Process for the preparation of amoxicillin sodium with high consistency

Abstract

The application relates to the field of drug preparation process control, in particular to a high-consistency amoxicillin sodium preparation process. The method constructs a working stability matrix through the rotation speed sequence and the temperature sequence of each crystallization tank in each crystallization process of each crystallization system in historical data, identifies abnormal crystallization systems and normal crystallization systems through the correlation between the working stability matrices, further trains a classification neural network, and judges whether an abnormality exists in a target crystallization system by using the classification neural network through real-time data. The abnormality degree of the abnormal target crystallization system is further judged through the local reachable density in the distance neighborhood of the abnormal characteristic descriptor of the abnormal target crystallization system, and the termination of the abnormal target crystallization system is controlled. The application collects preparation data in the preparation process, judges the abnormal crystallization system and the abnormal degree thereof through the correlation between the data, and controls the same, so that cost waste is avoided, the preparation efficiency is improved, and the high consistency of the overall output is ensured.

Description

Technical Field

[0001] This invention relates to the field of pharmaceutical manufacturing process control, specifically to a highly consistent amoxicillin sodium manufacturing process. Background Technology

[0002] Pharmaceutical manufacturers evaluate the quality of approved generic drugs during production, adhering to the principle of consistency with the original drug in quality and efficacy. For substandard drugs, they improve the manufacturing process to enhance overall quality. Amoxicillin sodium is a common generic drug. In its production, the crystallization process is the main process affecting drug quality. The crystallization process influences the crystallization efficiency and purity of amoxicillin sodium. Poor crystallization quality during the crystallization process significantly impacts the efficacy of the finished product.

[0003] To ensure the quality of amoxicillin sodium crystals, periodic testing is often incorporated into the crystallization process to check the crystal quality. However, periodic crystal quality checks increase the number of process steps, wasting time and costs, affecting drug preparation efficiency, and making it difficult to promptly identify abnormal crystallization systems. Summary of the Invention

[0004] To address the aforementioned technical problems, the present invention aims to provide a highly consistent amoxicillin sodium preparation process, the specific technical solution of which is as follows:

[0005] This invention proposes a highly consistent amoxicillin sodium preparation process, the process comprising:

[0006] Obtain the rotational speed and temperature of the crystallizer for each crystallization process in each crystallization system from historical data;

[0007] The operating stability matrix is ​​constructed by combining the rotational velocity sequence and temperature sequence over time; abnormal crystallization systems and normal crystallization systems are identified based on the correlation of the operating stability feature matrices among different crystallization systems.

[0008] A classification neural network is trained based on the operational stability matrices of the abnormal crystallization system and the normal crystallization system; the real-time operational stability matrix of the target crystallization system is input into the classification neural network to determine whether the real-time crystallization system is abnormal;

[0009] If the target crystallization system is an abnormal target crystallization system, then the dissolution temperature information of the sodium isooctanoate dissolution tank in all crystallization systems at this time is obtained, and the dissolution temperature information, the average rotation speed of the crystallizer during the crystallization process, and the average temperature of the crystallizer during the crystallization process are used as feature descriptors; based on the feature descriptors, the distance neighborhood of the abnormal feature descriptors of the abnormal target crystallization system is obtained using the K-nearest neighbor algorithm; the local reachability density of the abnormal feature descriptors is obtained according to the sample distance from the abnormal feature descriptors to all samples in the distance neighborhood; the crystallizer of the abnormal target crystallization system is controlled to continue to execute the crystallization process and the local reachability density is continuously updated; if the local reachability density is continuously less than a preset threshold within a preset time period, the crystallization process is terminated and an early warning signal is provided.

[0010] Furthermore, the identification of abnormal and normal crystallization systems based on the correlation of the operational stability feature matrices between different crystallization systems includes:

[0011] Using the operational stability feature matrix as training data in the isolated forest algorithm, the isolated forest anomaly detection algorithm is used to identify the anomalous crystallization system and the normal crystallization system in the historical data.

[0012] Further, training the classification neural network based on the operational stability matrices of the abnormal crystallization system and the normal crystallization system includes:

[0013] The classification neural network uses the OneClass SVM algorithm structure, and uses the high-dimensional vector obtained by expanding the working stability matrix of the normal crystallization system in historical data row by row as the training input data to train the classification neural network.

[0014] Furthermore, obtaining the dissolution temperature information of the sodium isooctanoate dissolving tanks in all crystallization systems during this process includes:

[0015] The maximum and minimum temperatures during the operation of the sodium isooctanoate dissolving tank are used as the dissolving temperature information.

[0016] Further, obtaining the local reachability density of the anomaly feature descriptor based on the sample distances from the anomaly feature descriptor to all samples in the distance neighborhood includes:

[0017] The cosine distance between the anomalous feature descriptor and all samples in the distance neighborhood is used as the sample distance; the reciprocal of the average sample distance between the anomalous feature descriptor and all samples in the distance neighborhood is used as the local reachability density.

[0018] The present invention has the following beneficial effects:

[0019] This invention collects preparation data on the rate and temperature generated by the crystallization tank from historical data of the amoxicillin sodium crystallization system. It analyzes the anomalies of the current process using the preparation data generated during the preparation process, and then automatically labels the preparation data of abnormal processes in the subsequent network, accelerating the network training process. A classification neural network analyzes the preparation feature data in the real-time crystallization system to determine whether the current target crystallization system meets the requirements. Furthermore, it combines data from the dissolution process to construct feature descriptors. By analyzing the sample distance between the abnormal feature descriptors of the abnormal target crystallization system and other feature descriptors, it obtains the local reachability density. This local reachability density characterizes the degree of abnormality of the current abnormal target crystallization system, and controls the start and stop of the current process based on the local reachability density, avoiding cost waste caused by unlimited continuous execution of the crystallization process. This invention can promptly identify abnormal processes based on the preparation data generated by the preparation process, making it convenient for staff to maintain or terminate abnormal processes, preventing defective products from abnormal processes from affecting the overall yield rate, improving the preparation efficiency of generic amoxicillin sodium, and ensuring high consistency of the overall output. Attached Figure Description

[0020] To more clearly illustrate the technical solutions and advantages 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.

[0021] Figure 1 This is a flow chart of a highly consistent amoxicillin sodium preparation process provided in one embodiment of the present invention. Detailed Implementation

[0022] To further illustrate the technical means and effects adopted by the present invention to achieve the intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a highly consistent amoxicillin sodium preparation process proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0023] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0024] The preparation process of amoxicillin sodium involved in the embodiments of the present invention specifically includes:

[0025] (1) Add 550L of anhydrous ethanol to the sodium isooctanoate dissolving tank, cool to 15-20℃, add 100-110Kg of sodium isooctanoate, stir until clear, add 1.5Kg of 767 activated carbon, stir for 15 minutes, and filter into the crystallization tank.

[0026] (2) Add 450L of anhydrous ethanol and 76L of diisopropylamine to the amoxicillin dissolving tank, cool to 0-5℃, add 150Kg of amoxicillin, stir for 10 minutes until completely dissolved, add 250L of anhydrous ethanol and 1.5Kg of 767 activated carbon, decolorize for 30 minutes, filter to crystallization tank, adjust the speed of crystallization tank to 30-35 rpm, add the filtered sodium isooctanoate solution, and crystallize at 0-5℃ for 6-7 hours.

[0027] (3) Filter by suction, wash the filter cake with 200-250L of anhydrous ethanol, filter by suction, wash the filter cake with 150-200L of anhydrous ethanol, dry by suction, and vacuum dry at 70-85℃.

[0028] Since the embodiments of this invention focus on the crystallization quality in the crystallization process, information collection and analysis are mainly conducted on the crystallization process in the crystallization tank during the second step of the preparation process.

[0029] The following describes in detail, with reference to the accompanying drawings, a specific scheme for the preparation process of highly consistent amoxicillin sodium provided by the present invention.

[0030] Please see Figure 1 It illustrates a process flow diagram for preparing highly consistent amoxicillin sodium according to an embodiment of the present invention. The preparation process includes:

[0031] Step S1: Obtain the rotation speed and temperature of the crystallizer for each crystallization process in each crystallization system from historical data.

[0032] The crystallization system is a preparation system consisting of all the devices used in the preparation of amoxicillin sodium.

[0033] In this embodiment of the invention, a speed sensing device is installed in the transmission mechanism of the crystallizer to obtain the rotational speed data of the crystallizer through sensor data; a temperature sensing device is installed inside the crystallizer to obtain the temperature data of the crystallizer through sensor data. In the industrial system, the preparation information generated by each preparation process can be stored in a historical database using Internet of Things (IoT) technology for easy retrieval during data analysis.

[0034] Step S2: Combine the rotational speed sequence and temperature sequence in the time series to form a working stability matrix; identify abnormal crystallization systems and normal crystallization systems based on the correlation of the working stability feature matrices between different crystallization systems.

[0035] The timing data can reflect the working stability of the process, thus obtaining the timing rotation speed sequence and temperature sequence.

[0036] Since the temperature range for temperature-controlled crystallization in the crystallizer is fixed, and the power of the refrigeration elements remains the same from the initial temperature drop to the subsequent temperature rise, a faster temperature drop in the crystallizer generally results in a slower temperature rise due to overshoot. Conversely, a slower temperature drop in the crystallizer, within the 0-5℃ temperature range, inevitably leads to a higher ambient temperature, making it difficult for the refrigeration unit to stabilize at a lower temperature. In this case, the faster temperature rise in the crystallizer is generally greater. This regular variation is observed in the time series of a normal crystallization process.

[0037] For the rotation speed sequence of the crystallizer, the speed should be kept constant during the crystallization process to achieve a smooth crystallization process. Therefore, the elements in the rotation speed sequence in a normal crystallization process should be evenly distributed.

[0038] For abnormal crystallization systems, abnormal preparation parameters in the process can lead to an unsatisfactory crystallization process, thus affecting drug quality. Specifically, abnormal crystallization systems may exhibit abnormal variations in the crystallizer rotation speed or temperature. Therefore, the operating stability matrix K is constructed from the rotation speed and temperature sequences, expressed as: Where H i x Let be the temperature of the crystallizer during the x-th crystallization process at the i-th sampling time. Let be the rotational speed of the crystallizer during the x-th crystallization process at the i-th sampling time.

[0039] The causes of crystallization system anomalies are multifaceted, including factors such as temperature controller environmental conditions, deterioration of the crystallizer's partition performance, and leaks at joints. Therefore, the operational stability matrix can represent the performance characteristics of the crystallization process in an amoxicillin sodium crystallizer. The operational stability matrix corresponding to an abnormal crystallization system is more specific than that of a normal crystallization system. Because abnormal crystallization systems are low-probability events—meaning the number of abnormal crystallization systems is less than the number of normal ones—abnormal and normal crystallization systems can be identified based on the correlation between the operational stability feature matrices of different crystallization systems. A stronger correlation indicates more similar operational stability matrices, suggesting a more normal crystallization system; a weaker correlation indicates a more specific operational stability matrix, suggesting a more abnormal crystallization system. Preferably, the operational stability feature matrix is ​​used as training data in the Isolation Forest algorithm, and the Isolation Forest anomaly detection algorithm is used to identify abnormal and normal crystallization systems in historical data.

[0040] It should be noted that the Isolation Forest algorithm is a well-known technique in the art, and will not be elaborated upon further. Only the implementation process in the embodiments of this invention will be briefly described here:

[0041] (1) Use the working stability matrix as the training data for constructing the isolated forest, randomly select a subsample in the training dataset as the root node in the isolated tree, and then randomly specify a dimension to generate a cut point p in the current node data. The cut point p is randomly generated between the maximum and minimum values ​​of the specified dimension of the current node.

[0042] (2) The current node's data space is divided into two subspaces by the cutting point p: data with a specified dimension less than the cutting point p is placed in the left subtree of the current node, and data with a dimension greater than or equal to the cutting point p is placed in the right subtree of the current node.

[0043] (3) Repeat steps (1) and (2) in the child nodes of the tree until there is only one data in the child node or the child node reaches the preset height to complete the construction of the isolated tree.

[0044] (4) Calculate the anomaly score S for each sample point x. x The anomaly score determines whether the preparation system corresponding to the sample point has an anomaly. Anomaly score S x The formulas for obtaining the formulas include: Where h(x) is the height of sample point x in each tree, a(n) is the mean path length for a given number of samples n, and E(h(x)) is the expected value of h(x); h(x) = ln x + §, where § ≈ 0.5772456649.

[0045] (5) Set a threshold value. The threshold value is between 0.4 and 0.6. If the abnormal score is greater than the threshold value, the corresponding crystallization system is identified as an abnormal crystallization system.

[0046] Step S3: Train a classification neural network based on the operational stability matrices of the abnormal crystallization system and the normal crystallization system; input the real-time operational stability matrix of the target crystallization system into the classification neural network to determine whether the real-time crystallization system is abnormal.

[0047] To enable rapid and accurate identification of abnormal results in subsequent implementation processes, a classification neural network is trained based on the operational stability matrices of both abnormal and normal crystallization systems. This allows the real-time operational stability matrix of the target crystallization system to be directly input into the classification neural network for rapid and accurate determination of whether the real-time crystallization system is abnormal.

[0048] Preferably, the classification neural network uses the OneClass SVM algorithm structure, and uses the high-dimensional vector obtained by expanding the working stability matrix of the normal crystallization system in historical data row by row as the training input data to train the classification neural network.

[0049] It should be noted that OneClass SVM is a classic single-classification algorithm network. By training a classification neural network with the OneClass SVM structure, the network can learn the working characteristics of a normal crystallization system. By learning the hyperplane represented by the distribution of most common output state vectors, it roughly fits the vector of the out-of-plane region that can distinguish unique states. Thus, in the later stage, it can determine whether a state is unique by only inputting data, without relying on a large amount of data for comparison. This enables fast and accurate determination of whether there is an abnormality in the working of the crystallization system in practical use.

[0050] Step S4: If the target crystallization system is an abnormal target crystallization system, obtain the dissolution temperature information of the sodium isooctanoate dissolution tank in all crystallization systems at this time, and use the dissolution temperature information, the average rotation speed of the crystallizer during the crystallization process, and the average temperature of the crystallizer during the crystallization process as feature descriptors; based on the feature descriptors, use the K-nearest neighbor algorithm to obtain the distance neighborhood of the abnormal feature descriptors of the abnormal target crystallization system; obtain the local reachability density of the abnormal feature descriptors according to the sample distance from the abnormal feature descriptors to all samples in the distance neighborhood; control the crystallizer of the abnormal target crystallization system to continue to execute the crystallization process and continuously update the local reachability density; if the local reachability density is continuously less than the preset threshold within a preset time period, terminate the crystallization process and provide an early warning signal.

[0051] For abnormal crystallization systems, due to their poor crystallization process, it is impossible to form amoxicillin sodium that meets the standards within the normal process time. Therefore, it is necessary to terminate and maintain the abnormal crystallization system promptly. However, for some minor anomalies, the resulting abnormal information will not lead to serious problems with the quality of the prepared drug. Furthermore, data analysis is inevitably affected by environmental factors or other noise factors. Blindly terminating crystallization systems with abnormal information will still affect the overall preparation efficiency.

[0052] Therefore, if the target crystallization system is determined to be an abnormal target crystallization system, the dissolution temperature information of the sodium isooctanoate dissolution tanks in all crystallization systems at that time is obtained. The dissolution temperature information, the average rotation speed of the crystallizer during the crystallization process, and the average temperature of the crystallizer during the crystallization process are used as feature descriptors. Based on the feature descriptors of all crystallization systems in real time, the degree of abnormality of the current abnormal target crystallization system is determined. Based on the feature descriptors, the distance neighborhood of the abnormal feature descriptors of the abnormal target crystallization system is obtained using the K-nearest neighbor algorithm. It should be noted that the K-nearest neighbor algorithm is a numerical technique used by those skilled in the art and will not be elaborated here.

[0053] Preferably, the maximum and minimum temperatures during the operation of the sodium isooctanoate dissolving tank are used as the dissolution temperature information. In this embodiment of the invention, a temperature sensor is also used to obtain the dissolution temperature information inside the sodium isooctanoate dissolving tank.

[0054] If the degree of anomalousness of the crystal system corresponding to the anomaly feature descriptor is high, its distribution in the corresponding distance neighborhood will be relatively sparse compared to other samples; if the degree of anomalousness of the crystal system corresponding to the anomaly feature descriptor is low, or if the anomalous information is generated by data fluctuations, its distribution in the corresponding distance neighborhood will be relatively concentrated compared to other samples. Therefore, the local reachability density of the anomaly feature descriptor is obtained based on the sample distance from the anomaly feature descriptor to all samples in the distance neighborhood, specifically including:

[0055] The cosine distance between the anomaly feature descriptor and all samples in its neighborhood is used as the sample distance. The reciprocal of the average sample distance between the anomaly feature descriptor and all samples in its neighborhood is used as the local reachability density. That is, the larger the local reachability density, the more concentrated the samples are in the neighborhood, and the smaller the degree of anomalousness of the corresponding anomalous target crystal system; the smaller the local reachability density, the sparser the samples are in the neighborhood, and the greater the degree of anomalousness of the corresponding anomalous target crystal system.

[0056] The crystallization tank of the abnormal target crystallization system continues to execute the crystallization process and continuously updates the local reachable density. If the local reachable density remains below a preset threshold for a preset time period, it indicates that the product of the abnormal target crystallization system is unlikely to meet high consistency standards, and the drug quality cannot meet the requirements. In this case, the crystallization process needs to be terminated in advance, and an early warning signal should be provided to notify personnel to maintain and repair the abnormal target crystallization system to ensure the efficiency of the overall production batch. In this embodiment of the invention, the local reachable density is updated every 5 minutes, and the preset time period is set to 30 minutes. The threshold for the local reachable density can be set according to the specific implementation scenario and is not limited here.

[0057] This invention constructs a working stability matrix by using the rotational speed and temperature sequences of the crystallizer for each crystallization process in historical data for each crystallization system. The correlation between these working stability matrices identifies abnormal and normal crystallization systems, which are then used to train a classification neural network. This network is then used to determine whether a target crystallization system exhibits anomalies based on real-time data. The degree of anomaly is further determined by the local reachability density within the distance neighborhood of the anomaly feature descriptor of the abnormal target crystallization system, and the abnormal target crystallization system is terminated accordingly. This invention, by collecting preparation data during the preparation process and using the correlation between data points to identify and control abnormal crystallization systems and their degrees of anomaly, avoids cost waste and improves preparation efficiency and production yield.

[0058] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0059] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0060] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A highly consistent amoxicillin sodium preparation process, characterized in that, The process includes: Obtain the rotational speed and temperature of the crystallizer for each crystallization process in each crystallization system from historical data; The operating stability matrix is ​​constructed by combining the rotational velocity sequence and temperature sequence over time. Abnormal and normal crystallization systems are identified based on the correlation between the operating stability feature matrices of different crystallization systems. The expression for the operating stability matrix K, composed of the rotational velocity sequence and temperature sequence, is: ,in Let be the temperature of the crystallizer during the x-th crystallization process at the i-th sampling time. Let be the rotational speed of the crystallizer during the x-th crystallization process at the i-th sampling time. A classification neural network is trained based on the operational stability matrices of the abnormal crystallization system and the normal crystallization system; the real-time operational stability matrix of the target crystallization system is input into the classification neural network to determine whether the real-time crystallization system is abnormal; If the target crystallization system is an abnormal target crystallization system, then the dissolution temperature information of the sodium isooctanoate dissolution tank in all crystallization systems at this time is obtained, and the dissolution temperature information, the average rotation speed of the crystallizer during the crystallization process, and the average temperature of the crystallizer during the crystallization process are used as feature descriptors; based on the feature descriptors, the distance neighborhood of the abnormal feature descriptors of the abnormal target crystallization system is obtained using the K-nearest neighbor algorithm; the local reachability density of the abnormal feature descriptors is obtained according to the sample distance from the abnormal feature descriptors to all samples in the distance neighborhood; the crystallizer of the abnormal target crystallization system is controlled to continue to execute the crystallization process and the local reachability density is continuously updated; if the local reachability density is continuously less than a preset threshold within a preset time period, the crystallization process is terminated and an early warning signal is provided.

2. The highly consistent amoxicillin sodium preparation process according to claim 1, characterized in that, The identification of abnormal and normal crystallization systems based on the correlation of the operational stability feature matrices between different crystallization systems includes: Using the operational stability feature matrix as training data in the isolated forest algorithm, the isolated forest anomaly detection algorithm is used to identify the anomalous crystallization system and the normal crystallization system in the historical data.

3. The highly consistent amoxicillin sodium preparation process according to claim 1, characterized in that, The step of training a classification neural network based on the operational stability matrix of the abnormal crystallization system and the normal crystallization system includes: The classification neural network uses a One Class SVM algorithm structure, and uses the high-dimensional vector obtained by expanding the working stability matrix of the normal crystallization system in historical data row by row as the training input data to train the classification neural network.

4. The highly consistent amoxicillin sodium preparation process according to claim 1, characterized in that, The process of obtaining the dissolution temperature information of the sodium isooctanoate dissolving tank in all crystallization systems at this time includes: The maximum and minimum temperatures during the operation of the sodium isooctanoate dissolving tank are used as the dissolving temperature information.

5. The highly consistent amoxicillin sodium preparation process according to claim 1, characterized in that, The step of obtaining the local reachability density of the anomaly feature descriptor based on the sample distances from the anomaly feature descriptor to all samples in the distance neighborhood includes: The cosine distance between the anomalous feature descriptor and all samples in the distance neighborhood is used as the sample distance; the reciprocal of the average sample distance between the anomalous feature descriptor and all samples in the distance neighborhood is used as the local reachability density.

Images