Method for determining a machine setting of a tablet press, computer program product for predicting at least one tablet parameter, and system comprising a tablet press and a computing unit
The method uses machine learning to predict tablet parameters during the pressing process, addressing the inefficiencies of current quality control methods by automating the determination of machine settings, thereby enhancing production efficiency and reducing waste.
Patent Information
- Application Number
- PCT/EP2026/063998
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Priority Date
- 2025-05-15
- Filing Date
- 2026-05-13
- Publication Date
- 2026-11-19
AI Technical Summary
Current methods for determining tablet quality in manufacturing are labor-intensive, time-consuming, and require skilled personnel, limiting the automation and efficiency of the production process, and often result in significant material waste and production delays.
A method using machine learning to predict tablet parameters during the pressing process, reducing the need for complex physical and chemical measurements by utilizing a machine learning model to determine machine settings based on production parameters, allowing for inline prediction and automation of quality control.
This approach significantly reduces the time and effort required for developing machine settings, minimizes material waste, and enhances production efficiency by enabling faster adjustment to desired tablet parameters, improving product development time and reducing the risk of production halts.
Smart Images

Figure IMGF000035_0001
Abstract
Description
[0001] Prediction of tablet and production parameters of a tablet press
[0002] DESCRIPTION
[0003] The invention relates to a method for determining a machine setting for a tablet press. A first data set comprising at least one tablet parameter is provided, which is intended to apply to a tablet that can be produced using the tablet press. A first pressing operation is carried out with a first machine setting. A first set of production parameters is recorded during the first pressing operation. At least one tablet parameter is predicted based on a machine learning model using the recorded first set of production parameters. A second set of production parameters is determined based on the at least one predicted tablet parameter. A second machine setting is determined based on the second set of production parameters. The second machine setting is then set on the tablet press.At least one tablet parameter is measured on a tablet produced by the tablet press with the second machine setting to verify that the at least one tablet parameter matches the first data set.
[0004] Furthermore, the invention relates to a computer program product and a system.
[0005] Background and state of the art
[0006] Ensuring high tablet quality during tablet compression presents a significant technical challenge, as it plays a crucial role in efficacy, stability, and handling. A high-quality tablet must be able to reliably release its active ingredients and withstand the mechanical stresses encountered during manufacturing, packaging, and transport. Therefore, it is essential that the physical and mechanical properties of tablets are monitored and optimized throughout the entire production process. These properties, which include porosity and tensile strength, are of central importance because they significantly influence the disintegration behavior and strength of the tablets.
[0007] For example, the porosity of a tablet directly influences its strength and disintegration behavior. Excessive porosity results in tablets with low strength, making them difficult to handle and prone to breakage or damage during packaging and use. Conversely, insufficient porosity can prolong the disintegration time of the tablets, delaying the release of the active ingredient and thus impairing the tablet's effectiveness. Therefore, achieving an optimal balance of porosity is essential for manufacturing stable and effective tablets.
[0008] The tensile strength of tablets is another important parameter that must ensure they can withstand the mechanical stresses throughout the entire manufacturing process. It must be selected such that the tablets do not break or crumble during pressing, packaging, and transport. This physical property is typically measured in the prior art using specific fracture strength tests, which determine the force required to break a tablet. The tensile strength is then calculated by determining the ratio of the measured force to the tablet dimensions.
[0009] Determining these tablet parameters is generally achieved through complex physical and / or chemical analyses requiring specialized equipment and / or expertise. Porosity is typically measured by comparing the apparent density with the true density of the pressing material, a process that requires prior, complex determination of the true density. Alternatively, porosity is measured using pycnometers, which in the past employed toxic substances such as mercury as the working medium. These methods are not only time-consuming but also raise a number of safety and environmental concerns.
[0010] The process of developing dedicated machine settings, also known as tablet formulations, requires continuous and often iterative monitoring of tablet parameters to achieve the desired quality standards. Experts with in-depth knowledge analyze the measurement results to optimize the formulation and the production process, which is typically a time-consuming and costly undertaking. Each compression operation must be monitored with specialized measuring instruments to ensure the tablets possess the required physical properties. This process demands not only precise technical execution but also significant financial investment in personnel, materials, and equipment.
[0011] A well-known standard process for determining tablet quality relies on the use of external testing equipment, such as the Kraemer UTS 4.1 tablet tester. This equipment allows for the measurement of tablet tensile strength through specific tests, such as the breaking strength test, thus providing insights into tablet quality. When determining porosity, tablet geometry, weight, and density are often measured to obtain the most accurate assessment of physical properties. However, these measurement methods still require skilled personnel to operate the testing equipment and manually perform the necessary measurements. Furthermore, conducting such tests is time-consuming, significantly reducing the efficiency of the overall production process.
[0012] The need for specialized personnel and the high technical effort associated with measuring and analyzing tablet parameters represent a significant limitation for the automation of the tablet production process.
[0013] Prior art also includes approaches for automatically predicting tablet parameters. These approaches could reduce the need for manual measures of tablet parameters.
[0014] In Hayashi et al. (2023), a data-driven model for predicting tablet parameters such as tensile strength, disintegration time, and density is revealed. The model is based on a comprehensive database utilizing the properties of 81 different active pharmaceutical ingredients (APIs). For each API, tablets with identical compositions and three different compression pressures were manufactured. By employing a boosted neural network, the model was able to predict the relevant tablet parameters with high accuracy. The study demonstrates that this data-driven approach is capable of identifying complex relationships between material properties and tablet characteristics and efficiently predicting tablet quality. This enables more precise and automated control of tablet manufacturing.However, the data is taken from tablets that have already been manufactured, so a large number of tablets must first be produced.
[0015] Persson and Göran (2018) describe a hybrid approach for predicting the relationship between tablet tensile strength and compression pressure, known as the Strength-Pressure Relationship (SPR). The aim of this approach is to develop a simplified method for determining this relationship. The proposed approach is based on two key parameters: the Kawakita parameter, which describes the packing density development during compression, and the Heckel yield stress, which indicates the point at which plastic deformation begins. Combining these two parameters enables a more precise and simplified prediction of tensile strength as a function of compression pressure. This method helps to simplify the complex processes of tablet manufacturing and provides a reliable way to predict tablet quality under different compression conditions.However, this approach relies on already established relationships that provide an approximation of the relationship between the different properties. A sufficient degree of precision in these relationships, particularly with regard to a desired machine setting, is not achieved.
[0016] DE 102023109039 A1 discloses a method for configuring the manufacturing process of a rotary press. To configure the rotary press's manufacturing process, the remaining two groups of parameter data are determined from at least two of four groups of parameter data based on a correlation with predefined parameter data. The four groups of parameter data comprise component parameter data for components assembled in the rotary press, process parameter data for process parameters during operation of the rotary press, powder parameter data for powder properties of the powder material to be compressed in the rotary press, and tablet parameter data for tablet properties of the tablets to be produced in the rotary press. Powder parameter data and / or tablet parameters can be determined within the manufacturing process. The groups of parameter data are stored in a database.The parameter data stored in the database can be determined experimentally through test series. Correlation of the parameter data stored in the database can involve a machine learning algorithm. However, prediction of tablet parameters during a pressing process is not performed.
[0017] CN 114334031 A discloses a method for processing the compression parameters of a tablet press. An input vector is provided, comprising parameters of the active ingredient and excipients, as well as parameters of the tablet press. Additionally, a predefined target output vector is defined, describing the desired tablet parameters. A predictive tablet parameter calculator is built based on historical data. This calculator is implemented as a deep convolutional neural network (CNN). The model learns a mapping from input vectors to output vectors. After training, the model can provide a predicted output vector for a given input vector. An optimizer, e.g., based on simulated annealing, is combined with the predictive model. The optimizer varies the variable part of the input vector, while a fixed part remains constant.The goal is to iteratively modify the input vector until the deviation between the predicted output vector and the target output vector falls below a predefined threshold. The input vector thus determined is defined as the target compression parameter and output. The execution of an initial actual compression process with the recording of production parameters during this process is not disclosed. CN 118780683 A discloses a quality control and optimization system for pharmaceutical manufacturing processes. Production and quality data from multiple batches are recorded, including active ingredient parameters, final product quality parameters, and production process parameters. The quality data defines target quality attributes that should fall within predefined tolerance limits. Based on the recorded data, a regression model is constructed: Y = f(X, Z), where X denotes the process parameters, Z the material attributes, and Y the target quality attributes.The model is based on a machine learning model, specifically a graph convolutional neural network, which processes time-dependent process parameters and predicts quality attributes or parameters. Based on these predictions, the process parameters are optimized. Batches are selected to perform multiple optimizations and derive optimized process parameter curves. The optimized process parameters are then automatically applied to the production process by a tuning module. During and after production, real-world production and quality data are measured and recorded. The model is updated iteratively, ensuring continuous optimization of the production process.
[0018] Currently, most solutions for determining tablet quality are either manual or only partially automated and are primarily used in research environments. In industrial practice, particularly in the mass production of tablets, the need for efficient, automated quality control methods remains high. Automation solutions that do not require skilled personnel to perform the measurements and simultaneously enable precise and reliable quality control are currently only available to a limited extent and involve significant technical effort.
[0019] Therefore, the prior art presents a need for improvement, namely to determine a method for determining a machine setting of a tablet press. Objective of the invention
[0020] The object of the invention is to provide a method for determining a machine setting of a tablet press, which can be carried out with little effort and enables reliable monitoring of the tablet parameters during production.
[0021] Summary of the invention
[0022] The invention relates to a method for determining a machine setting of a tablet press comprising the following steps:
[0023] a) Provision of an initial data set comprising at least one tablet parameter that should apply to a tablet that can be produced using a tablet press,
[0024] b) Performing a first pressing operation with a first machine setting, c) Recording a first set of production parameters during the first pressing operation,
[0025] d) Prediction of at least one tablet parameter based on a machine learning model using the recorded first set of production parameters from step c).
[0026] e) Determination of a second set of production parameters based on the at least one predicted tablet parameter from step d).
[0027] f) Determination of a second machine setting based on the second set of production parameters from step e).
[0028] g) Setting the second machine setting on the tablet press,
[0029] h) Measurement of the at least one tablet parameter on a tablet produced by the tablet press with the second machine setting to verify the agreement of the at least one tablet parameter with the first data set,
[0030] i) optionally repeat steps b) - h) until at least one tablet parameter of the tablet substantially matches the first data set, so that the machine setting for tablets is determined using the first data set.
[0031] A significant advantage of the method according to the invention is that the need to perform complex physical and / or chemical investigations to determine tablet parameters is eliminated in most phases of the production process. In the prior art, optimizing and controlling tablet quality requires extensive testing involving physical and / or chemical measurements of the manufactured tablets. These tests, which are often laboratory-intensive and time-consuming, represent a large portion of the overall effort involved in developing a tablet and / or machine settings. The method according to the invention, which utilizes machine learning and predictive modeling in particular, allows tablet quality to be derived directly from the production parameters and the predicted tablet parameters. This reduces the need for extensive physical and / or chemical measurements to a minimum.This is based, among other things, on the fact that a second set of tablet parameters is predicted using the machine learning model. Physical and / or chemical measurements may still be required to perform intermediate and / or final validations, which significantly speeds up and reduces the overall development process. This leads to a significant reduction in product development time and lower production costs.
[0032] Furthermore, the method according to the invention leads to a significant reduction in workload and better utilization of their expertise, allowing them to dedicate their time to the more critical and value-adding aspects of product development and process optimization. The increased productivity of the specialist personnel thus contributes directly to improving the efficiency of the entire production process.
[0033] Another advantageous effect of the method according to the invention is the increased availability of tablet presses and measuring instruments. In traditional production processes, repeated test runs require the continuous use of machines and measuring instruments, which can lead to longer downtimes and suboptimal utilization of available resources. Since the method according to the invention uses predictive modeling to determine the tablet parameters, the test cycle can be accelerated and the need to use tablet presses and measuring instruments in a continuous testing process can be reduced. The predictive models make it possible to achieve the desired results faster and with less effort, thus using these resources more efficiently and increasing their availability. This leads to better utilization of the machines and measuring instruments.
[0034] A further advantage of the method according to the invention lies in the increased automation of trials. In traditional tablet manufacturing processes, many steps depend on manual intervention, which limits the flexibility and efficiency of the process. Performing tests to determine tablet parameters and the subsequent adjustment of machine settings often takes place in iterative cycles that require intensive intervention from skilled personnel. The method according to the invention makes it possible to largely automate these steps, since an adjustment of the machine settings is based directly on the predicted values forecast by the machine learning model. This automation allows trials to be carried out faster and without the need for constant manual checks and interventions.Furthermore, automating processes can minimize potential human errors and increase the reproducibility of results, leading to greater process stability and reliability.
[0035] Automation and optimization of the process also significantly accelerates the availability of new tablet recipes, i.e., new machine settings. In known prior art manufacturing processes, developing and fine-tuning a new machine setting is often a lengthy process requiring numerous manual adjustments and extensive testing to achieve the desired tablet parameters. The inventive method, which utilizes a machine learning model for prediction, enables faster adjustment of the machine setting and more precise control of tablet quality without requiring manual validation of each individual step. This allows new tablet recipes to be developed and implemented in production more quickly.The faster availability of recipes not only improves the time-to-market for new products, but also makes it possible to react quickly to changes in market requirements or customer needs by allowing adjustments to machine settings to be made more quickly.
[0036] Advantageously, manual test measurements are largely avoided. This also reduces potential material waste and the number of possible interventions during the tablet manufacturing process. This further reduces the risk of causing a machine stoppage or a halt to the tablet production process due to the tablet press. Such a stoppage would be inefficient and could lead to significant economic losses.
[0037] Furthermore, during the production process, tablet parameters and thus tablet quality can be determined. This provides parameters that allow for monitoring the quality of the current production and, if necessary, enabling earlier quality control. According to current technology, significant quantities of rejects are sometimes generated because production fluctuations are detected too late. The inventive method avoids these issues.
[0038] In contrast to the prior art, production parameters are recorded during the pressing process. In other words, production parameters are preferably measured during the pressing process. This is done particularly in-line. The prediction of the tablet parameters is carried out using the recorded or measured production parameters. This means, in particular, that the prediction is based on the recorded or measured production parameters. Preferably, production parameters recorded during a pressing process thus provide a basis for predicting the tablet parameters.
[0039] In particular, the prediction of at least one tablet parameter takes place during a pressing operation of the tablet press. This significantly reduces the need for complex downstream experimental tests to measure the tablet parameters. The method according to the invention is particularly capable of determining a machine setting for a tablet press.
[0040] Machine setting preferably refers to a setting that can be adjusted on the tablet press to enable a manufacturing process or production procedure for tablets. Preferably, a machine setting relates to those parameters that can be adjusted by a user on the tablet press. Examples of machine setting parameters, but not limited to, include the speed of components of the tablet press, such as the rotor speed for the die disc, the fill height, the dosing height, the plunger depth, the filling wheel speed, the compression force, etc.
[0041] First, a preliminary data set is provided, comprising at least one tablet parameter intended to apply to a tablet that can be produced using a tablet press. A data set preferably refers to a collection of data containing at least one parameter, and preferably at least two, three, four, five, six, seven, eight, nine, ten, five, five, five, ten ...
[0042] A tablet press is a device or system used to form tablets by compressing a pressing material, which can be supplied in powder or granule form. The tablet press ensures the required strength and shape of the tablets. Basic components of a tablet press, such as an upper and lower punch, a die, a die plate, etc., are familiar to the average professional.
[0043] A first pressing process is carried out using the tablet press with an initial machine setting.
[0044] The pressing process primarily refers to the actual manufacturing process in which a pressing material is subjected to pressure in a tablet press to form a tablet. The pressing process comprises one or more steps in tablet manufacturing, in which the quality, consistency, and / or shape of the tablet are determined.
[0045] The initial machine setting corresponds to the first pressing operation. This initial machine setting can be adjusted manually, for example, by a user. The first pressing operation then takes place according to the configuration established after the initial machine setting.
[0046] During the first pressing process, an initial set of production parameters is recorded.
[0047] Production parameters are preferably parameters that relate to one or more settings measured and recorded during the manufacturing process. Specifically, production parameters can be those configured "internally" on the tablet press while the pressing process is being carried out, but which cannot necessarily be directly adjusted via the machine settings. While the machine settings can be directly adjusted via an interface on the tablet press, this is not necessarily the case for production parameters.
[0048] The first set of production parameters, however, is related to the first pressing operation and thus to the initial machine setup. Following this initial machine setup, a first pressing operation is performed, during which an initial set of production parameters is recorded. This first set of production parameters specifically concerns parameters of the tablet press, particularly parameters of tablet press components that are configured on the tablet press during the first pressing operation.
[0049] At least one tablet parameter is predicted based on a machine learning model using the acquired first set of production parameters. Preferably, the at least one tablet parameter is predicted non-destructively and / or inline based on the machine learning model using the acquired first set of production parameters.
[0050] Inline prediction preferably means that the prediction is performed during ongoing production and / or without interrupting the production process. This refers in particular to the prediction of at least one tablet parameter while the tablet press is running. In this context, non-destructive preferably means that no removal, damage, and / or destruction of the manufactured tablets is required to predict the tablet parameter.
[0051] Preferably, machine settings and production parameters can overlap. Thus, (at least some) parameters can be assigned to both machine settings and production parameters. However, machine settings and production parameters differ in that machine settings pertain to those parameters that are configured for carrying out a pressing operation. This can be done, for example, at a dedicated human-machine interface for operating the tablet press. Production parameters are, in particular, those parameters that are recorded or measured during a pressing operation. This can be done, for example, by measuring devices and / or sensors, which are preferably installed in the tablet press.
[0052] It is possible for there to be a discrepancy between a parameter in a machine setting and a corresponding production parameter. This means that the values for the machine setting parameters and the production parameters do not necessarily have to match exactly. It is also possible for a parameter value in a machine setting to be essentially identical to the value of the corresponding parameter in the production parameters. The latter can occur particularly with a compaction simulator.
[0053] The machine learning model is preferably a mathematical model based on algorithms for learning from data and making predictions or decisions. In the context of the invention, the machine learning model is preferably used to calculate the predicted tablet parameters based on the acquired production parameters.
[0054] The model preferably learns from past pressing operations and is used to predict tablet parameters based on production parameters. In this context, "predict" means "forecast," so a prediction is a forecast. The machine learning model enables the identification of patterns and / or relationships in large datasets, particularly between production parameters and tablet parameters, especially when combined with appropriate machine settings.
[0055] In particular, the machine settings, production parameters, and tablet parameters are interconnected. The machine settings define production parameters, which in turn determine the tablet parameters.
[0056] Tablet parameter prediction primarily involves using a machine learning model and the recorded production parameters to predict the expected physical properties of the tablets. These predictive calculations enable a fast and accurate assessment of tablet quality without the need for extensive testing.
[0057] Based on at least one predicted tablet parameter, a second set of production parameters is determined.
[0058] Determining the second set of production parameters refers to the process of setting the corresponding production parameters based on the predicted tablet parameters in order to optimize tablet production. Optimization here means approximating the tablet parameters that should apply to the tablets produced by the tablet press. This step aims to adjust the production parameters, and thus the machine settings, so that they correspond to the desired tablet parameter values predicted by the machine learning model. This is an iterative adjustment. The determination of the second set of production parameters serves, in particular, to identify those production parameters that achieve an approximation of the first set of tablet parameters, i.e., the target value for the tablets.
[0059] Accordingly, a second machine setting is determined based on the second set of production parameters.
[0060] Based on the second set of production parameters, a reference is made to a second machine setting. This second machine setting is determined by the second set of production parameters of the tablet press. Therefore, the second machine setting and the second set of production parameters correspond to each other. The second machine setting, which was determined previously, is adjusted accordingly on the tablet press. The tablet press can then produce one or more tablets using this second machine setting.
[0061] At least one tablet parameter is measured on a tablet produced by the tablet press using the second machine setting. This serves to verify the agreement of this at least one tablet parameter with the first set of tablet parameters. The first set of tablet parameters represents a target value.
[0062] This represents the target properties that should apply to the tablet.
[0063] The initial machine setting thus represents a "random guess" that serves as a first approximation of the initial data set of tablet parameters. In practice, it is unusual, especially for a user lacking the necessary experience, for the tablet parameters of the produced tablets to essentially match the initial data set of tablet parameters using the initial machine setting.
[0064] Terms such as "essentially," "approximately," "about," "approximately," etc., preferably describe a tolerance range of less than ±20%, more preferably less than ±10%, even more preferably less than ±5%, and particularly less than ±1%. Statements using "essentially," "approximately," "about," etc., always disclose and include the exact value stated. By measuring at least one tablet parameter of a tablet produced by the tablet press with the second machine setting, it can be determined how successful an approximation to the desired tablet parameters is. If the approximation to the first set of tablet parameters is sufficiently close, repeating the described steps is not necessary.If the approximation is not sufficiently close to achieving a substantial match between the measured tablet parameters and the specified tablet parameters from the first data set, the steps described above are repeated. These steps are repeated until a match between the measured tablet parameters and the tablet parameters of the first data set is achieved.
[0065] Preferably, steps b) - h) are repeated. It may also be preferred to repeat steps b) - g). It may also be preferred to perform steps b) - g) instead of repeating steps b) - h).
[0066] Furthermore, it may be preferable to repeat the sequence of steps b) - g). This allows for multiple pressing operations to be carried out using the tablet press. The number of measurement operations is significantly reduced, which is particularly efficient because the effort associated with measuring tablet parameters is eliminated or reduced.
[0067] Repeating steps b) - h) and / or b) - g) thus represents a further iteration to achieve a closer approximation of the manufactured tablet parameters to the predefined tablet parameters of the first data set. This reduces the need for physical and / or chemical measurements or tests to verify the extent to which a desired machine setting corresponds to the desired predefined tablet parameters. In this respect, a correlation between a machine setting and the first data set of tablet parameters is established. In particular, the method allows the determination of which machine setting must be used to produce tablets with tablet parameters that essentially correspond to the predefined tablet parameters from the first data set.
[0068] It may also be preferable for step h) to be optional. Measuring the at least one tablet parameter on a tablet produced by the tablet press with the second machine setting serves to verify the correspondence of the at least one tablet parameter with the first data set. This measurement thus allows the determination of the extent to which tablets with tablet parameters corresponding to the desired tablet parameters have been produced. The desired tablet parameters are, in particular, specified by the first data set.
[0069] In a further preferred embodiment, the method is characterized in that the tablet parameters are selected from a group including tensile strength, porosity, tablet mass, disintegration time and / or geometric design.
[0070] The tensile strength of a tablet preferably refers to the tablet's resistance to tensile stress before it breaks or fractures. This parameter is important for determining the tablet's mechanical stability during handling and storage. Tensile strength is a crucial factor for the tablet's robustness, especially when it undergoes a multi-stage manufacturing process.
[0071] The porosity of a tablet indicates the proportion of its volume occupied by cavities or pores, compared to its total volume. It is an important parameter because it affects the tablet's residence time in the stomach, for example, and thus the rate of drug release. Higher porosity can accelerate the tablet's disintegration rate, as more water can penetrate the interior. Porosity can also be expressed as a percentage. It can also be related to the tablet's stability or strength.
[0072] Tablet mass describes the weight of a single tablet. A consistent tablet mass across a batch is a quality indicator in tablet production, as deviations from the target weight can affect both the dosage of the active ingredient and the consistency of the product. Tablet mass is typically expressed in milligrams (mg) or grams (g).
[0073] The disintegration time of a tablet refers to the time it takes for a tablet to completely dissolve or disintegrate after coming into contact with a liquid. This property is particularly important for tablets with a rapid release profile or for those that require a specific time to take effect in the body. Disintegration time is often expressed in seconds or minutes. A disintegration time that is too long or too short can impair the tablet's effectiveness.
[0074] The geometric design of a tablet preferably refers to its shape and / or structure, including its dimensions (e.g., diameter, thickness) and contours (e.g., flat, convex, round, oval, geometric shape). These geometric parameters influence the handling and / or the release dynamics of the active ingredient. The geometric design preferably also includes the surface area or total area and / or volume.
[0075] The aforementioned preferred tablet parameters represent important factors for achieving a desired tablet design, as they directly influence quality, stability, and / or efficacy. Precise tensile strength ensures that the tablets do not break or crumble during production, packaging, and storage, thus guaranteeing tablet integrity. Porosity, for example, affects the drug release dynamics, as it controls liquid absorption and tablet dissolution in the body. A uniform tablet mass is essential for accurate drug dosing to achieve a consistent therapeutic effect. Disintegration time determines how quickly a tablet dissolves in the body and thus releases the drug, which is particularly important for tablets with specific release profiles.The tablet's geometric design also contributes to its ease of use and influences how it disintegrates. Therefore, efforts are made to optimize and / or determine these preferred parameters to ensure the tablets' quality and efficacy and to guarantee consistent product quality.In a further preferred embodiment, the method is characterized in that the production parameters are selected from a group comprising force, energy, pressing pressure and / or displacement, wherein the energy is preferably selected from a group comprising plastic energy, elastic energy, compression energy, wherein the force is preferably selected from a group comprising pre-pressing force, main pressing force, ejection force, stripping force, radial die wall pressure, wherein the displacement is preferably selected from a group comprising immersion depth, web height, layer height, venting thickness, recipe profile area, elastic rebound profile, compression thickness and / or compression area.
[0076] Force refers specifically to the force exerted on the tablet material during the pressing process. Force can be described as the product of the pressure exerted on the material and the area of the material over which the pressure acts. Several specific types of force can be distinguished in a tablet press.
[0077] The pre-compression force refers to the force applied at the beginning of the pressing process to bring the material being pressed into the desired shape and achieve initial compaction. This also serves to expel air from the die.
[0078] The main pressing force is the greatest force acting on the material during the pressing process. It is responsible for the final shaping and compaction of the tablet.
[0079] The ejection force refers to the force that occurs when the finished tablet is pushed out of the mold. The ejection force is the mechanical result of the ejection unit and is used to classify potential tablet defects as well as material properties, such as stickiness in tablets.
[0080] The stripping force refers to the force applied to remove the tablet material after forming, for example to detach remaining pressing material from the die.
[0081] Radial die wall pressure is the pressure exerted on the walls of the die, compressing the tablet material. It is a consequence of the vertically applied pressure. Depending on the powder properties, the material expands radially. The radial die wall pressure is a result of this pressure, and its magnitude can slightly alter the tablet's density.
[0082] The energy applied during the pressing process is recorded in various forms and determines the physical properties of the tablet. Different forms can also be distinguished based on the energy levels involved.
[0083] Plastic energy is applied when the material being pressed is plastically deformed, i.e., when the material is permanently compressed. It is a measure of the irreversible compression of the material.
[0084] Elastic energy arises when the compressed material is elastically deformed and partially returns to its original shape after the pressure is released. It plays a role in the tablet's recovery after compression. Compression energy refers to the total energy required to compress the material and bring the tablet into its final shape. It encompasses both plastic and elastic energy.
[0085] The compression force is the pressure exerted on the tablet material throughout the entire compression process. It is generally determined by the combination of the applied forces (especially the pre-compression and main compression forces) and the contact area between the tools and the tablet material. High compression force generally results in higher density and therefore a stronger tablet.
[0086] The path describes the movement of the pressing elements, especially the punch (upper and / or lower punch), during the pressing process and significantly influences the shape and density of the tablets. The path can be differentiated in various aspects.
[0087] The immersion depth describes how deeply a punch (especially the upper and / or lower punch) penetrates the die to compress the material. A deeper immersion depth results in greater compaction.
[0088] The rim height is the height of the remaining margins in a die and influences the shape and strength of the tablets. Specifically, the rim height is the distance between the two punch surfaces (upper punch and lower punch) in the tablet press during compression. The rim height refers to the height after compression.
[0089] The fill height refers to the height of the material layer placed in the die before the compression process begins. A uniform fill height is important for consistent tablet mass and / or density. The fill height specifically refers to the height before compression.
[0090] The venting thickness refers to the space left in the die to allow air and / or other gases to escape from the tablet. An appropriate venting thickness can help prevent printing errors and irregular tablet shapes.
[0091] The recipe profile area describes a design that concerns the material structure in the die, e.g. the number of layers and / or the thickness of the powder matrix.
[0092] The elastic rebound profile indicates how strongly the material rebounds elastically after being unloaded, which particularly affects the final shape and density of the tablet.
[0093] Compression thickness describes the thickness of the tablet after the compression process. It is crucial for determining tablet stability and strength.
[0094] The compression area refers to the area of the die that is compressed during the pressing process, thus determining the final shape of the tablet.
[0095] The aforementioned preferred production parameters have proven particularly helpful in configuring the machine and achieving appropriate settings to ensure consistent, high-quality tablet production. Precise adjustment of these parameters allows for optimal control and optimization of the tablets' physical properties, such as strength, disintegration time, and / or porosity.
[0096] In another preferred embodiment, the second set of production parameters is determined by trial and error.
[0097] Trial-and-error refers to the trial-and-error method. This involves experimenting with production parameters (for example, by adjusting machine settings) through test measurements to find the optimal parameters that most closely match the desired tablet parameters. After each measurement, the quality of the produced tablets is assessed, and based on these results, the production parameters are adjusted to most closely approximate the desired tablet parameters.
[0098] The trial-and-error method for determining the second set of production parameters has the advantage of allowing a direct, experimental approach to the optimal production parameters and thus also the machine settings. This method can be particularly beneficial in cases where complex or unpredictable relationships exist between the production parameters and the tablet parameters, which are difficult to capture using a machine learning model. By systematically testing different machine settings and subsequently verifying tablet quality, the best possible production range can be determined experimentally. This offers a high degree of flexibility in the process, as the settings can be adapted to the specific requirements of the respective tablet manufacturing process, even if no extensive historical data is available for the machine learning model.Furthermore, in some cases the trial-and-error method can achieve rapid adaptation to new or changed production conditions without relying on the need for modeling.
[0099] In a further preferred embodiment, the method is characterized in that the at least one tablet parameter is predicted with a confidence interval using the machine learning model, wherein the confidence level of the confidence interval is preferably at least 90%, more preferably at least 95%, and most preferably at least 97%. A confidence interval indicates the range within which the true value of the predicted tablet parameter lies with a certain probability. For example, a confidence interval with a confidence level of 95% means that the true value of the tablet parameter lies within this interval with a probability of 95%. The confidence interval thus indicates the uncertainty of the prediction and ensures that not only a single point value but also the reliability of the prediction is known.
[0100] Advantageously, the confidence interval ensures that not only the predicted value of a tablet parameter is known, but also the reliability or consistency of that value. For example, if the confidence level of the confidence interval is 95%, it can be assumed with a high degree of probability that the actual value of the tablet parameter lies within this range. This gives tablet manufacturers greater confidence in their machine adjustments and the quality of the tablets produced. By providing a confidence interval, deviations in tablet production can be detected earlier and better understood. For instance, if the confidence interval is large, this may indicate that the prediction is uncertain and that more precise fine-tuning of the machine and production parameters may be necessary.On the other hand, a smaller confidence interval can indicate stable and well-controlled production. This leads to better quality control.
[0101] A predicted tablet parameter with a high confidence level (e.g., 95% or 97%) allows for more precise machine settings, so that one can be more certain that these machine settings will lead to the desired tablet parameters.
[0102] In a further preferred embodiment, the method is characterized in that the machine learning model accesses a database in which tablet parameters, production parameters and / or a machine setting are stored, and the machine learning model assigns tablet parameters to the production parameters and / or the machine setting.
[0103] A database is a collection of data structured and organized for efficient retrieval, editing, and / or updating. Databases are typically managed using database software, which allows users to store, search, analyze, and / or manipulate data. Databases can have various formats and / or data structures, such as linked tables or even more complex structures that define relationships between different parameters.
[0104] By leveraging a database of mappings between tablet parameters, production parameters, and / or machine settings, the machine learning model can advantageously model the relationships and patterns that occur during the tablet manufacturing process more precisely. In particular, the database provides a historical collection of empirical data that helps the machine learning model make better predictions about the impact of specific machine settings and / or production parameters on tablet quality. Because the mappings have already been empirically determined, the model can make more accurate and relevant predictions for future production runs. This minimizes uncertainties and ensures more stable and reliable production.
[0105] The database containing mappings between tablet parameters and production parameters also helps to maintain consistent and standardized tablet manufacturing. By accessing this database, the machine learning model ensures that machine settings and production parameters are optimized using the same principles and parameters every time. This prevents variability in product quality and ensures that the tablets achieve the same high quality standard in every batch.
[0106] Access to the database is preferably read-only. This means that data can be automatically retrieved from the database (read) and new data can be added to the database or existing data can be automatically modified (write). "Read-only" access means that data is retrieved but not modified. Specifically, this means that information can be requested from the database, but nothing within the database itself is changed.
[0107] The term "write" primarily refers to inserting, updating, and / or deleting data from the database. This means that new information can be added or existing values can be changed.
[0108] This makes it possible to update the database at any time if new requests and / or conditions arise in order to produce tablets with desired tablet parameters using a tablet press.
[0109] Preferably, data comprising tablet parameters, production parameters, and / or machine settings are stored in the database. In particular, the data can be stored in the database in such a way that an assignment (or assignment) of tablet parameters to production parameters and / or machine settings is possible. By accessing the database, the machine learning model can thus achieve more optimal learning and, consequently, the development of further relationships between tablet parameters, production parameters, and / or machine settings. In a further preferred embodiment, the method is characterized in that the machine learning model accesses the database via an encrypted connection.
[0110] The term "encrypted connection" primarily refers to a connection where data exchange between the machine learning model and the database is encrypted. In other words, the data between the database and the machine learning model is encrypted. Examples of encryption methods that can be implemented include symmetric encryption (using the same key for both encryption and decryption) and asymmetric encryption (using a public key to encrypt the data and a private key to decrypt it), but these are not the only possibilities.
[0111] Using an encrypted connection ensures that data is not manipulated or altered during transmission. Encryption protects against potential data corruption and guarantees that information arrives unchanged and intact. This is particularly important in the context of tablet manufacturing and machine settings, where precise and reliable data transmission is essential for accurate predictions by the machine learning model. Encryption also protects against data theft and fraud. If sensitive production data or secrets about machine settings are transmitted unencrypted, they could be intercepted by malicious actors and used for fraudulent purposes.An encrypted connection ensures that even if data is intercepted during transmission, it remains unusable, as it cannot be read without the decryption key. The database contains information such as specific production parameters, machine settings, and / or tablet parameters, which can be considered highly sensitive. This data is crucial for optimal production control and could cause significant financial damage in the event of a data breach or unauthorized access. An encrypted connection protects the confidentiality of this data. Only authorized parties with the correct decryption keys can restore the data to its original, unreadable form.
[0112] In a further preferred embodiment, the method is characterized in that the machine learning model is based on supervised learning, unsupervised learning and / or reinforcement learning, wherein the supervised learning is preferably selected from a group comprising logistic regression, SVM (Support Vector Machines), Naive Bayes, decision trees, linear regression, kNN, random forest, boosting algorithms, wherein the unsupervised learning is preferably selected from a group comprising a k-means algorithm and / or hierarchical clustering.
[0113] Supervised learning is a machine learning method in which the model is trained using a labeled dataset. This means that both the input data (e.g., production parameters such as temperature, pressure, and speed) and the correct output data (e.g., tablet parameters such as tensile strength and porosity) are known. The goal of supervised learning is to learn a function that accurately maps the input data to the output data.
[0114] Supervised learning enables the model to make more accurate predictions because it works with a dataset where the correct results are known. The model can learn which machine settings lead to the desired tablet parameters. By using supervised learning, the machine settings can be adjusted to optimally achieve the desired tablet parameters, such as a desired tensile strength or porosity.
[0115] The preferred algorithms listed (e.g., logistic regression, SVM, Naive Bayes, decision trees, etc.) offer various methods for modeling the relationships between input and output data.
[0116] Support Vector Machines (SVMs) and Random Forests, for example, are well-suited for modeling nonlinear relationships and are robust against outliers in the data. Logistic regression is used particularly in binary classifications, e.g., to predict whether a tablet falls within an acceptable range of a tablet parameter, such as tensile strength.
[0117] Support Vector Machines (SVM) is particularly suitable for finding non-linear separations between different machine settings and tablet parameters.
[0118] The Random Forest algorithm is an ensemble algorithm that incorporates multiple decision trees, improving robustness against outliers and increasing the accuracy of predictions. Unsupervised learning refers to methods where the model is trained on an unlabeled dataset, meaning the output data is unknown. The goal is to discover hidden patterns or structures in the input data. Unsupervised learning is frequently used to segment data into clusters or groups.
[0119] Therefore, unsupervised learning can be used to identify hidden patterns in production parameters that correlate with specific tablet parameters. This is particularly useful when not all possible relationships between production processes and tablet quality are yet known. Unsupervised learning can also help detect similarities between different production batches without a predefined target, leading to automatic segmentation and optimization of production.
[0120] Various algorithms can also be used to implement unsupervised learning. The k-means algorithm is a clustering algorithm used to group similar machine settings or production parameters into clusters that result in similar tablet qualities.
[0121] Hierarchical clustering is an algorithm that helps identify the hierarchical structure in data and form clusters based on similar tablet production characteristics. It is particularly useful when the data does not have clearly defined categories but can still be analyzed in a structured way.
[0122] In a further preferred embodiment, the method is characterized in that the machine learning model assigns a porosity to a pressing pressure, among other things, by means of
[0123] (1 ) In (E / EO) = -k*P with E = 1 -D rei
[0124] achieved, where E is the porosity of a compressed material, Eo of an uncompressed material, D rei a relative density of the compressed material and k denotes a Heckel constant.
[0125] The relative density D reiRelative density is a measure of a material's compactness compared to its maximum possible density. It describes the ratio of the material's density after compression to its density in the uncompressed state (e.g., the uncompressed state). In this context, relative density refers to the degree to which the tablet material has been compacted after compression. It influences tablet porosity, as higher density is associated with lower porosity, and conversely, lower density with higher porosity.
[0126] The Heckel constant k is a parameter that describes the consolidation properties of the material during the pressing process. It is a measure of the material's response to the applied pressing pressure and represents the steepness of the compaction curve used in the Heckel equation. A high Heckel constant indicates that the material is compacted quickly, meaning it has a high compaction rate during pressing. A low Heckel constant indicates that the material is compressed more slowly, resulting in a lower compaction rate. In the Heckel equation (1), the Heckel constant k is a proportionality factor that indicates how sensitive the material's porosity is to changes in pressing pressure. The Heckel constant k is material-dependent, as different materials have different compaction properties.The Heckel constant can be determined through experimental tests (such as Heckel analyses).
[0127] Achieving this mapping preferably means that the machine learning model uses, among other things, the aforementioned relationship (or other mathematical relationships) to determine an initial approach to the connection between production parameters, machine settings, and / or tablet parameters. In particular, the machine learning model itself can determine further relationships that exist between the production parameters, machine settings, and / or tablet parameters.
[0128] Equation (1) is known in tablet manufacturing as the Heckel equation. Equation (1) is relevant in that it describes the relationship between compression pressure and porosity of a tablet material and provides a way to mathematically link these quantities.
[0129] By utilizing this well-known equation, the machine learning model is built on a robust theoretical foundation. This enables more precise and reproducible predictions of porosity in response to changes in compression pressure. One advantage is that the Heckel equation takes into account the solidification properties of tablet material at different compression pressures, leading to more efficient and targeted optimization of tablet manufacturing. The machine learning model is particularly well-suited to determining further mathematical relationships for different parameters, especially between production parameters and tablet parameters. Equation (1) can serve as a first step in this process.
[0130] Furthermore, it is advantageously possible to predict and thus optimize tablet porosity based on compression pressure. Precise porosity control is important because it influences the disintegration time, tensile strength, and / or release behavior of tablets. By mapping porosity to a specific compression pressure, it becomes possible to reliably select the optimal compression pressure to achieve the desired tablet quality without using excessive pressure that could lead to damaged tablets. The use of the Heckel equation (1) also ensures that the machine learning model takes material-dependent properties (such as the relative density of the material) into account. This ensures that the compression pressure is adjusted according to the specific behavior of the material, enabling the production of tablets with consistently high quality.It enables a more precise prediction of how different materials behave at different pressing pressures and leads to a better adjustment of the process parameter in production.
[0131] The Heckel equation (1) can be easily integrated into the machine learning model to enable even more precise predictions. For example, in addition to predictions based on the Heckel equation, the machine learning model can also consider other production parameters and data (such as temperature, humidity, mixing times, etc.) to recognize complex patterns and correlate corresponding machine settings, production parameters, and / or tablet parameters. In particular, it may be preferable for the machine learning model to be implemented in such a way that equation (1) is used as a basis for determining further relationships, especially relationships between a machine setting, production parameters, and / or tablet parameters. In particular, equation (1) can serve as a basis for determining the relationship between a machine setting, production parameters, and / or tablet parameters.The approach can be used to determine relationships between a machine setting, production parameters, and / or tablet parameters in the machine learning model. A corresponding preferred implementation of the machine learning model can be carried out by the specialist. In particular, the specialist knows how equation (1) can be implemented in the machine learning model as a basis for determining further relationships between a machine setting, production parameters, and / or tablet parameters.
[0132] In a further preferred embodiment, the method is characterized in that the machine learning model determines a pressing process, among other things, by means of the relationship (2) In (1 / 1-D) = k*P + AmitA = In (1 / 1-D) + B
[0133] modeled, with o ma x = 2*B / (ir*D*d), where o max denotes a maximum tensile stress, B a required breaking force, D a diameter and d a bridge height of a tablet that can be produced using the tablet press, P a pressing force and k a Heckel constant.
[0134] Equation (2) advantageously provides an approach for modeling the mechanical properties of a tablet, in particular the tensile stress and the breaking strength as a function of the compression force and the porosity. These mechanical properties are crucial because they determine the behavior of the tablet under load, which is of great importance for the stability and shelf life of tablets.
[0135] The maximum tensile stress o ma The coefficient x and the required breaking force B are important parameters to ensure that the tablet does not break or become damaged during manufacturing, storage, and use. The model helps to precisely determine and optimize these properties.
[0136] Using equation (2), production parameters (such as pressing force P) and tablet parameters (such as diameter D) can be adjusted to achieve the desired breaking strength and tensile stress. This allows for precise control of the tolerances for the mechanical stability of the tablets during the manufacturing process, resulting in higher product quality. The model enables optimized control of the pressing pressure to ensure that the tablets are neither too soft (low breaking strength) nor too hard (excessive tensile stress).
[0137] Equation (2) also takes into account the diameter D and the bridge height d of the tablet, meaning that the tablet's geometric properties are incorporated into the model. This is important because the tablet's shape influences the pressure distribution during the compression process, which in turn affects porosity and mechanical strength. More accurate modeling of the tablet's geometric properties ensures that the compression process is tailored not only to the material's chemical properties but also to the specific shape of the tablet, resulting in more uniform and stable tablets.
[0138] By incorporating statistical and mathematically sound models to calculate tensile stress based on breaking force, the predictive accuracy of the machine learning model is significantly improved. Considering relevant mechanical parameters enables more precise modeling of the pressing process, resulting in a more stable and reliable manufacturing process.
[0139] Preferably, equation (2) represents a further development of equation (1). Equation (2) can also be implemented in the machine learning model as a basis for determining further relationships between a machine setting, production parameters, and / or tablet parameters. It is preferable that equation (1), equation (2), or a combination of equations (1) and (2) is implemented as the basis in the machine learning model.
[0140] The implementation of mathematical relationships as an approach or basis in a machine learning model for determining further relationships, in this case between a machine setting, production parameters and / or tablet parameters, is within the knowledge range of the average specialist.
[0141] Advantageously, the machine learning model can be developed in such a way as to enable an optimal mapping between machine settings, production parameters and tablet parameters, in particular to be able to produce tablets with the desired tablet parameters (first data set comprising tablet parameters).
[0142] In a further preferred embodiment, the method is characterized in that the database was created on the basis of physical measurements on the tablet press, preferably involving the assignment of production parameters to tablet parameters within the framework of the physical measurements.
[0143] In this context, a physical measurement preferably refers to a measurement used to record quantities occurring during the pressing process of a tablet press. It is particularly preferred that production parameters are measured. In particular, the force, energy, pressing pressure, and / or displacement, possibly including their specifications, can be recorded as part of the physical measurement. It is also preferred that tablet parameters are measured. Thus, the physical measurement can encompass the measurement of production parameters and / or tablet parameters. For the physical measurement, measures are preferably taken and / or means, such as sensors, are provided within the tablet press to record the corresponding measured quantities.
[0144] Advantageously, using physical measurements to create the database ensures highly precise and realistic data. Directly collecting measurements from the actual pressing process during tablet press operation makes the basis for the machine learning model more realistic. This allows for more accurate predictions of tablet parameters. Consequently, the machine learning model can rely on empirically obtained values, which increases the reliability of the resulting optimizations. The physical measurements also advantageously establish a clear and verifiable correlation between production parameters and tablet parameters. This enables a sound correlation that supports the optimization of machine settings in a scientifically rigorous manner.This classification helps to develop a better understanding of the influence of specific production parameters on the quality of the tablets.
[0145] This allows for a particularly reliable determination of which production parameters and / or machine settings must be selected to produce tablets with the desired parameters. The physical measurements yield especially realistic, reproducible, and easily verifiable results.
[0146] Advantageously, the results are also reproducible. This, in turn, has a beneficial effect on the overall production process for tablet manufacturing.
[0147] In a further preferred embodiment, the method is characterized in that, for measuring a pressing force by at least one punch, a force sensor is operatively connected to the at least one punch, wherein preferably the force sensor is attached to a slide through which the at least one punch is guided during pressing, wherein preferably the force sensor is in the form of a strain gauge.
[0148] The functional connection of the force sensor with at least one punch, in particular an upper and / or lower punch, preferably refers to a connection through which force transmission between the punch and the force sensor can occur. The force sensor can detect the mechanical load generated by the punch. This can, for example, relate to the pressing force exerted by the punch on the material being compressed. Advantageously, the use of a force sensor enables a very precise measurement of the pressing force, since the force sensor detects the actual load exerted by the punch in real time. The pressing force is an important parameter for ensuring the quality and consistency of tablet manufacturing.
[0149] The force sensor is preferably mounted on a slide. The slide serves as a guide mechanism for the ram during the pressing process. The sensor is positioned in this way so that it can measure the force acting on the ram as it compacts the material being pressed.
[0150] By attaching the force sensor to the slide that guides the ram, a direct measurement of the pressing force is advantageously enabled, and continuous and long-lasting monitoring of the pressing force is ensured.
[0151] The use of a strain gauge as a force sensor is particularly advantageous here. It is known that the functionality of a strain gauge is based on the piezoresistive effect. When the strain gauge is subjected to a mechanical load, the material of the strain gauge deforms, leading to a change in its electrical resistance. This change in resistance is proportional to the strain of the material. When the carriage guides the ram during the pressing process, and when the ram presses against the material being pressed, a force is transmitted to the carriage. This force is then measured by the attached strain gauge. As the pressing process progresses, the applied pressing force causes a minimal deformation (stretching or compression) of the carriage to which the strain gauge is attached. The strain gauge detects this change in the shape of the material.The deformation of the strain gauge changes its resistance, as its length and / or cross-section change according to the applied force. This change in resistance is directly related to the force acting on the slide. Therefore, the change in resistance allows conclusions to be drawn about the applied pressing force.
[0152] In a further preferred embodiment, the method is characterized in that an upper punch and a lower punch of the tablet press are operatively connected to a force sensor and a force, a strain and / or a pressure on a material to be compressed and / or a movement of the upper punch and / or lower punch is detected by means of the force sensors.
[0153] The force sensor is capable of detecting the mechanical force exerted by the upper and lower punches on the material being pressed. For example, the force can be measured by using a strain gauge as a force sensor on the guide rail of one of the punches or by attaching it to the punch body. When the upper and lower punches press down on the material, the length of the strain gauge changes, resulting in a change in its electrical resistance.
[0154] The elongation particularly concerns the elongation of the pressing material, which results from the pressure exerted on the pressing material by the upper and lower dies.
[0155] When the sensor is mounted on the punch or its guide, it measures how the punch stretches or compresses due to the applied force. To measure the stretching or compression of the material being pressed, the force exerted on the material is used, as this correlates with the material's stretching or compression. The respective force sensor measures the force exerted on the material by the upper and lower punches. This applied force is proportional to the deformation experienced by the material being pressed. Therefore, the measured force also allows conclusions to be drawn about the stretching or deformation of the material being pressed.
[0156] The pressure on the material is calculated as the force per unit area. The force sensor measures the force and divides it by the contact area of the plunger with the material. This calculation allows the determination of the pressure on the material to be compressed. Pressure is an important factor because it influences the compaction properties of the material and thus the strength and homogeneity of the tablets.
[0157] The movement of the upper and lower punches can be directly detected, for example, by force sensors that monitor the punch's position (e.g., by measuring displacement or acceleration). This measurement provides valuable information about the dynamics of the pressing process, such as the speed at which the punches penetrate the material or the delay at the start of the pressure application. In a further preferred embodiment, the method is characterized in that one or more motion sensors are used to measure a linear punching motion of an upper and / or lower punch, to measure the respective positions of the upper and / or lower punch, the fill level of a die with material to be compressed, and / or the geometric shape of the tablet.
[0158] The linear punching motion refers to the linear movement of the upper punch (and, if applicable, the lower punch) along a straight line during the pressing process. Specifically, this means that the upper punch strikes the material being pressed in a straight vertical motion, compacting it accordingly. This type of movement is called a punching motion because the punch exerts a kind of punching effect on the material by compressing it to form the tablet.
[0159] Motion sensors can be used to monitor the position of both the upper and lower punches. These sensors measure the linear movement of the punches relative to their position throughout the entire pressing process. The upper punch moves downwards to press the material further into the die. A motion sensor can detect the precise vertical position of the upper punch. This measurement is crucial for determining the depth of penetration into the material. The lower punch typically has a supporting function, stabilizing the material within the die. A motion sensor can also measure the vertical movement of the lower punch, helping to monitor the entire movement sequence.
[0160] The fill height of a die describes how high the material being compressed is in the die before compression. The fill height thus primarily describes the amount of material within the die before the upper punch impacts the material to be compressed. This fill height is an important parameter because it influences the amount of material in the finished tablet and its density and strength.
[0161] The measurement of the fill level depends, among other things, on the movement of the upper piston.
[0162] As soon as the upper punch enters the die, the fill level of the material in the die changes due to the pressure exerted by the upper punch. A motion sensor can measure how deep the upper punch penetrates the die. From this, the fill level of the material can be determined, at least indirectly. A greater fill level means that the upper punch starts higher up and therefore more material is fed into the die. By comparing the upper punch position and the fill level, the exact amount of material in the die can be determined.
[0163] The position of the upper punch and the fill level of the material also influence the geometry of the tablet, or its geometric shape. The movement of the upper punch determines how deeply it penetrates the material and how much material is compacted, which in turn determines the thickness of the tablet. For example, a greater fill level can result in a thicker tablet, while a lesser fill level can result in a thinner tablet.
[0164] In a further preferred embodiment, the method is characterized in that one or more position sensors are used to measure the position of a lower and / or upper punch, wherein preferably the one or more position sensors are attached to a stop of the upper and / or lower punch.
[0165] A position sensor is a sensor used to measure the precise position of the upper and / or lower punch within a tablet press. Position sensors can be based on various principles; for example, they can be optical, inductive, capacitive, or magnetic field position sensors.
[0166] The precise position of the punches is a crucial parameter during the pressing process. The upper punch moves downwards towards the die to compress the material. Its position is monitored by a position sensor to ensure it reaches the correct depth and thus applies the correct pressure. The lower punch supports the material by stabilizing it within the die. Its position is also critical for ensuring the entire pressing process runs smoothly and the material is compressed evenly.
[0167] The position of the punches is monitored by position sensors, which can be mounted on either the upper and / or lower punch. The sensors measure the vertical movement of the punches and thus provide precise data about their position throughout the entire pressing process.
[0168] As the upper punch moves downwards during the pressing process, a position sensor mounted at the upper punch stop measures the lowest position the punch reaches. This information can be used to determine how much material has been compacted in the die. Similarly, the lower punch can be monitored to ensure it is correctly positioned and providing the necessary support for the material.
[0169] The stop is preferably a point or area where the punch stops or its position is limited during its movement cycle. Position sensors attached to these stops can precisely measure when the punch has reached its end position. For example, an upper punch stop could be the point where the punch has fully compacted the material in the die. The position sensor can then measure the maximum depth of the upper punch. Similarly, a lower punch stop could be the point where the lower punch supports the material from below or assists with the pressure. Again, the position sensor can measure the maximum height of the lower punch.
[0170] The position sensors enable precise and continuous monitoring of the punch positions and offer several advantages for efficient recording of the pressing process. For example, the precise measurement of the upper and / or lower punch position allows for a correlation to be established between the depth of penetration into the pressing material and / or die by the upper and / or lower punch and the shape and / or density of the tablet. This accurate position measurement using the position sensors thus provides optimal insights into the relationship between tablet parameters and production parameters. For instance, it allows for determining the maximum penetration depth of the upper punch into the die, as excessive penetration can lead to over-compression.
[0171] In a further preferred embodiment, the method is characterized in that tablet parameters are determined and assigned to the forces and / or movements by means of recording a temporal development of forces and / or movements of an upper and / or lower punch during pressing.
[0172] The temporal evolution of forces and / or movements can be advantageously used to establish a direct correlation between the forces and the resulting tablet parameters. This means that the forces and movements of the upper and lower punches measured by the sensors are recorded in real time to draw conclusions about the tablet parameters.
[0173] For example, it may be advantageous to continuously measure and record the forces and movements of the upper and lower punches at the beginning of the pressing process. This can be achieved using suitable sensors, such as force sensors (e.g., strain gauges) and position sensors to measure the punch movements. Over time, the force exerted on the material being pressed and the depth to which the punches penetrate the die are monitored. This measurement data can be recorded in real time and is helpful for accurately tracking the progress of the pressing process.
[0174] Based on the recorded data on the temporal development of forces and movements, the corresponding tablet parameters (such as density, strength, etc.) can be determined. For example, a stronger force over a specific period could correlate with a higher tablet density or strength, while a weaker or irregular force might indicate potential problems in the pressing process or the material itself.
[0175] In a further preferred embodiment, the method is characterized in that a lower and / or upper punch is controlled by means of external control sensors.
[0176] An external control sensor is a sensor that is not directly connected to the upper and / or lower punch of a tablet press, but is positioned elsewhere. The external control sensor can, for example, be located external to the tablet press. Nevertheless, when using an external control sensor, a data connection to a lower and / or upper punch exists, enabling control, particularly of the movement, of the lower and / or upper punch.
[0177] By means of one or more external control sensors, adjustments can advantageously be made to the operation of the tablet press, particularly with regard to the effect of the punches, without having to carry out maintenance on the tablet press itself. For example, if a different tablet shape is desired, the machine settings and thus also the production parameters can be changed by the external control sensors so that tablet parameters for the desired tablet shape are achieved. In a further preferred embodiment, the method is characterized in that the tablet press is a compaction simulator, a rotary press, or an eccentric press.
[0178] The aforementioned preferred tablet presses have become established in the state of the art for reliably producing tablets. The individual tablet press variants offer unique advantages.
[0179] A compaction simulator is a tablet press designed primarily for laboratory and experimental purposes. It allows for the simulation and analysis of the pressing process for tablet production. Specifically, it enables the determination of how various production parameters influence the tablet parameters. In this sense, the compaction simulator is a tablet press that replicates the compaction process of tablet manufacturing under controlled conditions, but on a smaller scale than an industrial tablet press.
[0180] Since the compaction simulator prefers to operate on a smaller scale, many tests and adjustments can be performed without requiring large production runs of tablets. This also reduces the costs incurred during tablet manufacturing.
[0181] A rotary tablet press is a special type of tablet press in which the dies are mounted on a rotating circle or rotary mechanism. Rotary tablets are particularly suitable for high-volume tablet production. The rotating mechanism ensures that each die passes continuously under the punches that form the tablets. This makes them ideal for the industrial production of large quantities of tablets.
[0182] An eccentric tablet press is a tablet press in which the pressing force is generated via an eccentric mechanism. This mechanism transfers the force to the upper punch and die to compress the material. These machines are often suitable for medium- to high-throughput tablet production and are characterized by stable and consistent compaction pressure. Eccentric tablet presses are highly adaptable, meaning they can process a wide variety of shapes, sizes, and materials.
[0183] In another aspect, the invention relates to a computer program product for determining a machine setting of a tablet press, comprising commands which, when the computer program product is executed by a computer, cause it to perform the following steps:
[0184] a) Input of a first data set comprising at least one tablet parameter that should apply to tablets that can be produced using a tablet press,
[0185] b) Recording of an initial set of production parameters during a pressing operation of the tablet press,
[0186] c) Prediction of at least one tablet parameter based on a machine learning model using the recorded first set of production parameters from step b), d) Determination of a second set of production parameters based on the at least one predicted tablet parameter from step c).
[0187] The computer program product can be written or programmed in any programming language or model-based development environment, e.g., in C / C++, C#, Objective-C, Java, Basic / VisualBasic, MATLAB, Simulink, StateFlow, Lab View, Python and / or Assembler.
[0188] The average person skilled in the art recognizes that the technical features, definitions and advantages of the method according to the invention and of preferred embodiments of the method also apply to the computer program product and its preferred embodiments, and vice versa.
[0189] A key advantage of the computer program product according to the invention is that it eliminates the need for complex physical and / or chemical investigations to determine tablet parameters in most production phases. Prior art methods require extensive testing that is laboratory-intensive and time-consuming. The computer program product, which utilizes machine learning, allows tablet parameters to be derived directly from the production parameters, thus minimizing the need for extensive measurements. Physical and / or chemical measurements are only necessary for interim or final validations, which significantly accelerates the development process and makes it more cost-effective.
[0190] Another advantage of the computer program is the increased availability of tablet presses and measuring instruments. In traditional processes, repeated tests lead to longer downtimes and poorer resource utilization. Through predictive modeling, the computer program enables faster test cycles and reduces the need for continuous machine use, thereby optimizing resource utilization and increasing machine availability.
[0191] The computer program also increases the automation of trials by reducing repetitive manual interventions. Tests and adjustments are based on the predicted values of the machine learning model, which accelerates the process and minimizes potential human error. This increases the reproducibility and stability of the production process.
[0192] The computer software accelerates the availability of new machine settings, i.e., new tablet formulations. In traditional processes, developing new machine settings requires numerous manual adjustments and tests. The computer software enables faster and more precise adjustment of machine settings, thus reducing time-to-market for new products or tablets and allowing for a quicker response to market demands or customer needs.
[0193] In a further preferred embodiment, the computer program product is characterized in that the input of the first data record is made via a user interface provided by the computer program product, preferably in the form of a graphical interface. The user interface is the interface through which a user can interact with the computer program. It includes all elements and functions that enable the user to input data, obtain results, and communicate with the computer program product. These input functions could, for example,
[0194] Includes text fields, dropdown menus, buttons and / or other input elements.
[0195] In particular, the first data set is entered via the user interface. This initial set of tablet parameters includes those parameters that should apply to the tablets produced by the tablet press. The user interface allows the user to enter this data efficiently and easily. The user interface makes data entry more intuitive and accessible, simplifying the user's initial setup of the computer program and the process of finding the correct machine settings.
[0196] Preferably, the user interface is in the form of a graphical user interface (GUI). A graphical user interface (GUI) uses visual elements such as windows, icons, buttons, text fields, and / or other graphical representations to facilitate user interaction with the computer. This type of interface is particularly user-friendly because it allows data to be entered in a visually intuitive way and results to be displayed in the form of graphs or tables. The graphical interface allows the user to enter the initial data set by simply filling in input fields and / or making selections in a visually structured environment.
[0197] In another aspect, the invention relates to a system comprising a tablet press and a computing unit, wherein a computer program product as described above is installed in the computing unit.
[0198] The average person skilled in the art recognizes that the technical features, definitions and advantages of the method and computer program product according to the invention and of preferred embodiments of the method and computer program product also apply to the system, and vice versa.
[0199] Preferred computing units include, without limitation, an integrated circuit (IC), an application-specific integrated circuit (ASIC), a programmable logic circuit (PLC), a field-programmable gate array (FPGA), a microprocessor, a microcomputer, a programmable logic controller and / or any other electronic circuit, preferably programmable.
[0200] In preferred embodiments, the computing unit can also be configured to process recorded and / or input data, e.g., production parameters and / or tablet parameters. Data processing preferably means that the computing unit is configured to perform calculations and / or computational steps, for example, based on recorded production parameters.
[0201] The term computing unit preferably refers to any device or unit that can be configured to perform arithmetic operations.
[0202] Preferably, the computing unit is, for example, a processor, a processor chip, a microprocessor, and / or a microcontroller. The computing unit can also preferably be a programmable circuit board. The computing unit can also preferably comprise a computer-usable or computer-readable medium, such as a hard drive, random access memory (RAM), read-only memory (ROM), flash memory, etc.
[0203] The processing unit may preferably also include a storage unit. A storage unit allows for the backup and / or temporary storage of data. Non-limiting examples of storage devices, preferably semiconductor storage devices, are volatile memory, (RAM) memory, or non-volatile memory such as ROM memory, EPROM memory, EEPROM memory, or flash memory, and / or other storage technologies.
[0204] In particular, the computer program can be installed on the processing unit. Using this program, tablet parameters can be predicted. This is done indirectly by determining the machine settings required to produce tablets with the desired parameters using a tablet press. The desired tablet parameters can be entered as an initial data set. Specifically, the computer program can predict at least one tablet parameter during a pressing operation on a tablet press.
[0205] This significantly reduces the need for costly downstream experimental tests to measure tablet parameters.
[0206] The characteristics of the computer program product, especially one installed on the processing unit, can be considered technical characteristics because they have a direct physical effect. This physical effect lies in establishing a correlation between machine settings, production parameters, and / or tablet parameters. Accordingly, the system itself also exhibits a direct physical effect and therefore includes technical characteristics.
[0207] The computing unit is preferably data-connected to the tablet press. A data connection preferably means a connection that allows the exchange of data or parameters (such as tablet parameters, production parameters and / or machine settings) between the computing unit and the tablet press.
[0208] The computing unit can preferably be internal or external. An internal computing unit is preferably integrated within the tablet press. An external computing unit is preferably located outside the tablet press.
[0209] The tablet parameters predicted by the machine learning model can preferably be output or provided as machine-readable data. It is also preferred that the predicted tablet parameters be displayed in a user interface. The user interface can, in particular, be a graphical interface. References
[0210] Hayashi, Yoshihiro, et al. "A data-driven approach to predicting tablet properties after accelerated test using raw material property database and machine learning." Chemical and Pharmaceutical Bulletin 71.6 (2023): 406-415.
[0211] Persson, Ann-Sofie, and Göran Alderborn. "A hybrid approach to predict the relationship between tablet tensile strength and compaction pressure using analytical powder compression." European Journal of Pharmaceutics and Biopharmaceutics 125 (2018): 28-37.
Claims
PATENT CLAIMS 1. Method for determining a machine setting of a tablet press comprising the following steps: a) Provision of an initial data set comprising at least one tablet parameter to be met for a tablet that can be produced using a tablet press, b) Performing a first pressing operation with a first machine setting, c) Recording a first set of production parameters during the first pressing operation, d) Prediction of at least one tablet parameter based on a machine learning model using the recorded first set of production parameters from step c). e) Determination of a second set of production parameters based on the at least one predicted tablet parameter from step d). f) Determination of a second machine setting based on the second set of production parameters from step e). g) Setting the second machine setting on the tablet press, h) Measurement of the at least one tablet parameter on a tablet produced by the tablet press with the second machine setting to verify the agreement of the at least one tablet parameter with the first data set, i) optionally repeat steps b) - h) until at least one tablet parameter of the tablet substantially matches the first data set, so that the machine setting for tablets is determined using the first data set.
2. Procedure according to the previous claim characterized by the fact that The tablet parameters are selected from a group including tensile strength, porosity, tablet mass, disintegration time and / or geometric shape and / or the production parameters are selected from a group including force, energy, pressure and / or distance, where the energy is preferably selected from a group comprising plastic energy, elastic energy, compression energy, wherein the force is preferably selected from a group comprising pre-compression force, main compression force, ejection force, stripping force, radial die wall pressure, wherein the path is preferably selected from a group comprising immersion depth, ridge height, layer height, venting thickness, recipe profile area, elastic rebound profile, compression thickness and / or compression area.
3. Method according to one or more of the preceding claims, characterized in that The determination of the second set of production parameters is carried out by means of trial and error.
4. Method according to one or more of the preceding claims characterized by the fact that that at least one tablet parameter is predicted with a confidence interval using the machine learning model, wherein preferably a confidence level of the confidence interval is at least 90%, more preferably at least 95%, and more preferably at least 97%.
5. Method according to one or more of the preceding claims characterized by the fact that The machine learning model accesses a database containing tablet parameters, production parameters and / or machine settings, and the machine learning model assigns tablet parameters to the production parameters and / or machine settings. The machine learning model preferably accesses the database via an encrypted connection.
6. Method according to one or more of the preceding claims characterized by the fact that the machine learning model is based on supervised learning, unsupervised learning and / or reinforcement learning, where supervised learning is preferably selected from a group including logistic regression, SVM (Support Vector Machines), Naive Bayes, decision trees, linear regression, kNN, random forest, boosting algorithms, where unsupervised learning is preferably selected from a group comprising a k-means algorithm and / or hierarchical clustering.
7. Method according to one or more of the preceding claims characterized by the fact that the machine learning model assigns a porosity to a pressure, among other things, by means of In (E / EO) = -k*P with E = 1 -D rei achieved, where E denotes a porosity of a compressed material, Eo of uncompressed material, Drei a relative density of the compressed material and k a Heckel constant.
8. Method according to one or more of the preceding claims characterized by the fact that The machine learning model performs a pressing process, among other things, using the relationship In (1 / 1-D) = kP + AmitA= In (1 / 1-D) + B modeled with a ma x= 2B / (irDd), where Omax denotes a maximum tensile stress, B a required breaking force, D a diameter and d a bridge height of a tablet that can be produced using the tablet press, P a pressing force and k a Heckel constant.
9. Method according to claim 5 characterized by the fact that the database was created based on physical measurements taken on the tablet press, preferably, within the framework of physical measurements, an assignment of production parameters to tablet parameters was carried out.
10. Method according to one or more of the preceding claims characterized by the fact that A force sensor is used to measure a pressing force by means of at least one punch, with which at least one punch is operatively connected. wherein preferably the force sensor is attached to a slide through which the at least one punch is guided during pressing, where preferably the force sensor is in the form of a strain gauge, wherein preferably an upper punch and a lower punch of the tablet press are operatively connected to a force sensor and a force, a strain and / or a pressure on a material to be compressed and / or a movement of the upper punch and / or lower punch is detected by means of the force sensors.
11. Method according to one or more of the preceding claims characterized by the fact that One or more motion sensors are used to measure a linear punching movement of an upper and / or lower punch, to measure the respective positions of the upper and / or lower punch, a fill level of a die with a material to be compressed and / or a geometric design of the tablet.
12. Method according to one or more of the preceding claims characterized by the fact that one or more position sensors are used to measure the position of a lower and / or upper punch, wherein preferably one or more position sensors are attached to a stop of the upper and / or lower punch.
13. Method according to one or more of the preceding claims characterized by the fact that Tablet parameters are determined and assigned to the forces and / or movements by recording the temporal development of forces and / or movements of an upper and / or lower punch during pressing.
14. Computer program product for predicting at least one tablet parameter comprising commands which, when executed by a computer, cause the computer to perform the following steps: a) Input of an initial data set comprising at least one tablet parameter that should apply to tablets that can be produced using a tablet press, b) Recording of an initial set of production parameters during a pressing operation of the tablet press, c) Prediction of at least one tablet parameter based on a machine learning model using the recorded first set of production parameters from step b). d) Determination of a second set of production parameters based on the at least one predicted tablet parameter from step c).
15. System comprising a tablet press and a computing unit, wherein a computer program product according to claim 14 is installed in the computing unit.
