[Key technologies and applications of industrial big data in manufacturing of Chinese medicine].

Along with the striding of the Chinese medicine(CM) manufacturing toward the Industry 4.0, some digital factories have accumulated lightweight industrial big data, which become part of the enterprise assets. These digital assets possess the possibility of solving the problems within the CM production system, like the Sigma gap and the poverty of manufacturing knowledge. From the holistic perspective, a three-tiered architecture of CM industrial big data is put forward, and it consists of the data integration layer, the data analysis layer and the application scenarios layer. In data integration layer, sensing of CM critical quality attributes is the key technology for big data collection. In data analysis and mining layer, the self-developed iTCM algorithm library and model library are introduced to facilitate the implementation of the model lifecycle methodologies, including process model development, model validation, model configuration and model maintenance. The CM quality transfer structure is closely related with the connection mode of multiple production units. The system modeling technologies, such as the partition-integration modeling method, the expanding modeling method and path modeling method, are key to mapping the structure of real manufacturing system. It is pointed out that advance modeling approaches that combine the first-principles driven and data driven technologies are promising in the future. At last, real-world applications of CM industrial big data in manufacturing of injections, oral solid dosages, and formula particles are presented. It is shown that the industrial big data can help process diagnosis, quality forming mechanism interpretations, real time release testing method development and intelligent product formulation design. As renewable resources, the CM industrial big data enable the manufacturing knowledge accumulation and product quality improvement, laying the foundation of intelligent manufacturing.

[Quality by design approaches for pharmaceutical development and manufacturing of Chinese medicine].

The pharmaceutical quality was built by design, formed in the manufacturing process and improved during the product's lifecycle. Based on the comprehensive literature review of pharmaceutical quality by design (QbD), the essential ideas and implementation strategies of pharmaceutical QbD were interpreted. Considering the complex nature of Chinese medicine, the "4H" model was innovated and proposed for implementing QbD in pharmaceutical development and industrial manufacture of Chinese medicine product. "4H" corresponds to the acronym of holistic design, holistic information analysis, holistic quality control, and holistic process optimization, which is consistent with the holistic concept of Chinese medicine theory. The holistic design aims at constructing both the quality problem space from the patient requirement and the quality solution space from multidisciplinary knowledge. Holistic information analysis emphasizes understanding the quality pattern of Chinese medicine by integrating and mining multisource data and information at a relatively high level. The batch-to-batch quality consistence and manufacturing system reliability can be realized by comprehensive application of inspective quality control, statistical quality control, predictive quality control and intelligent quality control strategies. Holistic process optimization is to improve the product quality and process capability during the product lifecycle management. The implementation of QbD is useful to eliminate the ecosystem contradictions lying in the pharmaceutical development and manufacturing process of Chinese medicine product, and helps guarantee the cost effectiveness.

[Application of quality by design in granulation process for ginkgo leaf tablet (⠡)： identification of critical quality attributes].

Quality by design (QbD) highlights the concept of "begin with the end", which means to thoroughly understand the target product quality first, and then guide pharmaceutical process development and quality control throughout the whole manufacturing process. In this paper, the Ginkgo biloba granules intermediates were taken as the research object, and the requirements of the tensile strength of tablets were treated as the goals to establish the methods for identification of granules' critical quality attributes (CQAs) and establishment of CQAs' limits. Firstly, the orthogonal partial least square (OPLS) model was adopted to build the relationship between the micromeritic properties of 29 batches of granules and the tensile strength of ginkgo leaf tablets, and thereby the potential critical quality attributes (pCQAs) were screened by variable importance in the projection (VIP) indexes. Then, a series of OPLS models were rebuilt by reducing pCQAs variables one by one in view of the rule of VIP values from low to high in sequence. The model performance results demonstrated that calibration and predictive performance of the model had no decreasing trend after variables reduction. In consideration of the results from variables selection as well as the collinearity test and testability of the pCQAs, the median particle size (D50) and the bulk density (Da) were identified as critical quality attributes (CQAs). The design space of CQAs was developed based on a multiple linear regression model established between the CQAs (D50 and Da) and the tensile strength. The control constraints of the CQAs were determined as 170 µm< D50<500 µm and 0.30 g•cm⁻³

[Application of quality by design in granulation process for ginkgo leaf tablet (â )： comprehensive characterization of granule properties].

In this paper, the granules intermediate prepared from the wet granulation process of ginkgo leaf tablet were taken as the research object, and then the stackability, homogeneity, flowability, compressibility and stability of granules were characterized by using micromeritics evaluation method. The physical fingerprint of granules were constructed by 16 indexes including bulk density, tapped density, span, width, relative homogeneity index, aspect ratio, Hausner ratio, angle of repose, granule flow time, inter-particle porosity, Carr index, specific surface area, pore volume, pore size distribution, loss on drying and hygroscopicity. Furthermore, compressibility parameters (i.e. index of parameter, index of parametric profile and index of good compression) were employed to analyze the compressibility characteristics of the granules. Two principal components (first principal component representing dimension parameter and second principal component representing morphology parameter), could be extracted from the physical fingerprint by the principal component analysis (PCA). The granules' physical fingerprint is of great importance to evaluate the batch-to-batch quality consistency of Ginkgo biloba granules and analyze the potential impacts of granules' quality attributes on product quality, which can provide guidance for the granules' quality control and process development..

[Application of quality by design in granulation process for Ginkgo leaf tablet (⠢)： process control strategy based on design space].

In this paper, under the guidance of quality by design (QbD) concept, the control strategy of the high shear wet granulation process of the ginkgo leaf tablet based on the design space was established to improve the process controllability and product quality consistency. The median granule size (D50) and bulk density (Da) of granules were identified as critical quality attributes (CQAs) and potential critical process parameters (pCPPs) were determined by the failure modes and effect analysis (FMEA). The Plackeet-Burmann experimental design was used to screen pCPPs and the results demonstrated that the binder amount, the wet massing time and the wet mixing impeller speed were critical process parameters (CPPs). The design space of the high shear wet granulation process was developed within pCPPs range based on the Box-Behnken design and quadratic polynomial regression models. ANOVA analysis showed that the P-values of model were less than 0.05 and the values of lack of fit test were more than 0.1, indicating that the relationship between CQAs and CPPs could be well described by the mathematical models. D50 could be controlled within 170 to 500 µm, and the bulk density could be controlled within 0.30 to 0.44 g•cm⁻³ by using any CPPs combination within the scope of design space. Besides, granules produced by process parameters within the design space region could also meet the requirement of tensile strength of the ginkgo leaf tablet..

[Application of quality by design in granulation process for ginkgo leaf tablet (⠣)： influence and control of raw materials' quality variation].

Raw materials' quality variation could affect the quality consistency of product and the clinical efficacy. In this paper, the high shear wet granulation (HSWG) process of the ginkgo leaf tablet was taken as the research object. Ginkgo biloba extracts and excipients microcrystalline cellulose collected from various sources and batches were used to simulate raw materials' quality variation. Real-time torque was recorded to analyze the viscosity of the wetting mass, and then by combining with physical fingerprint, the impact of physical quality variation of powders on granule properties could be investigated. Based on regime map thesis, whether the granules' nucleation mode was in mechanical dispersion regime was determined by calculating dimensionless parameters, which would lead to the unstable output in considerations of granule yield ratio and particle size distribution (PSD) curve. The orthogonal partial least square (OPLS) model was adopted to build the relationship between the micromeritic properties and the mediangranule size (D50) of Ginkgo biloba granules and then the critical material attributes (CMAs) were screened by variable importance in the projection (VIP) indexes. The results demonstrated that the properties of powders including hygroscopicity, angle of repose, Hausner ratio, Carr index, D10 and loss on drying affected the granule size. Besides, Ginkgo biloba granules were compressed into tablets. In view of tensile strength analysis, the raw materials' quality variation did not result in decrease of tensile strength of the ginkgo leaf tablets. The design space of critical quality attributes (CQAs) and the process design space which could cope with raw materials' quality variation were proved to be robust..

[Online endpoint detection algorithm for blending process of Chinese materia medica].

Blending process, which is an essential part of the pharmaceutical preparation, has a direct influence on the homogeneity and stability of solid dosage forms. With the official release of Guidance for Industry PAT, online process analysis techniques have been more and more reported in the applications in blending process, but the research on endpoint detection algorithm is still in the initial stage. By progressively increasing the window size of moving block standard deviation (MBSD), a novel endpoint detection algorithm was proposed to extend the plain MBSD from off-line scenario to online scenario and used to determine the endpoint in the blending process of Chinese medicine dispensing granules. By online learning of window size tuning, the status changes of the materials in blending process were reflected in the calculation of standard deviation in a real-time manner. The proposed method was separately tested in the blending processes of dextrin and three other extracts of traditional Chinese medicine. All of the results have shown that as compared with traditional MBSD method, the window size changes according to the proposed MBSD method (progressively increasing the window size) could more clearly reflect the status changes of the materials in blending process, so it is suitable for online application.

[Continual improvement of quantitative analytical method development of Panax notogineng saponins based on quality by design].

This study is aimed to propose a continual improvement strategy based on quality by design (QbD). An ultra high performance liquid chromatography (UPLC) method was developed to accomplish the method transformation from HPLC to UPLC of Panax notogineng saponins (PNS) and achieve the continual improvement of PNS based on QbD, for example. Plackett-Burman screening design and Box-Behnken optimization design were employed to further understand the relationship between the critical method parameters (CMPs) and critical method attributes (CMAs). And then the Bayesian design space was built. The separation degree of the critical peaks (ginsenoside Rg1 and ginsenoside Re) was over 2.0 and the analysis time was less than 17 min by a method chosen from the design space with 20% of the initial concentration of the acetonitrile, 10 min of the isocratic time and 6%•min⁻¹ of the gradient slope. At last, the optimum method was validated by accuracy profile. Based on the same analytical target profile (ATP), the comparison of HPLC and UPLC including chromatograph method, CMA identification, CMP-CMA model and system suitability test (SST) indicated that the UPLC method could shorten the analysis time, improve the critical separation and satisfy the requirement of the SST. In all, HPLC method could be replaced by UPLC for the quantity analysis of PNS.

[Physical fingerprint for quality control of traditional Chinese medicine extract powders].

The physical properties of both raw materials and excipients are closely correlated with the quality of traditional Chinese medicine preparations in oral solid dosage forms. In this paper, based on the concept of the chemical fingerprint for quality control of traditional Chinese medicine products, the method of physical fingerprint for quality evaluation of traditional Chinese medicine extract powders was proposed. This novel physical fingerprint was built by the radar map, and consisted of five primary indexes (i.e. stackablity, homogeneity, flowability, compressibility and stability) and 12 secondary indexes (i.e. bulk density, tap density, particle size<50 µm percentage, relative homogeneity index, hausner ratio, angle of repose, powder flow time, inter-particle porosity, Carr index, cohesion index, loss on drying, hygroscopicity). Panax notoginseng saponins (PNS) extract was taken for an example. This paper introduced the application of physical fingerprint in the evaluation of source-to-source and batch-to-batch quality consistence of PNS extract powders. Moreover, the physical fingerprint of PNS was built by calculating the index of parameters, the index of parametric profile and the index of good compressibility, in order to successfully predict the compressibility of the PNS extract powder and relevant formulations containing PNS extract powder and conventional pharmaceutical excipients. The results demonstrated that the proposed method could not only provide new insights into the development and process control of traditional Chinese medicine solid dosage forms.

[Near infrared analysis of blending homogeneity of Chinese medicine formula particles based on moving window F test method].

Blending uniformity is essential to ensure the homogeneity of Chinese medicine formula particles within each batch. This study was based on the blending process of ebony spray dried powder and dextrin(the proportion of dextrin was 10%),in which the analysis of near infrared (NIR) diffuse reflectance spectra was collected from six different sampling points in combination with moving window F test method in order to assess the blending uniformity of the blending process.The method was validated by the changes of citric acid content determined by the HPLC. The results of moving window F test method showed that the ebony spray dried powder and dextrin was homogeneous during 200-300 r and was segregated during 300-400 r. An advantage of this method is that the threshold value is defined statistically, not empirically and thus does not suffer from threshold ambiguities in common with the moving block standard deviatiun (MBSD). And this method could be employed to monitor other blending process of Chinese medicine powders on line.

[Research on modeling method to analyze Lonicerae Japonicae Flos extraction process with online MEMS-NIR based on two types of error detection theory].

To establish a rapid quantitative analysis method for online monitoring of chlorogenic acid in aqueous solution of Lonicera Japonica Flos extraction by using micro-electromechanical near infrared spectroscopy (MEMS-NIR). High performance liquid chromatography(HPLC) was used as reference method．Kennard-Stone (K-S) algorithm was used to divide sample sets, and partial least square(PLS) regression was adopted to establish the multivariate analysis model between the HPLC analysis contents and NIR spectra. The synergy interval partial least squares (SiPLS) was used to selected modeling waveband to establish PLS models. RPD was used to evaluate the prediction performance of the models. MDLs was calculated based on two types of error detection theory, on-line analytical modeling approach of Lonicera Japonica Flos extraction process was expressed scientifically by MDL. The result shows that the model established by multiplicative scatter correction(MSC) was the best, with the root mean square with cross validation(RMSECV), root mean square error of correction(RMSEC) and root mean square error of prediction(RMSEP) of chlorogenic acid as 1.707, 1.489, 2.362, respectively, the determination coefficient of the calibration model was 0.998 5, and the determination coefficient of the prediction was 0.988 1．The value of RPD is 9.468.The MDL (0.042 15 g•L⁻¹) selected by SiPLS is less than the original,which demonstrated that SiPLS was beneficial to improve the prediction performance of the model. In this study, a more accurate expression of the prediction performance of the model from the two types of error detection theory, to further illustrate MEMS-NIR spectroscopy can be used for on-line monitoring of Lonicera Japonica Flos extraction process.

[Quality evaluation method for Chinese medicine based on color grading].

Traditional identification method is an effective approach to evaluate the quality for Chinese herbal medicine (CHM). Color is one of the important indicators for quality evaluation due to high correlation with quality. Therefore, a new theory of quality control for CHM based on color grading was discussed in this article. The scientific nature of this theory was illustrated by investigating the relation between CHM color, medicinal properties and active compound contents. The effect of origins, collecting time, processing, and storage on the CHM color was also analyzed. To overcome the drawback of the traditional identification method, the novel objective color evaluation methods such as spectrocolorimeter and machine vision technology were reviewed, including the application, advantages and disadvantages in Chinese medicine field, and the significance of color sense digitalization was illustrated finally.

[Near infrared spectrum monitoring of scale up process for macroporous resin purification of Grardeniae extract].

To establish a fast detection method during the purifying process of the extracts from Grardeniae using macroporous resin based on near infrared spectroscopy. First, the ethanol eluent was collected from the purification process of small size sample; and near infrared (NIR) spectrum was collected. Then the content of the geniposide was determined by HPLC method, and partial least squares (PLS) method was used to establish the quantitative model to predict the content of geniposide by NIR spectrum. This model was used to supervise the changes of geniposide concentrations in ethanol eluent during medium scale process. Experimental results showed that the NIR small scale model can accurately predict the concentrations of geniposide in the production process of medium scale. However, with the proceeding of batch processes, the prediction performance of the model was decreased, so model updating method was employed to maintain the model. After twice updates, the NIR quantitative model can accurately predict the concentrations of the geniposide during medium scale process. Therefore, through model updates, the established NIR quantitative model can be applied in different scales of macroporous resin purification processes, to improve the data utilization efficiency of small scale process and save the cost of rebuilding the quantitative model of medium scale.

[SIC algorithm based model updating for near infrared analysis of Salvia miltiorrhiza alcohol extraction process].

The near-infrared (NIR) spectroscopy for offline monitoring of alcohol extraction process of Salvia miltiorrhiza was investigated, with high performance liquid chromatography (HPLC) determination of value for reference. The partial least squares method was adopted to establish the tanshinone ⅡA quantitative calibration model, so as to detect extraction process of Salvia miltiorrhiza. Because the differences between batches of raw materials may endanger the robustness of the original model, the simple interval calculation (SIC) was applied in updating the near-infrared quantitative model for traditional Chinese medicine extraction process for the first time, and compared with Random Selection (RS) method. SIC's final updating results showed that root mean square with cross validation (RMSECV), root mean square error of prediction (RMSEP) and residual predictive deviation (RPD) of tanshinone ⅡA were 0.006 8 g•L⁻¹, 0.005 4 g•L⁻¹ and 3.14, respectively; but RS' final updating results showed that RMSECV, RMSEP and RPD were 0.006 4 g•L⁻¹, 0.006 8 g•L⁻¹ and 2.50, respectively. This study suggested that SIC is superior to RS, and provided a research foundation for quality control and monitoring of S. miltiorrhiza extraction process in the future.

[Quality by design based high shear wet granulation process development for the microcrystalline cellulose].

The design space of the high shear wet granulation process was established and validated within the framework of quality by design (QbD). The system of microcrystalline cellulose-de-ioned water was used in this study. The median granule size and bulk density of granules were identified as critical quality attributes. Plackeet-Burmann experimental design was used to screen these factors as follows: dry mixing time, the impeller and chopper speed of dry mixing, water amount, water addition time, wet massing time, the impeller and chopper speed of wet massing and drying time. And the optimization was implemented with the central composite experimental design based on screened critical process parameters. The design space of the high shear wet granulation process was established based on the quadratic polynomial regression model. Since the P-values of both models were less than 0.05 and values of lack of fit were more than 0.1, the relationship between critical quality attributes and critical process parameters could be well described by the two models. The reliability of design space, illustrated by overlay plot, was improved with the addition of 95% confidence interval. For those granules whose process parameters were in the design space, the granule size could be controlled within 250 to 355 µm, and the bulk density could be controlled within a range of 0.4 to 0.6 g x cm(-3). The robustness and flexibility of the high shear wet granulation process have been enhanced via the establishment of the design space based on the QbD concept.

[NIR Assignment of Magnolol by 2D-COS Technology and Model Application Huoxiangzhengqi Oral Liduid].

Near infrared (NIR) spectroscopy assignment of Magnolol was performed using deuterated chloroform solvent and two-dimensional correlation spectroscopy (2D-COS) technology. According to the synchronous spectra of deuterated chloroform solvent and Magnolol, 1365~1455, 1600~1720, 2000~2181 and 2275~2465 nm were the characteristic absorption of Magnolol. Connected with the structure of Magnolol, 1440 nm was the stretching vibration of phenolic group O-H, 1679 nm was the stretching vibration of aryl and methyl which connected with aryl, 2117, 2304, 2339 and 2370 nm were the combination of the stretching vibration, bending vibration and deformation vibration for aryl C-H, 2445 nm were the bending vibration of methyl which linked with aryl group, these bands attribut to the characteristics of Magnolol. Huoxiangzhengqi Oral Liduid was adopted to study the Magnolol, the characteristic band by spectral assignment and the band by interval Partial Least Squares (iPLS) and Synergy interval Partial Least Squares (SiPLS) were used to establish Partial Least Squares (PLS) quantitative model, the coefficient of determination Rcal(2) and Rpre(2) were greater than 0.99, the Root Mean of Square Error of Calibration (RM-SEC), Root Mean of Square Error of Cross Validation (RMSECV) and Root Mean of Square Error of Prediction (RMSEP) were very small. It indicated that the characteristic band by spectral assignment has the same results with the Chemometrics in PLS model. It provided a reference for NIR spectral assignment of chemical compositions in Chinese Materia Medica, and the band filters of NIR were interpreted.

[Rapid discriminant analysis of sulfur fumigated Puerariae Lobatae Radix based on vertical and horizontal surfaces].

Near-infrared (NIR) was used as rapid analysis method to identify the sulfur fumigated Puerariae Lobatae Radix. NIR spectra of the cross-sectional and longitudinal selection of samples were acquired. Principal component analysis was conducted. The samples were randomly selected. The different pretreatment methods were compared. Discriminant models were established for every type of spectra to calculate the recognition rate. The orthogonal test and nonparametric test were used to test data normality. The result showed that absorbance values of different sections were different due to the different structure, and the raw spectra were analyzed by PCA method. The result founded that the cumulative contribution rate was arrived at 99.2% while the PC numbers were arrived at 3. The pretreatment method based on the MSC + 1D + Savitzky-Golay was the best to establish the model. For the 50 models constructed with cross-section and longitudinal spectra and total spectra, the recognition rate were (94.4 ± 0.66)%, (94.4 ± 0.66)%, (95.3 ± 0.65)%, respectively, and no difference was observed. The NIR method could be used to identify the sulfur fumigated Puerariae Lobatae Radix.

[Study on manufacturing process performance index of Qingkailing injection].

In this paper, the process performance indexes (PPIs) P(p) and P(pk) were introduced and applied to evaluate the process capability and quality consistence of Chinese medicine products. The historical quality analysis data of Qingkailing injection were collected and taken as the research object. The confidence intervals of P(p) and P(pk) were estimated based on the Bootstrap sampling methods. Results showed that the value and width of the confidence interval P(p) of were smaller that of P(pk) suggesting that the P(pk) index was more sensitive than P(p) in process capability analysis. Within the production period concerned, the P(pk) values estimated from different quality indexes of Qingkailing injection, such as baicalin, cholic acid, geniposide and total nitrogen, were 1.122, 2.055, 1.564 and 0.891, respectively. It could be found that the cholic acid had the highest process capability, followed by the geniposide and baicalin. The total nitrogen had the lowest process capability, indicating that it is necessary to reinforce the quality management of total nitrogen related manufacturing processes. The case studies demonstrate the effectiveness and feasibility of PPIs, which are convenient to be used in production practice of Chinese medicine.

[Rapid multi-elemental analysis on four precious Tibetan medicines based on LIBS technique].

The laser-induced breakdown spectroscopy (LIBS) was applied to perform a qualitative elementary analysis on four precious Tibetan medicines, i. e. Renqing Mangjue, Renqing Changjue, 25-herb coral pills and 25-herb pearl pills. The specific spectra of the four Tibetan medicines were established. In the experiment, Nd: YAG and 1 064 nm-baseband pulse laser were adopted to collect the spectra. A laser beam focused on the surface of the samples to generate plasma. Its spectral signal was detected by using spectrograph. Based on the National Institute of Standard and Technology (NIST) database, LIBS spectral lines were indentified. The four Tibetan medicines mainly included Ca, Na, K, Mg and other elements and C-N molecular band. Specifically, Fe was detected in Renqing Changjue and 25-herb pearl pills; heavy mental elements Hg and Cu were shown in Renqing Mangjue and Renqing Changjue; Ag was found in Renqing Changjue. The results demonstrated that LIBS is a reliable and rapid multi-element analysis on the four Tibetan medicines. With Real-time, rapid and nondestructive advantages, LIBS has a wide application prospect in the element analysis on ethnic medicines.

[Detection analysis of particle size effects in overtone and combination region of NIR based on multi-types of Chinese material medica].

Near-infrared (NIR) diffuse reflectance spectroscopy was used to analyze the impact of multi-class particle size of Chinese material medica (CMM) based on the spectral characteristics in overtone and combination region. Several types of CMM (60, 80, 100,120 mesh) were subjected to NIR spectra analysis. Spectral reproducibility was examined after sample repackage. The result showed that the effects of particle size on the NIR spectra were different according to different bands, in the combination region and first combination-overtone region. Spectroscopy intensity was proportional to the particle size and influence of particle size was greater as the wavelength increased. While in the second combination-overtone region, it was inversely proportional to particle size. To the sampling loading error, the result indicated that when the mesh number was larger than 60 mesh, the error was small. The appropriate particle size was clarified to guarantee the accuracy and reliability of NIR diffuse reflectance spectroscopy in CMM.