Method and automated assembly for analyzing material composition of test object

Abstract

The invention relates to a method for evaluating material components of a test object (40), in which an impedance spectrum (ImS) based on the amplitude and / or frequency (f) of an electrical test signal (PS) is obtained from the test object (40) by means of an impedance measuring unit (5) using the electrical test signal (PS) via a sensor (6), different material components (Mi) are stored in a distribution table (9), -providing a respective sample (Pi) for each material component (Mi),-using the respective sample (Pi) for the test object (40) and using an impedance measuring unit (5) to obtain a predeterminable number (N) of test spectra (TSik) from the respective sample (Pi) in a test phase (EP) and storing the number of test spectra (TSik) as material components (Mi) of a specific training data set (TDS) in a sample detection unit (10), in a learning phase (LP), a specific neural network (NNi) is trained for the number (N) of test spectra (TSik) of the respective material composition (Mi), and in an analysis phase (AP), a currently obtained impedance spectrum (ImS) of the test object (40) is analyzed using the material composition (Mi) of the specific neural network (NNi). And outputting the best matching material component (Mi) for the impedance spectrum (ImS) to be analyzed to the corresponding material component (Mi) of the particular neural network (NNi) as a result.

Description

Technical Field

[0001] The invention relates to a method for analyzing the material composition of a test object, in which method an impedance measuring unit uses an electrical test signal via a sensor to obtain an impedance spectrum from the test object that is dependent on the amplitude and / or frequency of the test signal.

[0002] The invention further relates to an automation component comprising an impedance measuring unit designed to obtain, using an electrical test signal via a sensor, an impedance spectrum from a test object that is dependent on the amplitude and / or frequency of the test signal. Background Art

[0003] By measuring the impedance of a test object at different frequencies, its properties can be revealed. This measurement method is called impedance spectroscopy.

[0004] For impedance measurement, suitable measuring devices already exist. These allow impedance measurements to be made at different frequencies. The resulting complex impedance values ​​can then be plotted against frequency, i.e. as an impedance spectrum. The impedance spectrum is usually determined with the aid of special sensors that are able to test different test objects or materials.

[0005] WO 2018 / 203055 A1 describes a method for determining the electrical conductivity of a material by means of impedance spectroscopy.

[0006] EP 0 782 692 B1 shows a special operation of a measuring coil with multiple frequency components simultaneously. Unlike electrical components, in which the impedance is mainly described by physical relations and can be calculated, there are also impedance curves that cannot be easily assigned to physical laws. Summary of the invention

[0007] The object of the invention is to provide, in particular for production automation technology, a method for analyzing the material composition of test objects, ie for test objects whose impedance curve cannot be verified by physical laws (eg by calculation with formulas).

[0008] The object is achieved by a method for analyzing the material composition of a test object, the method comprising: an impedance measuring unit using an electrical test signal to obtain an impedance spectrum from the test object via a sensor, which is dependent on the amplitude and / or frequency of the test signal, wherein different material compositions are stored in an assignment table, wherein a sample is provided for each material composition, wherein in a detection phase, the corresponding sample is used for the test object and a predetermined number of test spectra are obtained from the corresponding sample by means of the impedance measuring unit, and the number of test spectra is stored in a sample detection unit as a training data set specific to the material composition, and in a subsequent learning phase, the number of test spectra for the corresponding material composition is obtained. The spectrum trains a neural network for a specific material composition, and in a later analysis phase that is decisive for the manufacturing process, the currently obtained impedance spectrum of the test object is analyzed using the neural network dedicated to the material composition, and the material composition is output as a result, and the material composition is used for the impedance spectrum to be analyzed. It is proved that the best match with the corresponding neural network dedicated to the material composition is obtained, wherein the impedance spectrum is specified as a complex-valued frequency curve, and a separate neural network dedicated to the material composition is trained for the absolute value, phase angle, real part and imaginary part, which works in parallel in the analysis phase for comparison with the currently obtained impedance spectrum in order to improve the accuracy of the statement about the material composition to be determined.

[0009] In the sense of the present invention, a neural network, i.e. an artificial neural network, maps a neuronal structure with learned knowledge. In the field of artificial intelligence, there are database knowledge discovery methods, which are also called data mining. From this data mining, a general requirement is derived to discover unknown relationships from usually very large data sets. In this case, the algorithms used are run differently, i.e. without presetting a clear desired result, for analyzing the database. In particular, the materials used in automated manufacturing are subject to production-related or batch-related fluctuations.

[0010] The invention provides the advantage of analyzing test objects in which the relationship between object properties and the impedance spectrum cannot be described analytically by physical laws.

[0011] In a further development of the method, a target material composition of the test object is stored in an assignment table and known deviations in composition from the target material composition are selected for other material compositions, wherein, based on the material composition determined in the analysis phase, the current deviation of the material composition from the target material composition is determined with the aid of the target material composition.

[0012] The inventors believe that, in contrast to the known methods of impedance spectroscopy, it is now possible to combine an evaluation device based on artificial intelligence methods with an automated device to provide material analysis in manufacturing technology, especially for materials that cannot be described in physical or chemical form. The resulting advantage is that a particular impedance curve can be assigned to a specific characteristic of a test object during a training process. In a continuous manufacturing process, the classification of the characteristics can then be carried out in real time. Due to this classification, KMO testing and / or monitoring of the quality characteristics of the test object is feasible.

[0013] In cooperation with an automated device, a control or regulation process is then possible. For this purpose, the method according to the invention is provided in a refinement for use in a manufacturing process of a test object, making it possible to intervene in the manufacturing process using the knowledge of the material composition deviation so that the test object again has a target material composition.

[0014] A further improved method step provides that the detection phase is performed on the local automation component and a training data set specific to the material composition is sent to a superordinate computing system, wherein the learning phase is performed on the superordinate computing system and a learned or calculated neural network specific to the material composition is sent from the superordinate computing system back to the local automation component and stored in the automation component for the analysis phase.

[0015] According to the method, the automation device can then be constructed modularly and can even be designed as a mobile structure. This mobile structure can then be used specifically for teaching artificial intelligence applications. The data set can then be fed into a cloud service, for example, and the trained neural network is sent back as a result. The trained neural network is then then stored on the modular automation device and can then be integrated into the system located in the process. It is also conceivable to download the neural network during system operation.

[0016] When evaluating the impedance spectrum in the form of classification by means of a neural network, the material composition can be calculated in these individual representation forms, namely absolute value, phase, real part and imaginary part, even with separate networks working in parallel. The properties of the material to be examined can have different effects on these individual representation forms (absolute value, phase, real part and imaginary part), and therefore a more accurate hit rate for classification in the neural network can be achieved.

[0017] For an automation component, comprising an impedance measuring unit, which uses an electrical test signal via a sensor to obtain an impedance spectrum from a test object that depends on the amplitude and / or frequency of the test signal, wherein a processor module is designed to process a neural network and a data storage device in which the neural network is stored, the above-mentioned purpose is achieved by means of an assignment table in which different material components are stored, and a sample detection unit, which is designed to obtain a preset number of test spectra for the material components specified in the assignment table by means of the impedance measuring unit during a detection phase, and further a data manager, which is designed to store and manage the number of test spectra as training data sets specific to the material components in the sample detection unit, an interface is designed to establish a connection to a learning tool, wherein the learning tool is designed to provide a neural network specific to the material component for each corresponding training data set during the learning phase, and the data storage device is designed so that a specific neural network is stored in the data storage device, wherein the processor module is designed to enable the currently obtained impedance spectrum to be analyzed with the specific neural network during the analysis phase, and the material component is output as a result, and the material component for which the impedance spectrum to be analyzed is proved to be the best match for the corresponding neural network specific to the material component.

[0018] The automation component is improved thereby, storing a target material composition of the test object in an allocation table, and storing other material compositions having known deviations in composition from the target material composition, wherein a computing device is provided, which is designed to derive the deviation of the material composition based on the material composition determined in the analysis phase with the aid of the target material composition.

[0019] Due to the training process, the specific characteristics of the test object are assigned to a specific impedance curve in a subsequent learning phase. In an ongoing process, the characteristics can then be classified in real time. This classification is possible to control / monitor the quality characteristics of the test object. In conjunction with automated equipment, there is also the possibility of controlling or regulating the manufacturing process.

[0020] For this purpose, the automation component is designed for use in a manufacturing process for the test object and has a reporting unit which is again designed to inform a regulating unit and / or a control unit of deviations in the material composition, which in turn is designed to intervene in the manufacturing process so that the test object to be manufactured has a target material composition again.

[0021] In order to outsource the training data set or to process it, for example, in a cloud service, the automation component is designed with a data transmission unit which is designed to send the training data set specific to the material composition to a superordinate computing system using a data manager.

[0022] Due to the modular construction of the automation equipment, it is possible to create a mobile structure, especially using automation components. The mobile structure can then be dedicated to training the artificial intelligence application. The trained neural network can then be stored back to the automation component. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] The accompanying drawings illustrate exemplary embodiments of the present invention, in which:

[0024] Figure 1 shows an automated assembly for analyzing material composition,

[0025] Figure 2 shows the process flow with symbols representing different stages, and

[0026] Figure 3 The determination of the test spectrum is shown using a sample specially prepared in advance. DETAILED DESCRIPTION

[0027] Figure 1 An automation component 1 is shown which is designed to analyze a material composition Mi of a test object 40. The test object 40 is analyzed by means of a test signal PS via a sensor 6. An impedance measuring unit 5 is designed to obtain an impedance spectrum ImS which is dependent on the amplitude and / or frequency of the test signal PS from the test object 40 via the sensor 6. A sensor-specific sensor interface 4 is arranged between the sensor 6 and the impedance measuring unit 5.

[0028] For evaluating the neural network NN, the automation component 1 has a processor component 7. The processor component 7 is coupled to a data memory 8 in which one or more neural networks NNi are stored.

[0029] For the detection phase EP (see Figure 2 ), a sample Pi is used for each of the test objects i and O, and a predetermined number N = 100 test spectra TSi1, ..., 100 are obtained from the corresponding sample Pi by means of an impedance measurement unit 5. The number N = 100 test spectra are stored in the sample detection unit 10 as a training data set TDs dedicated to the material component Mi.

[0030] This means that in the first test phase EP, a first sample P1 corresponding to a first material composition Mi is provided for measuring a test spectrum corresponding exactly to this material composition. Thus, in the assignment table 9 there are stored as many material compositions Mi as samples Pi.

[0031] For the learning phase LP, the test spectra TSi1, ..., 100 of the corresponding material composition Mi are used to train the neural network NNi for forming the material composition Mi. For this purpose, the training data set TDs can be transferred to the learning tool 12 via the interface 11 or forwarded to the bus system 31 via the bus interface 32, so that the training data set TDs can be forwarded to the superordinate computing system 30. In the superordinate computing system 30, there is a learning algorithm 33, in which the training data set TDs is trained to form a neural network NNi specific to the material composition Mi.

[0032] In the assignment table 9, a target material composition SM for the test object 40 is also stored. For other material compositions Mi, known deviations in composition from the target material composition SM are also stored. A calculation device 13 is present, which is designed to derive the deviations of the material composition SM with the aid of the target material composition SM from the test spectrum for the material composition determined in the analysis phase AP.

[0033] Furthermore, a reporting unit 14 is present in the automation component, which is designed to inform a regulating unit and / or a control unit of deviations in the material composition, which in turn is designed to intervene in the manufacturing process so that the test object 40 to be manufactured has a target material composition again.

[0034] according to Figure 2 , basically shows the detection phase EP, the learning phase LP and the analysis phase AP of the method for analyzing the material composition Mi of the test object. In the detection phase EP, the test object 40 is provided with samples P1, P12, P3, ..., Pi-2, Pi-1, Pi respectively. The impedance measurement unit 5 is used to obtain a number N of impedance spectra ImS for each sample Pi. The obtained impedance spectrum ImS is then assigned to the sample Pi or the material composition Mi, and the obtained test spectrum TSik is generated from the corresponding sample Pi.

[0035] In the learning phase LP, a neural network NNi specific to the material composition Mi is trained for a corresponding number N of test spectra TSik of the material composition Mi. Thus, from a first test spectrum TS1k for a first sample P1, a first neural network NN1 specific to the material composition M1 is formed, and likewise, for an i-th sample Pi of an i-th material composition Mi, an i-th test spectrum TSik is combined and an i-th neural network NNi specific to the material composition is trained or generated or formed.

[0036] In the subsequent analysis phase AP, the currently acquired impedance spectrum ImS of the test object 40 is again analyzed using the neural network NNi dedicated to the material composition Mi, and the material composition Mi is output as a result, and the material composition used for the impedance spectrum ImS to be analyzed is proven to be the best match for the corresponding neural network NNi dedicated to the material composition Mi.

[0037] Figure 3 Once again, the relationship between the known deviations in the material composition of the samples Pi and the target material composition SM is shown. Using the fourth sample, a test spectrum TS4 is obtained corresponding to the target spectrum SS. Using the third sample P3, a third test spectrum TS3 is then obtained, which corresponds to the material composition SM of the fourth sample P4 minus the deviation D1. Using the fifth sample P5, a test spectrum TS is obtained, which corresponds to the material deviation of the fourth sample P4 plus the deviation D4, etc.

Claims

1. A method for analyzing the material composition of a test object (40), in which method, - an impedance measuring unit (5) uses an electrical test signal (PS) via a sensor (6) to obtain an impedance spectrum (ImS) from the test object (40) depending on the amplitude and / or frequency (f) of the test signal (PS), - storing different material compositions (Mi) in an allocation table (9), - provide a separate test specimen (Pi) for each material composition (Mi), - in a testing phase (EP), using a corresponding test sample (Pi) for the test object (40), obtaining a predeterminable number (N) of test spectra (TSik) from the corresponding test sample (Pi) by means of the impedance measuring unit (5), and storing the number of test spectra (TSik) as a training data set (TDS) specific to the material composition (Mi) in a sample testing unit (10), - in a learning phase (LP), training a specific neural network (NNi) for said number (N) of test spectra (TSik) corresponding to said material composition (Mi), - in an analysis phase (AP), the currently acquired impedance spectrum (ImS) of the test object (40) is analyzed using the neural network (NNi) specific to the material composition (Mi), and the material composition (Mi) which has proven to be the best match for the corresponding neural network (NNi) specific to the material composition (Mi) for the impedance spectrum (ImS) to be analyzed is output as a result, It is characterized in that The impedance spectrum (ImS) is specified as a complex-valued frequency curve, and a separate neural network (NNi) dedicated to the material composition (Mi) is trained for the absolute value, phase angle, real part and imaginary part respectively, and the neural network works in parallel during the analysis phase (AP) for comparison with the currently obtained impedance spectrum (ImS) to improve the accuracy of the report on the material composition to be determined.

2. The method according to claim 1, wherein: A target material composition (SM) of the test object (40) is stored in the allocation table (9) and known deviations in composition relative to the target material composition (SM) are selected for other material compositions (Mi), wherein the current deviation of the material composition is derived from the material composition (Mi) determined in the analysis phase (AP) with the aid of the target material composition (SM).

3. The method according to claim 1 or 2, wherein the method is used in a manufacturing process of the test object (40), wherein: The knowledge of the deviations from the material composition (Mi) is used for intervention in the production process so that the test object (40) to be produced again has the target material composition (SM).

4. The method according to claim 1 , wherein the detection phase (EP) is performed on a local automation component ( 1 ) and a training data set specific to the material composition (Mi) is sent to a superordinate computing system ( 30 ), wherein: The learning phase (LP) is performed on the superordinate computing system (30) and the learned or calculated neural network (NNi) specific to the material component (Mi) is sent back from the superordinate computing system (30) to the local automation component (1) and stored for the analysis phase (AP).

5. An automation component (1), comprising: - an impedance measuring unit (5) designed to use an electrical test signal (PS) via a sensor (6) to obtain from a test object (40) an impedance spectrum (ImS) depending on the amplitude and / or frequency of the test signal (PS), - a processor module (7) designed to process a neural network (NN), - a data memory (8) in which a neural network (NN) is stored, characterized in that: - an allocation table (9) in which different material compositions (Mi) are stored, a sample test unit (10) which is designed to obtain, in a test phase (EP)) by means of the impedance measuring unit (5), a predeterminable number (N) of test spectra (TSik) for the material components (Mi) specified in the allocation table (9), a data manager (2) designed to store and manage the number of test spectra (TSik) as a training data set (TDS) specific to a material composition (Mi) in the sample detection unit (10), - an interface (11) to a learning tool (12) designed to provide, in a learning phase (LP), for the respective training data set (TDS), a neural network (NNi) specific to the material composition (Mi), - a data memory (8) in which a specific said neural network (NNi) is stored, - An embodiment of a processor module (7), which implements the analysis of the impedance spectrum (ImS) currently obtained with a specific neural network (NNKi) in the analysis phase (AP) and outputs the material composition (Mi) as a result, which material composition has been proven to be the best match for the corresponding neural network (NNi) dedicated to the material composition (Mi) for the impedance spectrum (ImS) to be analyzed.

6. The automation component (1) according to claim 5, wherein: The target material composition (SM) of the test object (40) is stored in the allocation table (9), and other material compositions (Mi) having known deviations in the composition relative to the target material composition (SM) are stored, wherein a calculation device (13) is provided, which is designed to derive the deviation of the material composition based on the material composition (Mi) determined in the analysis phase (AP) with the help of the target material composition (SM).

7. The automation component (1) according to claim 6 is designed to be used in the manufacturing process of the test object (40), and the automation component includes a reporting unit (14), which is designed to notify a regulating unit and / or a control unit of the deviation of the material composition, and the regulating unit and / or the control unit are designed to intervene in the manufacturing process so that the test object (40) to be manufactured has the target material composition (SM) again.

8. The automation component (1) according to any one of claims 5 to 7, comprising a data transmission unit (14) which is designed to send the training data set (TDS) specific to the material composition (Mi) to a superordinate computing system (30) using a data manager (2).

Images