A multi-index quality assessment system for Bupleurum preparations

Abstract

This invention relates to the field of quality testing and process quality evaluation of traditional Chinese medicine preparations, specifically to a multi-index quality judgment system for Bupleurum preparations. This system forms a batch data matrix by collecting process node data from the same batch of Bupleurum preparations, constructs node paths, index characteristics, and transformation mappings, calculates transformation trigger vectors, generates gated quality scores, and outputs quality judgment results and process transformation risk results. This invention unifies the organization of node detection data, node exposure data, node interval data, and node reliable data through a batch data matrix, combines the index construction results and transformation trigger vectors to generate gated quality scores, and outputs the quality judgment results and process transformation risk results for Bupleurum preparations, thereby improving process traceability, transformation identification, and risk location capabilities.

Description

Technical Field

[0001] This invention relates to the field of quality testing and process quality evaluation of traditional Chinese medicine preparations, and more specifically, to a multi-index quality judgment system for Bupleurum preparations. Background Technology

[0002] Bupleurum preparations are common traditional Chinese medicine preparations, and their quality evaluation typically involves saikosaponin components, fingerprint similarity, total saponin levels, and other physicochemical quality indicators. With increasing requirements for process control in the production of traditional Chinese medicine preparations, the scope of quality evaluation has gradually expanded from single-point testing of the finished product to multiple processes such as extraction, concentration, drying, granulation, processing, and storage. By collecting and analyzing test data from different process stages of the same batch of Bupleurum preparations, a more comprehensive reflection of batch quality changes can be achieved, providing a data foundation for process quality control and risk traceability.

[0003] Existing quality control methods for Bupleurum preparations primarily rely on finished product testing results, single-process testing results, or multi-index content determination results to determine whether a batch meets quality standards. However, in actual production, prototype indicators such as saikosaponin a, saikosaponin c, and saikosaponin d, as well as secondary indicators such as saikosaponin b1, saikosaponin b2, and saikosaponin h, may undergo correlated changes due to acid-base environments, heating conditions, process intervals, and the intensity of process handling. An increase in certain secondary indicators may reflect changes in the level of active ingredients or may originate from the transformation of prototype indicators under specific process conditions. Therefore, judging quality contribution solely based on indicator increases or decreases can easily lead to a lack of correlation between quality evaluation results and actual process risks.

[0004] Furthermore, data from different process nodes exhibit variations in time intervals, process sequence, processing intensity, detection repeatability, peak signal-to-noise ratio, and sampling bias. Without standardized processing of node paths, node reliability, and indicator transformation relationships, it is difficult to establish comparable and traceable evaluation criteria for changes in indicators across multiple nodes. Especially in stages such as concentration, drying, processing, or storage, which are prone to acid and heat exposure, there may be a shared relationship between the decrease in prototype indicators, the increase in secondary indicators, and the intensification of exposure conditions. Existing evaluation methods still require further integration of process path characteristics and exposure features for risk identification.

[0005] Based on this, this application proposes a multi-index quality judgment system for Bupleurum preparations. Summary of the Invention

[0006] To address the shortcomings of existing technologies, the present invention aims to provide a multi-index quality assessment system for Bupleurum preparations.

[0007] To achieve the above objectives, the present invention provides the following technical solution: A multi-index quality assessment system for Bupleurum preparations includes a processor and a memory. The memory stores a computer program executed by the processor. The computer program includes a data matrix acquisition module, an index vector construction module, a path feature calculation module, a gating score generation module, and a risk result output module. The data matrix acquisition module is used to receive batch node data of the same batch of Bupleurum preparations and store the batch node data as a batch data matrix. The batch node data includes node detection data, node exposure data, node interval data and node reliability data in the process node sequence. The indicator vector construction module is used to generate indicator construction results based on batch node data. The indicator construction results include node path data, indicator feature data, and transformation mapping data. The node path data includes the set of adjacent node pairs, target adjacent node pairs, normalized node distance, and node confidence weight. The indicator feature data includes the prototype indicator set, secondary indicator set, auxiliary indicator set, and standard feature data. The path feature calculation module is used to generate a transformation trigger vector based on the index construction results and node exposure data. The transformation trigger vector includes the composite transition feature value, acid-heat exposure coupling value and transformation path consistency value of the target adjacent node pairs. The gating score generation module is used to generate gating quality scores based on conversion trigger vectors, indicator construction results, and preset trigger rules. The risk result output module is used to output the quality judgment result of Bupleurum preparation and the process transformation risk result based on the gate quality score, indicator construction result and preset trigger rules.

[0008] In one embodiment, the process node sequence includes at least three process nodes arranged in the order of preparation, processing, or storage, and the process nodes are selected from the extraction node, concentration node, drying node, granulation node, processing node, finished product node, and storage node.

[0009] In one embodiment, the node detection data includes content data or peak area data of at least two of saikosaponin a, saikosaponin c, and saikosaponin d, and at least two of saikosaponin b1, saikosaponin b2, and saikosaponin h, and includes at least one of total saikosaponin index, fingerprint spectrum similarity index, and non-saikosaponin quality index.

[0010] In one embodiment, the node exposure data includes pH data and thermal exposure data corresponding to each process node; the node interval data includes at least one of the time interval between adjacent process nodes, process sequence interval, and processing intensity interval; and the node reliability data includes at least one of the detection repeatability data, peak signal-to-noise ratio data, and sampling time deviation data.

[0011] In one embodiment, the set of adjacent node pairs includes at least two sets of adjacent node pairs, each set of adjacent node pairs including a predecessor node and a successor node; the target adjacent node pair is any adjacent node pair in the set of adjacent node pairs; the normalized node distance is determined based on the node interval data and corresponds to the adjacent node pair; the node trust weight is determined based on the node trust data and corresponds to each process node.

[0012] In one embodiment, the prototype index set includes at least two of saikosaponin a, saikosaponin c, and saikosaponin d; the secondary index set includes at least two of saikosaponin b1, saikosaponin b2, and saikosaponin h; the auxiliary index set includes at least one of total saikosaponin index, fingerprint spectrum similarity index, and non-saikosaponin quality index; and the standard feature data includes the prototype index vector, secondary index vector, and auxiliary index vector corresponding to each process node.

[0013] In one embodiment, the generation of composite transition feature values ​​includes: determining the predecessor and successor nodes in the target adjacent node pair, and generating composite transition feature values ​​corresponding to the target adjacent node pair based on the standard feature data corresponding to the predecessor and successor nodes, the normalized node distance corresponding to the target adjacent node pair, the transformation mapping data, and the node confidence weights corresponding to the predecessor and successor nodes.

[0014] In one embodiment, the acid-heat exposure coupling value is generated as follows: the acid-base deviation component and heat exposure component corresponding to the target adjacent node pair are generated based on the node exposure data corresponding to the preceding and following nodes in the target adjacent node pair; the interaction gain is generated based on the common change relationship between the acid-base deviation component and the heat exposure component in the target adjacent node pair; and the acid-heat exposure coupling value corresponding to the target adjacent node pair is generated based on the interaction gain.

[0015] In one embodiment, the generation of the gating quality score includes: configuring the indicators in the secondary indicator set as quality contribution attributes; when the conversion trigger vector meets the preset trigger rules, switching the indicators in the secondary indicator set that are higher in the subsequent nodes of the target adjacent node pair relative to the preceding nodes from quality contribution attributes to conversion risk attributes, and generating compensation gating data based on the conversion risk attributes; when the conversion trigger vector does not meet the preset trigger rules, maintaining the quality contribution attributes of the indicators in the secondary indicator set that are higher in the subsequent nodes of the target adjacent node pair relative to the preceding nodes; and generating the gating quality score based on the quality contribution attributes, conversion risk attributes, auxiliary indicator vectors, and compensation gating data.

[0016] In one embodiment, the process conversion risk result is generated based on the target adjacent node pair that meets the preset triggering rules. The process conversion risk result includes the risk process node, the risk indicator name, the risk triggering basis, and the risk formation type. The risk triggering basis includes the composite transition characteristic value, the acid-heat exposure coupling value, and the conversion path consistency value. The risk formation type includes at least one of the following: extraction conversion risk, concentration conversion risk, drying conversion risk, granulation conversion risk, processing conversion risk, and storage conversion risk.

[0017] Compared with the prior art, the present invention has the following beneficial effects: This invention generates index construction results including node path data, index feature data, and transformation mapping data through an index vector construction module, and generates transformation trigger vectors including composite transition feature values, acid-heat exposure coupling values, and transformation path consistency values ​​through a path feature calculation module. This enables changes in the prototype index set, secondary index set, and auxiliary index set to be combined with adjacent node pairs, target adjacent node pairs, normalized node distances, and node confidence weights, thereby improving the ability of Bupleurum preparations to identify process path changes and component transformation relationships in multi-index quality judgment. The gating scoring generation module generates gating quality scores based on the transformation trigger vector, indicator construction results, and preset trigger rules. The risk result output module outputs the quality judgment results of Bupleurum preparations and the process transformation risk results. This allows the quality judgment to not only reflect the overall quality status of the batch, but also to associate risk process nodes, risk indicator names, risk trigger basis, and risk formation type. This facilitates the location and description of transformation risks in processes such as extraction, concentration, drying, granulation, processing, and storage. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the overall system structure of a multi-index quality determination system for Bupleurum preparations according to the present invention; Figure 2 This is a schematic diagram illustrating the formation of the batch data matrix in this invention; Figure 3 This is a schematic diagram showing the generation of the indicator construction results of the present invention; Figure 4 This is a schematic diagram illustrating the generation of the gate control quality score value according to the present invention. Detailed Implementation

[0019] Reference Figure 1 A multi-index quality assessment system for Bupleurum preparations includes a processor and a memory. The memory stores a computer program executed by the processor. The computer program includes a data matrix acquisition module, an index vector construction module, a path feature calculation module, a gating score generation module, and a risk result output module. Reference Figure 2The data matrix acquisition module is used to receive batch node data of the same batch of Bupleurum preparations and store the batch node data into a batch data matrix. The batch node data includes node detection data, node exposure data, node interval data and node reliability data in the process node sequence. It centrally receives and matrix-organizes the detection, exposure, interval and reliability related data generated by the same batch of Bupleurum preparations at different process nodes. Batch data scattered across extraction, concentration, drying, granulation, processing, finished product, and storage stages are incorporated into a unified data structure, enabling comparability between data from different process stages within the same batch. The batch data matrix retains detection information, environmental or process exposure information, interval information between stages, and detection reliability information for each process stage, providing a unified data source for subsequent indicator construction and path feature calculation.

[0020] The same batch of Bupleurum preparations includes no fewer than three process steps arranged according to the order of preparation, processing, or storage, which are denoted as follows: ,in For the first Each process node. The node data for each process node includes prototype index detection values, secondary index detection values, auxiliary index detection values, pH, original heat exposure records, adjacent node interval data, and reliable detection data. Among these, This refers to the number of prototype indicators. For the number of secondary indicators, For the number of auxiliary indicators, For the first For each process node, the original heat exposure record includes at least one of the following: node temperature profile, average temperature, maximum temperature, and duration of heat exposure. For the first From the first process node to the... The time interval between each process node For process sequence intervals, To handle intensity intervals, To detect the relative standard deviation of repeatability, For peak signal-to-noise ratio. This represents the sampling time deviation. Processing intensity interval. It is formed by the absolute value or standardized variation range of at least one of the following: extraction power, concentration vacuum degree, concentration temperature, drying air temperature, drying time, granulation pressure, processing firepower, and storage temperature. The set of process parameter indices involved in the calculation of the intensity interval in a group of adjacent node pairs is denoted as , No. The standardized variation range of the process parameter is denoted as: The corresponding processing intensity weight is denoted as Furthermore, all processing intensity weights involved in the calculation are non-negative and total 1. The processing intensity interval is calculated using the following formula: When the parameter configuration record specifies the use of the maximum change range rule. Pick The maximum value of each standardized variation range; during the same batch of the same product, the processing intensity interval adopts either the weighted total rule or the maximum variation range rule, and the rule type is fixed in the parameter configuration record. When a process node has no adjacent subsequent nodes. and Recorded as null, it is not included in the calculation of adjacent node pairs.

[0021] A process node sequence is established for the same batch of Bupleurum preparations, and at least three process nodes are selected according to the preparation sequence, processing sequence, or storage sequence. These process nodes include several nodes from the following: extraction node, concentration node, drying node, granulation node, processing node, finished product node, and storage node. Taking a batch of Bupleurum granule preparations as an example, data can be collected sequentially from the extraction node, concentration node, drying node, and finished product node. Node detection data, node exposure data, and node reliability data are recorded at each process node, and node interval data are recorded between adjacent process nodes. Node detection data includes Bupleurum soap. The data includes the content or peak area data of at least two of saikosaponin a, saikosaponin c, and saikosaponin d, and at least two of saikosaponin b1, saikosaponin b2, and saikosaponin h. It also includes at least one of the total saikosaponin index, fingerprint spectrum similarity index, and non-saikosaponin quality index. For example, at the extraction node, the peak area data of saikosaponin a, saikosaponin d, saikosaponin b1, and saikosaponin b2 are recorded, and at the concentration node, the content data of the same index are recorded. The total saikosaponin index and fingerprint spectrum similarity index are recorded simultaneously to maintain the consistency of the detection caliber between different nodes of the same batch. Batch node data collected from each process node within the same batch are organized into a batch data matrix. Each row in the matrix corresponds to a process node, and each column corresponds to a type of node detection data, node exposure data, node interval data, or node reliability data. Node exposure data includes pH data and thermal exposure data corresponding to each process node. For example, the extraction node records the pH of the extract and the duration of the extraction temperature, the concentration node records the pH of the concentrate and the heating temperature range, and the drying node records the heating time and drying temperature of the material. Node interval data includes at least one of the time interval between adjacent process nodes, process sequence interval, and treatment intensity interval. For example, the transfer waiting time between the extraction node and the concentration node, and the change in treatment intensity between the concentration node and the drying node. Node reliability data includes at least one of the following: detection repeatability data, peak signal-to-noise ratio data, and sampling time deviation data. For example, the relative deviation of parallel detection results at the same node, the chromatographic peak signal-to-noise ratio, and the offset of the sampling time relative to the set sampling time. The batch data matrix described above preserves the component changes, acid and heat exposure, node intervals, and detection reliability of the same batch of Bupleurum preparations at different process nodes, providing a complete, continuous, and traceable data foundation for multi-indicator quality judgment.

[0022] Reference Figure 3 The indicator vector construction module is used to generate indicator construction results based on batch node data. The indicator construction results include node path data, indicator feature data, and transformation mapping data. The node path data includes the set of adjacent node pairs, target adjacent node pairs, normalized node distance, and node confidence weight. The indicator feature data includes the prototype indicator set, secondary indicator set, auxiliary indicator set, and standard feature data. Based on the batch node data, an indicator construction result suitable for quality judgment calculation is formed. Node path data is established around the process node sequence, and combined with node detection data to form indicator feature data. The set of adjacent node pairs, target adjacent node pairs, normalized node distance, and node reliability weight in the node path data reflect the path relationships, distance relationships, and data reliability between adjacent process nodes of the same batch of Bupleurum preparations. The prototype indicator set, secondary indicator set, auxiliary indicator set, and standard feature data in the indicator feature data group and standardize Bupleurum saponin components and related quality indicators, enabling different nodes and indicators to be included in the same judgment framework. Transformation mapping data reflects the transformation correlation between prototype indicators and secondary indicators, providing indicator correspondence for identifying process transformation risks.

[0023] Establish node path data based on batch node data of the same batch of Bupleurum preparations, and combine two adjacent process nodes in the process node sequence into a set of adjacent node pairs. Each set of adjacent node pairs includes a preceding node and a following node. For example, when the extraction node, concentration node, drying node and finished product node are arranged in sequence, the set of adjacent node pairs includes three sets of adjacent node pairs: extraction node to concentration node, concentration node to drying node, and drying node to finished product node. In the same batch The process node and the first The process node constitutes the first A pair of adjacent nodes is denoted as ,in For the previous node, This is the subsequent node. At least one of the following must be collected: time interval, process sequence interval, and processing intensity interval. Uncollected items are assigned a value of 0 and do not participate in the actual distance weight allocation. The base value for distance weight is denoted as... and All three are non-negative and their sum is 1. Actual distance weights and The weights are normalized from the base weights corresponding to the collected time intervals, process sequence intervals, and processing intensity intervals. When the sum of the base weights corresponding to the collected items is greater than 0, the actual distance weight is equal to the base weight of the corresponding collected item divided by the sum of the base weights of all collected items; when the sum of the base weights corresponding to the collected items is 0, the collected items are equally weighted. The normalized node distance of a group of adjacent node pairs is denoted as: in, For the first The normalized node distance between pairs of adjacent nodes. As a reference for time interval normalization, As a reference for normalizing the interval between process sequences, To handle the intensity interval normalization benchmark, all three normalization benchmarks are positive numbers. and and and They should all use the same unit of measurement or the same standardized caliber.

[0024] Meanwhile, the node credibility weight is calculated based on the detection repeatability data, peak signal-to-noise ratio data, and sampling time deviation data corresponding to the preceding and subsequent nodes. For example, the smaller the detection repeatability deviation, the higher the peak signal-to-noise ratio, and the lower the sampling time deviation, the higher the node credibility weight corresponding to that process node. At least one of the following should be collected: repeatability data, peak signal-to-noise ratio data, and sampling time deviation data. Uncollected data is assigned a value of 0 and is not included in the actual reliability weight coefficient allocation. The base value of the reliability weight is denoted as... and All three are non-negative and their sum is 1. Actual credible weight coefficient and The weights are normalized from the basic weights corresponding to the collected repeatability data, peak signal-to-noise ratio data, and sampling time deviation data. When the sum of the basic weights corresponding to the collected items is greater than 0, the actual reliable weight coefficient is equal to the basic weight of the corresponding collected item divided by the sum of the basic weights of all collected items; when the sum of the basic weights corresponding to the collected items is 0, the collected items are equally weighted. The node trust weight of each process node is denoted as: in, For the first The node trust weight of each process node The maximum permissible relative standard deviation for detection repeatability, This is a reference value for peak signal-to-noise ratio. This represents the maximum allowable sampling time deviation, and all three are positive numbers. and Use percentages or decimals uniformly. When multiple detection indicators exist at the same node, Take the maximum value of the relative standard deviation of each detection index at this node. Take the minimum peak signal-to-noise ratio of each detection index at that node. The trusted weight of a pair of adjacent nodes is denoted as . Take the trusted weight of the preceding node. With the trusted weight of subsequent nodes The geometric mean is calculated using the following formula: Based on the node detection data, indicator feature data is constructed, forming a prototype indicator set, a secondary indicator set, an auxiliary indicator set, and standard feature data. Among them, the prototype indicator set selects at least two of saikosaponin a, saikosaponin c, and saikosaponin d; the secondary indicator set selects at least two of saikosaponin b1, saikosaponin b2, and saikosaponin h; and the auxiliary indicator set selects at least one of the total saikosaponin index, fingerprint spectrum similarity index, and non-saikosaponin quality index. Taking the concentration node as an example, the content data of saikosaponin a and saikosaponin d are used to form the prototype index vector corresponding to the node, the content data of saikosaponin b1 and saikosaponin b2 are used to form the secondary index vector corresponding to the node, and the total saikosaponin index and fingerprint spectrum similarity index are used to form the auxiliary index vector corresponding to the node. The standardization is performed according to the upper and lower limits of the detection index to obtain the standard feature data corresponding to each process node. When the node detection data is peak area data, according to the detection indicators The standard curve is converted to content data, and the slope of the standard curve is denoted as . The standard curve intercept is denoted as ,and For positive numbers, the conversion formula is: in, It can be any of the prototype indicators, secondary indicators, or auxiliary indicators. For the first Inspection indicators at each process node peak area, This is the converted content data. When a relative peak area is used instead of an external standard curve, the detected value is the ratio of the peak area of ​​the detected index to the peak area of ​​the standard within the same process node or the peak area of ​​a selected common peak. This ratio is denoted as... Detected values ​​below the limit of quantitation but already detected should be handled according to the quantitation limit conversion rules recorded in the test method document; if the test method document does not record conversion rules, half of the quantitation limit should be taken as the limit of quantitation. The data is recorded as below the limit of quantitation in the output. Undetected indicators are not treated as below the limit of quantitation data and are recorded as missing data. The amplitude limiting function in the standardization process is denoted as: in, The value to be limited. As the lower limit, The upper limit is defined as the limit, and the lower limit is less than the upper limit. The positive or negative attributes of a detection indicator are determined by pharmacopoeia limits, company internal control standards, testing method documents, or product parameter configuration records. Positive indicators are calculated using the following formula: The contrarian indicator is calculated using the following formula: in, For the first Inspection indicators for each process node The standardized value, For detection indicators The lower limit benchmark, For detection indicators The upper limit benchmark, and Greater than , To prevent positive numbers with a denominator of zero, the range of values ​​is [value missing]. . and External benchmarks are derived from pharmacopoeia limits, internal control limits, methodological validation data, the 5th and 95th percentiles of historical qualified batches, or the same batch testing range. When pharmacopoeia limits, internal control limits, methodological validation data, or historical qualified batch statistics exist, these external benchmarks are preferred; only when these benchmarks have not yet been established is the same batch testing range used as a temporary normalization benchmark. When determined according to the same batch testing range... and If the results are equal, the historical qualified batch benchmark, methodology validation benchmark, or enterprise internal control benchmark should be used. If a valid upper and lower limit still cannot be obtained, the test indicator is marked as undifferentiated data, not included in the batch calculation, and recorded in the missing indicator set. When the test indicator... When it belongs to the prototype indicator, secondary indicator, or auxiliary indicator, They are respectively denoted as and After standardization, both positive and negative auxiliary indicators are converted into indicators where higher values ​​indicate better quality.

[0025] When a certain detection index is missing at a single non-endpoint process node, the arithmetic mean of the standardized values ​​corresponding to the preceding and following process nodes is taken as the substitute standardized value for that node. The substituted value is still based on... Recording: When a missing node is located at an endpoint, the standardized value of the nearest detected node to that endpoint is used. When two or more consecutive process nodes are missing, the median standardized value of the same detection index for the same process node in historical qualified batches is used. If there is no corresponding node data in historical qualified batches, or if the number of historical batches available for calculation is insufficient, no substitution is performed, the detection index is not included in the calculation involving the corresponding node pair, and it is recorded in the missing index set. When all detection indices are missing in the same batch, the detection index is not included in the calculation of composite transition characteristic value, transformation path consistency value, and gate quality score value, and is recorded as a missing index in the output results.

[0026] The transformation mapping data records the correspondence between the prototype indicator set and the secondary indicator set. For example, the correlation between the decrease of saikosaponin a, saikosaponin c or saikosaponin d and the increase of saikosaponin b1, saikosaponin b2 or saikosaponin h. After combining the normalized node distance and node confidence weight, the indicator construction result containing node path data, indicator feature data and transformation mapping data is formed.

[0027] The transformation mapping data is denoted as the transformation mapping matrix. .in, For the first The prototype indicator and the first The conversion correlation weights between the secondary indicators range from 0 to 1. The original conversion correlation weights are denoted as... , It is a non-negative number. The data is derived from the results of thermal conversion simulation experiments of saikosaponins, adjacent node data from historical batches of the same product, conversion relationships of saikosaponins recorded in publicly available literature, or internal control verification results of the enterprise. When quantitative experimental data is unavailable, a clear conversion correlation exists. A range of 0.8 to 1.0 indicates a weak correlation. Take values ​​between 0.1 and 0.3; no association confirmed. Set to 0. At least one of the original conversion association weights corresponding to each secondary indicator is greater than 0. The conversion mapping matrix is ​​obtained as follows: In the saikosaponin index, saikosaponin a, saikosaponin c, and saikosaponin d are classified as primary indices, while saikosaponin b1, saikosaponin b2, and saikosaponin h are classified as secondary indices. Once the transformation mapping matrix is ​​determined, it is associated with and stored in relation to the product name, formulation type, detection method version, and process node sequence, remaining unchanged throughout the calculation period for the same batch.

[0028] The path feature calculation module is used to generate a transformation trigger vector based on the indicator construction results and node exposure data. The transformation trigger vector includes the composite transition feature value, acid-heat exposure coupling value, and transformation path consistency value corresponding to the target adjacent node pair. Based on the indicator construction results and node exposure data, feature extraction is performed on the quality change path in the target adjacent node pair. This module focuses on the changing relationships of prototype, secondary, and auxiliary indicators between adjacent process nodes, and combines pH data, thermal exposure data, normalized node distance, and node confidence weights to form a transformation trigger vector. The composite transition feature value in the transformation trigger vector reflects the comprehensive transition of multiple indicator changes between adjacent target node pairs; the acid-thermal exposure coupling value reflects the combined influence of pH and thermal exposure conditions on adjacent target node pairs; and the transformation path consistency value reflects the degree of agreement between the direction of indicator change and the transformation mapping relationship. This module expands a single detection result into a change feature with process path implications.

[0029] For each target adjacent node pair, the preceding and subsequent nodes are determined. The standard feature data corresponding to the preceding and subsequent nodes, the normalized node distance corresponding to the target adjacent node pair, the transformation mapping data, and the node confidence weights corresponding to the preceding and subsequent nodes are retrieved. The change range of the prototype index vector, secondary index vector, and auxiliary index vector between the two nodes is calculated. No. Group of adjacent nodes In this context, the set of prototype index numbers participating in the calculation is denoted as... The set of secondary index numbers involved in the calculation is denoted as The set of auxiliary indicator numbers involved in the calculation is denoted as . Each by the first The pair of adjacent nodes consists of prototype indices, secondary indices, and auxiliary indices, where both the preceding and succeeding nodes have standardized values, or where standardized values ​​are obtained after replacing missing values. The decreasing component of each prototype indicator is denoted as: No. The increase component of each secondary indicator is denoted as: After standardization, the auxiliary indicators are all indicators where higher values ​​indicate better quality. The loss components of each auxiliary indicator are denoted as: The above three formulas respectively apply to and The corresponding indicators are executed. Among them, For the first The first process node The standardized value of a prototype indicator. For the first The first process node The standardized value of a prototype indicator. For the first The first process node Standardized values ​​of secondary indicators For the first The first process node Standardized values ​​of secondary indicators For the first The first process node The standardized values ​​of the auxiliary indicators, For the first The first process node Standardized values ​​of each auxiliary indicator. Based on the transformation mapping matrix. Calculate the first The predicted increase component of each secondary indicator: in, For the first The predicted increase component of each secondary indicator is obtained by mapping the decreasing component of the prototype indicator. This formula applies to... The execution of various secondary indicators. When some prototype indicators have not yet entered... At that time, transform the associated weight No re-unification, not entered The prototype indicator corresponding to the missing item is not included in the calculation of the predicted elevation component, and the corresponding missing prototype indicator is recorded in the missing indicator set. The consistency value of the transformation path for a group of adjacent node pairs is denoted as: in, For the first The consistency value of the transformation path for adjacent node pairs. When or When empty, If a transition is deemed uncalculationable, the transition weight corresponding to the consistency value of the transformation path is set to 0 in the adjacent node pair, and the remaining valid transition weights are normalized proportionally. and When all values ​​are not empty, and the total value of the secondary indicator increase components is 0, . The closer it is to 1, the more consistent the upward direction of the secondary indicator is with the downward direction of the prototype indicator as defined by the transformation mapping matrix.

[0030] Taking the preceding node as the concentration node and the following node as the drying node as an example, in the concentration node, saikosaponin a and saikosaponin d are at a high level, while in the drying node, saikosaponin b1 and saikosaponin b2 increase, while saikosaponin a and saikosaponin d decrease. Then, based on the transformation mapping data, it is determined whether this change conforms to the corresponding relationship of the transformation from the prototype index to the secondary index. No. In a group of adjacent node pairs, the average decrease component of the prototype index is denoted as... The average increase in secondary indicators is denoted as The average loss component of the auxiliary indicator is denoted as The aforementioned average components are taken as the arithmetic mean of the decreasing component of the prototype indicator, the increasing component of the secondary indicator, and the losing component of the auxiliary indicator, respectively. When calculating the average component, the denominator is the number of indicators actually participating in the calculation from the corresponding indicator sequence set; missing indicators not included in the corresponding indicator sequence set are not counted in the denominator. When the corresponding indicator sequence set is empty, the average component is recorded as 0, and the corresponding transition weight is reset to 0; other transition weights are normalized proportionally. When there is no data to calculate for the prototype indicator, secondary indicator, and auxiliary indicator, the composite transition characteristic value and the consistency value of the transformation path of the adjacent node pair are both recorded as uncalculated, the adjacent node pair does not enter the trigger judgment, and is recorded as insufficient data in the output result. The composite transition eigenvalue of a group of adjacent node pairs is denoted as: in, For the first Composite transition eigenvalues ​​of pairs of adjacent nodes. For the first The trusted weight of a group of adjacent node pairs. For the first The normalized node distance between pairs of adjacent nodes. These are the transition weights for the prototype indicator's decreasing component, the secondary indicator's increasing component, the auxiliary indicator's losing component, and the consistency value of the transformation path. All transition weights are non-negative and total 1; when there are uncalculated items or components corresponding to empty sets, the corresponding transition weights are reset to 0, and the remaining effective transition weights are normalized proportionally. The higher the value of , the more concentrated the changes in the indicators of the adjacent node pair are in the transition state where the prototype decreases, the secondary increases, the auxiliary indicator loss and the transformation path are consistent.

[0031] Calculate the acid-base deviation component and the heat exposure component based on the node exposure data corresponding to the preceding and following nodes in the target adjacent node pair. In the assessment of bufotenosides conversion risk, the acid-base deviation component is calculated based on the acidic deviation component. The alkaline deviation is recorded as background exposure data and is not included in the positive gain for conversion risk. When product validation shows that alkaline conditions trigger specific quality risks, the alkaline risk item is recorded separately in the parameter configuration. The acid-base pairing value of adjacent node pairs is denoted as When both the preceding and following nodes collect pH data, Take the smaller of the two values; when only one node collects pH data, Collect the pH value of this node; if pH value data for both nodes is missing. If not found, the acid deviation component is recorded as 0, and the missing acid / alkalinity value is recorded in the missing index set. The paired acid / alkalinity value, equivalent average temperature, and high-temperature duration should correspond to the process records of the same batch of material within the same adjacent node interval. When multiple acid / alkalinity detection values ​​exist, the value higher than the specified value should be used preferentially. The measured values ​​are taken during the heating period; if no pH value is available during the heating period, the value closest to that heating period is used, and the time deviation between the pH value and the heat exposure record is recorded. Only... If it exists, calculate the first one according to the following formula. Acidity deviation components of adjacent node pairs: in, For the first The acid deviation component of adjacent node pairs, The target pH or neutral reference pH for the process. As an acidic risk benchmark, and Greater than . No. The thermal exposure components of adjacent node pairs are denoted as: in, For the first Thermal exposure components of adjacent node pairs For the first The average temperature of the corresponding materials for adjacent nodes in a group. This represents the heating duration for the adjacent node pair. The starting temperature for heat exposure. As a normalization benchmark for heat exposure, and It is a positive number. When collecting continuous temperature curves, it will be higher than the starting temperature of the heat exposure. The temperature range is converted into an equivalent average temperature using a time-weighted method, and this equivalent average temperature is used as... Using the duration of the corresponding high temperature as Enter the above formula. When the continuous temperature curve, average temperature, and heating duration are all missing, the heat exposure component is recorded as 0, and the missing heat exposure is recorded in the missing index set. When or If any item is missing and cannot be converted from the process record, Furthermore, the recorded heat exposure data was incomplete. and This information originates from product process specifications, acid-heat simulation conversion tests, historical qualified batch statistics, or enterprise internal control verification records. The acid-heat exposure coupling value of a pair of adjacent nodes is denoted as: in, This is the interaction coefficient between the acidity deviation component and the heat exposure component, with a value ranging from 0 to 1. For the first The acid-heat exposure coupling value of adjacent node pairs. When the acid-heat exposure coupling value is included in the transformation trigger data, it is calculated together with the composite transition characteristic value and the transformation path consistency value.

[0032] The transformation path consistency value is determined according to the degree of matching between the decreasing direction of the prototype index and the increasing direction of the secondary index in the transformation mapping data. When the decreasing changes of saikosaponin a, saikosaponin c, or saikosaponin d and the increasing changes of saikosaponin b1, saikosaponin b2, or saikosaponin h maintain a corresponding relationship in the target adjacent node pair, the transformation path consistency value is taken at a higher level. If the secondary index increases but the prototype index does not show a corresponding decrease, the value is reduced. This forms a transformation trigger vector that includes the composite transition characteristic value, acid-heat exposure coupling value, and transformation path consistency value.

[0033] Reference Figure 4The gating score generation module generates gating quality scores based on the transformation trigger vector, indicator construction results, and preset trigger rules. This module does not simply improve quality evaluation by increasing the content of secondary indicators; instead, when the transformation trigger vector meets the preset trigger rules, it switches the indicators in the subsequent nodes of the target adjacent node pair that increase relative to the preceding nodes from quality contribution attributes to transformation risk attributes, and forms compensation gating data based on the transformation risk attributes. For cases where the preset trigger rules are not met, the relevant indicators retain their quality contribution attributes. The gating quality score integrates quality contribution attributes, transformation risk attributes, auxiliary indicator vectors, and compensation gating data to reflect the gating relationship between the quality contribution of multiple indicators and the transformation risk of the process.

[0034] At least two of the saikosaponin b1, saikosaponin b2 and saikosaponin h in the secondary index set are pre-configured as quality contribution attributes, and the secondary index vectors corresponding to the preceding and following nodes in the target neighbor node pair are combined to identify the secondary index that increases in the following node relative to the preceding node. Taking the concentration node to the drying node as the target adjacent node pair, when the standardized values ​​of saikosaponin b1 and saikosaponin b2 in the current sequence node are 0.28 and 0.31 respectively, and the standardized values ​​of saikosaponin b1 and saikosaponin b2 in the subsequent sequence node rise to 0.52 and 0.57 respectively, the increase is recorded as the quality change item to be determined, and the composite transition characteristic value, acid-heat exposure coupling value and transformation path consistency value in the transformation trigger vector are read simultaneously. The preset triggering rules include at least two of the following: the composite transition characteristic value reaches the transition threshold, the acid-heat exposure coupling value reaches the exposure threshold, and the transformation path consistency value reaches the path threshold; or the three items are combined in a weighted manner to reach the comprehensive triggering threshold. No. Group of adjacent node pairs with composite transition eigenvalues Acid-heat exposure coupling value Consistency value with conversion path This data serves as the trigger determination data. Calculated trigger determination items are considered valid trigger items. Uncalculable trigger determination items are not counted as 0 in the overall trigger score; their corresponding overall trigger weights are reset to 0. The overall trigger weights corresponding to valid trigger items are normalized proportionally, and the normalized overall trigger weights are denoted as... and . No. The combined trigger score for a group of adjacent node pairs is denoted as: in, The comprehensive triggering base weights are the composite transition characteristic value, acid-heat exposure coupling value, and transformation path consistency value, respectively. All comprehensive triggering base weights are non-negative and sum to 1. When the sum of the base weights corresponding to a valid triggering item is greater than 0, the normalized comprehensive triggering weight is equal to the base weight of that valid triggering item divided by the sum of the base weights of all valid triggering items; when the sum of the base weights corresponding to a valid triggering item is 0, the valid triggering items are equally weighted. When there are fewer than two valid triggering items, the adjacent node's triggering determination result is recorded as insufficient data and is not included in the gating variable assignment. When there are at least two valid triggering items, the adjacent node is determined to be triggered when at least two of the three items—the composite transition characteristic value reaching the transition threshold, the acid-heat exposure coupling value reaching the exposure threshold, and the transformation path consistency value reaching the path threshold—are met, or when the comprehensive triggering score reaches the comprehensive triggering threshold; otherwise, it is determined as not triggered. For the transition threshold, For the exposure threshold, For path threshold, This is the comprehensive trigger threshold. The trigger rule type is recorded according to the trigger conditions. When at least two individual thresholds are met and the comprehensive trigger score does not reach the comprehensive trigger threshold, it is recorded as a single threshold combination trigger. When the comprehensive trigger score reaches the comprehensive trigger threshold and less than two individual thresholds are met, it is recorded as a comprehensive threshold trigger. When both trigger methods are met, it is recorded as a double trigger. When the preset trigger rules are not met, the trigger rule type is recorded as not triggered.

[0035] Thresholds are derived from historical qualified batches, historical risky batches, acid-heat simulation conversion tests, or enterprise internal control verification data. When only historical qualified batches are available, the transition threshold and exposure threshold are taken as the 95th percentile of the corresponding score for the historical qualified batches. When historical risky batches or acid-heat simulation conversion test data exist, the path threshold is taken as the 5th percentile of the path consistency value in the risk data. When only historical qualified batches are available and no risky batches or acid-heat simulation conversion test data are available... Use a preset value between 0.70 and 0.85; if no product-specific confirmation record is generated, The default value is 0.75. If no historical qualified batches, historical high-risk batches, or acid-heat simulation conversion test data are available, and The default values ​​are used from the parameter configuration records; these default values ​​are updated after verification by the company's internal controls or after review by at least three batches of trial production data. The default threshold is updated based on the verification results after the product completes acid-heat simulation conversion tests or historical risk batch verification. (Comprehensive trigger threshold) When risky data exists, the candidate boundary point with the highest distinguishability between qualified and risky batches is selected. When there is no risky data, the comprehensive trigger threshold is the weighted value calculated by the transition threshold, exposure threshold, and path threshold according to their corresponding comprehensive trigger base weights. The distinguishability of the candidate boundary point is evaluated by subtracting 1 from the sum of sensitivity and specificity, and the candidate boundary point with the highest distinguishability is used as the threshold candidate value. When there are multiple candidate boundary points with the same distinguishability, the lower value among them is selected as the trigger threshold.

[0036] When saikosaponin a and saikosaponin d decrease in the same target adjacent node pair, while saikosaponin b1 and saikosaponin b2 increase, and the acid-base deviation component and heat exposure component are enhanced together, it is determined that the increase has the meaning of process transformation risk, and the corresponding secondary index is switched from quality contribution attribute to transformation risk attribute. Compensation gating data is generated based on conversion risk attributes, and quality contribution attributes, conversion risk attributes, auxiliary indicator vectors, and compensation gating data are all incorporated into the calculation of gating quality score. No. In the pair of adjacent nodes, the first The gating variable for each secondary indicator is denoted as: When the first The group of adjacent node pairs meets the preset triggering rules, the first When the increase component of a secondary indicator is greater than 0, and the predicted increase component of that secondary indicator is greater than 0, Take 1; otherwise Take 0. When At that time, the secondary indicator maintains its quality contribution attribute; when At that time, the secondary indicator switched from a quality contribution attribute to a transformation risk attribute. The mass loss term for a pair of adjacent nodes is denoted as: No. The secondary index contribution of a group of adjacent node pairs is denoted as: Auxiliary correction items are denoted as: The compensation limiting parameter is denoted as In this embodiment The compensation limit item is denoted as: No. The gating quality score of a pair of adjacent nodes is denoted as: in, For quality loss items, As a contributor to secondary indicators, As an auxiliary correction item, To compensate for the amplitude limit, This is the gate control quality score. For the first The loss weights of each prototype metric, For the first The contribution weight of each secondary indicator For the first The loss weights of each auxiliary indicator, For the first Risk deduction coefficient for each secondary indicator For the first The contribution inhibition coefficient of each secondary indicator, For the first The auxiliary adjustment coefficients for each auxiliary indicator. All weights and coefficients mentioned above are non-negative, and the risk deduction coefficient and contribution suppression coefficient are both values ​​between 0 and 1. and All values ​​are non-negative. The sum of all loss weights and contribution weights in the same scoring configuration does not exceed 1, and the sum of auxiliary correction coefficients does not exceed [a certain value]. The risk deduction coefficient, contribution inhibition coefficient, and auxiliary correction coefficient are derived from historical batch scoring verification, acid-heat simulation conversion tests, and confirmation records of enterprise internal control standards or expert rules. These coefficients remain unchanged within the same batch calculation period. and When both are empty, the adjacent node pair If it is marked as not calculable, it will not be included in the batch scoring.

[0037] For example, if the levels of saikosaponin b1 and saikosaponin b2 increase during the drying process, but the fingerprint similarity decreases and the total saikosaponin index does not improve synchronously, the compensation gating data applies a high deduction to the contribution of this increase, and the gating quality score is not directly adjusted upwards according to the increase of secondary indicators. When the conversion trigger vector does not meet the preset trigger rules, the increase of saikosaponin b1, saikosaponin b2, or saikosaponin h in subsequent nodes relative to the previous nodes maintains the quality contribution attribute, and is included in the gating quality score according to the contribution value corrected by the auxiliary indicator vector. This results in a scoring result that takes into account the quality contribution of multiple indicators and the risk of process conversion.

[0038] The risk outcome output module is used to output the quality judgment results of Bupleurum preparations and the process transformation risk results based on the gated quality score, indicator construction results, and preset trigger rules. It combines the gated quality score, indicator construction results, and preset trigger rules to form output results targeting the batch quality and process risk of Bupleurum preparations. The module outputs the quality assessment results and process transition risk results for Bupleurum preparations. The quality assessment results reflect the quality status of the batch under multi-index gating scoring, while the process transition risk results reflect potential process transition anomalies between adjacent target nodes. This module ensures that the quality assessment results are not limited to a single endpoint indicator, but simultaneously present both the overall batch quality status and the risk information between process nodes.

[0039] The system reads the gate quality score, indicator construction results, and preset trigger rules, and verifies each target adjacent node pair of the same batch of Bupleurum preparations item by item according to the batch number. When the target adjacent node pair meets the preset trigger rules, the system records the preceding and succeeding nodes in the target adjacent node pair, and extracts the corresponding risk indicator name, risk trigger basis, and risk formation type. Among them, the risk trigger basis includes the composite transition characteristic value, acid-heat exposure coupling value, and transformation path consistency value. The risk formation type is determined according to the process link in which the target adjacent node pair is located, including at least one of the following: extraction transformation risk, concentration transformation risk, drying transformation risk, granulation transformation risk, processing transformation risk, and storage transformation risk. For example, between the concentration node and the drying node, when saikosaponin b1 and saikosaponin b2 significantly increase in subsequent nodes, while saikosaponin a and saikosaponin d decrease synchronously, and when the composite transition characteristic value, acid-heat exposure coupling value, and transformation path consistency value all reach the thresholds corresponding to the preset triggering rules, the concentration node and the drying node are recorded as risk process nodes, saikosaponin b1 and saikosaponin b2 are recorded as risk indicator names, and the risk formation type is recorded as drying transformation risk; The quality judgment results of Bupleurum preparations are generated based on the gated quality score value, and the adjacent node pairs of targets that meet the preset trigger rules are organized into process transformation risk results. The quality judgment results of Bupleurum preparations are formed into quality statuses such as qualified, warning or unqualified according to the judgment interval of the gated quality score value. The process transformation risk results are recorded in a structured manner according to the risk process node, risk indicator name, risk trigger basis and risk formation type. Batch scoring incorporates adjacent node pairs with gating quality scores. The batch gating quality score for the same batch of Bupleurum preparations is denoted as follows: , Take the minimum gating quality score among all adjacent node pairs included in the batch scoring. If no adjacent node pair is included in the batch scoring, If a data point cannot be calculated, the batch quality assessment result is recorded as insufficient data. Valid data coverage is the proportion of valid adjacent node pairs included in the batch scoring to the total number of adjacent node pairs to be tested. Valid data coverage must not be lower than the minimum coverage threshold, which is 80% in this embodiment. If any preset key adjacent node pair cannot be calculated, or if the valid data coverage is lower than the minimum coverage threshold, the batch quality assessment result is recorded as insufficient data. Preset key adjacent node pairs are determined in the parameter configuration record and include at least one of the following: concentration to drying, drying to granulation, pre-processing to post-processing, and pre-storage to post-storage. The quality assessment threshold includes a pass / fail threshold. and warning threshold The warning threshold is lower than the qualified threshold, and both are between 0 and 100. In this embodiment, When a batch's quality control score reaches the acceptable threshold, the batch quality is deemed acceptable; if it falls below the acceptable threshold but reaches the warning threshold, the batch quality is deemed warning; if it falls below the warning threshold, the batch quality is deemed unacceptable. If the company has at least twenty historical acceptable batch score data, Take the 5th percentile of the historical gate quality score for qualified batches; when historical risky batch score data exists. The candidate cutoff point with the highest distinguishability between qualified and risky batches is selected. The distinguishability of the candidate cutoff point for the quality scoring threshold is evaluated by subtracting 1 from the sum of sensitivity and specificity; when multiple candidate cutoff points with the same distinguishability exist, the higher value among them is used for the quality scoring threshold. If the empirically calculated warning threshold is not lower than the qualified threshold, this result is not directly adopted, but re-determined according to the parameter confirmation record, while keeping the warning threshold lower than the qualified threshold. The batch quality output includes the batch gating quality score. Effective data coverage and qualified threshold Warning threshold And batch quality judgment results.

[0040] The results of process transformation risk are output in a field format, including the target adjacent node pair. Risk indicator name, risk triggering basis, and compound transition characteristic value Transition threshold Acid-heat exposure coupling value Exposure threshold Conversion path consistency value Path threshold Comprehensive trigger score Comprehensive trigger threshold The system includes trigger rule types, process risk formation types, chemical risk subtypes, and a set of missing indicators. Risk triggering is based on the recorded threshold-reaching criteria, comprehensive trigger score, corresponding threshold, and trigger rule type. Process risk formation types are recorded by adjacent nodes to the target process step as extraction / conversion risk, concentration / conversion risk, drying / conversion risk, granulation / conversion risk, processing / conversion risk, or storage / conversion risk. Chemical risk subtypes are recorded as prototype conversion risk or secondary accumulation risk. When no... When secondary indicators are generated, the set of results for process transformation risk is recorded as an empty set, and the risk indicator name, risk triggering basis, and risk formation type are all recorded as not triggered.

[0041] Parameter configuration records are organized by category. Detection conversion parameters include the slope of the standard curve. Standard curve intercept The rules for converting the limit of quantitation (LOQ), the method for calculating the relative peak area, and the rules for selecting internal standard peaks or common peaks; the indicator configuration parameters include the name of the detection indicator, the serial number of the detection indicator, the classification results of the prototype indicator, secondary indicators, and auxiliary indicators, as well as the positive or negative attributes of each detection indicator; the standardization parameters include... Indiscriminate data processing rules; missing data handling parameters include endpoint missing replacement rules, consecutive missing replacement rules, minimum quantity requirements for historical qualified batches, and missing indicator set recording rules; processing intensity parameters include and processing intensity interval rule types; transformation mapping parameters include and Node distance parameters include and The node trust weight parameters include: and Acid-heat exposure parameters include In accordance with the rules governing the deviation between pH and heat exposure recording time, when an alkaline risk item exists, the acid-heat exposure parameters also include the name of the alkaline risk item, its triggering conditions, and the scoring method; transition weight parameters include... and Trigger threshold parameters include And threshold candidate boundary point discrimination evaluation rules; gating scoring parameters include and Batch scoring parameters include A minimum coverage threshold and a list of preset key adjacent node pairs are provided. Parameter configuration records are saved in association with product name, formulation type, process node sequence, test method version, and parameter confirmation date. Parameter configuration records remain unchanged during the same batch calculation period. When parameters are adjusted, a new parameter configuration record is created, while the previous parameter record is retained.

[0042] For example, if the gated quality score of the same batch of Bupleurum granules is in the warning range, and there is an increase in saikosaponin B1 and saikosaponin B2 from the concentration node to the drying node, accompanied by enhanced acid and heat exposure, the output content includes the batch's quality judgment result as a warning, the risk process node as the concentration node to the drying node, the risk indicator names as saikosaponin B1 and saikosaponin B2, the risk triggering basis as the composite transition characteristic value reaching the transition threshold, the acid and heat exposure coupling value reaching the exposure threshold, and the conversion path consistency value reaching the path threshold, and the risk formation type as drying conversion risk, so that the batch quality status and the specific process risk location are presented simultaneously.

[0043] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A multi-index quality assessment system for Bupleurum preparations, characterized in that, It includes a processor and a memory. The memory stores a computer program executed by the processor. The computer program includes a data matrix acquisition module, an indicator vector construction module, a path feature calculation module, a gated score generation module, and a risk result output module. The data matrix acquisition module is used to receive batch node data of the same batch of Bupleurum preparations and store the batch node data as a batch data matrix. The batch node data includes node detection data, node exposure data, node interval data and node reliability data in the process node sequence. The indicator vector construction module is used to generate indicator construction results based on batch node data. The indicator construction results include node path data, indicator feature data, and transformation mapping data. The node path data includes the set of adjacent node pairs, target adjacent node pairs, normalized node distance, and node confidence weight. The indicator feature data includes the prototype indicator set, secondary indicator set, auxiliary indicator set, and standard feature data. The path feature calculation module is used to generate a transformation trigger vector based on the index construction results and node exposure data. The transformation trigger vector includes the composite transition feature value, acid-heat exposure coupling value and transformation path consistency value of the target adjacent node pairs. The gating score generation module is used to generate gating quality scores based on conversion trigger vectors, indicator construction results, and preset trigger rules. The risk result output module is used to output the quality judgment result of Bupleurum preparation and the process transformation risk result based on the gate quality score, indicator construction result and preset trigger rules.

2. The multi-index quality judgment system for Bupleurum preparations according to claim 1, characterized in that, The process node sequence includes at least three process nodes arranged in the order of preparation, processing, or storage. The process nodes are selected from the extraction node, concentration node, drying node, granulation node, processing node, finished product node, and storage node.

3. The multi-index quality judgment system for Bupleurum preparations according to claim 2, characterized in that, The node detection data includes content data or peak area data of at least two of saikosaponin a, saikosaponin c, and saikosaponin d, and at least two of saikosaponin b1, saikosaponin b2, and saikosaponin h, and includes at least one of the following: total saikosaponin index, fingerprint spectrum similarity index, and non-saikosaponin quality index.

4. The multi-index quality judgment system for Bupleurum preparations according to claim 3, characterized in that, The node exposure data includes pH and thermal exposure data corresponding to each process node; the node interval data includes at least one of the time interval between adjacent process nodes, process sequence interval, and treatment intensity interval; the node reliability data includes at least one of the detection repeatability data, peak signal-to-noise ratio data, and sampling time deviation data.

5. The multi-index quality judgment system for Bupleurum preparations according to claim 4, characterized in that, The set of adjacent node pairs includes at least two sets of adjacent node pairs, each set of adjacent node pairs includes a predecessor node and a successor node; the target adjacent node pair is any adjacent node pair in the set of adjacent node pairs; the normalized node distance is determined based on the node interval data and corresponds to the adjacent node pair; the node credibility weight is determined based on the node credibility data and corresponds to each process node.

6. The multi-index quality judgment system for Bupleurum preparations according to claim 5, characterized in that, The prototype index set includes at least two of saikosaponin a, saikosaponin c, and saikosaponin d; the secondary index set includes at least two of saikosaponin b1, saikosaponin b2, and saikosaponin h; the auxiliary index set includes at least one of total saikosaponin index, fingerprint spectrum similarity index, and non-saikosaponin quality index; and the standard feature data includes the prototype index vector, secondary index vector, and auxiliary index vector corresponding to each process node.

7. The multi-index quality judgment system for Bupleurum preparations according to claim 6, characterized in that, The generation of composite transition feature values ​​includes: determining the predecessor and successor nodes in the target adjacent node pair, and generating composite transition feature values ​​corresponding to the target adjacent node pair based on the standard feature data corresponding to the predecessor and successor nodes, the normalized node distance corresponding to the target adjacent node pair, the transformation mapping data, and the node confidence weights corresponding to the predecessor and successor nodes.

8. The multi-index quality judgment system for Bupleurum preparations according to claim 1, characterized in that, The acid-heat exposure coupling value is generated as follows: Based on the node exposure data of the preceding and following nodes in the target adjacent node pair, the acid-base deviation component and heat exposure component corresponding to the target adjacent node pair are generated. An interactive gain is generated based on the common variation relationship between the acid-base deviation component and the heat exposure component in the target adjacent node pair, and the acid-heat exposure coupling value corresponding to the target adjacent node pair is generated based on the interactive gain.

9. A multi-index quality assessment system for Bupleurum preparations according to any one of claims 1-8, characterized in that, The generation of the gating quality score includes: configuring the indicators in the secondary indicator set as quality contribution attributes; when the conversion trigger vector meets the preset trigger rules, switching the indicators in the secondary indicator set that increase in the successor node relative to the predecessor node of the target adjacent node pair from quality contribution attributes to conversion risk attributes, and generating compensation gating data based on the conversion risk attributes; when the conversion trigger vector does not meet the preset trigger rules, maintaining the quality contribution attributes of the secondary indicator set that increases in the successor node relative to the predecessor node of the target adjacent node pair; and generating the gating quality score based on the quality contribution attributes, conversion risk attributes, auxiliary indicator vectors, and compensation gating data.

10. The multi-index quality judgment system for Bupleurum preparations according to claim 9, characterized in that, The process transformation risk results are generated based on the target adjacent node pairs that meet the preset triggering rules. The process transformation risk results include the risk process node, the risk indicator name, the risk triggering basis, and the risk formation type. Risk triggering criteria include composite transition characteristic values, acid-heat exposure coupling values, and transformation path consistency values; risk formation types include at least one of extraction transformation risk, concentration transformation risk, drying transformation risk, granulation transformation risk, processing transformation risk, and storage transformation risk.

Images