Method for process automation of production process

Abstract

The invention relates to a computer-implemented method (200) for process automation of a production process involving an extrusion foaming thermoplastic polyurethane process for producing polyurethane foams. The method (200) includes providing (210) process data indicative of at least a portion of a production process performed for producing polyurethane foam. Further, a material property of the produced polyurethane foam is determined using (220) a property determination model, where the material property is indicative of at least one characteristic of the produced polyurethane foam, and where the property determination model is a data-driven model adapted to determine the material property based on the process data. Based on the determined material properties, a control action for controlling the production process of the polyurethane foam is generated (230). The method allows for improved automatic control of production processes involving extrusion foaming thermoplastic polyurethane processes.

Description

Technical Field

[0001] The present invention relates to a computer-implemented method, apparatus and computer program product for process automation of a production process. Furthermore, the present invention relates to a system for automatic production using any of the above methods, apparatus or computer program products. Furthermore, the present invention relates to a training method, training apparatus and computer program product for parameterizing a characteristic determination model used in the above methods, apparatus and computer program products. Furthermore, the present invention relates to control data generated according to the method, apparatus and computer program product for process automation. Background Art

[0002] Controlling the execution of the extrusion foaming thermoplastic polyurethane process in automatic process control may be a challenging task. In general, the process parameters that control the execution of the process are usually fixed during the commissioning of the corresponding production plant and are set according to the predetermined design specifications of the foam to be produced. Then the compliance with these set process parameters is usually controlled by a process automation system, which allows the control of corresponding systems, such as heaters, mixers, etc., to provide the corresponding conditions required for foam production. In the case of changes in certain precursors for producing specific foams, for example, if the quality of the prepolymer used to produce the foam changes, the adjustment of the process variables is necessary and is usually implemented manually based on the experience of the corresponding experts. However, the foam produced by the process may still deviate from the target specification, wherein the process of perceiving such deviations, finding out the reasons thereof and adjusting the process variables may be time-consuming, costly and often result in the production of waste before adjusting the process variables. Therefore, it would be advantageous if a method that allows automatic adjustment of process variables could be provided, which is objective (i.e. not based on the experience of experts) and improves foam production in general, such as reducing the waste of foam and reducing production downtime. Summary of the invention

[0003] It is an object of the present invention to provide a computer-implemented method, apparatus and computer program product that can achieve improved automatic control of a production process, the production process involving an extrusion foamed thermoplastic polyurethane process for producing polyurethane foam. In addition, it is another object of the present invention to provide a method, a training apparatus and a computer program product that allow for the training of a parameterized property determination model that can be used in the method, apparatus and computer program product to improve the production process of polyurethane foam.

[0004] In a first aspect of the present invention, a computer-implemented method for process automation of a production process is proposed, the production process relating to an extrusion foaming thermoplastic polyurethane process for producing polyurethane foam, wherein the method comprises: a) providing process data indicating at least a portion of the production process performed for producing polyurethane foam, b) determining material properties of the produced polyurethane foam using a property determination model, wherein the material property indicates at least one feature of the produced polyurethane foam, and wherein the property determination model is a data-driven model suitable for determining material properties based on process data, and c) generating control actions for controlling the production process of the polyurethane foam based on the determined material properties.

[0005] Since current process data are used in the property determination model to predict at least one material property of the produced polyurethane foam, deviations from the corresponding target properties of the polyurethane foam can be detected very quickly and control actions can be generated to correct these deviations. In particular, the possibility of directly predicting the foam properties and controlling the production accordingly becomes important. Even if the properties are measured online, the measurement does not provide direct information about the influence of process parameters like the model does. In practice, the measurement is not carried out directly after production, but after the production life cycle of the product using the produced foam. In this case, the present invention prevents the detection after the production of an entire batch of products that do not meet the target properties. As a result, the control process becomes more efficient and in particular much faster, because it is not necessary to wait for the corresponding measurement of the foam material properties and / or the corresponding waste foam production can be avoided. In addition, the method can reduce the role of human experience in controlling the production process.

[0006] In general, the computer-implemented method may be implemented and performed by any hardware and / or software or a combination thereof that provides the corresponding functionality of the method as defined above. In particular, the method may be performed by one or more computing devices, for example, may be performed by only one general or special computing device, or may be performed in distributed computing on multiple computing devices, for example in the form of cloud computing or a distributed network application. Utilizing distributed computing has the advantage that data may be received from anywhere in the network and the process may also be provided at least partially at all communication coupling devices that have access to the distributed computing. This may make it possible to also include data, for example, target characteristic values ​​or measured characteristic values ​​from the customer of the produced foam. Furthermore, the customer may even directly access the method via the distributed computing network, for example to directly determine control actions based on his preferred foam characteristics.

[0007] The production process relates to an extrusion foaming thermoplastic polyurethane process for producing polyurethane foam. In general, the foaming process can consist of an extrusion step, in which one or more blowing agents, chain extenders and nucleating additives are mixed with the polymer material. Optionally, after this step, the foamed polymer is granulated and decompressed to produce granulated polyurethane foam. In particular, this extrusion foaming thermoplastic polyurethane process includes a mixing step, an extrusion step, a decompression step and a cooling step. However, depending on the corresponding settings of the extrusion foaming thermoplastic polyurethane process, additional steps may also be part of the process, for example, an additional granulation step may be provided to produce granulated foam.

[0008] In a first step, the method includes providing process data indicating at least a portion of a production process for producing polyurethane foam being performed. In particular, the process data may refer to any data indicating a currently performed production process, for example, control data for controlling the performed production process, for example, by defining process variables, measurement data from measurements performed during the production process, and soft sensor data involving predicted measurement values. In this context, process variables refer to variables that can be manipulated and set during the process to obtain polyurethane foam with predetermined characteristics. For example, process variables may refer to temperature, ratios between specific materials, mixing speed, pressure distribution, etc. The measurement data may generally be measured by any sensor provided as part of a production device or in the environment of a production device, wherein the measured quantity may have an impact on the production process. The soft sensor data is preferably derived from the measured sensor data based on a known or learned function between the measured quantity and the derived quantity. Such derived measurement data is particularly used in situations where it is impossible or at least costly to set a corresponding sensor for the derived quantity. Preferably, the process data indicates a process performed during at least one of a mixing step, an extrusion step, a decompression step, and a cooling step of the production process. Optionally, the process data may also indicate the process performed by the granulator. In particular, it is preferred that the process data provided include quantities that are generally known to affect the material properties of the produced polyurethane foam. However, since in a complex production environment, such as in the production of polyurethane foam, not all of the complex relationships between the quantities and the material properties of the foam are predictable, the process data may also include other quantities whose relationship with the material properties of the foam is not yet known. In this case, it is possible to determine the relevant process data quantities independently of previous knowledge during the learning process. In general, the process data provided include such quantities, that is, the characteristic determination model has been trained with these quantities to determine the material properties. Accordingly, the provision of process data may also include selecting appropriate process data from all available process data so that the characteristic determination model can determine the material properties of the produced polyurethane foam. However, the selection may also be performed when providing process data as input to the characteristic determination model, and all available process data are provided in the providing step. The process data may be provided, for example, via an input unit, which is configured to receive corresponding process data, for example, from a storage device or as an input of a user. However, preferably, the process data are provided via an interface unit, which is configured to interface directly with, for example, a process automation system controlling the production process and / or with one or more sensors (e.g. also soft sensors) to directly receive the corresponding measured process data.In a preferred embodiment, the process data comprises at least one of the following: the ratio between blowing agent and polymer, the ratio between talc and polymer, the rotation speed of the extruder, the temperature distribution inside the extruder and the pressure distribution inside the extruder. In addition, if a granulator is present, the process data may also comprise at least one of the flow rate through the granulator and the rotation speed of the cutting disc. The inventors have found that providing at least one of the above quantities as part of the process data allows a particularly accurate determination of the material properties of the polyurethane foam.

[0009] In another step, the method comprises determining at least one material property of the produced polyurethane foam using at least one property determination model. In general, the material property indicates at least one feature of the produced polyurethane foam. In particular, it is preferred that the determined material property refers to at least one of the foam bulk density, tensile strength and shape parameter. These quantities have the greatest impact on the technical application possibilities of the produced foam and are therefore often used to characterize the produced foam. However, the material property may also refer to the properties of the polyurethane foam in the final product. For example, if the polyurethane foam is to be used in a sole, the material property may also refer to the material properties of the sole that are at least partially affected by the polyurethane foam utilized.

[0010] Utilization of the property determination model may include, for example, accessing the property determination model, providing process data as an input to the property model, and receiving at least one material property as an output of the property model.

[0011] The characteristic determination model is a data-driven model suitable for determining material characteristics based on process data. In general, the characteristic determination model can refer to any function that can determine the material characteristics of the corresponding polyurethane foam produced based on the provided process parameters. However, the characteristic determination model can also be a more complex model, for example, it can be based on a known machine learning algorithm, for example, the characteristic determination model can use linear regression, neural network, polynomial, or for automated algebraic model optimization learning (ALAMO) regression algorithm as a machine learning algorithm. In addition, decision tree regression or random forest algorithm can also be used. The characteristic determination model has been derived from historical training data, for example, during the training process, and the historical training data includes a plurality of process data and the corresponding measured material characteristics of the previously produced polyurethane foam. In general, the data-driven characteristic determination model can be trained based on historical data, in particular, based on historical measurements of the material characteristics of the produced polyurethane foam associated with the corresponding process data of the production process of the corresponding polyurethane foam. This allows the characteristic determination model to be trained specifically for the corresponding production site and the specific settings of the corresponding production site, so that the characteristic determination model can achieve the characteristics of the foam produced by the corresponding production site to be determined very accurately.

[0012] In addition, the method includes generating a control action for controlling the production process of the polyurethane foam based on the determined material properties. In particular, a control action may refer to control data that can be used directly in a process automation system to automatically control the production process, for example, for setting one or more process variables. However, in some cases, a control action may also refer to a recommended action that can be provided to a user, in particular for situations where it is necessary to manually adjust a process variable, for example, if the corresponding process variable is not controlled by the process automation system, or if the adjustment can only be performed by manually adjusting one or more components of the production system. In addition, the generated control action may be generated in such a way that it results in direct control, such as directly interfering with the control process of the production process; or indirect control, such as before the control action is implemented, the control action may first be interpreted by the process automation system or may first be presented to the user for verification. In general, the control of a production process refers to adjusting or modifying one or more process variables of a production process, which results in a modification of the corresponding production process and, therefore, also in a change in the material properties of the produced polyurethane foam.

[0013] In particular, it is preferred to generate control actions such that the determined material property of the polyurethane foam meets a predetermined criterion. Thus, the generation of control actions can be based on predetermined rules, which indicate, for example, which process variables have to be adjusted for which case of the determined material property if the determined material property does not meet the respective predetermined criterion. In particular, these predetermined rules can be based on theoretical considerations, human experience, or can preferably refer to learned predetermined rules. In particular, such rules can be learned based on previous controls of the production process, for example based on previous measurements of the material property of the produced polyurethane foam associated with the corresponding control action.

[0014] In a preferred embodiment, the method further comprises: a) providing a target material property that the produced polyurethane foam must meet, b) comparing the determined material property with the target material property, and c) generating a control action based on the comparison result. In particular, the target material property may refer to a predetermined standard that the determined material property should meet. The target property may be any material property of the polyurethane foam, or may be derived from one or more material properties of the polyurethane foam. In particular, the target material property may be provided by a customer of the polyurethane foam based on the expected product in which the foam should be used. For example, the customer may provide the target material property via a communication network (such as the Internet) or using a local communication method. The measured material property generally refers to the same quantity as the determined material property. The comparison preferably includes: determining whether the determined material property or a property that can be derived from one or more determined material properties meets the target property within a predetermined limit. In particular, if the comparison results in a deviation between the determined material property and the target property, a control action is then generated, wherein the control action is then generated so that, for example, the deviation is minimized by using a predetermined rule. In a preferred embodiment, generating a control action based on the comparison includes using an optimization routine that is configured to minimize the deviation between the determined material properties and the target material properties for one or more process variables of the production process. Preferably, the optimization routine allows optimization not only for one process variable at a time, but also for more than one process variable (preferably, all process variables of the production process). In addition, it is preferred that the optimization routine allows more than one optimization target. In particular, an optimization target can refer to minimizing the deviation between the determined material properties and the target properties. However, additional optimization targets can also be used, such as minimizing the resources utilized during production, maximizing the output of the produced polyurethane foam, etc. Therefore, it is preferred that the optimization routine refers to a multivariate and multi-objective optimization routine, which allows optimization not only for one target, but also for more than one target. For example, the optimization routine can refer to an optimization objective function, wherein if more than one target is utilized, the optimization can refer to a Pareto optimization. In particular, during the iterative optimization algorithm, the optimization routine can utilize a characteristic determination model in order to predict material properties based on modified process data. Preferably, the optimization is subject to constraints, and the values ​​of the constraints are given by technical and safety limitations. Constrained optimization generally refers to adding corresponding conditions in the optimization algorithm, which should be met to a predetermined degree during the optimization. In particular, these technical and safety limitations can be predetermined based on technical specifications of the production site, manuals of one or more of the machines used in the production and / or specifications of materials used during the production. Constraints can be implemented in the objective function, for example, as penalty terms, or can be used to select between different solutions a solution that most closely meets the corresponding constraints.Suitable optimization techniques for solving constrained optimization problems are for example the method of Lagrange multipliers, the quadratic programming method or the interior point method etc. In general, known algorithms and methods for solving optimization problems, in particular using the above-mentioned methods, can be used.

[0015] In one embodiment, the method further comprises providing a measured material property of the produced polyurethane foam and comparing the measured material property with the determined material property, and generating a control signal causing a retraining of the property determination model if the determined material property is outside a predetermined limit around the measured material property. For example, the measured material property may be provided according to a regular product quality check, wherein one or more properties of the product are measured at predetermined time intervals (e.g., every hour, every day, or every few days). Thus, such a quality check may also check the accuracy of the utilized property determination model at regular intervals, and cause a retraining of the property determination model if the quality and the corresponding accuracy are not met. Furthermore, as part of a customer quality check, the material property may be measured on-site or off-site, for example after the customer has received the foam. The customer may then provide the results of the quality check (particularly the measured material property) to the device as user feedback, for example, using a communication network that enables the customer to communicate with the device. Thus, the production of the product may not only incorporate on-site data, but also incorporate data measured by subsequent customers of the product, so that the subsequent measured data is not lost and can be used to optimize the foam production. In particular, the property determination model may be retrained taking into account the measurements that lead to the retraining of the property determination model. Additionally, retraining can also utilize historical training data that was not previously used to train the characterization model. Thus, if new situations arise during the production of the polyurethane foam that may not have been represented in the previously utilized training data, resulting in an inaccurate characterization model, these situations can be taken into account in the retraining process, thereby improving the accuracy of the characterization model in an increasing number of situations over the long term.

[0016] In one embodiment, the method also includes, for example, providing precursor data via an input unit, the precursor data indicating the precursor properties of at least one polymer material utilized in the production process for producing polyurethane foam, and wherein the utilized property determination model is suitable for further determining material properties based on the precursor data. Precursor in this context refers to the polymer material utilized during the production of polyurethane foam. In particular, it is preferred that the precursor properties include at least one of the following items: the melting point of the polymer material, the polymer melt flow rate, the melt flow ratio, the molecular weight distribution, the density, the bulk density, the amount of free isocyanate (NCO), the tear strength, the elongation at break and the tensile strength. In general, if, for example, the quality of the polymer material utilized changes, this precursor property can change. The quality change of the polymer material can be based on the production source of the polymer material, but can also be caused by the storage or environmental conditions of the polymer material. In some cases, the precursor properties can have a strong impact on the results of the production process of polyurethane foam, so that considering the precursor data allows more accurate determination of material properties, especially in the case where the precursor properties are expected to change. Preferably, the precursor data are provided based on measurements of the respective polymer material utilized in a production process, but may also be derived, for example, based on production information about the production of the polymer material or based on environmental information about the storage environment of the polymer material.

[0017] In one embodiment, the method also includes providing environmental data, for example, via an input unit, which indicates the environmental conditions to which the polymer material utilized in the production process is exposed for a certain period of time, and the characteristic determination model utilized is suitable for further determining material characteristics based on the environmental data. Since, as already mentioned above, environmental conditions can affect the characteristics of the polymer material utilized, environmental conditions can also be directly used as inputs to the characteristic determination model to determine material characteristics accordingly. In general, environmental data can refer to any one of the amount of specific gases in the temperature, the humidity, gas exchange, etc. In a preferred embodiment, the humidity conditions to which the polymer material utilized in the production process is exposed. Humidity can have a strong impact on polymer materials, and therefore can also strongly affect the production process of polyurethane foam. For example, in a preferred production process, two precursors, i.e. polymer materials, are utilized, one of which is non-reactive, and the other contains NCO reactive groups. In particular, these groups are very sensitive to water, so that the humidity in the environment of the precursor can have a strong impact on the characteristics of the precursor, and therefore also has a strong impact on the foam production process. Therefore, in particular if such reactive precursors are utilized, it is preferred that the property determination model is configured to also take into account humidity for predicting the material properties.

[0018] In another aspect of the present invention, a training method for a parameterized property determination model is proposed, wherein the training method includes: a) receiving historical training data, the historical training data including a plurality of process data and corresponding measured material properties of previously produced polyurethane foams; b) using the received historical training data to parameterize the property determination model, so that the parameterized property determination model is suitable for determining the material properties of the polyurethane foam based on the process data; and c) outputting the parameterized property determination model.

[0019] In another aspect of the present invention, an apparatus for process automation of a production process is proposed, the production process being an extrusion foaming thermoplastic polyurethane process for producing polyurethane foam, wherein the apparatus comprises: i) an input interface unit configured to provide process data indicative of at least a portion of the production process for producing polyurethane foam being performed; ii) a processor configured to: a) determine material properties of the produced polyurethane foam using a property determination model, wherein the material properties are indicative of characteristics of the produced polyurethane foam, and wherein the property determination model is a data-driven model suitable for determining material properties based on process data; and b) generate control actions for controlling the production process of polyurethane foam based on the determined material properties; and iii) an output interface unit configured to provide the generated control actions to a distributed control system of a plant. Preferably, the generation of the control actions is based on solving an optimization problem, which minimizes the difference between one or more target material properties and the values ​​of the determined corresponding material properties for process variables of the production process (e.g. considering predefined ranges of these process variables as corresponding constraints).

[0020] In another aspect, a system for automatically producing polyurethane foam in an extrusion foaming thermoplastic polyurethane process is proposed, wherein the system includes: a) an apparatus as described above, and b) a production system configured to produce polyurethane foam in an extrusion foaming thermoplastic polyurethane process based on control actions received from the apparatus.

[0021] In another aspect, a training device for a parameterized property determination model is proposed, wherein the training device includes: a) a receiving unit configured to receive historical training data, the historical training data including a plurality of process data sets and corresponding measured material properties of previously produced polyurethane foams; b) a processor configured to parameterize at least one property determination model using the received historical training data, so that the parameterized property determination model is suitable for determining at least one material property of the polyurethane foam based on the process data; and c) an output unit configured to output the parameterized property determination model.

[0022] In another aspect, a computer program product for process automation of a production process is proposed, wherein the computer program product comprises program code means for causing an apparatus as described above to execute a method as described above.

[0023] In another aspect, a computer program product for parameterizing a property determination model is proposed, wherein the computer program product comprises program code means for causing a training device as described above to perform a training method as described above.

[0024] In another aspect, a control action generated according to the method as described above is proposed.

[0025] In another aspect, it is proposed to use the method as described above and / or the control actions generated by the method as described above in an extrusion foaming thermoplastic polyurethane process.

[0026] In another aspect, a computer-implemented method for process automation of an extrusion foaming thermoplastic polyurethane process is proposed, the method comprising at least a mixing step in an extruder and a decompression step, wherein the method comprises the following steps: a) providing process data of the extrusion step via an input unit, preferably relating to the ratio of blowing agent to polymer, the ratio of talc to polymer, the rotation speed of the extruder and / or the temperature and pressure distribution inside the extruder, b) providing precursor data of the polymer material to be processed via an input unit, preferably relating to melting point, polymer melt flow rate, molecular weight distribution, density and / or tensile strength, c) providing at least one predicted material property via a simulation module using a property determination model, preferably relating to foam bulk density, tensile strength and / or shape parameter, d) determining via a comparison module whether the predicted material property is within an expected threshold relative to physical and / or soft sensor measurements of the material property or within a predetermined range of one or more material properties, wherein if the comparison results in a condition not being met, the property determination model is retrained until the condition is met, i.e. the accuracy is obtained, and steps c) and d) are continued with the retrained model.

[0027] In another aspect, an apparatus for process automation of an extrusion foaming thermoplastic polyurethane process is provided, wherein the apparatus comprises: a) a simulation module, which utilizes a property determination model to determine at least one material property, wherein the at least one material property is preferably foam bulk density, tensile strength and / or shape factor; b) a training module, which uses historical data and has executable instructions for training and validating the property determination model for determining the material property, wherein these property determination models preferably refer to pure data models based on machine learning algorithms, or hybrid models based on a combination of data-driven models and first principles models; c) a comparison module, which is configured to determine whether the determined material property is within an expected region or within a range relative to the material property measured by a physical and / or soft sensor within a predetermined range of acceptance of the material property, wherein such comparison provides a validation check on the property determination model to trigger a model retraining process; d) an optimization module having executable instructions for determining optimal process variables within technically feasible limits, wherein the optimization module preferably reduces the difference of the product property from one or more target values ​​of the determined material property by simultaneously finding appropriate process variable values ​​preferably for extruder temperature and / or pressure profile, rotation speed, temperature and / or pressure in the underwater pelletizer, wherein the optimization module is configured to send the determined process variable values ​​to the control module; and e) a control module configured to control the current operating state of the production process based on the determined optimal conditions, in particular based on the determined temperature and temperature-pressure in the underwater pelletizer.

[0028] It shall be understood that the method as described above, the apparatus as described above, the computer program product as described above, and a system comprising an apparatus as described above have similar and / or identical preferred embodiments, in particular as defined in the dependent claims.

[0029] It shall be understood that a preferred embodiment of the invention can also be the dependent claims or any combination of the above-mentioned embodiments with corresponding dependent claims.

[0030] These and other aspects of the invention will be apparent from and elucidated with reference to the embodiments described hereinafter. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 A system for automatically producing polyurethane foam according to the present invention is schematically and exemplarily shown.

[0032] Figure 2 A flow chart of a method for automatically producing polyurethane foam according to the present invention is schematically and exemplarily shown,

[0033] Figure 3A training method for training a characteristic determination model that can be utilized in the present invention is schematically and exemplarily shown, and

[0034] Figure 4 A flow chart schematically and exemplarily shows a more detailed embodiment of the present invention,

[0035] Figure 5 A typical extrusion foaming thermoplastic polyurethane process is shown schematically and exemplarily. DETAILED DESCRIPTION

[0036] Figure 1 A system 100 for automatically producing polyurethane foam in an extrusion foaming thermoplastic polyurethane process is schematically and exemplarily shown. The system 100 includes a production system 120, which is configured to produce polyurethane foam in an extrusion foaming thermoplastic polyurethane process provided, for example, in a production plant. In addition, the system 100 includes an apparatus 110 for process automation of a production process for producing polyurethane foam 121. Optionally, the system 100 may also include a training device for training and / or retraining a characteristic determination model that can be used in the device 130.

[0037] The production system 120 may refer to any production system that allows the execution of an extrusion foaming thermoplastic polyurethane process to produce polyurethane foam. Examples of such processes are described in documents EP1266928A1, US5605937, and WO2013153190. In addition, Figure 5 A typical extrusion foaming thermoplastic polyurethane process is schematically and exemplarily shown. In a typical extrusion foaming thermoplastic polyurethane process, nucleating agents, chain extenders and other additives are fed into the extruder in appropriate proportions. A typical extruder configuration starts with a melting zone, followed by an alternating conveying zone and a mixing zone. Based on the requirements of the material, a blowing agent is injected into the extruder, where the blowing agent is mixed with the polymer material. This is a physical drive step, in which factors such as blowing agent type, temperature, pressure, energy input and residence time can have a strong impact on the foam produced. Optionally, after the mixing step, a granulator is provided, in which the polymer material is cut into small pieces, while being decompressed and cooled. The combination of all these factors determines the density, morphology and closed surface of the foam particles.

[0038] The device 110 includes an input interface unit 111, a processor 112 and an output interface unit 113. The input interface unit can be implemented in the form of any type of interface unit that can be docked with the device 110. In particular, the interface unit is configured to provide the processor with data that has been received from a source outside the device 110. In particular, the input interface unit is configured to receive and provide process data indicating at least a portion of a production process for producing polyurethane foam, such as performed as described above. In particular, the process data can be based on or include control data for controlling the production process of polyurethane foam, measurement data measured by a sensor during the production process of polyurethane foam, and / or virtual sensor data provided from virtual sensors, which derive sensor data from the measurement data of other sensors. Preferably, the process data includes at least one of the following items: the ratio between a blowing agent and a polymer, the ratio between talc and a polymer, the rotation speed of an extruder, the temperature distribution inside the extruder, and the pressure distribution inside the extruder. Preferably, the process data includes process information from a feed, an extruder, and optionally also from a granulator unit for foam production. For example, corresponding sensors can be provided that can measure one or more process parameters as part of the process data. In addition, it is preferred that precursor data characterizing the precursor properties are used as additional parameters. Preferably, the production system performing the extrusion foaming thermoplastic polyurethane process includes at least two process parameters that can be manipulated, i.e. controlled, to run a stable process and obtain stable foam properties, such as density or bead size. However, more than two or only one parameter can also be used to control the process and thus as a process variable.

[0039] In general, the process data received via the data stream 123 is provided to the processor 112. The processor 112 is then configured to determine one or more material properties of the produced polyurethane foam 121 using a property determination model. In particular, the determined material properties indicate the characteristics of the produced polyurethane foam. Preferably, the determined material properties refer to at least one of the foam bulk density, tensile strength and shape parameters of the produced polyurethane foam 121. The property determination model utilized is a data-driven model suitable for determining material properties based on process data. In general, the property determination model may refer to any parameter-driven model that can be parameterized based on historical data. However, preferably, the data-driven property determination model refers to a model based on machine learning, and may be based on any known machine learning algorithm, for example, may be based on a neural network, a decision tree algorithm, a random forest algorithm, a regression algorithm, etc. Preferred algorithms are linear regression and ALAMO constrained regression algorithms.

[0040] The data-driven characteristic determination model can then be trained using, for example, a training device 130. The training device 130 is suitable for training the characteristic determination model and includes a receiving unit 131, a processor 132 and an output unit 133. The receiving unit 131 can refer to a corresponding storage unit storing historical training data or data that can be used as historical training data or can be communicatively coupled to a corresponding storage unit and can be suitable for providing and optionally selecting corresponding historical training data. In addition, optionally, the receiving unit 131 can also directly receive measured material property data that has been directly measured from the produced polyurethane foam 121 via an optional data stream 124. Such measured material property data can then be used again as training data together with corresponding process data for training new characteristic determination models or for retraining already existing characteristic determination models, as will be described in more detail later. In general, the training data for training the characteristic determination model includes a plurality of data sets, wherein one of the data sets includes at least one process data value and a corresponding measured or otherwise derived material property of the produced polyurethane foam. In general, the training data can be selected so that different characteristic determination models are trained and then utilized in order to determine different material properties. However, the training data may also be selected so that the trained characteristic determination model is subsequently suitable for determining more than one material property. The processor 132 is then adapted to train the data-driven characteristic determination model. In particular, training is performed by providing historical training data as input to the trainable characteristic determination model until the trainable characteristic determination model is trained to determine the material property based on the corresponding provided process data. In general, the characteristic determination model can be trained using all known training methods for training corresponding trainable machine learning-based models. For example, the training algorithm can be selected based on the amount of process data or the presence of variables with integer values ​​in the trainable characteristic determination model. The preferred method used is based on Newton's method, in particular based on the gradient descent method. The gradient descent method is most suitable for trainable characteristic determination models with a large number of trainable parameters. Such a trained characteristic determination model can then be stored, for example, on a model database together with an identification indicating which polyurethane foam, which production system, which process data and which material property the model can be used for. The output unit 133 can then be configured to provide the corresponding trained model, for example, to a storage device or directly to the device 110. The model so trained is then used by the processor 112 to determine the material properties of the polyurethane foam 121 based on the provided process data.

[0041] In addition, the processor 112 is configured to generate a control action for controlling the production process 120 of the polyurethane foam based on the determined material properties. In particular, a predetermined or learned rule can be used to generate a control action based on the determined material properties. In particular, the processor can be configured to determine whether the determined material properties deviate from a predetermined standard, in particular from a predetermined target property of the polyurethane foam. In the event of a deviation, the processor can be configured to call, for example, an optimization routine, which is configured to optimize at least one process variable of the production system 120 to minimize the deviation, in particular so as to allow the production of a polyurethane foam that meets the target material properties within predetermined limits. The optimization routine can optimize at least one process variable of the determined properties of the polyurethane foam using, for example, a Lagrange multiplier method, a quadratic programming method, or an interior point method. In general, during optimization, known constraints can also be considered, for example, based on corresponding system specifications or information from historical constraints.

[0042] The control actions thus generated can then be provided by the output interface unit 113 to the production plant, for example via the data stream 122, and can be implemented into the production process 120. In general, the control actions can be implemented directly, for example via the production automation system managing the production process 120, or can first be provided to the user for verification, wherein the user can then decide whether to implement the corresponding control action. The polyurethane foam 121 is then produced according to the implemented production process. Optionally, as with regard to Figure 4 As will also be described in more detail, at regular time intervals, a measurement process can be performed on the produced polyurethane foam 121, for example in the form of a quality check, in order to measure or derive the corresponding material properties of the polyurethane foam 121. These measurement data of the material properties can be provided again together with the corresponding process data for producing the measured polyurethane foam to, for example, a training device 130 or device 110 for training a model. The processor 112 can be configured to compare the accuracy of a property determination model for determining the material properties of the polyurethane foam based on the measured material properties of the provided polyurethane foam 121. For example, if the determined material properties deviate from the measured material properties, in particular if the deviation is above a predetermined limit, the processor 112 can be configured to initialize a retraining process for retraining the property determination model. Retraining can then be performed again, for example, by the device 130 based on the received measured material property data that caused the property determination model to be retrained. More details on this optimization process will be referred to. Figure 4 Give a description.

[0043] Figure 2 Schematically and exemplarily, a flow chart of a method for process automation of a production process is shown, which production process relates to an extrusion foaming thermoplastic polyurethane process for producing polyurethane foam. In particular, method 200 can be as described in relation to Figure 1 The method 200 includes providing process data in step 210, for example, as described above with respect to the production process 120. In addition, the method 200 includes determining the material properties of the produced polyurethane foam using a property determination model in step 220. The property determination model can be, for example, as described with respect to Figure 1 The described device 130 is trained, in particular according to the following Figure 3 The described training method is trained. In a further step 230, the method 200 then comprises generating control actions for controlling the production process of the polyurethane foam based on the determined material properties.

[0044] Figure 3 A flow chart of a method 300 for training, in particular parameterizing, a property determination model that can be used by an apparatus or method as described above is schematically and exemplarily shown. In a first step 310, the method 300 comprises receiving historical training data, wherein the historical training data comprises a plurality of data sets. Each data set comprises at least one material property of a polyurethane foam associated with process data, i.e., values, measured values, etc. of one or more process variables associated with a production process of the polyurethane foam. In a next step 320, for example, based on information about Figure 1 The process described in the apparatus 130 in the embodiment of the present invention trains a material property determination model based on historical training data. In particular, any known training algorithm for training a corresponding data-driven model can be utilized. In a final step 330, such a trained property determination model is then provided to, for example, a storage unit for storage, or can be directly provided to, for example, the apparatus 110 for use in the description of the material property determination model. Figure 2 used in the described method.

[0045] Figure 4 An exemplary embodiment of an optimization algorithm that can be used in the process automation according to the present invention is shown schematically and exemplarily. Figure 4In the example shown, the production process is performed under the control of a controller using corresponding control data, for example indicating which values ​​the process parameters are to be set to. In general, the production process involves an extrusion foaming thermoplastic polyurethane process for producing polyurethane foam. One or more sensors, in particular soft sensors, measure one or more quantities associated with the production process. For example, in addition to the control data used, the measurement data thus determined can be provided to a storage device as process data. In addition, the sensor can also be used to measure the material properties of the produced polyurethane foam, for example during a quality inspection. In addition, the process data including the control data and / or the measurement data of the production process can be utilized by a characteristic determination model (i.e., a product characteristic model), which is configured to determine the material properties of the polyurethane foam, for example, according to the above-mentioned principles. The predicted material properties can then be compared with the measured material properties using the measured material properties provided by the sensor to determine whether the prediction of the characteristic determination model is within a predetermined range, in particular whether a predetermined accuracy is provided. If the characteristic determination does not provide the desired accuracy, that is, if the deviation between the determined material properties and the measured material properties is higher than a predetermined threshold value ( Figure 4 ), the characteristic determination model can be retrained using historical data, for example using more training data sets for training the characteristic determination model, and in particular also in combination with training data corresponding to situations where the characteristic determination model is inaccurate. If the comparison results in that the characteristic determination model is sufficiently accurate, i.e. if the answer is "yes", the optimizer uses the material characteristic determination model to optimize the production process. In particular, the determined material property is compared with the target material property by the optimizer, and if a deviation is determined, the optimizer uses an optimization routine to minimize the deviation until the set point (i.e. the target material property) is met. The control data determined in this way can then be provided to the controller in the form of control actions for implementation in the production process.

[0046] The execution of controlling the extrusion foaming thermoplastic polyurethane process can be realized by utilizing a process automation system, which executes an algorithm that can consider, for example, precursor characteristics, process variables, historical data and / or soft sensor information. In general, process variables can include at least one of the following items: the ratio of blowing agent to polymer, the rotation speed of the extruder, the residence time, the temperature distribution, the temperature at the template and the ratio of the decompression, and the temperature of the cooling bed immediately after the template. However, process variables can also refer to other parameters or components of the production process, for example, the parameters of the heat exchanger, pump, etc. used can be referred to. The value of the process variable is usually fixed during the commissioning of the corresponding production plant and is set according to the design specification. For example, this process is described in detail in US 9884947 B2, US10137601B2 and WO 2015123961 A1. Generally, the process variable value set can be periodically changed by manually reprogramming the process automation system, which may be necessary, for example, due to changes in the quality of the polymer material (i.e., the precursor used in the process). However, such processes often do not result in the quality of the foam produced meeting the desired results or lead to unnecessary waste production if the process variables are not adapted quickly enough when conditions of the production process change.

[0047] The present invention therefore relates in particular to being able to provide a more advanced type of process automation system, which can use historical data, such as information from physical and / or soft sensors, process variables, etc. associated with the corresponding material properties of the foam produced, to train a property determination model. The property determination model so trained can then be used to estimate appropriate process conditions, i.e. process variables, to achieve the desired product properties of the foam produced, for example to cope with changed precursor properties. The property determination model is typically trained until an accuracy criterion is met. The property determination model can then be used in a process that can estimate new values ​​of process variables involving flow, temperature, pressure, etc. to optimize product properties. In addition, the process automation method according to the present invention may also include a step that can learn from current process data and / or precursor data and adjust its operating point to optimize product properties.

[0048] In a preferred embodiment of the invention, the computer-implemented method for process automation as described above includes using process data (particularly corresponding to the extrusion step and corresponding to the amount of the extruder), precursor data (particularly the amount associated with the raw material) and optionally measured data of material properties to check the effectiveness of the currently used characteristic determination model online. In general, the foaming process consists of an extruder, in which one or more blowing agents, chain extenders and nucleating additives are mixed with the polymer material. After this step, the foaming polymer is granulated and decompressed to produce polymer foam. The foam properties at this point in the process can be measured to correlate with the characteristic determination model, which can associate, for example, precursor properties and / or process conditions with product properties. In a preferred embodiment, the characteristic determination model is based on a machine learning algorithm, which is a linear or quadratic regression. In particular, since these algorithms can reduce the computing resources used during the optimization step. In general, the characteristic determination model can first be trained offline based on historical data, and the process is automatically performed in a dedicated unit, which can also be used for future retraining of the model. Once the characteristic determination model is implemented, current and past process conditions, such as in the form of process data, can be obtained and can be used to predict the value of one or more target characteristics, such as product bulk density. In a preferred embodiment, a comparison between the determined current one or more characteristics and the predicted (i.e. determined) characteristics is performed, wherein such comparison can be regularly completed before the characteristic is predicted using the characteristic determination model, and therefore, if necessary, the characteristic determination model can be adjusted accordingly. If one or more models include accuracy within a preset range, their predictions can be compared with the characteristic set point (i.e. the target value of one or more material characteristics). If a difference is detected and it is based on the characteristic model, the optimizer can reduce the difference by modifying the process conditions (i.e. process variables) until the optimal conditions are preferably achieved within the constraints given to the process variables. These new optimal process conditions, i.e. process variables, can then be sent to the controller to perform the required process adjustments. The above process can then be repeated while the production process is running to provide optimal control of production, thereby producing foam with improved quality.

[0049] In the following, a specific example for implementing the above method and device is given with respect to controlling an underwater pelletizer using a closed-loop optimization device utilizing the above method. In particular, the underwater pelletizer is used to control the bulk density and size of the foam beads produced. In this case, it can be determined that for a particular foam, the target bulk density of the foaming material and the minimum size of the polymer beads should be achieved. Based on historical training data, a linear regression model, preferably a linear elastic net machine learning model, can be trained as a characteristic determination model that correlates the bulk density and the corresponding process data, and thus the bulk density of the foam can be determined based on the process data. As a basis for historical training data, for example, process data and corresponding bulk densities collected over a suitable long period of time (e.g., one or two years) for executing the corresponding production process can be utilized. However, if the measurement is, for example, more intensive, i.e., provides a higher data density, data collected over a shorter period of time (e.g., one month or one week) can also be utilized. In the specific implementation of the characteristic determination model, for example, the characteristic determination model realizes 16 parameters collected in this way, wherein 12 parameters in these parameters refer to process parameters provided by process data, and 4 parameters correspond to precursor parameters provided by precursor data. Then an optimization routine can be utilized, which uses an objective function to reach a target bulk density, i.e., a set point. In particular, the objective function can be created as a set point and the square difference of the bulk density determined by the characteristic determination model. In addition, constraints in the form of equations describing the minimum and maximum values ​​of one or more process variables (e.g., pressure at a given temperature, and / or polymer throughput) can be determined from the same historical training data and / or equipment specifications. Preferably, these constraints include feasible values ​​of process parameters (e.g., temperature, pressure, and polymer throughput). An optimization function can be created based on one or more parameters in these parameters, which can be manipulated during the production process and therefore can be used as process variables. Preferably, an optimization function is created to optimize the foam properties about at least two process variables.

[0050] The loop optimization of this example can then be started by entering a set point (i.e., target bulk density) in the optimization. The optimization program then compares the set point with the current bulk density determined by the characteristic determination model. Based on this comparison, the optimization routine can then search for process variable values ​​that can achieve the set point within the feasible region (i.e., satisfying constraints such as temperature, pressure, and throughput) using known optimization algorithms. The values ​​of these process parameters can then be continuously adjusted to keep the target bulk density as constant as possible and as equal to the set point as possible.

[0051] Other variations to the disclosed embodiments can be understood and effected by those skilled in the art in practicing the claimed invention, from a study of the drawings, the disclosure, and the appended claims.

[0052] For the processes and methods disclosed herein, the operations performed in the processes and methods may be implemented in different orders. In addition, the operations outlined are provided only as examples, and some of these operations may be optional, combined into fewer steps and operations, supplemented with further operations, or expanded into additional operations without departing from the essence of the disclosed embodiments.

[0053] In the claims, the word "comprising" does not exclude other elements or steps, and the indefinite article "a" or "an" does not exclude a plurality.

[0054] A single unit or device may fulfill the functions of several items recited in the claims. The mere fact that certain measures are recited in mutually different dependent claims does not indicate that a combination of these measures cannot be used to advantage.

[0055] The processes performed by one or several units or devices, such as providing process data, determining models using characteristics, generating control actions, etc., may be performed by any other number of units or devices. These processes may be implemented as program code components of a computer program and / or dedicated hardware.

[0056] The computer program product may be stored / distributed on a suitable medium, such as an optical storage medium or a solid-state medium, supplied together with or as part of other hardware, but may also be distributed in other forms, such as via the Internet or other wired or wireless telecommunication systems.

[0057] Any unit described herein may be a processing unit as part of a classical computing system. The processing unit may include a general-purpose processor and may also include a field programmable gate array (FPGA), an application-specific integrated circuit (ASIC), or any other dedicated circuit. Any memory may be a physical system memory, which may be volatile, non-volatile, or some combination of the two. The term "memory" may include any computer-readable storage medium, such as a non-volatile mass storage device. If the computing system is distributed, the processing and / or memory capabilities may also be distributed. The computing system may include multiple structures as "executable components". The term "executable component" is a structure that is well understood in the computing field as a structure that can be software, hardware, or a combination thereof. For example, when implemented in software, a person of ordinary skill in the art will understand that the structure of the executable component may include software objects, routines, methods, etc. that can be executed on the computing system. This may include executable components in the stack of the computing system or on a computer-readable storage medium. The structure of the executable component may exist on a computer-readable medium so that when interpreted by one or more processors of the computing system (e.g., by a processor thread), the computing system performs a function. Such a structure may be directly computer readable by a processor, for example, as in the case where an executable component is binary, or it may be structured to be, for example, interpretable and / or compilable, whether in a single stage or in multiple stages, so as to generate such binary that can be directly interpreted by a processor. In other instances, the structure may be a hard-coded or hard-wired logic gate that is implemented exclusively or nearly exclusively in hardware, such as in a field programmable gate array (FPGA), an application specific integrated circuit (ASIC), or any other dedicated circuit. Therefore, the term "executable component" is a term for a structure that is well understood by a person of ordinary skill in the field of computing, whether implemented in software, hardware, or a combination. Any embodiment of the present invention is described with reference to actions performed by one or more processing units of a computing system. If such actions are implemented in software, one or more processors direct the operation of the computing system in response to having executed the computer executable instructions constituting the executable component. The computing system may also include a communication channel that allows the computing system to communicate with other computing systems, such as through a network. "Network" is defined as one or more data links that enable electronic data to be transmitted between computing systems and / or modules and / or other electronic devices. When information is transferred or provided to a computing system over a network or another communications connection (e.g., hardwired, wireless, or a combination of hardwired or wireless), the computing system properly views the connection as a transmission medium. Transmission media may include networks and / or data links that may be used to carry desired program code components in the form of computer-executable instructions or data structures and which may be accessed by a general-purpose or special-purpose computing system or combination thereof.Although not all computing systems require a user interface, in some embodiments, a computing system includes a user interface system for interfacing with a user. The user interface serves as an input or output mechanism to the user, such as via a display.

[0058] Those skilled in the art will appreciate that at least parts of the present invention can be practiced in a network computing environment with many types of computing system configurations, including personal computers, desktop computers, laptop computers, message processors, handheld devices, multiprocessor systems, microprocessor-based or programmable consumer electronics, network PCs, minicomputers, mainframe computers, mobile phones, PDAs, pagers, routers, switches, data centers, wearable devices (such as glasses), etc. The present invention can also be practiced in a distributed system environment, where local and remote computing systems linked, for example, by hardwired data links, wireless data links, or by a combination of hardwired and wireless data links, all perform tasks over a network. In a distributed system environment, program modules can be located in both local and remote memory storage devices.

[0059] Those skilled in the art will also appreciate that at least a portion of the present invention can be practiced in a cloud computing environment. A cloud computing environment can be distributed, although this is not required. When distributed, a cloud computing environment can be distributed internationally within an organization and / or have components owned across multiple organizations. In this specification and the following claims, "cloud computing" is defined as a model for implementing on-demand network access to a shared pool of configurable computing resources (e.g., networks, servers, storage devices, applications, and services). The definition of "cloud computing" is not limited to any of the other numerous advantages that can be obtained from such models when deployed. The computing system of the accompanying drawings includes various components or functional blocks that can implement various embodiments disclosed herein as explained. Various components or functional blocks can be implemented on a local computing system, or can be implemented on a distributed computing system that includes elements residing in the cloud or implementing various aspects of cloud computing. Various components or functional blocks can be implemented as software, hardware, or a combination of software and hardware. The computing system shown in the figure may include more or fewer components than those shown in the figure, and some of the components can be combined when the environment guarantees.

[0060] Any reference signs in the claims should not be construed as limiting the scope.

[0061] The present invention relates to a method for process automation of a production process, the production process involving an extrusion foaming thermoplastic polyurethane process for producing polyurethane foam. The method comprises providing process data indicating at least a portion of the production process performed for producing polyurethane foam. In addition, a material property of the produced polyurethane foam is determined using a property determination model, wherein the material property indicates at least one feature of the produced polyurethane foam, and wherein the property determination model is a data-driven model suitable for determining the material property based on the process data. Based on the determined material property, a control action for controlling the production process of the polyurethane foam is generated. The method allows for improved automatic control of a production process involving an extrusion foaming thermoplastic polyurethane process.

Claims

1. A computer-implemented method (200) for process automation of a production process, the production process being an extrusion foaming thermoplastic polyurethane process for producing polyurethane foam, wherein the method (200) comprises: providing (210) process data indicative of at least a portion of the production process performed for producing the polyurethane foam, determining a material property of the produced polyurethane foam using (220) a property determination model, wherein the material property is indicative of at least one characteristic of the produced polyurethane foam, and wherein the property determination model is a data driven model adapted to determine the material property based on the process data, and A control action for controlling the production process of the polyurethane foam is generated (230) based on the determined material properties.

2. The method according to claim 1, wherein the method further comprises: Providing measured material properties of the produced polyurethane foam, comparing the measured material properties with the determined material properties, and generating a control signal causing retraining of the property determination model if the determined material properties are outside of predetermined limits around the measured material properties.

3. The method according to any one of claims 1 or 2, wherein the method further comprises: a) providing a target material property that the produced polyurethane foam must meet, b) comparing the determined material property with the target material property, and c) generating the control action based on the result of the comparison.

4. The method of claim 3, wherein generating the control action based on the comparison comprises: An optimization routine is used, the optimization routine being configured to minimize a deviation between the determined material property and the target material property for one or more process variables of the production process. The method according to claim 4 , wherein the optimization is subject to constraints, and the values ​​of the constraints are given by technical and safety limitations.

6. A method according to any one of the preceding claims, wherein the method further comprises providing precursor data indicating precursor properties of at least one polymer material utilized in the production process for producing the polyurethane foam, and wherein the property determination model utilized is suitable for further determining the material properties based on the precursor data.

7. The method of claim 6, wherein the precursor properties include at least one of the following: melting point of the polymer material, polymer melt flow rate, melt flow ratio, molecular weight distribution, density, bulk density, amount of free isocyanate (NCO), tear strength, elongation at break, and tensile strength.

8. A method according to any one of the preceding claims, wherein the method further comprises providing environmental data indicative of environmental conditions to which the polymer material utilized in the production process is exposed for a certain period of time, and wherein the property determination model utilized is suitable for further determining the material properties based on the environmental data.

9. The method of claim 8, wherein the environmental data is indicative of humidity conditions to which the polymer material utilized in the production process is exposed.

10. A training method (300) for a parameterized property determination model, wherein the training method (300) comprises: receiving (310) historical training data, the historical training data comprising a plurality of process data and corresponding measured material properties of previously produced polyurethane foams, parameterizing a property determination model using (320) the received historical training data, such that the parameterized property determination model is suitable for determining material properties of the polyurethane foam based on process data, and A parameterized property determination model is output (320).

11. An apparatus (110) for process automation of a production process, the production process relating to an extrusion foaming thermoplastic polyurethane process for producing polyurethane foam, wherein the apparatus (110) comprises: an input interface unit (111) configured to provide process data indicative of at least a part of the production process performed for producing the polyurethane foam, A processor (112), the processor being configured to: a) using a property determination model to determine the material properties of the produced polyurethane foam, wherein the material property is indicative of a characteristic of the produced polyurethane foam, and wherein the property determination model is a data driven model adapted to determine the material property based on the process data, and b) generating control actions for controlling the production process of the polyurethane foam based on the determined material properties, and An output interface unit (113) is configured to provide the generated control action to a distributed control system of the factory.

12. A system (100) for automatically producing polyurethane foam in an extrusion foaming thermoplastic polyurethane process, wherein the system include: The device (110) according to claim 11, and A production system (120) is configured to produce the polyurethane foam (121) in an extrusion foaming thermoplastic polyurethane process based on control actions received from the device.

13. A training device (130) for a parameterized characteristic determination model, wherein the training device (130) comprises: a receiving unit (131) configured to receive historical training data, the historical training data comprising a plurality of process data sets and corresponding measured material properties of previously produced polyurethane foams, one or more processors (132) configured to parameterize at least one property determination model using the received historical training data, such that the parameterized property determination model is suitable for determining at least one material property of the polyurethane foam based on the process data, and An output unit (133) is configured to output the parameterized characteristic determination model.

14. A computer program product for process automation of a production process, wherein the computer program product comprises program code means for causing the apparatus according to claim 11 to execute the method according to any one of claims 1 to 9.

15. A control action generated by the method according to any one of claims 1 to 9.

Images