guchang Zhixie Pills. The mixing process involves numerous material properties, and variations in mixing time can impact the quality of intermediates and the final product. Fluctuations in these critical properties directly affect the uniformity and stability of subsequent production stages and the final product's quality. Traditional mixing processes are typically conducted under fixed process parameters, making it impossible to monitor quality changes in real time. Furthermore, quality differences between different batches are difficult to accurately identify and analyze. Current production relies primarily on process parameter management and quality standard indicator analysis, lacking rapid analytical methods. This makes it impossible to obtain real-time information on changes in material quality during the mixing process and between batches, and hinders timely responses to situations arising during mixing. Therefore, there is an urgent need to establish a mixing process monitoring method. Summary of the Invention

[0005] To address the aforementioned issues, this invention discloses a quality control method for the mixing process of Guchangzhixie Pills based on NIRS spectroscopy combined with MSPC technology. By collecting NIRS spectra of multiple batches of Guchangzhixie Pills during the mixing process, information is extracted using MSPC technology to establish a quality monitoring method for the mixing process of Guchangzhixie Pills based on NIRS spectroscopy, providing a reference for achieving dynamic quality control of the Guchangzhixie Pills production process.

[0006] The objective of this invention is achieved through the following technical solution.

[0007] A quality control method for the mixing process of a traditional Chinese medicine for treating diarrhea based on NIRS spectroscopy combined with MSPC technology, characterized by the following steps: (1) Raw material pretreatment: Weigh out the following medicinal materials according to the prescription ratio: dried plum pulp, Coptis chinensis, dried ginger, costus root, Corydalis yanhusuo and poppy shell. The dried plum pulp is soaked in water and then steamed. All medicinal materials are pulverized into powder of the target particle size. (2) Mixing process: The medicinal powders were mixed and periodically inverted. Samples were collected at different time points and spatial locations during the mixing process. (3) NIRS spectral acquisition: using a diffuse reflectance sampling device at 4000–10000 cm⁻¹ -1 Collect spectral data of the sample within the spectral range; (4) MSPC modeling: The three-dimensional spectral data matrix X(I×J×K) is reduced to a two-dimensional matrix X(IJ×K) by variable expansion, where I represents the batch, J represents the time point, and K represents the spectral variables. An MSPC model was built using training batches B1–B6, and principal component scores PC1 and Hotelling T were calculated. 2 and DModX statistic; The control limits for the three types of statistics were set at the mean ± 3 times the standard deviation. (5) Process monitoring: Real-time acquisition of new batch spectral data and input into the MSPC model. If PC1 and Hotelling T... 2 If any statistic in DModX exceeds the control limit, the process is deemed abnormal.

[0008] Furthermore, in the above quality control method, step (1): The amount of water added to the dried plum pulp is 15–30% of the weight of the dried plum pulp; The soaking time is 0.5–2 hours; Steaming time is 20–40 minutes.

[0009] Furthermore, in the above quality control method, step (3): NIRS resolution set to 4–64 cm -1 ; The number of scans is set to 16–128.

[0010] Furthermore, in the above quality control method, step (4): The control limits are set by the mean ± k times the standard deviation, where k ≥ 2; The principal component score uses the first principal component PC1 as the core monitoring indicator.

[0011] Furthermore, in the above quality control method, the abnormality types in step (5) include: The particle size of the pulverized medicinal materials is abnormal and the mixing time deviates from the set range.

[0012] Furthermore, the aforementioned quality control methods also include a PLS quantitative prediction model for key components: Spectral preprocessing methods are used to eliminate interference from powder physical properties; A dynamic content prediction model for new chlorogenic acid, chlorogenic acid, magnoflorine, noscapine hydrochloride, berberine, and berberine hydrochloride was established based on the PLS algorithm.

[0013] Furthermore, in the above-mentioned quality control method, the spectral preprocessing includes the SNV or MSC method; The number of latent variable factors is 5–14.

[0014] This invention also discloses a quality control system for the mixing process of a traditional Chinese medicine for constipation and diarrhea, comprising: NIRS spectrometer, equipped with diffuse reflectance sampling device; The data processing module is used to execute any of the above quality control methods.

[0015] The present invention also discloses a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of any of the above-described quality control methods.

[0016] The present invention also discloses an electronic device, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the program to implement the steps of any of the above-described quality control methods. Compared with existing technologies, the present invention has the following advantages and beneficial effects: (1) By using NIRS spectroscopy and MSPC technology to monitor the mixing process of Guchangzhixie pills, it is possible to monitor the changes in the particle size of the raw materials and the abnormal fluctuations in the process caused by mixing failures in a timely manner. This helps to promptly identify batch-to-batch quality abnormalities and process problems in the mixing process, guide the new batch to make corresponding adjustments, and provide real-time feedback for process optimization.

[0017] (2) An innovative NIRS quantitative prediction model based on the PLS algorithm was established, which realized the dynamic prediction of key chemical quality attributes in the mixing process of Guchangzhixie Pill, and made up for the lack of quantitative analysis of the mixing process of a single batch. Attached Figure Description

[0018] Figure 1 A schematic diagram illustrating two methods of dimensionality reduction from 3D data X to 2D data: batch expansion and variable expansion. Figure 2 Original NIRS plots (B1 ~ B6) of samples from the mixing process of Guchangzhixie Pills; Figure 3 Principal component score diagram of the B4 mixing process; Figure 4 Monitoring charts for the mixing process of Guchangzhixie Pills: A. PC1 control chart; B. Hotelling T 2 Control charts; C.DModX control charts; Figure 5 Normal batch B7 ~ B8 monitoring chart, A. PC1 control chart; B. Hotelling T 2 Control charts; C.DModX control charts; Figure 6 Monitoring charts for batches with abnormal material feeding (B9~B12): A. PC1 control chart; B. Hotelling T 2 Control charts; C.DModX control charts; Figure 7 Monitoring charts for batches B13~B14 with process anomalies: A. PC1 control chart; B. Hotelling T 2 Control charts; C.DModX control charts; Figure 8 Curves showing the changes in the content of each chemical component during the mixing process (B3); Figure 9 Latent variable factors for different spectral preprocessing methods; Figure 10 Correlation diagram between predicted and reference values; Figure 11 Comparison of NIRS validation results with reference method results. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below. However, it should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of the invention. Furthermore, descriptions of well-known structures and technologies are omitted in the following description to avoid unnecessarily obscuring the concept of the invention. All raw materials used in the embodiments of this invention are commercially available.

[0020] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. The present invention will now be described in detail with reference to the embodiments.

[0021] Definitions: NIRS stands for Near-Infrared Spectroscopy, a technique that analyzes the composition of a sample by analyzing the absorption of near-infrared light by the substance.

[0022] MSPC stands for Multivariate Statistical Process Control, which uses multivariate statistical models to monitor abnormal situations in the production process.

[0023] PLS stands for Partial Least Squares, a partial least squares regression algorithm used to establish a quantitative relationship model between spectral density and chemical composition content.

[0024] DModX: The full English name is Distance to Model in X-space, which is used to characterize the degree to which a sample deviates from the principal component model.

[0025] Hotelling T 2 The Chinese translation is "Hotlin T". 2 Statistics are used to measure the degree of variation of a sample in the principal component space.

[0026] SNV stands for Standard Normal Variate, which can eliminate spectral scattering interference.

[0027] MSC stands for Multiplicative Scatter Correction, which reduces the influence of powder physical properties on the spectrum.

[0028] Number of latent variable factors: The number of latent variables in the PLS model determines the model complexity (optimized to 5-14 in the example).

[0029] 3D variable expansion: A method to reorganize the three-dimensional matrix of batch (I) × time (J) × spectral variable (K) into a two-dimensional matrix (IJ×K).

[0030] Demixing phenomenon: the stratification of materials due to density differences during the mixing process.

[0031] 1. Types and sources of raw materials: 1.1 Test Drugs: Ume plum, Coptis chinensis, dried ginger, Aucklandia lappa, and Corydalis yanhusuo were all purchased from the Hongxinglin Traditional Chinese Medicine Wholesale Market in Xi'an. Poppy shells were provided by the Shaanxi University of Traditional Chinese Medicine Pharmaceutical Factory. All drugs were identified by Professor Yan Yonggang of Shaanxi University of Traditional Chinese Medicine as meeting the relevant requirements of the 2020 edition of the Chinese Pharmacopoeia. The pits of the ume plums were removed, and the dried pulp was stored in resealable bags for later use. (Note: The processing method for the ume plum pulp was based on previous research by the research group. The group previously used entropy weight method combined with analytic hierarchy process and backpropagation neural network to optimize the processing technology of ume plum pulp: adding 22.5% water, soaking for 1 hour, and steaming for 30 minutes.) 1.2 Instruments Antaris™ II FT-NIR Analyzer (Thermo Fisher Scientific), Agilent 1200 HPLC (Agilent Technologies, Inc.), NewCLassic MS-S analytical balance (Shanghai Mettler Toledo Instruments Co., Ltd.), KQ-250DE CNC ultrasonic cleaner (Kunshan Ultrasonic Instruments Co., Ltd.), JY502 electronic balance (Shanghai Puchun Metrology Instruments Co., Ltd.), GZX-DH.500BS electric thermostatic drying oven (Shanghai Yuejin Medical Instruments Co., Ltd.), high-speed multi-functional pulverizer (Yongkang Boou Hardware Products Co., Ltd.), Microtrac S3500 laser particle size analyzer (Dachong Huajia Commercial Co., Ltd.).

[0032] 2. Raw Material Dosage and Process Flow: Weigh out the following medicinal materials according to the prescribed proportions: dried plum pulp, Coptis chinensis, dried ginger, Aucklandia lappa, Corydalis yanhusuo, and poppy shell. Grind each material separately into a fine powder and place it in a sample cup with a uniform bottom thickness. Mix for 30 minutes, inverting the cup 5 times per minute at a uniform speed. Collect samples at different time points and sampling points (upper, middle, and lower) every 5 minutes of mixing, for a total of 18 sampling points per batch. (Note: This experiment simulates the actual production process of Gu Chang Zhi Xie Wan on a laboratory scale.) 3. Experimental Design A total of 14 batches of experiments were designed for the mixing process of the Guchangzhixie Pill (Table 1), including normal operation batches (B1 ~ B8), batches with abnormal feeding (B9 ~ B12), and batches with abnormal processes (B13, B14). Among them, B1 ~ B6 (training set) were used to build the MSPC model, and B7 and B8 were normal validation batches. Based on actual production experience, manual settings may affect the raw material properties and operating parameters of abnormal processes, which were used to test the monitoring capability of the model. Other operations were consistent with those of the normal operation batches.

[0033] 3.1 Establishment of the MSPC Model 3.1.1 Data Processing: The multi-batch mixing process data of Guchangzhixie Pills consists of a three-dimensional data matrix X (I×J×K) composed of three dimensions: batch (I), time (J), and process variables (number of spectral variables, K). The MSPC model is based on two-dimensional data; therefore, before constructing the model, the three-dimensional data X needs to be reduced to two dimensions. This is commonly done through batch-by-batch expansion and variable-by-variable expansion. Figure 1 As can be seen from the matrix form, the batch-based expansion method can show the differences between batches at a macro level, but it cannot show the differences of the same batch at any point in the entire production process; the variable-based expansion method can monitor all points in time, thus enabling monitoring of the entire production process.

[0034] Monitoring any point in time during batch production, the three-dimensional matrix is ​​expanded using variable expansion, retaining the process variable dimension. The batch and time dimensions are merged to obtain a two-dimensional matrix X (IJ×K). Each row represents the spectral variable at a specific time point for a specific batch, and each column represents the absorbance at the corresponding spectral variable at all sampling times for all batches. Using the spectral sampling time point within a batch as a process progress indicator, a PLS regression is constructed between X and the corresponding Y variable (time). An MSPC model is then established, and principal component scores and Hotelling T are calculated. 2Three types of monitoring charts using the DModX statistic as an indicator are used to calculate the statistics for batches under normal operating conditions at each time point and plot the trend of the statistics over time. Among them, the principal component of the model explains the process variable information more completely and can intuitively represent the process change trend. In practical applications, the first principal component (PC1) is often used to monitor the process. The mean and standard deviation (Std.Dev) of the score of the batch under normal operating conditions at each time point are calculated, and the mean ± 3Std.Dev is used as the upper and lower limits of the score control chart.

[0035] Data analysis was performed using SIMCA 14.1 and MATLAB R2023b software, and visualization was performed using Orign software.

[0036] 3.1.2 Model Establishment: Calculate PC1 and HotellingT at each batch time point during the training set mixing process. 2 Using the DModX statistic, plot the trajectory of the statistic over time, monitor new batches by setting control limits, and establish an MSPC model.

[0037] 3.2 Application of the MSPC Model The robustness of the established MSPC model was verified by monitoring new normal operating batches (B7 and B8). Batches with abnormal raw material properties (B9 to B12) were then incorporated into the established MSPC model to examine whether the model could detect abnormalities in the particle size of the fed medicinal materials. The established MSPC model was used to monitor batches with abnormal processes (B13 and B14) to examine whether the model could detect the abnormalities.

[0038] 3.3 Establishment of the Prediction Model Taking the mixing process of Guchangzhixie Pill as the research object, a NIRS quantitative prediction model for new chlorogenic acid, chlorogenic acid, magnoflorine, noscapine hydrochloride, berberine hydrochloride, and berberine hydrochloride was established based on the PLS algorithm. This model enables dynamic monitoring and prediction of chemical indicators in the mixing process, which can effectively identify and adjust quality fluctuations in the production process and reduce the occurrence of unqualified products.

[0039] Example 1. Instrument resolution evaluation: 2.0 g of powder intermediate sample was placed in a rotating cup, and spectral resolutions of 4, 8, 16, 32, and 64 cm⁻¹ were set. -1 The number of scans was 32, and other scanning parameters remained constant, ranging from 4000 to 10000 cm. -1 Within the range, NIRS measurements were collected at different spectral resolutions, repeated six times, and the relative standard deviation (RSD) of the average absorbance of the samples was calculated. The results showed that when the resolution was 16 cm⁻¹... -1At that time, the sample's RSD value was the lowest at 0.8011%. The smaller the RSD, the higher the precision of the analytical test results. Therefore, the optimal resolution for the sample was determined to be 16 cm⁻¹. -1 (Table 2).

[0040] 2. Examination of the number of instrument scans: The number of spectral scans were set to 16, 32, 64, and 128, with a resolution of 16 cm⁻¹. -1 Other scanning parameters remained constant, ranging from 4000 to 10000 cm⁻¹. -1 Within the range, NIRS at different spectral resolutions was collected and repeated 6 times to calculate the RSD of the average absorbance of the sample. The results showed that the RSD value of the sample was the lowest at 0.6684% when the number of scans was 16. Therefore, the optimal number of scans for the sample was determined to be 16 (Table 3).

[0041] 3. NIRS Data Acquisition: An appropriate amount of sample was taken from the mixing process of the Guchang Zhixie Pills. NIRS data of the powder from the Guchang Zhixie Pills mixing process was acquired using a near-infrared spectroscopy (NIRS) instrument with air as the background. The sample was scanned in integrating sphere mode, equipped with an integrating sphere diffuse reflectance sampling device, and the resolution was set to 16 cm⁻¹. -1 The number of scans was 16, and the spectral scanning range was 4000 ~ 10000 cm⁻¹. -1 Each sampling point was scanned three times to obtain the average spectrum, which was then used to build the model.

[0042] 4. Raw Spectroscopic Analysis: NIRS carries valuable chemical information about powder samples, which may be related to light scattering and the compounds that affect these properties. This information allows for qualitative or quantitative analysis of the sample. Figure 2 The raw NIRS spectra of samples from the mixing process of the Guchangzhixie Pills during normal operation are shown, with the wavelength range of 4000–5700 cm⁻¹. -1 The absorbance changes significantly throughout the mixing process, directly reflecting the changes in the physical state of the system. To eliminate interference from factors such as powder sample size and uniformity, the SNV method was used to preprocess the original spectrum.

[0043] 5. Process spectral PCA: Taking batch B4 as an example, process spectral PCA analysis was performed, and the results are shown in [the table below]. Figure 3Cross-validation yielded two principal components with a cumulative contribution rate of 93.9%. The first principal component (PComponent I) contributed 82.9%, and the second PComponent I contributed 11.0%. The principal component data formed a Z-shaped trajectory, reflecting the trend of NIRS changes over time during the mixing process. This is likely due to the combined effects of multiple components in the mixture system. The mixing time points for this batch mainly clustered into two categories: the upper left region (first 10 minutes) and the lower right region (15-30 minutes). Significant differences were observed in the first principal component, indicating a significant difference in the quality of the Gu Chang Zhi Xie Wan (a traditional Chinese medicine formula) with increasing mixing time. The larger changes in PC1 scores during the first 10 minutes of mixing suggest substantial spectral differences, likely related to the properties of the raw materials. From 10 to 25 minutes, the PC1 scores increased gradually but less significantly, indicating decreasing spectral differences, possibly related to mixing uniformity, leading to a gradual stabilization of the mixing process. From 25 to 30 minutes, the larger changes in PC1 scores suggest increased spectral differences, possibly indicating demixing.

[0044] 6. Establishment of the MSPC Model 6.1. Model Establishment: An MSPC model was established using normal batches B1 to B6 as the training set and B7 to B14 as the validation set. PC1 and Hotelling T were calculated at each batch time point during the training set mixing process. 2 Using the DModX statistic, plot the trajectory of the statistic over time. Monitor new batches by setting control limits. Three types of control charts are available. Figure 4 The principal component score control plot shows the trend of a batch of samples' scores on PC1 over time. PC1 explains 56.2% of the spectral variables, containing the vast majority of spectral information and intuitively reflecting the spectral changes throughout the mixing process. Hotelling T 2 Control charts and DModX control Figure 2 The two have complementary effects; when anomalies in process variables cannot be expressed through score plots and Hotelling T, 2 When using control charts for monitoring, the DModX statistic may be considered.

[0045] 6.2. Control Chart: Principal component score control charts reflect the spatial distribution of the principal component score vectors, providing a detailed picture of the fluctuations of each principal component with each batch. If the information contained in the principal component explanatory variables is relatively complete, it can intuitively characterize process trend changes.

[0046] 6.2.1. Hotelling T 2 Control chart: T 2The statistic represents the distance between the sampling point and the mean in the feature space of the principal components, indicating the degree to which each sample deviates from the principal component model in terms of amplitude and trend. The sampling point Hotelling T at time i... 2 The calculation formula is as follows: 6.2.2 DModX Control Chart: The DModX statistic represents the distance of the sampled data from the independent variable X space to the principal component model. It is a measure of changes in external data to the model and represents the changes in the sampled points that are not explained by the model. The formula for calculating the DModX value at time k is as follows: 7. Application of the MSPC model 7.1. Normal Batch Monitoring: The established MSPC model was used to monitor new normal operation batches (B7 and B8) to verify the robustness of the model and examine whether false alarms would occur during the monitoring process. The results are shown in […]. Figure 5 Both B7 and B8 fall within the control limits of the Class 3 control charts, indicating that the changes in the mixed process fluctuate within an acceptable range. The batch operation is normal, and no false alarms have occurred, demonstrating that the model has good robustness.

[0047] 7.2. Monitoring of Abnormal Feed Batches: The abnormal batches (B9 ~ B12) of raw materials were input into the established MSPC model, and the PC1 score and Hotelling T of these four abnormal batches at each time point during the mixing process were calculated. 2 Values ​​and DModX values, see control chart. Figure 6 The results showed that as the mixing time increased, B9 to B12 all exceeded the control limits of the three types of control charts. The MSPC model detected the abnormality of the particle size of the fed medicinal materials, indicating that the three types of control charts are all quite sensitive to the feeding abnormalities of the mixing system, and the particle size of the medicinal materials has a significant impact on the mixing process.

[0048] 7.3 Process Anomaly Batch Monitoring: B13 and B14 represent batches where mixing was terminated after 10 min and 15 min, respectively. Figure 7 These are the PC1 score control charts for these two batches, and Hotelling T. 2Control charts and DModX control charts. As can be seen from the charts, at the 10-minute mixing termination point, all three control charts detected anomalies. The PC1 score control chart exceeded the control limit around 20 minutes, which was delayed compared to the actual mixing termination time. This may be because the interference from mixing termination on the process data changes slowly, exhibiting a certain lag. Around 25 minutes, the PC1 score trajectory returned to within the control limit, indicating that the change in the system's material composition was relatively small after mixing termination. Hotelling T 2 The control chart exceeded the control limits at approximately 13 minutes, earlier than the actual mixing termination time, indicating that Hotelling T 2 The control charts are quite sensitive to abnormal feed rates in the mixing system. The DModX control chart exceeded the control limits around 15 minutes, which is close to the actual mixing termination time. At the 15-minute mixing termination time, the PC1 score control chart and the Hotelling T... 2 All control charts exceeded the control limits at 20 minutes, while the DModX control chart exceeded the control limits around 25 minutes. Compared with the B13 control chart, this indicates that there are differences in the composition changes of materials during the mixing process in different batches. When using the MSPC model to monitor the mixing process, it is necessary to analyze all three control charts together. If all three statistics are within the control limits, the batch is considered to be under control; otherwise, it is considered an abnormal situation.

[0049] 8. Establishment of the prediction model 8.1. Determination of Indicator Components: Samples were collected from different sampling points during the mixing process of the Guchangzhixie Pill. NIRS data were collected under optimal sampling conditions. A total of 108 average spectra were collected from 6 batches of normal operating samples (B1 ~ B6) for modeling. The contents of neochlorogenic acid, chlorogenic acid, magnoflorine, noscapine hydrochloride, berberine, and berberine hydrochloride in each sample during the mixing process were determined by HPLC.

[0050] 8.2. High-performance liquid chromatography (HPLC) conditions: Column: Agilent 5 TC-C18(2) (250 × 4.6 mm, 5 μm); Mobile phase: 0.1% phosphoric acid aqueous solution (A) - acetonitrile (B); gradient elution (0 ~ 15 min, 4% ~ 25% B; 15 ~ 25 min, 25% B; 25 ~ 31 min, 25% ~ 30% B; 31 ~ 35 min, 30% ~ 12% B; 35 ~ 40 min, 12% ~ 4% B); flow rate: 1 mL·min -1 The injection volume was 10 μL, the column temperature was 20 ℃, and the detection wavelength was 220 nm.

[0051] 8.3. Establishment of the PLS Quantitative Model: Using MATLAB R2023b software, the Kennard-Stone (KS) algorithm was used to divide the 108 modeling data collected during the mixing process into 75 training sets and 33 prediction sets. The training sets were used to establish a quantitative analysis model of the mixing process of the Guchang Zhixie Pill, and the prediction sets were used to predict the accuracy of the established model. The content distribution range of each component in the training and prediction sets is shown in Table 4. As can be seen from the table, the content distribution range of the 6 components in the prediction set is within the training set, which meets the requirements for establishing the prediction model. The change curves of the content of each chemical component in the mixing process are shown in Table 4. Figure 8 As can be seen from the figure, as the mixing time increases, the content of each chemical component gradually decreases overall, and gradually stabilizes around 20 minutes.

[0052] 8.4. Selection of Spectral Preprocessing Method: A NIRS quantitative model of the mixing process of Guchang Zhixie Pills was established using the PLS algorithm. To improve the model performance, Unscrambler×10.4 software was used for spectral data preprocessing and model calculation. The original spectra of the Guchang Zhixie Pills mixing process samples are shown below. Figure 2 This study examines the impact of preprocessing methods such as multivariate scattering correction (MSC), standard normal transformation (SNV), convolutional smoothing (SG), first derivative (1stD), and second derivative (2ndD) on model performance. The root mean square error of the correction set (RMSEC), the correlation coefficient of the correction set (Rc), the root mean square error of the prediction set (RMSEP), the correlation coefficient of the prediction set (Rp), and the root mean square error of cross-validation (RMSECV) are used as evaluation metrics for the model. These metrics provide a comprehensive evaluation of the quantitative model, reflecting important information related to the model's fitting or predictive performance, thereby establishing the optimal predictive model. The formulas for calculating Rc, RMSEC, Rp, and RMSEP are as follows: The effects of different preprocessing methods on the quantitative models of each component are shown in Tables 5 and 5 (continued). By comparing different preprocessing modeling methods, the results show that after SNV preprocessing, neochlorogenic acid, chlorogenic acid, noscapine hydrochloride, berberine, and berberine hydrochloride have relatively small RMSEC, RMSECV, and RMSEP, with Rc all greater than 0.85 and Rp all greater than 0.81, indicating good PLS model performance. Magnolia flower alkaloid also showed good PLS model performance after MSC preprocessing. Data comparison revealed that after 1stD and 2ndD preprocessing, each component had the smallest RMSEC and the largest Rp, but a larger RMSEP and a smaller Rp, leading to a decrease in model quality. This may be because derivative preprocessing amplifies high-frequency noise in the spectrum, causing the model to overfit the training set data and perform poorly on the prediction set. Therefore, SNV preprocessing is recommended for neochlorogenic acid, chlorogenic acid, noscapine hydrochloride, berberine, and berberine hydrochloride, while MSC is the optimal preprocessing method for magnolia flower alkaloid to improve model prediction performance.

[0053] 8.5. Screening of Latent Variable Factors: The number of latent variable factors in the models was determined through cross-validation. The results showed that the optimal number of latent variable factors for the chlorogenic acid and magnoflorine PLS quantitative models was 14, with corresponding RMSECV values ​​of 0.3427 and 0.3608, respectively; the optimal number of latent variable factors for the chlorogenic acid PLS quantitative model was 11, with a corresponding RMSECV value of 0.1323; the optimal number of latent variable factors for the noscapine hydrochloride PLS quantitative model was 12, with a corresponding RMSECV value of 0.0852; and the optimal number of latent variable factors for the berberine and berberine hydrochloride PLS quantitative models was 13, with corresponding RMSECV values ​​of 3.2781 and 4.6869, respectively. (See details...) Figure 9 .

[0054] 8.6. Establishment of PLS ​​Models: Based on the optimal spectral preprocessing method and the optimal number of latent variable factors selected above, PLS quantitative models for neochlorogenic acid, chlorogenic acid, magnoflorine, noscapine hydrochloride, berberine, and berberine hydrochloride were established. The quantitative models are shown below. Figure 10As shown in the figure, the predicted values ​​of the quantitative models have a good fit with the actual reference values. Except for chlorogenic acid (RC, 0.8536), the RCs of the quantitative calibration models for the other five chemical components are all greater than 0.9, indicating that the quantitative model for chlorogenic acid has a good fit to the calibration set data and moderate prediction accuracy on the calibration set. The quantitative models for the other five chemical components have a good fit to the calibration set data and high prediction accuracy on the calibration set. The RPs of the quantitative calibration models for all six chemical components are greater than 0.8, indicating that the models have good predictive performance and can be used for quantitative analysis of actual samples. In addition, the RMSEC values ​​of the six chemical components are small, indicating that the prediction error of the models on the calibration set is small and the fitting effect is good. Moreover, the RMSEP and RMSECV are close, indicating that the performance of the models on the prediction set is basically consistent with the cross-validation, and they have a certain degree of robustness.

[0055] 8.7. Validation of the PLS model: Data from a new batch of samples (B7) collected during the mixing process of the Guchangzhixie pills were substituted into the established PLS quantitative model to predict the content of chemical components. The results are shown in [Figure number missing]. Figure 11 It can be seen that the predicted values ​​of the samples given by the model are basically consistent with the trend of the reference values. The average relative errors of neochlorogenic acid, chlorogenic acid, magnoflorine, noscapine hydrochloride, berberine, and berberine hydrochloride are 0.8444%, 1.3609%, 5.7570%, 6.2769%, 9.8170%, and 9.3683%, respectively, all less than 10%, indicating that the prediction results of the established model are reliable. This shows that the combination of NIRS spectroscopy and PLS model can be used to determine the content of six chemical components in the mixing process of Guchangzhixie Pill.

[0056] In summary, the effectiveness of the method was verified through 14 batches of mixed experiments (including batches with normal operation, abnormal feeding, and process abnormalities): First, the NIRS parameters were optimized (resolution 16 cm⁻¹, 16 scans) and SNV / MSC preprocessed the spectra; based on batches B1-B6, an MSPC model was established, successfully detecting statistical out-of-limits in batches B9-B12 (particle size abnormality) and B13-B14 (mixing termination); at the same time, a PLS quantitative model (6 components including neochlorogenic acid) was established, and the average relative error between the predicted value and the HPLC reference value was <10% (e.g., the error of 9.37% for berberine hydrochloride in batch B7), confirming the reliability of dynamic quality control.

[0057] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the scope of protection of the present invention. Therefore, any changes and modifications made to the embodiments described herein based on the innovative concept of the present invention, or equivalent structural or procedural transformations made using the content of the present invention specification, directly or indirectly applying the above technical solutions to other related technical fields, are all included within the scope of protection of the present invention patent.

Claims

1. A quality control method for the mixing process of a traditional Chinese medicine for treating diarrhea based on NIRS spectroscopy combined with MSPC technology, characterized in that... Includes the following steps: (1) Raw material pretreatment: Weigh out the following medicinal materials according to the prescription ratio: dried plum pulp, Coptis chinensis, dried ginger, costus root, Corydalis yanhusuo and poppy shell. The dried plum pulp is soaked in water for 0.5-2 hours, the amount of water is 15-30% of the mass of the dried plum pulp, and the soaking time is 0.5-2 hours. Then it is steamed for 20-40 minutes. All medicinal materials are pulverized into powder of the target particle size. (2) Mixing process: Mix the medicinal powder and invert it periodically. During the mixing process, samples are collected at different time points and spatial locations, including the upper, middle and lower parts. (3) NIRS spectral acquisition: using a diffuse reflectance sampling device, in Within the spectral range, with The spectral data of the sample were acquired using high resolution and 16–128 scan parameters. (4) MSPC modeling: The three-dimensional spectral data matrix X (I×J×K) is reduced to a two-dimensional matrix X (IJ×K) by variable expansion, where I represents the batch, J represents the time point, and K represents the spectral variables. An MSPC model is built using the training set batches, and the principal component score PC1, Hotelling T², and DModX statistic are calculated. PC1 is the first principal component and explains more than 56.2% of the spectral variables, serving as the core monitoring indicator. The control limits for the three types of statistics are set at mean ± 3 standard deviations. (5) Process monitoring: Real-time acquisition of spectral data of new batches and input into the MSPC model. If any of the statistics of PC1, Hotelling T² and DModX exceeds the control limit, the process is judged to be abnormal. Abnormality types include abnormal particle size of medicinal materials and deviation of mixing time from the set range.

2. The method according to claim 1, characterized in that, It also includes PLS quantitative prediction models for key components: Spectral preprocessing methods are used to eliminate interference from powder physical properties; A dynamic content prediction model for new chlorogenic acid, chlorogenic acid, magnoflorine, noscapine hydrochloride, berberine, and berberine hydrochloride was established based on the PLS algorithm.

3. The method according to claim 2, characterized in that: The spectral preprocessing includes SNV or MSC methods; The number of latent variable factors is 5–14.

4. A quality control system for the mixing process of a traditional Chinese medicine for intestinal congestion and diarrhea, characterized in that, include: NIRS spectrometer, equipped with diffuse reflectance sampling device; A data processing module for performing the method according to any one of claims 1-3.

5. A computer-readable storage medium storing a computer program, characterized in that, When the program is executed by a processor, it implements the steps of the method according to any one of claims 1-3.

6. An electronic device comprising a memory, a processor, and a computer program stored in the memory, characterized in that, When the processor executes the program, it implements the steps of the method according to any one of claims 1-3.

Images