Method and apparatus for determining the material properties of fuels or fuel components or for classifying fuels, as well as application of the method and use of the apparatus

Abstract

The invention relates to a method and a device for determining material properties ( S 4 ∗ , S 5 ∗ , S 6 ∗ ) of fuels (50) or fuel components or for classifying fuels, in particular aviation fuels or aviation fuel components, in which a sensor system (1) is used which provides comparison data by means of a software system (3) including existing database data containing material properties and acquires physical measurement data by means of a data acquisition device (2). ( S 1 P , S 2 P , S 3 P ) of a fuel or fuel component under investigation and the material properties ( S 4 ∗ , S 5 ∗ , S 6 ∗ ) or the classification by combining the comparison data and the measurement data ( S 1 P , S 2 P , S 3 P ) and determined by performing a model calculation. A flexible, real-time determination of fuel properties is made possible by the fact that the material properties ( S 4 ∗ , S 5 ∗ , S 6 ∗ ) or a classification of the fuel is determined during an ongoing manufacturing process or a continuous use process of the fuel, whereby the determination of the material properties ( S 4 ∗ , S 5 ∗ , S 6 ∗ ) or the classification in the software system (3) is based on at least one machine learning model.

Description

[0001] The invention relates to a method for determining the material properties of fuels or fuel components or for classifying fuels, in particular aviation fuels or aviation fuel components, in which a sensor system is used which provides comparative data by means of a software system, including existing database data containing material properties, and acquires physical measurement data of a fuel or fuel component to be investigated by means of a data acquisition device, and determines the material properties or classification by combining the comparative data and the physical measurement data and by performing a model calculation, as well as to a device for carrying out the method and furthermore to the application of the method and the use of the device.

[0002] A method and apparatus of this type are described in US Patent 8,781,757 B2. In this known method and sensor system, a class, a subgroup, and properties of fuel samples are determined spectroscopically using spectra measured in a selected spectral range and mathematical models developed from a database. Properties of fuels from at least two classes are compared with the measured spectra, with both a coarse and a fine analysis being performed. To enable rapid analysis without sample pretreatment, it is proposed to integrate the method into a fuel production or distribution plant. Spectral analysis of fuel samples in conjunction with correlation models constitutes a complex sensor system, making rapid, flexible fuel classification and determination of fuel properties in an ongoing process complex and difficult to achieve.

[0003] CN 116776285A also discloses a method and a device for determining fuel properties, using a multi-sensor system.

[0004] In an article by Bolf, N., G. Galinec, and M. Invandić, "Soft sensors for kerosene properties estimation and control in crude distillation unit," published in Chemical and Biochemical Engineering Quarterly 23.3 (2009): 277-286, soft sensors based on a neural network are described for determining kerosene properties. The distillation point of kerosene is determined based on temperature and flow rate measurements. Soft sensor models are developed using linear regression techniques and neural networks. The developed soft sensors are intended for online determination of kerosene properties; however, no flexible determination of various material properties or application for determining material properties after the production of the final product is mentioned.

[0005] Determining detailed material properties, such as aromatic content, energy content, and / or hydrogen content of aviation fuels, is essential in aviation practice, particularly for verifying compliance with existing fuel quality standards. This testing is necessary at several points in fuel production and logistics, as well as at the airport's fuel infrastructure. For example, testing is required directly after production of each batch at the refinery or during the blending of different batches, which occurs multiple times during fuel transport to the airport. The determination of fuel properties is typically carried out using standardized test procedures in a separate analytical laboratory, making it both time-consuming and costly.

[0006] Existing sensor systems for the direct determination of detailed fuel properties using various methods, such as those mentioned above, are often either unsuitable or difficult to integrate into an existing system for in-flow measurement (online characterization), or they are complex and expensive, which hinders their widespread use in fuel logistics and fuel infrastructure.

[0007] Determining fuel properties in an external analysis laboratory is time-consuming and expensive, and always involves a time lag compared to online characterization. This currently hinders innovative utilization strategies for sustainable synthetic aviation fuels (SAF), as the exact fuel properties are unknown during the refueling process at the aircraft, particularly due to mixing within the airport's refueling infrastructure. One such innovative utilization strategy could be, for example, determining the refueling quantity based on the energy content of the specific fuel instead of using average values.

[0008] Previously known machine learning models for determining fuel properties typically determine the desired fuel property depending on the detailed chemical composition of the fuel, such as by breaking down the mass fraction via chemical groups and chain lengths, and are therefore complex.

[0009] The present invention is based on the objective of providing a method for determining the material properties of fuels or fuel components, or for classifying fuels, and a corresponding device, which enables the rapid and flexible classification and determination of material properties of fuels or fuel components, in particular aviation fuels or aviation fuel components, with minimal effort. Furthermore, various application possibilities of the method and uses of the device are to be specified.

[0010] The method provides for the determination of the material properties or a classification of the fuel during an ongoing manufacturing or usage process of the fuel in flow, wherein the determination of the material properties or the classification is carried out in the software system based on at least one machine learning model.

[0011] According to claim 16, the device for carrying out the method comprises a software system and a data acquisition device, wherein - comparison data is recorded or can be recorded in the software system, which includes database data from a database assigned to, connected to, or contained in the software system, - physical measured quantities can be captured by means of the measurement acquisition device in an ongoing manufacturing or usage process of the fuel and supplied to the software system as physical measurement data obtained therefrom, - in the software system the comparison data and the physical measurement data can be combined and - the software system contains at least one machine learning model by means of which, based on the comparison data and the supplied physical measurement data, the determination of the material properties of the fuel or fuel components or the classification of the fuel can be carried out or is carried out.

[0012] A key component of the data acquisition system is a physical sensor. In addition, the data acquisition system includes, where necessary, controllable actuators or control elements, and may also utilize, in whole or in part, actuators already present in a peripheral device.

[0013] The device for determining the material properties of fuels or fuel components, or for classifying fuels, thus comprises a hybrid sensor system, where "hybrid" means a combination of physical sensors integrated into the data acquisition device and at least one machine learning model integrated into the software system. The data acquisition device uses the physical sensors to measure selected physical properties of the fuel or fuel components in a flow-through environment or in real time (online). These measurements are then available—e.g., pre-processed—as input variables for the at least one machine learning model and are used, or can be used, to determine further material properties. This enables flexible, online-based evaluation with relatively little effort.

[0014] Advantageous embodiments of the method are specified in claims 2 to 15.

[0015] An advantageous design for the rapid acquisition of the physical quantities of the fuel or fuel components and the provision of the physical measurement data consists in selecting physical quantities that can be measured quickly and easily, which in particular include at least one of the quantities density, viscosity, temperature, electrical conductivity.

[0016] For many applications, it is advantageous to determine at least one of the following material properties: aromatic content, energy content, and hydrogen content.

[0017] The measures contribute to a reliable determination of the material properties or classification and to flexible adaptation options for different fuels or fuel components by training (learning) at least one machine learning model on the basis of detailed material properties of the database data and, if necessary, retraining and updating it when the database data is expanded.

[0018] For the reliable and, if possible, accurate determination of the material property(ies) of a fuel or fuel component, it is also advantageous to create at least one machine learning model with a structure of such high flexibility that the material property to be determined can be ascertained with a given, and preferably high, reliability. This highly flexible structure also offers the advantage that the models can be easily extended to include further material properties to be determined, provided the corresponding training data for the model is available.

[0019] Furthermore, to determine the material properties of the fuel or fuel component or to classify the fuel, it is advantageously provided that, after sending the physical measurement data to the software system, at least one optimal machine learning model is selected for the respective material property or classification of the fuel based on a classification with database data from the database, and the determination of the material property or the classification of the fuel is carried out.

[0020] Flexible adaptation of the procedure with the highest possible reliability and / or accuracy in determining fuel properties or classifying fuel is facilitated by the software system for classifying fuel or determining material properties sending a control signal to the measurement acquisition device and / or to a peripheral device in order to change the measurement condition, which includes a measurement process and a state of the physical quantity to be measured, in order to provide adapted physical measurement data and combine it with the comparison data.

[0021] Advantageous measures consist of changing the measurement condition by altering a sensitivity range, a measurement duration, a measurement interval and / or an averaging time of the measured value acquisition and / or by changing the physical measured quantity, such as temperature, via an actuator or the peripheral device or an actuating element thereof.

[0022] Further advantageous embodiments of the method consist in the fact that, for the classification or characterization of a fuel, e.g. as conventional or SAF fuel, exactly two or at least two measurements are carried out at two or more temperatures specified by the software system, in particular with a specified temperature difference ΔT, and made available to the software system, in particular to the at least one machine learning model.

[0023] The reliability of determining the material properties of a fuel or its classification is also enhanced by measures whereby the software system, in accordance with a material property or classification of a fuel to be determined, uses software elements to check the quality and integrity of available database data and measurement data, and in particular also determines the compatibility of the physical measurement data, the database data and the at least one machine learning model.

[0024] Further advantageous measures in carrying out the procedure consist of preprocessing the physical measurement data in the software system and feeding the preprocessed data to at least one machine learning model.

[0025] Further advantageous measures for carrying out the procedure consist of a first process step in which, using database data depending on the current state of a database, required machine learning models are trained, and subsequently versioning and provision for the classification of the fuel and / or for the determination of the substance properties to be ascertained in a further subsequent process step.

[0026] For the execution of the procedure, it is further advantageously provided that, in order to determine the material properties, a preliminary decision is first made in at least one classification model regarding the parameters to be specified for at least one estimation model and / or the measurement acquisition device.

[0027] This facilitates a reliable determination of the material properties by determining the material property to be determined through a weighting of different estimation models.

[0028] For the process flow, it is further advantageously provided that the at least one machine learning model, in particular at least one estimation model, is selected by means of a control unit assigned to, in particular contained in, the software system, taking into account user input and / or - in the case of implementation according to claim 13 or 14 - on the basis of the preliminary decision by means of the at least one classification model.

[0029] An advantageous embodiment of the device for carrying out the method consists in the software system comprising, as machine learning models, at least one classification model and at least one estimation model, and also a control unit of an associated control device for controlling the machine learning models and the measurement acquisition device.

[0030] For a user-friendly design and execution of the method, a computer program product in which the method according to one of claims 1 to 15 is implemented is advantageously provided according to claim 18.

[0031] Various advantageous applications of the method or the use of the device exist in the airport sector, such as for monitoring fuel properties in a fuel infrastructure, in particular for identifying fuel properties, for using the determined material properties to adjust the refueling quantity, especially taking into account the energy content of the fuel, or for using the determined material properties for targeted refueling during flights through climate-sensitive regions, especially with fuel containing low aromatics.

[0032] In fuel logistics, the method or device can be advantageously applied for the continuous validation of the fuel with regard to conformity with existing standards, especially after mixing with one or more fuels of different material properties or classes, or for use in the identification of a product batch.

[0033] Further advantageous applications of the method or uses of the device arise in fuel production for validating an end product with regard to conformity with existing standards or for validating an end product at decentralized, remote production sites, and furthermore in fuel production for process monitoring within the production process.

[0034] The invention is explained in more detail below with reference to exemplary embodiments and the drawings. The drawings show: Fig. 1 An embodiment of a device for determining the material properties of a fuel or fuel components in a schematic representation, Fig. 2 An exemplary embodiment of operating possibilities of a device for determining material properties of fuels or fuel components or for classifying fuels and Fig. 3 an embodiment of the device for determining material properties of fuels or fuel components or a classification of fuels in more detail, in particular an architecture of machine learning models of the device.

[0035] Fig. Figure 1 shows an embodiment of a device for determining material properties of fuels 50 or fuel components or for classifying fuels 50, wherein the fuels 50 are in particular aviation fuels.

[0036] The device comprises as an essential component a hybrid sensor system 1 as a combination of physical sensors 20 located in a data acquisition device 2 and machine learning models 33 located in a software system 3, optionally with further software elements. In the illustrated embodiment, the data acquisition device 2 comprises, in addition to the physical sensors 20, at least parts of a peripheral device 21 and provides physical quantities acquired by the physical sensors 20 as physical measurement data. S1P, S2P, S3P The data are available and are transmitted to the software system 3 via a wireless or wired transmission link from a control unit 4. The control unit 4 with the transmission link is advantageously designed for bidirectional data transmission. The data acquisition unit 2 records selected material properties of the fuel 50 as physical quantities, particularly in flow or in real time (online), and is, for example, assigned to a fuel line 51 of a fuel infrastructure. The data acquisition unit 2 may also include, as necessary, controllable actuators and / or utilize actuators already present in a peripheral device 21 for measurement.

[0037] The software system 3 is also connected to a database 6 in order to receive database data, in particular concerning material properties, via a corresponding transmission path, or the database 6 can at least partially be created in the software system 3.

[0038] In software system 3, data is generated based on database data and the physical measurement data supplied by the data acquisition device 2. S1P, S2P, S3P using machine learning models 33 material properties S4⋆,S5⋆,S6⋆ determined and used as output data, e.g., together with the physical measurement data S1P, S2P, S3P issued.

[0039] The device 1 with the hybrid sensor system serves for the rapid online characterization of fuels 50 or fuel components, wherein the physical sensor 20 measures the selected material properties of the fuel 50 in the flow as physical measurands, which are then recorded as physical measurement data. S1P, S2P, S3P or input variables for the machine learning model 33 are used to determine the material properties to be determined. S4⋆,S5⋆,S6⋆ to determine. Physical parameters used include, for example, density, viscosity, temperature, and / or conductivity of the fuel under investigation 50, which are easily measurable physical quantities. From these, machine learning models 33 are used to determine, for example, aromatic content, energy content, and / or hydrogen content. Other combinations of the physical parameters and the machine learning models used 33 are also possible to determine other material properties. S4⋆,S5⋆,S6⋆ or Sx⋆ to determine.

[0040] When selecting combinations, particular consideration is given to the following: - the physical quantities measured in the flow can be measured with existing, integrable and cost-effective physical sensors 20; - the target variables determined by the machine learning models 33, in particular material properties and classification of the fuels, can be reliably determined on the basis of the physical quantities (“Property-Property Machine Learning Models”); - the target variables determined by the machine learning models 33 are difficult or impossible to determine in the flow using physical sensors 20 or analytics; - the machine learning models 33 are created (“trained”) on the basis of an existing database of detailed material properties and are retrained and updated (“retraining”) when the database is expanded in order to improve the accuracy of the machine learning models 33 and to qualify the machine learning models 33 for novel aviation fuels; - the underlying structure of the machine learning models (e.g. gradient boosting, neural network or the like) has a high degree of flexibility, which allows the machine learning models 33 to be selected and optimized for the respective material property; - the calculation of the machine learning models 33 is possible in principle both on hardware directly at the physical sensor technology (“edge computing”) and spatially separated by transferring the measurement data to a server (“cloud computing”).

[0041] The machine learning models 33 provided for the device or method according to the invention flexibly determine the fuel properties based on easily measurable physical quantities (bulk properties, in particular density, viscosity, temperature, conductivity, and the like). The design and training of such property-property machine learning models 33 are possible using a broad database of conventional and alternative aviation fuels, such as those already available or accessible to the inventors. Furthermore, the present concept advantageously includes a classification of the fuel or aviation fuel based on the database 6 in order to generate at least one optimal machine learning model 33 for the respective fuel and fuel property. S4⋆,S5⋆,S6⋆ or Sx⋆ to select. This is made possible by the present device with the flexibly designed hybrid sensor system 1, which - unlike previously commercially available systems - is not limited to conventional fuels according to ASTM D1655, but can also reliably determine alternative aviation fuels according to ASTM D7566.

[0042] The inventive combination of physical sensors 1 with flexible machine learning models 33 (in contrast to complex sensors combined with chemometric models) offers maximum flexibility in the design of the sensor system for the respective application. Depending on the determining parameter, physical sensors can be integrated into the system and the resulting physical measurement data can be used as input for the machine learning models 33. The machine learning models 33 can be continuously updated and improved based on the measurements taken by the sensor system, thus functioning as a self-learning system. This is advantageously achieved through the inventive combination of the physical sensors 20 and the machine learning models 33.

[0043] One with a Fig. The method executable according to the schematically depicted device 1 is based on the integration of the physical sensors 20 into the fuel line 51 through which the fuel 50 flows, for example at an airport. The physical measurement data S1P, S2P, S3P (e.g., density, viscosity, temperature, conductivity) are sent to the connected software system 3, which contains the machine learning models 33 and provides an interface to the database 6 of already known aviation fuels. Based on a classification using database data from database 6, the optimal machine learning model 33 is selected for the respective material property to be determined. S4⋆,S5⋆,S6⋆ for example, energy content, aromatic content, hydrogen content, or classification of the aviation fuel, are selected and the target parameters, especially material properties, are chosen. S4⋆,S5⋆,S6⋆ or classification of aviation fuel, are carried out using at least one machine learning model 33 and the physical measurement data S1P, S2P, S3P determined. As a result, the hybrid sensor system 1 or the software system 3 presents the physical measurement data. S1P, S2P, S3P and the specific material properties S4⋆,S5⋆,S6⋆ and, if necessary, also the classification of the aviation fuel.

[0044] Fig. Figure 2 shows, as an exemplary embodiment, a more detailed configuration of the device, wherein the hybrid sensor system 1 with the measurement acquisition device 2 is connected to a fuel system 5, for example a fuel line 51, in order to detect a physical measurement parameter of the fuel 50. This configuration of the device can be advantageously used when an adjustment of the physical sensor system 20 or of the settings of the peripheral device 21 is required, for example for one or more of the target parameters to be determined, such as material properties. S4⋆,S5⋆,S6⋆, Should be required or expedient. In this case, the hybrid sensor system 1 or the software system 3 can send a control signal to the measurement acquisition device 2, in particular the physical sensor 20, wherein the hybrid sensor system 1 or the software system 3 functions as an active control element in the device.

[0045] According to the in Fig. In the embodiment of the device shown in Figure 2, the measurement acquisition device 2 for acquiring at least one physical measurement parameter of the fuel 50 is connected to the fuel system 5, for example the fuel line 51, at an upstream and at a downstream branch point of the fuel system 5 or the fuel line 51. In a parallel branch to the section of the fuel line 51 located between the two branch points, a main branch of the measurement acquisition device 2 is located between an upstream first valve unit 211.0 and a downstream second valve unit 211.1. The main branch comprises components arranged one after the other in the direction of flow, namely a filter 212, a third valve unit 211.2, a heat exchanger 210, and a fourth valve unit 211.3 and a throttle 213 as well as with sensor elements of the physical sensor technology 20, connected to capture the desired physical measured quantities and the physical measurement data. S1P, S2P, S3P to obtain. The heat exchanger 210 can be bypassed by means of a parallel bypass. An orifice 216 is arranged in the fuel line 51 to influence the flow rate. Furthermore, for fuel recirculation, a return branch 215 with a pump 214 arranged therein is connected with its inlet to the second valve unit 211.1 and its outlet to the first valve unit 211.0.

[0046] With the in Fig. In the embodiment shown in Figure 2, advantageous operating options are provided for the aforementioned adaptation of the physical sensor system 20 or for settings of the peripheral device 21 in order to obtain the target variables to be determined with high reliability and accuracy. Such adaptations may be necessary or expedient, for example, if - for the machine learning models 33, a certain setting of the physical sensor 20 or the peripheral device 21 should be available in order to provide, for example, the sensitivity range, the measurement duration, an averaging time or the like by means of the measurement acquisition device 2; - for the machine learning models 33 physical quantities or physical measurement data S1P, S2P, S3P under certain boundary conditions, e.g., a specific temperature, the desired sample temperature should be available. This can be achieved through a temperature control option within the physical sensor, e.g., a controllable heat exchanger 210. Alternatively, the hybrid sensor system 1 could achieve a desired sample temperature by controlling the peripheral device 21; - For the characterization of a fuel (e.g., SAF versus a conventional fuel) or for determining the composition of a fuel, two measurements at two different predefined temperatures with the largest possible temperature difference ΔT should be available to the machine learning models 33. This could be achieved by controlling the heat exchanger 210 and corresponding valve units to reach the desired temperatures. This regulation can also take place in a flow configuration by performing two measurements, e.g., one with heat exchanger 210 and one with its bypass; - To increase accuracy, measurements at multiple temperatures should be available to the machine learning models 33. For this purpose, the hybrid sensor system 1 or the software system 3 could simultaneously control the relevant valve units, the pump, and the heat exchanger to recirculate a fuel sample and obtain temperature profiles.

[0047] A simple operating option arises, for example, when the hybrid sensor system 1 is used for a rapid determination of a target quantity or material property. S4⋆,S5⋆,S6⋆ If conventional fuel is used with an accuracy or relative error of X%, then a control signal "NOT" is sent to the hardware of the data acquisition device 2, because the machine learning models 33 implicitly perform the necessary conversion or because preselected sensor settings are sufficient.

[0048] Fig. Figure 3 shows an embodiment of the device with an exemplary architecture of in Fig. The machine learning models 33 shown in Figure 1 comprise estimation models 300, 301, 302 and classification models 303, 304. A model training facility 30 connected to the database 6 contains a statistical evaluation unit 305, for example for performing an analysis of variance, classification models 303, 304 and estimation models 300, 301, 302 to be trained. A model provision 31, connected to the model training facility 30 for versioning machine learning models, contains a matching unit for matching to the variance in the database 6, a model weighting unit 311, a control unit 313, and classification models 303, 304 and estimation models 300, 301, 302. The comparison process involves identifying measured variables (input parameters) that deviate significantly from the experience range recorded in the database ("outlier detection").The measured quantity is thus compared with the value range and variance in the database. The control unit 313 is in data transmission communication with a user input device 312 for entering user settings and with the data acquisition device 2 for controlling the physical sensors and / or the peripheral device 21. The model provider 31 is also connected on the input side to a unit for data checking 320 with regard to quality and integrity, as well as to a unit for data preprocessing 321. On the output side, the target quantities to be determined or output, along with the material properties, are available at the software system 3, or, in the illustrated embodiment, at the model provider 31. Sx⋆, Information on the uncertainties σ x , of the model for a material property Sx⋆, Information on model versioning and further details on model and sensor parameters are available.

[0049] The in Fig. The three exemplary devices shown, utilizing the architecture of the machine learning models, offer diverse classification and calculation possibilities. Essentially, the architecture of the machine learning models employs 33 schemes and software models that allow for a variety of approaches, depending on the available computing capacity and the target variables to be determined, such as material properties. Sx⋆ or the classification, both to scale (provide supplementary information) and to simplify. The respective software elements do not have to run on a single hardware or computing unit, but can, for example, run on different distributed computing units. The software elements can include the following: - Verification of data quality and integrity: these processes also ensure the compatibility of the physical measured quantities and physical measurement data. S1P, S2P, S3P with the database or with the database data as well as the machine learning models 33 fixed; - Preparation or preprocessing of the data: this includes functions (e.g. for unit conversion) or physical models and relationships, such as those relating to physical measurements. SXP scale to a desired temperature or pressure; - Training and deployment of the various machine learning models: this includes training, versioning and deployment of the machine learning models depending on the current state of database 6; - Different machine learning models: depending on the quantity to be determined, different machine learning models or combinations thereof can be applied. The example implementation according to Fig.Figure 3 shows a case in which a classification model is first applied to make a preliminary decision regarding the parameters of the estimation models 300, 301, 302 and the physical sensors 20 and / or the peripheral device 21, as well as their settings. The target variable to be determined, such as the material properties, is then used. Sx⋆, can then be obtained by weighting the estimation models 300, 301, 302; - Control of the machine learning models, in particular estimation models 300, 301, 302, and the hardware, in particular the data acquisition device 2: These software elements can use user input as well as the preliminary decision of classification models 303, 304 for control.

[0050] The presented device with the hybrid sensor system 1 and the associated method offer diverse application possibilities. At an airport, for example, fuel properties can be monitored within a complex infrastructure, and anomalies in fuel properties, such as those resulting from contamination or product degradation, can be identified. The data acquired by the hybrid sensor system 1 can be used to adjust the refueling quantity to the actual energy content of the fuel instead of relying on average values, thus enabling fuel savings.

[0051] The data obtained using the hybrid sensor system 1 can be used for targeted refueling of flights through climate-sensitive regions (contrail formation) with fuel with low aromatic content.

[0052] In fuel logistics, continuous validation of the fuel or a fuel component can be performed with regard to conformity with existing standards, especially after blending with other fuels (SAF or conventional kerosene). The hybrid sensor system 1 can be used for marking ("fingerprinting") a product batch.

[0053] In fuel production, the final product or the manufactured fuel can be validated for conformity with existing standards (e.g., ASTM D1655, ASTM D7566). Remote validation of the final product at decentralized, smaller production sites is also possible. Furthermore, in fuel production, the material properties of intermediate products can be monitored within the production process. For example, individual components in the intermediate product can be identified to replace physical sampling. QUOTES INCLUDED IN THE DESCRIPTION

[0000] This list of documents cited by the applicant was automatically generated and is included solely for the reader's convenience. The list is not part of the German patent or utility model application. The DPMA accepts no liability for any errors or omissions. Cited patent literature

[0000] US 8 781 757 B2

[0002] CN 116776285A

[0003] Cited non-patent literature

[0000] Bolf, N., G. Galinec, and M. Invandić “Soft sensors for kerosene properties estimation and control in crude distillation unit” Chemical and Biochemical Engineering Quarterly 23.3 (2009): 277-286

[0004]

Claims

[1] Methods for determining material properties (S4⋆,S5⋆,S6⋆) of fuels (50) or fuel components or for classifying fuels, in particular aviation fuels or aviation fuel components, in which a sensor system (1) is used which provides comparison data by means of a software system (3) including existing database data containing material properties and acquires physical measurement data by means of a data acquisition device (2). (S1P,S2P,S3P) of a fuel (50) or fuel component under investigation and the material properties (S4⋆,S5⋆,S6⋆) or the classification by combining the comparison data and the physical measurement data (S1P,S2P,S3P) and determined by performing a model calculation, characterized by that the material properties (S4⋆,S5⋆,S6⋆) of the fuel (50) or fuel component or the classification of the fuel (50) during an ongoing manufacturing process or use process of the fuel (50) as it flows in real time, whereby the determination of the substance properties (S4⋆,S5⋆,S6⋆) or the classification in the software system (3) is based on at least one machine learning model (33). [2] Method according to claim 1, characterized by , that as physical measurement data (S1P,S2P,S3P) Quickly and easily measurable physical quantities are selected, which in particular include at least one of the following quantities: density, viscosity, temperature, electrical conductivity. [3] Method according to claim 1 or 2, characterized by that as material properties (S4⋆,S5⋆,S6⋆) At least one of the properties aromatic content, energy content and hydrogen content must be determined. [4] Method according to any one of the preceding claims, characterized by , that the at least one machine learning model (33) is trained on the basis of detailed material properties of the database data and, if necessary, is retrained and updated when the database data is expanded. [5] Method according to any one of the preceding claims, characterized by that at least one machine learning model is created with a structure of such high flexibility that the material property to be determined can be determined (S4⋆,S5⋆,S6⋆) can be determined with a predetermined, preferably high, reliability. [6] Method according to any one of the preceding claims, characterized by that after sending the physical measurement data (S1P,S2P,S3P) to the software system (3) based on a classification with database data from the database (6) the at least one optimal machine learning model (33) for the material property to be determined in each case (S4⋆,S5⋆,S6⋆) or the classification of the fuel (50) selected and the determination of the substance property (S4⋆,S5⋆,S6⋆) or the classification of the fuel (50) is carried out. [7] Method according to any one of the preceding claims, characterized by , that the software system (3) for classifying the fuel (50) or for determining the substance property (S4⋆,S5⋆,S6⋆) sends a control signal to the data acquisition device (2) and / or a peripheral device (21) to provide adapted physical measurement data for the measurement condition, which includes a measurement process and a state of the physical quantity to be measured. (S1P,S2P,S3P) and merge with the comparison data to change. [8] Method according to claim 7, characterized by, that to change the measurement condition a change in a sensitivity range, a measurement duration, a measurement interval and / or an averaging time of the measurement data acquisition and / or a change in the physical measured quantity, such as temperature, is made via an actuator or the peripheral device (21). [9] Method according to claim 7 or 8, characterized by , that to classify or characterize a fuel (50), e.g. as a conventional or SAF fuel, exactly two or at least two measurements are carried out at two or more temperatures specified by the software system (3), in particular with a specified temperature difference ΔT, and made available to the software system (3), in particular to the at least one machine learning model (33). [10] Method according to any one of the preceding claims, characterized by , that the software system (3) in coordination with a material property to be determined (S4⋆,S5⋆,S6⋆) or classification of a fuel (50) using software elements a verification of the quality and integrity of available database data and physical measurement data (S1P,S2P,S3P) undertakes and, in particular, ensures the compatibility of the physical measurement data (S1P,S2P,S3P) the database data and the at least one machine learning model (33). [11] Method according to any one of the preceding claims, characterized by , that in the software system (3) a preprocessing of the physical measurement data (S1P,S2P,S3P) is carried out and the pre-processed data are fed to at least one machine learning model (33). [12] Method according to any one of the preceding claims, characterized by, that in a first process step using database data depending on a current state of a database (6) required machine learning models are trained (33) and subsequently versioned and made available for the classification of the fuel (50) and / or the determination of the substance properties to be determined is carried out in a subsequent process step (S4⋆,S5⋆,S6⋆) This has been done. [13] Method according to any one of the preceding claims, characterized by that to determine the material properties (S4⋆,S5⋆,S6⋆) first, a preliminary decision is made in at least one classification model (303, 304) regarding parameters to be specified for at least one estimation model (300, 301, 302) and / or the measurement acquisition device (2). [14] Method according to claim 13, characterized by , that the material property to be determined (S4⋆,S5⋆,S6⋆) is determined by weighting different estimation models (300, 301, 302). [15] Method according to any one of the preceding claims, characterized by , that the at least one machine learning model (33), in particular at least one estimation model (300, 301, 302), is selected by means of a control unit (313) assigned to, in particular contained in, the software system (1) by means of user inputs and / or - in the case of implementation according to claim 13 or 14 - on the basis of the preliminary decision by means of the at least one classification model (303, 304). [16] Device for determining material properties (S4⋆,S5⋆,S6⋆) of fuels (50) or fuel components or for classifying fuels (50), in particular aviation fuels or aviation fuel components, for carrying out the method according to one of the preceding claims, comprising a software system (3) and a measurement acquisition device (2), wherein - comparison data are recorded or can be recorded in the software system (3), which includes database data from a database (6) assigned to, connected to, or contained in the software system (3), - physical measurement quantities can be captured by means of the measurement acquisition device (2) in an ongoing manufacturing or usage process of the fuel (50) and made available to the software system (3) as physical measurement data obtained therefrom (S1P,S2P,S3P) are feedable - in the software system (3) the comparison data and the physical measurement data (S1P,S2P,S3P) can be combined and - in the software system (3) at least one machine learning model (33) is available, by means of which, on the basis of the comparison data and the supplied physical measurement data, (S1P,S2P,S3P) the determination of material properties (S4*,S5*,S6*) of the fuel (50) or fuel components or the classification of the fuel (50) is feasible or is carried out. [17] Device according to claim 16, characterized by , that the software system (3) comprises, as machine learning models (33), at least one classification model (303, 304) and at least one estimation model (300, 301, 302), and also a control unit (313) of an associated control device (4) for controlling the machine learning models (33) and the measurement acquisition device (2). [18] Computer program product in which the method according to any one of claims 1 to 15 is implemented. [19] Application of the method according to any one of claims 1 to 15 or the device according to claim 16 or 17 in an airport area for - Monitoring of fuel properties in a fuel infrastructure, in particular for the identification of fuel properties, - Use of the determined material properties (S4*,S5*,S6*) for an adjustment of the refueling quantity, in particular taking into account the energy content of the fuel (50), or - Use of the determined material properties (S4*,S5*,S6*9) for targeted refueling during flights through climate-sensitive regions, especially with fuel (50) with low aromatic content. [20] Application of the method according to one of claims 1 to 15 or the device according to claim 16 or 17 in fuel logistics for - continuous validation of the fuel (50) with regard to conformity with existing standards, in particular after mixing with one or more fuels of other material properties or classes, or - Used for labeling a batch of fuel product. [21] Application of the method according to any one of claims 1 to 15 or the device according to claim 16 or 17 in fuel production for validating an end product with regard to conformity with existing standards or for validating an end product at decentralized, remote production sites. [22] Application of the method according to any one of claims 1 to 15 or the device according to claim 16 or 17 in fuel production for process monitoring within the production process.

Images