An on-line monitoring method for the tabletting quality of fumaric acid tablets

Abstract

The application discloses an online monitoring method for tabletting quality of fumaric acid tablets, and relates to the technical field of pharmaceutical preparation quality monitoring. Part of the tablets are periodically shunted as samples into a detection channel at a preset frequency period according to a tablet ejection slide of a tablet machine, physical index detection, near-infrared spectrum acquisition and appearance image acquisition are simultaneously performed on the tablets in the samples, the data in the same sampling period are respectively statistically processed to obtain physical index statistical characteristics, spectrum statistical characteristics and appearance defect statistical characteristics, the three types of statistical characteristics are used to construct a multivariate input feature vector, and a multivariate statistical process control model is used to calculate a judgment statistic, when the judgment statistic exceeds a control limit, abnormal root cause analysis is performed and a processing operation is executed. Through synchronous acquisition and collaborative analysis of multidimensional quality data, comprehensive judgment and early warning of the tabletting quality are realized, the accuracy and timeliness of abnormal detection are improved, and the quality controllability of the tabletting production of the fumaric acid tablets is effectively improved.

Description

Technical Field

[0001] This invention relates to the field of online monitoring technology for pharmaceutical preparation quality, and in particular to an online monitoring method for the tableting quality of fumaric acid tablets. Background Technology

[0002] Direct powder compression (DPCC) is increasingly used in the production of fumaric acid tablets due to its simple process, energy and time savings, and its ability to protect the stability of heat- and moisture-sensitive drugs. However, this process has stringent requirements for the powder properties of raw materials and excipients. In actual production, quality fluctuations such as uneven material mixing or poor tablet appearance may occur, placing higher demands on online monitoring of the tableting process.

[0003] To address the quality control needs of the tablet compression process, the industry has developed three mature online detection technologies: The first is online physical index detection, which uses an automated sampling mechanism to measure physical parameters such as tablet weight, hardness, and thickness in real time and provides feedback to adjust tablet compressor parameters. The second is online chemical detection based on near-infrared spectroscopy, a technology formally recognized by the Process Analytical Technology (PAT) framework, enabling non-destructive and rapid determination of the active pharmaceutical ingredient content in tablets. The third is visual defect detection based on machine vision, which can automatically identify common appearance problems such as cracked tablets, missing corners, and discoloration, and reject defective products. These technologies have been widely applied in pharmaceutical production, providing a fundamental guarantee for tablet compression quality control.

[0004] Nevertheless, existing technologies still have significant limitations in practical applications: various detection methods are typically deployed independently on different devices, resulting in asynchronous data acquisition and an inability to perform collaborative processing under the same analytical model; even in a few solutions that achieve online integration of physical and chemical indicators, the two types of data are analyzed independently, without inputting multidimensional quality data into the same analytical model for comprehensive judgment; most existing online quality control solutions are based on threshold deviations of single quality indicators to trigger parameter adjustments, making it difficult to fully utilize multidimensional data for comprehensive judgment and early warning. For fumarate tablets, this means that monitoring methods relying on single-dimensional indicators are unable to promptly detect process anomalies indicated by changes in multidimensional data combinations, ultimately making it difficult for existing methods to meet the requirements of high-quality production. Summary of the Invention

[0005] This invention provides an online monitoring method for the tableting quality of fumaric acid tablets, which solves the technical problem that existing online monitoring methods for the tableting quality of fumaric acid tablets are unable to fully utilize multidimensional data for comprehensive judgment and early warning.

[0006] The first aspect of this invention provides an online monitoring method for the tableting quality of fumaric acid tablets, comprising:

[0007] In the tablet ejection chute of the tablet press, a portion of the tablets are periodically diverted from the continuously falling tablets at a preset time frequency and sent into the testing channel as samples.

[0008] Within the detection channel, physical index detection, near-infrared spectral acquisition, and appearance image acquisition are performed simultaneously on the tablets in the sample to obtain physical index data, spectral data, and appearance image data for each tablet.

[0009] The physical index data, spectral data and appearance image data of all tablets obtained in the same sampling period are statistically processed to obtain the physical index statistical characteristics, spectral statistical characteristics and appearance defect statistical characteristics of the sample.

[0010] The physical index statistical features, the spectral statistical features, and the appearance defect statistical features are used to construct the multivariate input feature vector corresponding to the sample. Based on the multivariate statistical process control model, the multivariate input feature vector is calculated to obtain the judgment statistic.

[0011] When the determination statistic exceeds the control limit, an anomaly root cause analysis is performed based on the determination statistic, and corresponding processing operations are executed according to the analysis results.

[0012] Optionally, the step of constructing the multivariate input feature vector corresponding to the sample using the physical index statistical features, the spectral statistical features, and the appearance defect statistical features includes:

[0013] The physical index data and spectral data of each tablet in the same sample are matched one by one according to the tablet, and the single-particle correlation features between the physical index features and spectral features corresponding to each tablet are constructed respectively.

[0014] Based on the single-pill correlation characteristics corresponding to each of the aforementioned tablets, a comprehensive deviation distance is calculated to characterize the deviation between the physical index statistical characteristics and the spectral statistical characteristics. The comprehensive deviation distance is the degree of deviation between the physical index statistical characteristics and the spectral statistical characteristics caused by material stratification when the relative standard deviation of tablet weight difference does not increase.

[0015] The comprehensive deviation distance is used as a cross-dimensional correlation feature, and together with the physical index statistical features, the spectral statistical features, and the appearance defect statistical features, it is used to construct a multivariate input feature vector.

[0016] Optionally, the multivariate input feature vector may further include at least one of the following: the ratio of sheet weight to hardness, the ratio of sheet weight to thickness, or the ratio of hardness to thickness.

[0017] Optionally, the multivariate statistical process control model is a principal component analysis model, and the decision statistics include the Hotelling statistic and the Q residual statistic.

[0018] Optionally, the step of performing anomaly root cause analysis based on the judgment statistic when the judgment statistic exceeds the control limit, and executing corresponding processing operations according to the analysis results, includes:

[0019] When the judgment statistic exceeds the control limit, the key variables that cause the judgment statistic to exceed the limit are identified by contribution plot analysis.

[0020] When the key variable is the comprehensive deviation distance, the abnormal root cause type is determined to be the raw material stratification mode, and an adjustment command is generated and executed to increase the speed of the feeding paddle in the tablet press hopper or to increase the frequency of the vibrator in the feeding channel.

[0021] When the key variable is the edge peeling defect rate in the appearance defect statistical features, the abnormal root cause type is determined to be the sticking and punching early warning mode, and an adjustment command to issue a cleaning early warning prompt is generated and executed, or an adjustment command to fine-tune the main pressure is generated and executed.

[0022] Optionally, when the determined abnormal root cause type is a sticking and impact warning mode, and the adjustment command for fine-tuning the main pressure is generated, the fine-tuning main pressure is to increase the current main pressure by a preset percentage.

[0023] The second aspect of this invention provides an online monitoring system for the tableting quality of fumaric acid tablets, comprising:

[0024] The sampling and diversion module is used to periodically divert a portion of the tablets from the continuously falling tablets in the tablet ejection chute of the tablet press at a preset time frequency and send them into the detection channel as samples.

[0025] The synchronous detection module is used to simultaneously perform physical index detection, near-infrared spectral acquisition, and appearance image acquisition on the tablets in the sample within the detection channel, so as to obtain physical index data, spectral data, and appearance image data for each tablet.

[0026] The feature extraction module is used to perform statistical processing on the physical index data, spectral data and appearance image data of all tablets obtained in the same sampling period to obtain the physical index statistical features, spectral statistical features and appearance defect statistical features corresponding to the sample.

[0027] The statistical control module is used to construct a multivariate input feature vector corresponding to the sample by using the physical index statistical features, the spectral statistical features and the appearance defect statistical features, and to calculate the multivariate input feature vector based on the multivariate statistical process control model to obtain the judgment statistic.

[0028] An anomaly response module is used to perform anomaly root cause analysis based on the judgment statistic when the judgment statistic exceeds the control limit, and to perform corresponding processing operations based on the analysis results.

[0029] A third aspect of the present invention provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor performs the steps of the online monitoring method for the tableting quality of fumaric acid tablets as described above.

[0030] The fourth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed, implements the online monitoring method for the tableting quality of fumaric acid tablets as described above.

[0031] The fifth aspect of the present invention provides a computer program product comprising a computer program stored on a non-transitory computer-readable storage medium, the computer program comprising program instructions, wherein when the program instructions are executed by a computer, the computer performs the online monitoring method for the tableting quality of fumaric acid tablets as described above.

[0032] As can be seen from the above technical solutions, the present invention has the following advantages:

[0033] The periodic shunting sampling method employed in this invention can acquire representative samples with uniform time distribution without interfering with the continuous production rhythm of the tablet press. This avoids sample bias that may be caused by random sampling, ensuring that the monitoring results accurately reflect the quality fluctuations during continuous production. Simultaneous acquisition of physical indicators, near-infrared spectra, and appearance images is completed within the same detection channel, and precise correlation of single-particle data is achieved through unified identification. This fundamentally solves the problems of data timing misalignment and sample matching errors that exist when multiple systems are deployed independently, providing an accurate and reliable data foundation for subsequent comprehensive analysis. By inputting three different dimensions of quality statistical features into the same multivariate statistical process control model for collaborative calculation, this invention overcomes the limitations of existing technologies based on single-indicator threshold judgments. It can comprehensively consider the overall changing trends of multiple quality dimensions, effectively reducing the probability of false alarms caused by normal random fluctuations of a single indicator. Simultaneously, it can identify early quality anomaly signals of multi-dimensional collaborative changes that are difficult to detect with single detection methods, issuing warnings before quality deviations develop to a significant exceeding stage. Based on this, anomaly root cause analysis is performed using judgment statistics, and targeted processing is implemented. This changes the passive mode of existing technologies, which can only issue general alarms and require manual investigation one by one. It can provide operators with clear anomaly indications, shorten anomaly response time, and reduce the number of defective products caused by persistent anomalies. The overall technical solution meets the basic requirements of process analysis technology in the pharmaceutical industry. It can be implemented without large-scale modification of the existing tableting machine structure. While controlling modification costs, it effectively improves the quality controllability and production stability of the fumaric acid tablet compression process. Attached Figure Description

[0034] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0035] Figure 1 A flowchart illustrating the steps of an online monitoring method for the tableting quality of fumaric acid tablets provided in this embodiment of the invention;

[0036] Figure 2 A flowchart illustrating the steps for calculating the comprehensive deviation distance in an embodiment of the present invention;

[0037] Figure 3 A flowchart illustrating the steps of anomaly root cause analysis and processing provided in this embodiment of the invention;

[0038] Figure 4A structural block diagram of an online monitoring system for the tableting quality of fumaric acid tablets provided in an embodiment of the present invention;

[0039] Figure 5 This is a structural block diagram of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0040] This invention provides an online monitoring method for the tableting quality of fumaric acid tablets, which addresses the technical problem that existing online monitoring methods for the tableting quality of fumaric acid tablets cannot perform multi-dimensional comprehensive prediction and proactive adjustment in the early stages of abnormalities.

[0041] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. It should be noted that in the optional embodiments of the present invention, the object information and other related data involved require the permission or consent of the object when the embodiments of the present invention are applied to specific products or technologies, and the collection, use, and processing of related data must comply with the relevant laws, regulations, and standards of the relevant countries and regions. That is to say, if the embodiments of the present invention involve data related to the object, it needs to be obtained with the authorization and consent of the object, the authorization and consent of the relevant departments, and in compliance with the relevant laws, regulations, and standards of the country and region. If personal information is involved in the embodiments, the acquisition of all personal information requires the consent of the individual. If sensitive information is involved, the separate consent of the information subject is required, and the embodiments also need to be implemented with the authorization and consent of the object.

[0042] Please see Figure 1 , Figure 1 The flowchart illustrates the steps of an online monitoring method for the tableting quality of fumaric acid tablets provided in this embodiment of the invention.

[0043] This invention provides an online monitoring method for the tableting quality of fumaric acid tablets, comprising:

[0044] Step 101: In the tablet ejection chute of the tablet press, a portion of the tablets are periodically diverted from the continuously falling tablets at a preset time frequency and sent into the detection channel as samples.

[0045] It should be noted that the preset time frequency is not a fixed value, but is dynamically adjusted according to the real-time rotation speed of the tablet press. Specifically, the controller reads the spindle speed signal of the tablet press in real time and calculates the number of tablets produced per minute. (Slices / minute). Sampling period Calculate using the following formula:

[0046] ;

[0047] in, This is the sampling ratio coefficient, with a value ranging from 0.001 to 0.01. For example, when the tablet press is running at a speed of 8000 tablets / minute, take... =0.005, then =60 / (8000×0.005)=1.5 seconds, meaning that a split is performed every 1.5 seconds, and approximately 10 samples are taken each time.

[0048] In this embodiment of the invention, a diversion device is installed downstream of the tablet ejection chute of a tablet press (e.g., a FETTE 2090i rotary tablet press). This diversion device consists of a rotary paddle and a Y-shaped guide groove. The rotary paddle is driven by a servo motor and its surface is coated with a medical-grade polyurethane elastic layer to prevent mechanical damage to the tablets. The tablet press operates continuously at a constant speed, and the compressed fumaric acid tablets slide continuously down the tablet ejection chute under gravity. The controller sends trigger signals to the servo motor at a preset time frequency, driving the rotary paddle to swing periodically. When the paddle swings to the diversion position, a portion of the continuously sliding tablets is diverted into one branch of the guide groove, which connects to the detection channel; the remaining tablets continue to enter the finished product collection container along the other branch of the guide groove (the main chute). One diversion action lasts approximately 0.2 seconds and can divert approximately 10 tablets as samples.

[0049] Using periodic shunting instead of random sampling ensures a uniform distribution of samples over time, avoiding sample bias caused by fluctuations in tablet press speed or uneven material flow. Simultaneously, by dynamically adjusting the sampling period, it can adapt to the statistical representativeness requirements at different production speeds, ensuring sufficient data for quality analysis while avoiding excessive shunting that could impact production efficiency. After shunting, the sample tablets are fed into the testing channel via a guide groove.

[0050] Step 102: In the detection channel, the physical index detection, near-infrared spectral acquisition and appearance image acquisition are performed on the tablets in the sample simultaneously to obtain the physical index data, spectral data and appearance image data of each tablet.

[0051] In this embodiment of the invention, the detection channel adopts a linear layout design, in which tablets pass through each detection station sequentially at a constant speed v (e.g., 5-10 mm / s). To ensure data synchronization, all detection stations are synchronously triggered by the same controller using a unified clock source. Each tablet is assigned a unique ID when entering the detection channel, and the data collected by each subsequent station is associated with this ID and timestamped.

[0052] (1) Physical index testing: Physical index testing includes three sub-items: sheet weight testing, hardness testing and thickness testing, which are completed by the corresponding workstations.

[0053] The weighing station is located at the entrance of the testing channel, and its core component is a high-precision electromagnetic force balance sensor. The tablets are first fed onto the weighing pan at this station, and after stabilization, their weight is recorded at a sampling frequency of 1000Hz. After weighing, the tablets are transported to the next station by a conveyor belt or pusher. This type of online weighing device is standard technology in the field; for example, the SOTAX AT50 automatic tablet testing system integrates online weighing functionality with a weighing accuracy of ±0.1mg; the PCT-100 multi-parameter intelligent hardness tester also incorporates a Mettler high-precision weighing sensor with a range of 120g and an accuracy of 0.1mg.

[0054] The hardness and thickness testing station employs an integrated testing device. After the tablet enters this station, a laser displacement sensor first emits a laser beam from above to measure the tablet thickness, with a measurement range of 0–10 mm and an accuracy of ±0.01 mm. After thickness measurement, the tablet is sent to the hardness testing station, where a pressure sensor applies a constant-speed compressive force until the tablet breaks. The maximum pressure value at the moment of breakage is recorded as the hardness value, with a measurement range of 0–300 N and an accuracy of ±1 N. This integrated hardness and thickness testing device is common in the field. For example, the SOTAX AT50 thickness gauge with an integrated TouchControl™ sensor can avoid sample deformation during measurement, obtaining high-precision measurement results; the PCT-100 intelligent hardness tester can also complete both thickness and hardness testing in one operation. The tablet weight, hardness, and thickness test data are recorded individually for each tablet, constituting the physical property data for each tablet, and are represented by vectors. It means that, among them, For film weight, For hardness, For thickness.

[0055] (2) Near-infrared spectral acquisition

[0056] After physical property testing is completed, the tablets are transported to the near-infrared spectroscopy acquisition station. This station uses a diffuse reflectance near-infrared spectrometer, with the spectrometer probe vertically aligned with the tablet surface and connected to the main unit via optical fiber. The spectral acquisition parameters are as follows: wavelength range 900–1700 nm, spectral resolution… The integration time is 100ms, and diffuse reflectance spectral data of the tablet is acquired each time. Spectral acquisition is synchronously triggered by the controller and physical index detection, ensuring precise temporal correspondence between the spectral data and physical index data of the same tablet. Several manufacturers in this field offer near-infrared spectrometers and diffuse reflectance probes suitable for online tablet detection. For example, the AOTF near-infrared tablet analyzer has a spectral range of 850–2400 nm and a resolution of 2–10 nm; the Thermo Scientific SabIR diffuse reflectance probe enables remote, non-destructive sampling of drug samples via optical fiber, and its stainless steel structure and sapphire window are suitable for pharmaceutical manufacturing environments. The above near-infrared spectral data are recorded one tablet at a time, constituting the spectral data for each tablet, and vectorized. This indicates that k is the number of wavelength points (e.g., 120).

[0057] (3) Appearance image acquisition

[0058] After near-infrared spectral acquisition, the tablets are transported to the appearance image acquisition station. This station is equipped with four high-frame-rate industrial cameras, which simultaneously capture images of the tablets from four directions: top, bottom, left, and right. The camera parameters are as follows: resolution 1280×1024 pixels, frame rate 30fps. Each tablet is captured in four images, covering its top surface, bottom surface, and four sides, providing a complete view of the tablet's appearance without blind spots. Mature tablet appearance visual inspection systems already exist in this field. For example, ACG's VISITAB MH series uses 10 next-generation color cameras to cover the top, bottom, sides, and edges of the tablet through a multi-view system, achieving 360° inspection of tablet appearance defects. Hanlin Aerospace's intelligent inspection solution also integrates high-precision micro-weighing technology and image recognition technology, which can detect defects such as damage, black spots, poor coating, and adhesion, and automatically reject them. It should be noted that appearance image acquisition is a relatively mature technology in the pharmaceutical field. This invention does not aim to limit a specific image acquisition device, but rather to use this method to obtain appearance image data for subsequent statistical analysis. The aforementioned appearance image data is recorded one by one for each tablet, and combined with the extraction of appearance defect statistical features in subsequent step 103, statistical quantities such as edge peeling defect rate can be calculated.

[0059] It is worth noting that simultaneous detection in this invention refers to the simultaneous detection of various test items for the same tablet within the same sampling period, with the detection items being sufficiently close in time and correlated, rather than requiring each station to complete the measurement at the same nanosecond. In this embodiment, because the tablet's movement speed within the detection channel is constant and the distance between each station is fixed, the time difference between the same tablet entering the weighing station and completing all tests is... As a fixed value, the controller sets the trigger delay of each station based on this time difference to ensure that the physical index data, spectral data and appearance image data of each tablet can be accurately matched and associated with the same tablet ID.

[0060] Step 103: Perform statistical processing on the physical index data, spectral data and appearance image data of all tablets obtained in the same sampling period to obtain the physical index statistical characteristics, spectral statistical characteristics and appearance defect statistical characteristics of the corresponding samples.

[0061] In this embodiment of the invention, one sampling period includes m tablets (m is 10, but can also be adjusted within the range of 5 to 20 according to the actual production speed). The controller summarizes the data of each tablet obtained in step 102 according to the tablet ID to form a data set within the sampling period, and then performs statistical processing on the three types of data respectively.

[0062] (1) Statistical processing of physical index data

[0063] Physical parameters include the weight of each tablet. ,hardness ,thickness Statistical processing was performed on the physical property data of all tablets within the same sampling period, and the following statistical characteristics were calculated:

[0064] 1) Central tendency feature: Mean, which is used to characterize the average level of the sample slice weight.

[0065] ;

[0066] ;

[0067] ;

[0068] in, The arithmetic mean of the sample piece weights; This represents the total number of tablets collected during the current sampling period (m=10 in this example). The weight of the i-th tablet is expressed in mg. This represents the arithmetic mean of the sample hardness. The hardness of the i-th tablet is expressed in N. The arithmetic mean of the sample thickness; The thickness of the i-th tablet is expressed in mm.

[0069] 2) Standard Deviation (SD) and Relative Standard Deviation (RSD)

[0070] ;

[0071] ;

[0072] in, The sample standard deviation of slice weight reflects the degree of dispersion in slice weight; The relative standard deviation of tablet weight, in % %. The weight of the i-th tablet is expressed in mg. This is the arithmetic mean of the sample weights. Similarly, calculate the standard deviation of hardness. and thickness standard deviation The relative standard deviation of thickness .

[0073] 3) Extreme value characteristics

[0074] ;

[0075] in, This represents the maximum slice weight in the sample. This represents the minimum slice weight in the sample. This refers to the weight of the m-th tablet, in mg. Hardness Minimum hardness in the sample and thickness Minimum thickness in the sample Similarly.

[0076] The above statistical characteristics together constitute the set of statistical characteristics of physical indicators, denoted as When constructing multivariate input feature vectors, some features can be selected according to the analysis needs (usually the mean or RSD is sufficient).

[0077] (2) Statistical processing of spectral data

[0078] The spectral data consists of a vector of absorbance values ​​for each tablet at K wavelengths (K=120 in this example). Before performing statistical processing on the spectral data of all tablets within the same sampling period, spectral preprocessing is required. The preprocessed spectral data is denoted as... .

[0079] 1) Average spectrum

[0080] ;

[0081] in, Let be the spectral vector of the i-th tablet after preprocessing; This is the average spectrum of the sample.

[0082] 2) Spectral residuals

[0083] ;

[0084] in, For Euclidean distance (L2 norm); The spectral residual (scalar) represents the degree of deviation between the spectrum of each tablet and the average spectrum.

[0085] 3) Spectral similarity

[0086] ;

[0087] in, For spectral similarity; This represents the total number of tablets in the current sampling period. , For the tablet index number; For the first Spectral vector of pretreated tablets; For the first Spectral vector of pretreated tablets; It is the dot product of two spectral vectors.

[0088] 4) Principal component score statistics

[0089] Principal component analysis was performed on the preprocessed spectral data, retaining the first few principal components (e.g., the first three principal components), and the score vector for each tablet was calculated. (Dimension 3), where, Let represent the scores of the i-th tablet on the first, second, and third principal components, respectively. Calculate the mean and variance of the scores:

[0090] ;

[0091] in, The sample mean of the scores of the first principal component; The score of the i-th tablet on the first principal component; This represents the sample variance of the scores for the first principal component. Similarly, the mean score for the second principal component can be calculated. Sample variance of the second principal component score and the mean score of the third principal component. Sample variance of the third principal component score .

[0092] (3) Statistical processing of appearance image data and acquisition of statistical features of appearance defects

[0093] The appearance image data consists of four images of each tablet: the top surface, bottom surface, and four sides. This embodiment employs a machine vision-based defect detection method. First, the images are preprocessed: color images are converted to grayscale, median filtering is used to remove noise, and histogram equalization is applied to enhance contrast. Then, the Canny operator is used to extract the tablet outline. For edge peeling defects, the curvature of each point on the outline is calculated; when the curvature exceeds a certain value... Furthermore, the area of ​​the inward concavity in that contour is greater than... When the area difference exceeds 5%, it is determined to be an edge peeling defect. For corner missing defects, the tablet outline is compared with the area of ​​a standard template; if the area difference exceeds 5%, it is determined to be a corner missing defect. For cracked tablet defects, a length greater than [a certain value] is detected in the grayscale image. Linear dark lines. For color spot defects, the image is converted to the HSV color space to detect areas of abnormal hue or saturation, with an area exceeding [a certain size]. It is then determined to be a pigmentation spot.

[0094] For all m tablets within the same sampling period, each tablet is assessed for defects such as edge peeling, missing corners, cracks, and discoloration using the algorithm described above. The frequency of each type of defect is then calculated. The edge peeling defect rate is calculated as the number of tablets with edge peeling divided by m and then multiplied by 100%. Similarly, the missing corner defect rate, cracked tablet defect rate, and discoloration defect rate are calculated. These four defect rates together constitute the statistical feature set of appearance defects, with the edge peeling defect rate serving as a key variable for subsequent determination of the adhesion and impact warning mode.

[0095] Through the statistical processing described above, this method can obtain statistical information on the appearance defects of each batch of samples online and in real time, especially for edge peeling defects that are prone to occur in fumaric acid tablets due to their high viscosity, achieving quantitative monitoring. Compared with general visual inspection models (whose training data mostly consists of common defects, resulting in low recognition rates for special defects such as edge peeling), this method improves detection accuracy and reduces false negative and false positive rates by specifically targeting the curvature analysis and area loss judgment of edge peeling.

[0096] Step 104: Construct a multivariate input feature vector corresponding to the sample using physical index statistical features, spectral statistical features, and appearance defect statistical features. Based on the multivariate statistical process control model, calculate the multivariate input feature vector to obtain the judgment statistic.

[0097] In this embodiment of the invention, physical, spectral, and appearance quality data are fused into a unified multivariate feature vector. Two comprehensive judgment statistics are then calculated using a statistical model to determine whether the tableting process deviates from normal conditions. First, the physical and spectral data of each tablet in the same batch are mapped one-to-one to establish a single-tablet correlation. Then, based on a mapping model established from normal production data, a comprehensive index reflecting the deviation between physical and spectral consistency is calculated. Subsequently, physical statistical features, spectral statistical features, appearance defect statistical features, and cross-dimensional deviation indicators are integrated into a multivariate input feature vector and standardized to eliminate dimensional differences. Next, a principal component analysis statistical process control model is pre-established based on normal production samples to obtain the control limits corresponding to the normal fluctuation range. During online operation, the feature vector of the current batch is input into the model, and the Hotelling statistic and Q-residual statistic are automatically calculated as the core judgment statistics for determining whether the production process is abnormal.

[0098] Furthermore, step 104 includes the following steps:

[0099] S11. Match the physical index data and spectral data of each tablet in the same sample one by one, and construct the single-particle correlation features between the physical index features and spectral features corresponding to each tablet.

[0100] In this embodiment of the invention, the controller has assigned a unique ID to each tablet and recorded the timestamps of each detection station. Because the tablet's movement speed within the detection channel is constant (e.g., ...), Since the distance between each workstation is fixed, the correspondence between the physical index data and spectral data of the same tablet can be determined based on the time difference.

[0101] Assume there are a total of Tablets ( =10). For the first Tablets ( ), whose physical index data constitute a vector (Sheet weight, hardness, thickness), its near-infrared spectral data constitute a vector. (After pretreatment) 3D spectrum Then the first... Single-particle association features of tablets Defined as The combination of physical parameters and spectral data of the tablet is used to record the pairings. By constructing single-particle association features, this method preserves the original correspondence between cross-dimensional information within each tablet.

[0102] S12. Based on the single-particle correlation characteristics of each tablet, calculate the comprehensive deviation distance between the statistical characteristics of the physical indicators and the statistical characteristics of the spectral indicators.

[0103] It should be noted that the comprehensive deviation distance refers to the degree of deviation between the statistical characteristics of physical indicators and the statistical characteristics of spectra due to material stratification, when the relative standard deviation of the tablet weight difference does not increase. Specifically, this distance is calculated using the prediction error method based on partial least squares (PLS) regression: first, a PLS regression model between physical indicators and spectral data is established using normal production batch data; during online detection, the corresponding spectra are predicted using the measured physical indicators; and the average Euclidean distance between the predicted spectrum and the measured spectrum is calculated, which is the comprehensive deviation distance.

[0104] In embodiments of the present invention, such as Figure 2 As shown, under normal tableting conditions, there is a stable correlation between physical indicators such as tablet weight and hardness and API content (reflected by near-infrared spectroscopy). When micro-stratification occurs in the premix within the hopper, the API content initially shifts. However, due to the macroscopic balance of the total tableting material, the RSD of tablet weight and hardness may still remain within the acceptable range specified in the pharmacopoeia (e.g., tablet weight RSD < 3%). At this point, the cross-dimensional consistency between physical indicators and spectral data is disrupted, and the overall deviation distance significantly increases. Therefore, this distance can capture stratification signals at an early stage when traditional physical indicator monitoring fails.

[0105] This embodiment uses a prediction error method based on partial least squares (PLS) regression to calculate the overall deviation distance. The method consists of two stages: offline modeling and online calculation.

[0106] (1) Offline modeling stage: Data from 30 batches totaling 3000 tablets under normal production conditions was collected, i.e., normal batch data was collected. This was based on the physical index matrix. With (3000×3) as the independent variable and the spectral matrix Y (3000×120) as the dependent variable, a PLS regression model was established, and the model parameters were saved to the controller. This model describes the mapping relationship between physical indicators and spectra under normal conditions. The optimal number of principal components was determined to be 8 through leave-one-out cross-validation, which explained a cumulative variance of 92.5%, and the regression coefficient matrix B (3×120) was obtained.

[0107] (2) Online calculation stage: For each tablet i in the current sampling period: the measured physical indicators are calculated. Input the PLS model for single-particle data matching and predict the spectral vector. ,in, The PLS regression coefficient matrix, Let i be the physical index vector of the i-th tablet; calculate the predicted spectral vector. With the measured spectral vector The Euclidean distance between them is used as the single-particle deviation distance:

[0108] ;

[0109] in, The number of wavelength points (K=120). and These are the predicted absorbance and the measured absorbance at the k-th wavelength, respectively.

[0110] Then, the average single-pill deviation distance of all tablets within this sampling period is calculated to obtain the comprehensive deviation distance D:

[0111] ;

[0112] In the above formula, The units are the same as those for spectral absorbance (dimensionless). This reflects the degree of deviation between the overall physical properties and the spectrum of the sample. When When the time exceeds the control limit determined based on normal historical data statistics, it indicates that material stratification may exist.

[0113] S13. The comprehensive deviation distance is used as a cross-dimensional correlation feature, and together with the physical index statistical features, spectral statistical features and appearance defect statistical features, it is used to construct a multivariate input feature vector.

[0114] In this embodiment of the invention, the multivariate input feature vector includes the following four types of features:

[0115] (1) Statistical characteristics of physical indicators (6 dimensions in total): mean weight of sheet, RSD of sheet weight, mean hardness, RSD of hardness, mean thickness, RSD of thickness;

[0116] (2) Spectral statistical characteristics (5 dimensions in total): mean score of the first principal component, variance of the score of the first principal component, mean score of the second principal component, variance of the score of the second principal component, and spectral residual E;

[0117] (3) Statistical characteristics of appearance defects (3 dimensions in total): edge peeling defect rate, corner missing defect rate, and chipping defect rate;

[0118] (4) Cross-dimensional correlation features (1 dimension in total): namely, the comprehensive deviation distance D calculated in step S12.

[0119] The four types of features are concatenated to form a 15-dimensional multivariate input feature vector. To eliminate the influence of differences in the dimensions and numerical ranges of the features, each feature is standardized before being input into the MSPC model, so that its mean is 0 and its variance is 1. The standardization formula is:

[0120] ;

[0121] in, These are the original eigenvalues; This is the mean value of this feature under normal operating conditions; The standard deviation is denoted as .

[0122] It should be noted that the "cross-dimensional correlation feature" in this step, namely the comprehensive deviation distance D, has the physical meaning of capturing the cross-dimensional consistency deviation between physical indicators and spectral data. It is the core feature for achieving early hierarchical detection in this invention. By inputting this feature together with traditional statistical features into a multivariate statistical process control model, more sensitive anomaly warnings can be achieved using cross-dimensional information.

[0123] Furthermore, this embodiment also incorporates at least one of the following into the multivariate input feature vector: the ratio of tablet weight to hardness, the ratio of tablet weight to thickness, and the ratio of hardness to thickness. These ratios can eliminate the influence of tablet size variations on quality indicators, more accurately reflecting the intrinsic quality of the tablets. For example, when the tableting depth changes slightly, both tablet weight and thickness will change simultaneously, but the tablet weight / thickness ratio remains essentially unchanged. Inputting these ratios as additional features into the MSPC model can further improve the sensitivity and accuracy of anomaly detection.

[0124] It is worth mentioning that the multivariate statistical process control model is a principal component analysis model, and the decision statistics include the Hotelling statistic and the Q-residual statistic. Principal component analysis is a classic multivariate dimensionality reduction technique that can transform high-dimensional, correlated original features into a set of linearly independent principal components, simplifying calculations while retaining most of the information. Specifically, at least 30 batches of samples under normal production conditions are first collected. For each sample, a multivariate input feature vector (after standardization) is constructed according to the aforementioned steps to form a reference dataset. Principal component analysis is performed on this dataset, and principal components are extracted through singular value decomposition. Based on the principle that the cumulative explained variance is greater than or equal to 85%, the top A principal components are retained (A=6 in this embodiment), thereby obtaining the loading matrix P, the score matrix T, and the covariance matrix Λ of the principal components (a diagonal matrix, with elements representing the variance of each principal component).

[0125] For samples collected online, their standardized feature vectors are denoted as x, and the model calculates two decision statistics. The Hotelling T² statistic, or Hotelling statistic, measures the degree to which a sample deviates from its normal operating region within the principal component space; its formula is as follows: The Q-residual statistic measures the portion of the residuals that cannot be explained by the principal component model, detecting random noise or novel anomalies. The calculation formula is as follows: , where e is the residual vector; Let Q be the residual statistic; This is the standardized feature vector of the current sample; The load matrix; This is the transpose of the load matrix; This is the transpose of the residual vector. The control limits for Q are based on the F distribution and The distribution is determined (confidence level is 95%). The T² and Q values ​​calculated online are compared with their control limits. If either statistic exceeds the corresponding control limit, an anomaly is determined to have occurred in the production process. At this point, the system proceeds to step 105 to perform anomaly root cause analysis and processing.

[0126] Step 105: When the judgment statistic exceeds the control limit, perform anomaly root cause analysis based on the judgment statistic and execute corresponding processing operations according to the analysis results.

[0127] Furthermore, step 105 includes the following steps:

[0128] S21. When the judgment statistic exceeds the control limit, the key variables that cause the judgment statistic to exceed the limit are identified through contribution plot analysis.

[0129] In embodiments of the present invention, such as Figure 3 As shown, in step 104, Hotling is calculated. After calculating the statistics and the Q-residual statistics, the controller compares each statistic with its corresponding control limits in real time.

[0130] When Hotling Statistic ≤ When the control limit is reached and the Q residual statistic is less than or equal to the Q control limit, the current production process is determined to be in a normal and stable state, and the normal production state processing procedure is executed.

[0131] When Hotling Statistics > When the control limit or Q residual statistic is greater than the Q control limit, an anomaly is determined to have occurred in the current production process, and this step and the subsequent anomaly handling procedures S22-S23 are immediately executed. Specifically, when an abnormal production state is determined, the following operations are performed first:

[0132] (1) Abnormal alarm trigger: Immediately issue an audible and visual alarm through the human-machine interface, and display the type of statistical quantity exceeding the limit in the alarm information ( Or Q), the time exceeding the limit, and the current statistical value.

[0133] (2) Key variable identification: Key variables leading to the exceeding of the judgment statistic are identified through contribution plot analysis. Specifically:

[0134] 1) For Q residual statistics exceeding the limit: calculate the square of the corresponding component of each input feature in the residual vector as the contribution value of that feature;

[0135] 2) Regarding Hotelling Statistics Exceeding Limits: Based on the loading matrix, the principal component score contributions are back-projected onto the original input feature space to obtain the contribution value of each feature;

[0136] 3) Sort all features by their contribution values ​​from largest to smallest. The feature with the largest contribution value is the core key variable causing this anomaly.

[0137] (3) Anomaly root cause pre-classification: Based on the identified core key variable types, the corresponding anomaly handling branch is automatically matched:

[0138] 1) If the key variable is the comprehensive deviation distance, proceed to step S22 to perform API stratification mode processing;

[0139] 2) If the key variable is the edge peeling defect rate, proceed to step S23 to execute the sticking and punching early warning mode processing;

[0140] 3) If the key variable is another physical indicator or spectral statistical characteristic, execute the optional general alarm procedure: generate an adjustment instruction to issue a general alarm prompt, requiring the operator to perform manual inspection, without the need for automatic adjustment of process parameters.

[0141] S22. When the key variable is the comprehensive deviation distance, the abnormal root cause type is determined to be the raw material stratification mode, and an adjustment command is generated and executed to increase the speed of the feeding paddle in the tablet press hopper or to increase the frequency of the vibrator in the feeding channel.

[0142] In this embodiment of the invention, this step is executed after identifying the core key variable as the comprehensive deviation distance in step S21. The comprehensive deviation distance is a cross-dimensional correlation feature calculated in step S12, which characterizes the degree of deviation between physical indicators (sheet weight, hardness, thickness) and spectral data (reflecting API content). Under normal production conditions, this distance is small and fluctuates smoothly; when micro-stratification occurs in the premixed material in the hopper, the API content shifts first while the statistical characteristics of the physical indicators have not yet changed significantly, and the comprehensive deviation distance will increase significantly first. Therefore, exceeding this distance limit is a specific direct signal of material stratification.

[0143] Once the drug substance is identified as stratified, the controller generates and executes corresponding adjustment commands. Specifically, one or both of the following measures can be selected: increasing the rotational speed of the feed paddle in the tablet press hopper, or increasing the frequency of the vibrator in the feeding channel. For example, the feed paddle rotational speed can be increased by 10% to 20% (e.g., by about 15%), and the vibrator frequency can be increased by 10% to 15% (e.g., by about 12%). These ranges are determined based on a conventional understanding of powder mixing processes: too small an increment is insufficient to effectively disrupt the existing stratified structure, while too large an increment may cause excessive disturbance of the material, resulting in new stratification or increasing the equipment load.

[0144] After the adjustment command is executed, the system continuously monitors the change in the overall deviation distance: if the distance recovers to below the control limit and remains stable within the next 1-3 sampling periods, it indicates that the adjustment is effective and production can continue with the adjusted process parameters; if it continues to exceed the limit or continues to deteriorate, the system will automatically trigger the shutdown protection and issue a serious alarm, prompting the operator to check the material status in the hopper, troubleshoot the mixing equipment, or intervene manually.

[0145] It should be noted that the specific incremental values ​​of the feed paddle speed and vibrator frequency can be determined through conventional experiments based on the actual production equipment model, material powder characteristics, and severity of stratification. This step does not limit the precise values.

[0146] S23. When the key variable is the edge peeling defect rate in the statistical characteristics of appearance defects, the abnormal root cause type is determined to be the sticking and punching early warning mode, and an adjustment command to issue a cleaning early warning prompt is generated and executed, or an adjustment command to fine-tune the main pressure is generated and executed.

[0147] In this embodiment of the invention, this step is executed after the core key variable, edge peeling defect rate, is identified in step S21. The edge peeling defect rate is statistically obtained from the appearance image in step 103 and reflects the proportion of peeling and chipping at the edge of the tablet. Fumaric acid APIs contain carboxyl groups in their molecular structure, making them highly polar. During tableting, they easily adhere to the punch surface to form a drug film, leading to edge peeling defects at the corresponding positions in the subsequently pressed tablets. Therefore, an increase in the edge peeling defect rate is an early and specific warning signal of punch adhesion, earlier than traditional monitoring indicators such as punch pressure fluctuations.

[0148] Upon determining that the die sticking warning mode has been activated, the controller generates and executes an adjustment command. One of the following two options, or a combination thereof, can be selected: First, a cleaning warning is issued to remind the operator to prepare for die cleaning; second, an adjustment command to fine-tune the main pressure is generated and executed. The cleaning warning can be displayed via text on the human-machine interface or by issuing an audible and visual alarm, informing the operator that the die cleaning is planned for an appropriate time. For example, if only a cleaning warning is issued without immediately fine-tuning the main pressure, the operator can plan the cleaning in advance based on the warning information, avoiding the deterioration of the die sticking and resulting in a large number of defective products. If fine-tuning the main pressure is selected, it will be executed in the manner specified later in this step.

[0149] It is worth mentioning that when the abnormal root cause type is determined to be the sticking and impact warning mode, and an adjustment command for fine-tuning the main pressure is generated, the fine-tuning of the main pressure is to increase the preset percentage based on the current main pressure.

[0150] It should be noted that the reasonable range for this preset percentage is 3% to 8%, and the specific value can be determined through conventional process experiments based on the specific formulation of fumaric acid tablets, the model of the tablet press, and the severity of sticking. For example, when the edge peeling defect rate is at the warning level of 0.5% to 1%, the preset percentage can be 3% to 5%; when the defect rate exceeds 1%, it can be appropriately increased to 5% to 8%, and the cleaning warning should be implemented first.

[0151] In this embodiment of the invention, when the adjustment command for fine-tuning the main pressure is selected, this step further defines the specific method of fine-tuning the main pressure: adding a preset percentage based on the current main pressure. This preset percentage can be preset according to the severity of sticking or product characteristics. For example, when the edge peeling defect rate is at a low warning level (e.g., about 0.5% to 1%), the preset percentage can be around 3% to 5%; when the defect rate is high (e.g., exceeding 1%), the preset percentage can be appropriately increased and the cleaning warning can be prioritized. Those skilled in the art can determine the appropriate preset percentage range for a specific product and equipment through simple pre-experiments: an increment that is too small (e.g., less than 3%) may not effectively alleviate sticking, while an increment that is too large (e.g., exceeding 8%) may cause tablet cracking or over-compression. For example, when the current main pressure of the tablet press is 15 kN, the preset percentage can be 4%, i.e., an increase of 0.6 kN to 15.6 kN.

[0152] After executing the adjustment command, the system continuously monitors changes in the edge peeling defect rate: if the defect rate decreases significantly and returns to normal levels, it indicates that the adjustment is effective and production can continue under the main pressure; if the defect rate does not improve or continues to rise, the system can automatically stop the machine and issue an alarm for forced cleaning of the die. It should be noted that the specific values ​​of the preset percentages are illustrative and not intended to limit the scope of protection of this invention.

[0153] Please see Figure 4 , Figure 4 This is a structural block diagram of an online monitoring system for the tableting quality of fumaric acid tablets provided in an embodiment of the present invention.

[0154] This invention provides an online monitoring system for the tableting quality of fumaric acid tablets, comprising:

[0155] The sampling and diversion module 401 is used to periodically divert a portion of the tablets from the continuously falling tablets in the tablet ejection chute of the tablet press at a preset time frequency and send them into the detection channel as samples.

[0156] The synchronous detection module 402 is used to simultaneously perform physical index detection, near-infrared spectral acquisition, and appearance image acquisition on the tablets in the sample within the detection channel, so as to obtain physical index data, spectral data, and appearance image data for each tablet.

[0157] The feature extraction module 403 is used to perform statistical processing on the physical index data, spectral data and appearance image data of all tablets obtained in the same sampling period to obtain the physical index statistical features, spectral statistical features and appearance defect statistical features corresponding to the sample.

[0158] The statistical control module 404 is used to construct a multivariate input feature vector corresponding to the sample by using physical index statistical features, spectral statistical features and appearance defect statistical features, and calculate the decision statistic based on the multivariate statistical process control model.

[0159] The anomaly response module 405 is used to perform anomaly root cause analysis based on the judgment statistic when the judgment statistic exceeds the control limit, and to perform corresponding processing operations based on the analysis results.

[0160] Since the above system corresponds one-to-one with an online monitoring method for the tableting quality of fumaric acid tablets, and its implementation principle is consistent with that of the online monitoring method for the tableting quality of fumaric acid tablets, for the sake of convenience and brevity, those skilled in the art can clearly understand that the specific working process of the system and modules described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0161] Please see Figure 5 , Figure 5 This is a structural block diagram of an electronic device provided in an embodiment of the present invention.

[0162] An electronic device according to an embodiment of the present invention includes: a memory 501 and a processor 502. The memory 501 stores a computer program. When the computer program is executed by the processor 502, the processor 502 performs an online monitoring method for the tableting quality of fumaric acid tablets as described in any of the above embodiments.

[0163] Memory 501 may be an electronic memory such as flash memory, EEPROM (Electrically Erasable Programmable Read-Only Memory), EPROM, hard disk, or ROM. Memory 501 has storage space 503 for program code 513 for performing any of the method steps described above. For example, storage space 503 for program code may include various program codes 513 for implementing the various steps in the methods described above. This program code can be read from or written to one or more computer program products. These computer program products include program code carriers such as hard disks, CDs, memory cards, or floppy disks. The program code may be compressed, for example, in a suitable form. When run by a computing processing device, this code causes the computing processing device to perform the various steps in the methods described above. This program code can be read from or written to one or more computer program products. These computer program products include program code carriers such as hard disks, CDs, memory cards, or floppy disks. The program code may be compressed, for example, in a suitable form. When these codes are run by a computing device, the computing device causes the device to perform the various steps in the online monitoring method for the quality of fumaric acid tablet compression described above.

[0164] This invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements an online monitoring method for the tableting quality of fumaric acid tablets as described in any of the above embodiments.

[0165] This invention also provides a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium. The computer program includes program instructions, wherein when the program instructions are executed by a computer, the computer performs an online monitoring method for the tableting quality of fumaric acid tablets as described in any of the above embodiments.

[0166] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0167] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces, or indirect coupling or communication connection between apparatuses or units, and may be electrical, mechanical, or other forms.

[0168] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0169] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0170] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0171] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An online monitoring method for the tableting quality of fumaric acid tablets, characterized in that, include: In the tablet ejection chute of the tablet press, a portion of the tablets are periodically diverted from the continuously falling tablets at a preset time frequency and sent into the testing channel as samples. Within the detection channel, physical index detection, near-infrared spectral acquisition, and appearance image acquisition are performed simultaneously on the tablets in the sample to obtain physical index data, spectral data, and appearance image data for each tablet. The physical index data, spectral data and appearance image data of all tablets obtained in the same sampling period are statistically processed to obtain the physical index statistical characteristics, spectral statistical characteristics and appearance defect statistical characteristics of the sample. The physical index statistical features, the spectral statistical features, and the appearance defect statistical features are used to construct the multivariate input feature vector corresponding to the sample. Based on the multivariate statistical process control model, the multivariate input feature vector is calculated to obtain the judgment statistic. When the determination statistic exceeds the control limit, an anomaly root cause analysis is performed based on the determination statistic, and corresponding processing operations are executed according to the analysis results.

2. The online monitoring method for the tableting quality of fumaric acid tablets according to claim 1, characterized in that, The step of constructing the multivariate input feature vector corresponding to the sample using the physical index statistical features, the spectral statistical features, and the appearance defect statistical features includes: The physical index data and spectral data of each tablet in the same sample are matched one by one according to the tablet, and the single-particle correlation features between the physical index features and spectral features corresponding to each tablet are constructed respectively. Based on the single-pill correlation characteristics corresponding to each of the aforementioned tablets, a comprehensive deviation distance is calculated to characterize the deviation between the physical index statistical characteristics and the spectral statistical characteristics. The comprehensive deviation distance is the degree of deviation between the physical index statistical characteristics and the spectral statistical characteristics caused by material stratification when the relative standard deviation of tablet weight difference does not increase. The comprehensive deviation distance is used as a cross-dimensional correlation feature, and together with the physical index statistical features, the spectral statistical features, and the appearance defect statistical features, it is used to construct a multivariate input feature vector.

3. The online monitoring method for the tableting quality of fumaric acid tablets according to claim 1, characterized in that, The multivariate input feature vector also includes at least one of the following: the ratio of sheet weight to hardness, the ratio of sheet weight to thickness, or the ratio of hardness to thickness.

4. The online monitoring method for the tableting quality of fumaric acid tablets according to claim 1, characterized in that, The multivariate statistical process control model is a principal component analysis model, and the decision statistics include the Hotelling statistic and the Q residual statistic.

5. The online monitoring method for the tableting quality of fumaric acid tablets according to claim 2, characterized in that, The steps of performing anomaly root cause analysis based on the judgment statistic when it exceeds the control limit, and executing corresponding processing operations based on the analysis results, include: When the judgment statistic exceeds the control limit, the key variables that cause the judgment statistic to exceed the limit are identified by contribution plot analysis. When the key variable is the comprehensive deviation distance, the abnormal root cause type is determined to be the raw material stratification mode, and an adjustment command is generated and executed to increase the speed of the feeding paddle in the tablet press hopper or to increase the frequency of the vibrator in the feeding channel. When the key variable is the edge peeling defect rate in the appearance defect statistical features, the abnormal root cause type is determined to be the sticking and punching early warning mode, and an adjustment command to issue a cleaning early warning prompt is generated and executed, or an adjustment command to fine-tune the main pressure is generated and executed.

6. The online monitoring method for the tableting quality of fumaric acid tablets according to claim 5, characterized in that, When the determined abnormal root cause type is sticking and impact warning mode, and the adjustment command for fine-tuning the main pressure is generated, the fine-tuning main pressure is to increase the current main pressure by a preset percentage.

7. An online monitoring system for the tableting quality of fumaric acid tablets, characterized in that, include: The sampling and diversion module is used to periodically divert a portion of the tablets from the continuously falling tablets in the tablet ejection chute of the tablet press at a preset time frequency and send them into the detection channel as samples. The synchronous detection module is used to simultaneously perform physical index detection, near-infrared spectral acquisition, and appearance image acquisition on the tablets in the sample within the detection channel, so as to obtain physical index data, spectral data, and appearance image data for each tablet. The feature extraction module is used to perform statistical processing on the physical index data, spectral data and appearance image data of all tablets obtained in the same sampling period to obtain the physical index statistical features, spectral statistical features and appearance defect statistical features corresponding to the sample. The statistical control module is used to construct a multivariate input feature vector corresponding to the sample by using the physical index statistical features, the spectral statistical features and the appearance defect statistical features, and to calculate the multivariate input feature vector based on the multivariate statistical process control model to obtain the judgment statistic. An anomaly response module is used to perform anomaly root cause analysis based on the judgment statistic when the judgment statistic exceeds the control limit, and to perform corresponding processing operations based on the analysis results.

8. An electronic device, characterized in that, The device includes a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor performs the steps of the online monitoring method for the tableting quality of fumaric acid tablets as described in any one of claims 1-6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed, it implements the online monitoring method for the tableting quality of fumaric acid tablets as described in any one of claims 1-6.

10. A computer program product, characterized in that, The computer program product includes a computer program stored on a non-transitory computer-readable storage medium, the computer program including program instructions, wherein when the program instructions are executed by a computer, the computer performs the online monitoring method for the tableting quality of fumaric acid tablets as described in any one of claims 1-6.

Images