Method and apparatus for evaluating measurement signals of thermal analysis, method and apparatus for generating training data, and method and apparatus for generating training data for artificial intelligence modules

Abstract

This provides the simplest possible option for automatically evaluating thermal analysis measurement signals as much as possible. [Solution] The present invention relates to a method and apparatus (100) for evaluating a measurement signal of thermal analysis. The apparatus (100) comprises a data interface (110) configured to receive a measurement signal of thermal analysis, the measurement signal showing a measurement curve based on a temperature series, and a processing logic (120). The processing logic (120) is configured to determine a plurality of sliding windows based on the measurement signal, assign each sliding window to a corresponding section in a measurement curve having a plurality of measurement points, and is executed by the processing logic, and is configured to determine whether a thermal effect of the sample material underlying the thermal analysis exists for each of the plurality of sliding windows, the artificial intelligence module is configured to determine a continuous section of the thermal effect based on the measurement curve.

Description

[Technical Field]

[0001] The present invention relates to thermal analysis, and more particularly to the evaluation of measurement signals in thermal analysis. The present invention relates in particular to a method and apparatus for evaluating measurement signals in thermal analysis, a method and apparatus for generating training data for an artificial intelligence module, a corresponding computer-readable medium having such training data, a computer program, and a corresponding computer-readable medium. [Background technology]

[0002] Thermal analysis is a range of techniques that measure the physical and / or chemical properties of substances, mixtures of substances, and / or reaction mixtures as a function of temperature or time, in which the sample to be analyzed is processed according to a specified and / or controlled temperature program.

[0003] One example of a thermal analysis method is differential scanning calorimetry (DSC), in which the heat flow difference between a reference crucible and a crucible containing the sample is measured as a function of temperature and / or time. After performing this measurement, one or more energy effects that become prominent in the measurement, such as the appearance of a peak due to material melting or steps during glass transition, can be evaluated based on the measurement data.

[0004] The evaluation of energy effects in such measurements is usually performed manually. For example, the standard DIN EN ISO 11357.2 "Differential scanning calorimetry (DSC) - Part 2: Measurement of glass transition temperature and glass transition step height" for manual measurement of glass transition temperature specifies the equal-area method, the inflection point method, and the half-step height method. To manually evaluate peaks that may indicate the melting of the material during heating of the sample, for example, the area of ​​the peak can be determined, and from this area, the energy effect, i.e., the amount of heat absorbed or released in this process, can be determined.

[0005] For example, Patent Document 1 (European Patent Application Publication No. 2 824 450) relates to the automation of evaluation in thermal analysis measurements. In this case, by using a program-controlled computing device, at least the probability that the measurement result matches at least one dataset pre-stored in the computing device is calculated. This calculation is based on a comparison between effect data pre-extracted from the thermal analysis measurement curve and corresponding effect data stored in the dataset. In contrast, a more data-driven approach is desirable for evaluation.

[0006] Patent Document 2 (European Patent Application Publication No. 4 209 781) proposes providing first data to a first software module in the thermal analysis of a material sample. The first software module is configured to calculate a thermal analysis measurement curve from the first data, and this thermal analysis measurement curve outputs second data suitable for displaying the measurement curve, enabling the identification of thermal effects. This method further includes providing the second data to a second software module, which includes artificial intelligence configured to automatically identify thermal effects. The second software module is configured to output third data representing thermal effects, which is automatically identified by the artificial intelligence. The artificial intelligence includes at least one neural network. In contrast, an alternative approach that does not use a neural network is desirable. [Prior art documents] [Patent Documents]

[0007] [Patent Document 1] European Patent Application Publication No. 2824450 [Patent Document 2] European Patent Application Publication No. 4209781 [Overview of the Initiative] [Problems that the invention aims to solve]

[0008] The objective of this invention is to provide the simplest possible option for automatically evaluating thermal analysis measurement signals as much as possible. [Means for solving the problem]

[0009] This problem is resolved by the provisions of the independent claim. Advantageous further improvements to the present invention are as described in the dependent claims.

[0010] According to the first embodiment, a method for evaluating a measurement signal of thermal analysis is proposed. This method includes the step of receiving a measurement signal of thermal analysis, in which case the measurement signal represents a measurement curve based on a temperature series. This method further includes the step of determining a plurality of sliding windows based on the measurement signal, in which case each sliding window is assigned to a corresponding section in a measurement curve having a plurality of measurement points. This method further includes the step of determining, using an artificial intelligence module configured for classification, whether the thermal effect of the sample underlying the thermal analysis is present in each of the plurality of sliding windows (or one thereof), in which case the artificial intelligence module determines a continuous section of the thermal effect based on the measurement curve.

[0011] The proposed method enables a high degree of automation when evaluating thermal analysis measurement signals using technically easily implementable means. Based on an artificial intelligence module configured for classification, this method reliably detects thermal effects contained in measurements or measurement curves and automatically determines their effect limits as accurately as possible. By using multiple sliding windows, not only information from a single measurement point but also the measurement curve in the surrounding area of ​​the measurement point to be classified is observed for classification of individual measurement points. Simultaneously, thermal effects are detected or determined (judged) by classifying only individual curve sections and / or points, rather than the entire measurement curve. The evaluation of the measurement signal can be provided and / or output as corresponding generated data including the detected thermal effects.

[0012] The methods and apparatus described herein can be used for a variety of technical applications, such as laboratory use, product manufacturing, and product processing.

[0013] This method can be implemented and / or executed by the apparatus described herein, or by a general computing device, computer, processor, CPU, etc. Therefore, this method can be implemented by a computer. This method can be implemented, for example, in the form of a computer program, i.e., computer instructions. The measurement signal exists and can be received electronically, optionally in digital format. The determination results by the artificial intelligence module, i.e., the thermal effect determined as a section, can be provided and / or output in the form of output data generated to correspond to it. The determination results can be output, for example, in text format, graph format, etc.

[0014] In this specification, the general term "thermal analysis" (which may also be called thermoanalysis) can be understood as a variety of methods for measuring the physical and / or chemical properties of substances, mixtures of substances, and / or reaction mixtures as a function of temperature or time, in which case the sample to be analyzed is processed according to a specified and / or controlled temperature program. The methods and apparatus described herein are explained using differential scanning calorimetry (DSC) as an example, but it goes without saying that the principles of the methods and apparatus described herein are also applicable to other thermal analysis methods, such as differential thermal analysis (DTA), thermogravimetric analysis (TGA), evolved gas analysis (EGA), thermomechanical analysis (TMA), and dynamic mechanical analysis (DMA).

[0015] Differential scanning calorimetry (DSC), which is illustratively used in the illustrated embodiments, is a type of thermal analysis method that measures the difference in heat flow between a reference material, such as a reference crucible, and a sample, such as a crucible with a sample body, as a function of temperature and / or time. By this method, the amount of heat output or absorbed in the sample body can be measured under a defined temperature program, particularly under cooling and / or heating. For example, the temperature difference ΔT between a sample having a temperature T1 and a reference material having a temperature T2, for example, is caused by executing a temperature program, such as heating by a heater in a surrounding furnace. This temperature difference can be captured by at least one measurement element and a corresponding measurement circuit. The temperature difference ΔT is due to the difference in heat capacity between the sample and the reference material. The heat flow difference proportional to the temperature difference can be applied, for example, with respect to the reference temperature T2, time, etc. By differential scanning calorimetry, at least one of the following exemplary determinations can be made, namely, melting point, glass transition temperature, crystallinity, dynamic observation of chemical reactions, specific heat capacity, phase transition, decomposition point, and plasticity measurement.

[0016] As used herein, the "thermal effect" refers to any kind of thermal and / or energy effect that occurs in the measurement signal during measurement as part of thermal analysis, for example, during differential scanning calorimetry. The thermal effect can be caused, for example, by a temperature change of the sample material or the reference material, and / or by a physical and / or chemical transformation of the sample material or the reference material. The thermal effect can include and / or be caused by the glass transition, melting point, crystallization process, etc. of each material. The thermal effect can be reflected as a corresponding feature in the measurement signal and / or the measurement curve. For example, the glass transition of each material can be represented by a step (stage) present in the measurement curve, and only the height difference causes the measurement signal or the corresponding measurement curve to deviate from the baseline corresponding to the DSC signal without a thermal effect. Further, for example, the melting point can be represented by a peak on the measurement curve, that is, a relatively steep maximum value in the positive direction of the measurement curve. The crystallization process can be represented by a peak on the measurement curve, that is, a relatively steep maximum value in the negative direction of the measurement curve. However, as is apparent from the present disclosure, the principles of the methods and apparatuses described herein are suitable for detecting any thermal effect that can occur during thermal analysis.

[0017] The "artificial intelligence module" can be understood as any computer implementation method in the field of artificial intelligence suitable for the evaluation of the measurement signals described in this specification. The artificial intelligence module can be, for example, a software module executable by a computer, a processor, etc. Alternatively or additionally, the artificial intelligence module can also be implemented in hardware. The artificial intelligence module can be based on at least one algorithm and / or computer model in the field of machine learning and / or can include at least one of these. In order to adapt the artificial intelligence module for classification tasks, it can be learned by learning data. The determination of the thermal effect by the artificial intelligence module can be understood or referred to as the classification and / or prediction of the thermal effect.

[0018] According to a further configuration, the specific section and / or evaluation of the thermal effect or its result can be supplied to a production planning and / or control system and / or quality management system. In such applications, in particular, the following steps can be provided, namely, the step of at least randomly performing a thermal analysis of the manufactured or processed product, and the step of evaluating the measurement results of the thermal analysis by the method of the type described above, where each measurement result is assigned to one of a plurality of classes, for example, a quality class or a material class. In at least some exemplary embodiments, it is conceivable that, depending on the evaluation result, the intervention in each process, for example, manufacturing and / or processing, is automatically promoted, i.e., by a computer or the like. This intervention can be, for example, a controlled change (including, for example, stopping the machine) of at least one operating parameter of the at least one machine used in the process. Alternatively or additionally, the intervention can include, for example, that a specific product manufactured or processed is discarded, discharged, etc. from the process as waste.

[0019] In a further configuration, the thermal effect section can be determined based on the temperature series by lower and upper effect limits. The upper and lower effect limits can separate the thermal effect section into different thermal effects within the measurement curve, or sections without thermal effects within the measurement curve. This allows for accurate evaluation and / or identification of thermal effects. Multiple thermal effects, particularly those distinct from each other, may appear or exist within the measurement curve. Corresponding sections with lower and upper effect limits can be determined and / or identified for each individual thermal effect. Different sections and / or thermal effects may have, or can be identified by, corresponding identifiers, such as corresponding labels.

[0020] In a further configuration, the determination of thermal effects by the artificial intelligence module may further include extracting at least one measurement point from each of several sliding windows. Each of the at least one measurement points extracted from the multiple sliding windows can be supplied to the artificial intelligence module. The artificial intelligence module can determine whether or not a thermal effect exists for each of the extracted measurement points. Based on the determination of the thermal effect, the artificial intelligence module can determine the section of the thermal effect for each of the measurement points extracted across the multiple sliding windows.

[0021] For example, each of the multiple sliding windows may have a defined window size. The window size can be selected, for example, according to the number of measurement points that make up the measurement curve. Thus, each sliding window can contain the same number of measurement points. Within each sliding window, at least one feature of a curve section in the measurement curve can be extracted. At least one feature is used for classification or determination and / or detection. At least one feature may include, for example, a statistical feature such as the arithmetic mean, a more complex statistical feature, and / or other features suitable for characterizing a curve section in the measurement curve. For example, at least one feature can be selected from standard deviation, median, empirical loop, mirror symmetry index, coefficient of variation, point symmetry index, deviation from linear regression, curvature, absolute energy, C3 statistic, first position of the maximum value, data points above the mean, area below the linear connection, minimum or maximum difference, time-reversal asymmetric statistic, centroid, etc. The measurement curve can be traversed with a specified step width or a specified step width and a defined window size. The sliding window can slide along the measurement curve until it reaches the final step. At each step, at least one feature can be extracted and a corresponding label can be assigned. Label assignment can be based, for example, on whether the center point of the sliding window is within the effective limit of the thermal effect. For example, based on the corresponding at least one feature, each center point or each measurement point X Tj A label can be assigned to it. This is, for example, X Tj It can be expressed as ∈ [GT, peak, N], in which case the label GT is the limit of the glass transition GT at the corresponding measurement point x min and x maxThis indicates that the measurement point is within the range. The label peak indicates that the corresponding measurement point is within the peak limit. Label N indicates, for example, that the measurement point is outside the region of the thermal effect, i.e., in a region where the thermal effect does not exist. The thermal effect can be characterized substantially by its shape. A glass transition can cause a step in the measurement signal and / or measurement curve. In the case of a peak, a local maximum may exist. The artificial intelligence module can determine a label corresponding to the center point of the observed sliding window based on at least one extracted feature. This can also be understood and / or referred to as point-based classification. Information on individual measurement points as well as the curve in the region surrounding the measurement point to be classified are observed.

[0022] In a further configuration, the thermal effect section can be determined based on whether the same thermal effect was measured for adjacent and extracted measurement points across multiple sliding windows. Adjacent and extracted measurement points having the same thermal effect can be grouped together as a thermal effect. As described above, at least one feature can be extracted for each sliding window, and based on this, the artificial intelligence module determines (e.g., predicts, estimates) a label corresponding to the center point of the observed sliding window. Based on this, adjacent measurement points characterized by the same label can be combined as a continuous section of thermal effects. By combining the measurement points into thermal effects and further performing post-processing steps such as validation and section merging, it is possible to determine one or more thermal effects for the entire sample.

[0023] In a further configuration, the determination for each of the at least one extracted measurement points can be made for the central region or center point of each sliding window in a plurality of sliding windows. As described above, the sliding windows can slide along the measurement curve until the final step is reached. At each step, at least one feature can be extracted and a corresponding label can be assigned. The label assignment can be made based on whether the center point of the sliding window is within the effective limit of the thermal effect.

[0024] In a further configuration, the artificial intelligence module can determine whether or not a thermal effect exists for each of multiple sliding windows with respect to different sliding window sizes. The thermal effect section can be determined based on whether or not the same thermal effect was predicted for different sliding window sizes. This can also be understood and / or referred to as effect-based classification. Thus, the thermal effect section can be determined or detected as a whole.

[0025] By performing determinations and / or detections with different sliding window sizes, multiple determinations, such as predictions, can be obtained for the same thermal effect. These can then be combined for the corresponding common thermal effect.

[0026] In a further configuration, the determinations for each measurement point obtained with different sliding window sizes can be counted. Based on this count, the thermal effect section can be determined. For example, the thermal effect can be determined by majority vote. In other words, for each individual measurement point, the determination of the thermal effect, i.e., the effect judgment or effect prediction count, can be made.

[0027] In further configurations, thermal effects can be assigned to the glass transition, melting, or crystallization processes of the sample material in thermal analysis. As described above, thermal effects can be characterized substantially by their shape. Glass transitions can cause steps in the measured signal and / or measurement curve. In the case of peaks resulting from melting or crystallization, local maxima may exist. These can be determined, predicted, and / or classified by the artificial intelligence module.

[0028] According to a second embodiment, an apparatus for evaluating measurement signals of thermal analysis is proposed. This apparatus includes a data interface configured to receive measurement signals of thermal analysis, in which case the measurement signals represent a measurement curve based on a temperature series. The apparatus further includes processing logic, which is configured to determine a plurality of sliding windows based on the measurement signals, in which case each sliding window is assigned to a corresponding section in a measurement curve having a plurality of measurement points, and is configured to determine whether a thermal effect of the sample material underlying the thermal analysis exists for each of the plurality of sliding windows, executed by the processing logic and configured for classification by an artificial intelligence module. The artificial intelligence module is configured to determine a continuous section of the thermal effect based on the measurement curve.

[0029] This device can be configured to carry out or perform the method according to the first embodiment. The processing logic may include, or can be configured as, a computing device, processor, CPU, GPU, etc. The processing logic may be coupled with a data interface, data storage, etc. For further configuration of this device, please refer to the method according to the first embodiment.

[0030] Further configurations may include at least one of a support vector machine and a random forest method. Both support vector machines and random forests are machine learning techniques suitable for data classification. While it is conceivable to implement the artificial intelligence module with at least one neural network, such as a convolutional neural network (CNN), support vector machines and random forests have been found to be particularly easy, computationally efficient, and reliable to implement in determining thermal effects. Compared to neural networks, support vector machines and / or random forests require, for example, smaller datasets, shorter execution times, and lower-performance hardware to perform training within a reasonable timeframe.

[0031] According to a third aspect, a method is proposed for generating training data for an artificial intelligence module to be trained to evaluate measurement signals of thermal analysis. The method includes the step of receiving a training dataset, each of a plurality of samples, each containing a measurement signal of thermal analysis and at least one thermal effect assigned to each measurement signal of a sample underlying the thermal analysis, in which case each measurement signal represents a measurement curve based on a temperature series or time series. The method further includes the step of applying a plurality of sliding windows to the plurality of samples. The method further includes the step of assigning a label to each sliding window, in which case each label represents the corresponding thermal effect as a continuous section of the thermal effect based on each measurement curve. The method further includes the step of generating training data based on the training dataset and each assigned label.

[0032] The number of samples in the training dataset (e.g., multiple) is, for example, from "Sample1 (measurement data, effect)" to "Sample NThe data can be provided in the form of "(measurement data, effect)". Each sample may include measurement data in the form of a measurement curve with lower and upper effect limits. For classification, not only information from individual measurement points but also the measurement curve in the area surrounding the measurement point to be classified can be observed using multiple sliding windows. As described above, at least one feature can be extracted from each curve section within each sliding window. The extracted at least one feature can optionally be preprocessed, such as scaling.

[0033] The assignment of each label for the point-based classification described above can be performed by sliding multiple sliding windows, each having a defined window size and a defined step width, along the measurement curve. At each step, at least one feature can be extracted from the sliding window, and a corresponding label can be assigned depending on whether the center point of the observed sliding window is within the effect limits of the thermal effect, i.e., for example, the peak limit, the glass transition limit, or outside the thermal effect. The window size used during measurement may depend on how many measurement points the measurement curve consists of. Alternatively, the assignment of each label for the effect-based classification described above, i.e., the assignment of each label when the thermal effect is detected as a whole, can be performed not by sliding multiple sliding windows point by point, i.e., from measurement point to measurement, along the measurement curve with a fixed window size, but by selecting multiple sliding windows based on the effect appearing on the measurement curve.

[0034] In a further configuration, the method may further include supplying the generated training data to an artificial intelligence module. In this way, the artificial intelligence module can be trained to carry out the method according to the first embodiment.

[0035] According to the fourth aspect, a computer-readable medium is provided. Training data generated according to the third aspect can be stored in the computer-readable medium. Alternatively, the computer-readable medium is a data carrier signal that transfers the training data generated according to the method of the third aspect.

[0036] According to a fifth aspect, an apparatus is proposed for generating training data for an artificial intelligence module to be trained to evaluate measurement signals of thermal analysis. The apparatus comprises a data interface configured to receive a training dataset, each comprising a plurality of samples, each containing a measurement signal of thermal analysis and at least one thermal effect assigned to each measurement signal of a sample underlying the thermal analysis, in which case each measurement signal represents a measurement curve based on a temperature series. The apparatus further comprises processing logic, which is configured to apply a plurality of sliding windows to the plurality of samples and to assign a label to each sliding window, in which case each label represents the corresponding thermal effect as a continuous section of the thermal effect based on each measurement curve, and is configured to generate training data based on the training dataset and each assigned label.

[0037] This device can be configured to carry out and / or perform the method according to the third embodiment. The processing logic may have, or be configured as, a computing device, processor, CPU, etc. The processing logic may be coupled to a data interface. For further configurations of this device, please refer to the method according to the third embodiment.

[0038] In a further configuration, the device may be further configured to provide and / or supply the generated training data to an artificial intelligence module.

[0039] According to the sixth aspect, a computer program is provided. This computer program includes instructions (commands) that cause the computer to execute the method according to the first aspect and / or the method according to the third aspect when the computer program is executed by the computer.

[0040] According to the seventh aspect, a computer-readable medium is provided. This computer-readable medium includes instructions (commands) that cause the computer to perform the method according to the first aspect and / or the method according to the third aspect when executed by a computer.

[0041] It goes without saying that the above-described embodiments and further configurations can be combined arbitrarily, unless any combination is explicitly excluded.

[0042] The present invention will be described in more detail with reference to exemplary embodiments shown in the accompanying drawings.

[0043] The accompanying drawings are included for a further understanding of the present invention and constitute part of this specification. The drawings illustrate embodiments of the present invention and are used together with the specification to illustrate the principles of the present invention. Other embodiments of the present invention and many advantages of the present invention can be readily understood by referring to the detailed description below. The elements in the drawings are not necessarily drawn to the same scale as each other. Accordingly, the same reference numerals indicate similar parts. [Brief explanation of the drawing]

[0044] [Figure 1] A schematic diagram showing an exemplary assembly for performing thermal analysis and an apparatus for evaluating the measurement signals of thermal analysis according to one embodiment. [Figure 2] This is an explanatory diagram illustrating an exemplary measurement signal for thermal analysis in the form of a measurement curve with thermal effects. [Figure 3] This is an explanatory diagram showing an exemplary apparatus for generating training data for an artificial intelligence module according to one embodiment. [Figure 4]This is an explanatory diagram illustrating an exemplary first learning process for an artificial intelligence module based on a measurement signal, according to one embodiment. [Figure 5] A block diagram showing an exemplary first learning process for an artificial intelligence module according to one embodiment. [Figure 6] This is a block diagram illustrating an exemplary first determination or detection process for a thermal effect according to one embodiment. [Figure 7] A block diagram showing an exemplary second learning process for an artificial intelligence module according to one embodiment. [Figure 8] This flowchart shows a method for evaluating measurement signals in thermal analysis according to one embodiment. [Figure 9] This flowchart shows a method for generating training data for an artificial intelligence module according to one embodiment.

[0045] Unless otherwise specified, identical reference numerals in the figures indicate identical or functionally similar components. All directional terms, such as “up,” “down,” “left,” “right,” “above,” “downward,” “horizontal,” “vertical,” “back,” and “front,” are used for illustrative purposes only, and each embodiment is not limited to the specific arrangement shown in the drawings. [Modes for carrying out the invention]

[0046] Figure 1 schematically shows an exemplary assembly 10 for performing thermal analysis. Figure 1 further shows an exemplary apparatus 100 for evaluating the measurement signals of the thermal analysis.

[0047] Differential scanning calorimetry (DSC), as exemplary in the illustrated embodiment, is a type of thermal analysis method that measures the difference in heat flow between a reference material, e.g., a reference crucible, and a sample, e.g., a crucible containing the sample, as a function of temperature and / or time. This method allows for the measurement of the amount of heat output or absorbed in the sample under a defined temperature program, particularly under cooling and / or heating. Assembly 10 comprises a heating device 14, e.g., a furnace 12 having a heating coil. The assembly further comprises a sample 16 to be analyzed, e.g., a crucible containing the sample, and a reference material 18, e.g., a reference crucible. The assembly further comprises at least one measuring element for measuring the temperatures T1 and T2 of the sample 16 and the reference material 18, and a corresponding measuring circuit 20 configured to determine the temperature difference ΔT from the temperature measurements. The sample 16 and the reference material 18 have different heat capacities. The difference in heat flow between the sample 16 and the reference material 18 is at least substantially proportional to the temperature difference ΔT. The heat flow difference can be applied, for example, to a reference temperature, T2, or time. DSC can perform at least one exemplary determination (assessment) of the following: melting point, glass transition temperature, degree of crystallinity, dynamic observation of chemical reactions, specific heat capacity, phase transition, decomposition point, and plasticity measurement. Assembly 10 is configured to generate and / or provide the corresponding measurement signal, which shows a measurement curve based on a temperature series.

[0048] The apparatus 100 is configured to receive and evaluate thermal analysis measurement signals. The apparatus 100 includes a data interface 110 configured to receive thermal analysis measurement signals. The apparatus 100 further includes processing logic 120 coupled to the data interface 110 and configured to process, in particular, evaluate, the received measurement signals. The processing logic 120 may have, or be configured as, a computer, processor, CPU, GPU, etc. In the illustrated embodiment, for the sake of clarity, the apparatus 100 is shown as a computer, for example, a desktop computer or workstation. However, the apparatus 100 may also be the apparatus of assembly 10, or may be configured by, for example, dedicated equipment.

[0049] The processing logic 120 is configured to determine multiple sliding windows based on the measurement signal. Each of these sliding windows is assigned to a corresponding section (interval) in a measurement curve having multiple measurement points. The processing logic 120 is further configured, along with an artificial intelligence module configured for classification, to determine whether or not the thermal effect of the sample material, which forms the basis of the thermal analysis, exists for each of the multiple sliding windows. The artificial intelligence module is configured to determine a continuous section of the thermal effect based on the measurement curve.

[0050] The artificial intelligence module may include, for example, at least one of a support vector machine and a random forest method. Both support vector machines and random forest methods are machine learning techniques and are suitable for data classification. While it is also conceivable to implement the artificial intelligence module with at least one neural network, such as a convolutional neural network (CNN), support vector machines and random forest methods have been found to be particularly easy, computationally efficient, and reliable to implement in determining thermal effects.

[0051] To determine at least one thermal effect, two exemplary approaches, namely point-based classification and effect-based classification, are further described below.

[0052] Figure 2 shows an exemplary measurement signal for thermal analysis in the form of a measurement curve with thermal effects. The measurement signal originates from assembly 10 and can be processed or evaluated by apparatus 100.

[0053] The measurement signal includes two exemplary thermal effects. The glass transition of sample 16 material is indicated as "GT" and appears as a step in the measurement signal, or is characterized by such a step. "Peak" refers to the melting of sample 16 material in the graph shown, in which case a similar feature may appear in the negative direction during the crystallization process of sample 16 material, which can also be similarly determined by the apparatus 100. The melting point is characterized by a peak in the measurement curve, i.e., a relatively steep maximum value.

[0054] The device 100 is configured to automatically evaluate such thermal effects in the measurement signal.

[0055] Figure 3 shows an exemplary apparatus 200 for generating training data for an artificial intelligence module in apparatus 100. Apparatus 200 is connectable to apparatus 100 or connected to apparatus 100 according to Figure 3.

[0056] The device 200 comprises a data interface 210 and processing logic 220 that are coupled together. The processing logic 220 may have, or be configured as, a computing device, processor, CPU, GPU, etc. In the illustrated embodiment, for the sake of ease of explanation only, the device 200 is shown as a computer, for example, a desktop computer or workstation.

[0057] The data interface 210 is configured to receive a training dataset, which has multiple samples, each containing a measurement signal of the thermal analysis and at least one thermal effect assigned to each measurement signal of the sample underlying the thermal analysis. Each measurement signal represents a measurement curve based on a temperature series. The processing logic 220 is configured to apply multiple sliding windows to the multiple samples. The processing logic 220 is further configured to assign a label to each sliding window, each label representing the corresponding thermal effect as a continuous section of the thermal effect based on each measurement curve. The processing logic 220 is further configured to generate training data based on the training dataset and each assigned label.

[0058] Furthermore, the device 200 is configured to provide the generated training data to the artificial intelligence module, for example, via the data interface 210, and / or to supply the generated training data to the artificial intelligence module of the device 100.

[0059] In the following, two exemplary approaches for training the artificial intelligence module of device 100 using device 200, namely the point-based approach and the effect-based approach, will be further described.

[0060] Figure 4 shows an exemplary first learning process for the artificial intelligence module of device 100, based on a measured signal. The learning process in Figure 4 can be understood or referred to as a point-based approach or point-based classification.

[0061] The measurement signal exists as a measurement curve with multiple measurement points. Measurement points are generally represented as circular points. Multiple sliding windows are applied to the measurement points. In other words, the multiple sliding windows slide along the measurement curve and / or measurement points. For ease of explanation, in Figure 4, each center point of the sliding window is shown as a diamond. The sliding windows are indicated by square brackets.

[0062] In the case of the point-based learning process, the detection process of the thermal effect region is due to the classification of individual measurement points on the measurement curve. Specifically, each measurement point X Tj is assigned a label. This can be expressed, for example, as X Tj ∈ [GT, peak, N]. In this case, the label GT means that the point is within the limits of the glass transition or the effect limits x min and x max . Therefore, the label peak means that it is within the limits of the peak. The label N means that the measurement point is outside the region where the thermal effect or energy effect occurs. As described above, the thermal effect is substantially characterized by its shape. In the case of glass transition, a step can be recognized in the measurement signal. In the case of a peak, there is a local maximum in the curve progression of the measurement curve. In the illustrated curve section, different features are extracted within each sliding window. These are used for the learning of the classifier. Examples of features that can be extracted within each sliding window include, for example, the standard deviation or the arithmetic mean of the observation section, but other features and / or additional features are also conceivable. Each sliding window slides on the measurement curve with a window size N W during the learning process of point-based classification and reaches the final step. The window size N W determines the number of measurement points present within each sliding window. The features of the illustrated sliding window are extracted at each step and the corresponding labels are assigned depending on whether the center point of the observation window is within the peak limits, within the glass transition limits, or outside the effect.

[0063] The sliding window size N W = 5 in FIG. 4 is just an example. The sliding window size used during measurement can be selected, for example, according to the number of measurement points the measurement curve has. This can be achieved, for example, by a coefficient that divides the number of measurement points. For example, for a measurement with 1500 measurement points over a temperature range of 300 K, the sliding window coefficient w kIf = 5, the number of points observed within the sliding window is N. w =3000, and the sliding window spans a temperature range of 60 K. According to Figure 4, in the first step S1, the second step S2, and the final step SN, the label "N" can be assigned as the sliding window slides along the measurement curve. The sliding window in step Sk is assigned the label "GT" because its center point is within the limits of the glass transition. As mentioned above, the label assignment can be made based on whether the center point of each sliding window is within the effective limits of the thermal effect.

[0064] Figure 5 shows an exemplary first learning process for the artificial intelligence module of the device 100 in Figure 4, represented by a block diagram 300 having blocks 310-360. Thus, in this case as well, the learning process is point-based.

[0065] In block 310, a training dataset is received via the data interface 210 of the instrument 200, and this dataset has N samples, each containing a measurement signal MD of the thermal analysis and at least one thermal effect EFF assigned to each measurement signal MS of the sample underlying the thermal analysis.

[0066] Each measurement signal MD shows a measurement curve based on a temperature series, as shown in block 320.

[0067] According to block 330, the processing logic 220 is configured to apply multiple sliding windows to multiple samples. For example, each of the multiple sliding windows may have a defined window size. The window size can be selected, for example, according to the number of measurement points that make up the measurement curve. Thus, each sliding window may contain the same number of measurement points.

[0068] According to block 340, at least one feature of a curve section in the measurement curve contained within each sliding window is extracted within each sliding window. At least one feature is used for classification or determination and / or detection. The at least one feature may include, for example, statistical features such as the arithmetic mean, more complex statistical features, and / or other features suitable for characterizing a curve section in the measurement curve. For example, the at least one feature can be selected from standard deviation, median, empirical loop, mirror symmetry index, coefficient of variation, point symmetry index, deviation from linear regression, curvature, absolute energy, C3 statistic, first position of maximum value, data point above mean, area below linear connection, minimum or maximum difference, time-reversal asymmetric statistic, centroid, etc. The measurement curve can be traversed with a specified step width or a specified step width and a defined window size. The sliding window can slide along the measurement curve until it reaches the final step. At each step, at least one feature can be extracted and assigned a corresponding label. Label assignment can be based, for example, on whether the center point of the sliding window is within the effective limit of the thermal effect. For example, each center point or each measurement point X can be assigned based on at least one corresponding feature. Tj A label can be assigned to it. This is, for example, X Tj It can be expressed as ∈ [GT, peak, N], in which case the label GT is the limit of the glass transition GT at the corresponding measurement point x min and x max This indicates that the measurement point is within the range. The label peak indicates that the corresponding measurement point is within the peak limit. Label N indicates, for example, that the measurement point is outside the region of the thermal effect, i.e., in a region where the thermal effect does not exist. The thermal effect section can be determined based on whether the same thermal effect was measured for adjacent and extracted measurement points across multiple sliding windows. Adjacent and extracted measurement points having the same thermal effect can be specifically grouped together as a thermal effect.

[0069] According to block 350, at least one extracted feature can be preprocessed, such as scaling.

[0070] According to block 360, the processing logic 220 is configured to generate training data based on the training dataset and each assigned label. This training data can be supplied to the artificial intelligence module.

[0071] Figure 6 shows an exemplary first determination or detection process for thermal effects in apparatus 100, represented by block diagram 400 having blocks 410-460. The determination or detection of thermal effects is, in this case as well, performed on a point basis, similar to the exemplary first learning process.

[0072] The thermal effects included in the measurement curve can be detected or determined during the detection process based on the artificial intelligence module learned by the exemplary first learning process of the artificial intelligence module in the device 100, and its limit x min and x max This can be determined. The procedure for detecting the effect is the same as the procedure for the learning process shown in Figures 4 and 5.

[0073] Referring further to Figure 6, the measurement signal is received or acquired in block 410. According to block 420, the measurement curve is passed through a specified step width and a defined sliding window. According to block 430, features of each sliding window are extracted and transformed by preprocessing steps adapted during learning. According to block 440, based on these features, the artificial intelligence module makes a prediction or decision regarding the label of the center point in the observed sliding window. In block 450, post-processing of the predicted labels of the measurement points groups adjacent points identified with the same effect label as an effect region. According to block 460, the determination or prediction of the entire sample is obtained, along with the effect limits, by combining the thermal effects of each point and optional further post-processing steps.

[0074] Figure 7 shows an exemplary second learning process for an artificial intelligence module, represented by block diagram 500, which has blocks 510-560. The learning process in Figure 7 can be understood or referred to as an effect-based approach or effect-based classification. In contrast to the point-based approach described above, in the effect-based approach, the sliding window is selected based on the effects appearing on the measurement curve, rather than sliding point by point on the measurement curve with a fixed window size. The difference from the point-based approach is the number of learning examples. This is because the number of thermal effects is significantly less than the number of measurement points included in the measurement curve. In the example of block 530, three learning samples are obtained, two of which do not contain thermal effects, and one contains thermal effects, i.e., effect limit x min and x max This is a glass transition between two states.

[0075] In block 510, a training dataset is received via the data interface 210 of the instrument 200, and this dataset has N samples, each containing a measurement signal MD of the thermal analysis and at least one thermal effect EFF assigned to each measurement signal MS of the sample underlying the thermal analysis. In block 520, each measurement signal MD shows a measurement curve based on a temperature series. According to blocks 530 and 540, features are extracted from sliding windows of thermal effect regions or regions without thermal effect, and corresponding labels are assigned. In the example shown in Figure 7, the label "N" represents the two outer regions, i.e., the effect limit x min and x max It is assigned to two outer regions outside of the central region, because there are no thermal effects in these regions. The label "GT" is for the central region, i.e., the effect limit x min and x maxIt is assigned to an internal region because this region is involved in the glass transition. According to block 550, the extracted features can be preprocessed, such as scaling, normalization, etc. According to block 560, the processing logic 220 is further configured to generate training data based on the training dataset and each assigned label. This training data can be supplied to the artificial intelligence module.

[0076] The effect-based determination or detection process is performed with respect to different sliding window sizes because the length of the thermal effect region varies significantly in parts. By detecting with different sliding window sizes, multiple predictions can be expected for the same effect. These effects can then be combined as a common thermal effect. For this purpose, the effect predictions at individual measurement points can be counted based on the predictions obtained with different sliding window sizes. The type of thermal effect to be detected can be determined by majority vote.

[0077] Figure 8 shows a flowchart of an exemplary method 600 for evaluating the measurement signal of thermal analysis. Method 600 can be carried out, for example, by apparatus 100.

[0078] Method 600 includes a step 610 of receiving a measurement signal for thermal analysis, in which case the measurement signal represents a measurement curve based on a temperature series. Method 600 further includes a step 620 of determining a plurality of sliding windows based on the measurement signal, in which case each sliding window is assigned to a corresponding section in the measurement curve having a plurality of measurement points. Method 600 further includes a step 630 of determining, using an artificial intelligence module configured for classification, whether the thermal effect of the sample underlying the thermal analysis is present in each of the plurality of sliding windows. In this case, the artificial intelligence module determines a continuous section of the thermal effect based on the measurement curve.

[0079] Figure 9 shows a flowchart of method 700 for generating training data for an artificial intelligence module. Method 700 can be carried out, for example, by apparatus 200.

[0080] Method 700 includes a step 710 of receiving a training dataset, each of which has multiple samples, each containing a measurement signal of a thermal analysis and at least one thermal effect assigned to each measurement signal of a sample underlying the thermal analysis, and each measurement signal shows a measurement curve based on a temperature series or time series. Method 700 further includes a step 720 of applying multiple sliding windows to the multiple samples. Method 700 further includes a step 730 of assigning labels to each sliding window, each label indicating the corresponding thermal effect as a continuous section of the thermal effect based on each measurement curve. Method 700 further includes a step 740 of generating training data based on the training dataset and each assigned label. [Explanation of Symbols]

[0081] 10. Assembly for performing thermal analysis 12 Furnace 14 Heating device 16 samples 18 Reference object 20 Measurement circuit 100 devices 110 Data Interface 120 Processing Logic 200 equipment 210 Data Interfaces 220 Processing Logic 300 Block Diagram 310-360 blocks / process steps 400 Block Diagram 410-460 blocks / process steps 500 Block Diagram 510-560 blocks / process steps 600 ways 610-630 Method Steps 700 methods 710-740 Method Steps

Claims

1. A method for evaluating the measurement signal of thermal analysis (600), wherein the method is The step (610) is to receive the measurement signal of the thermal analysis, wherein the measurement signal shows a measurement curve based on a temperature series, and the step (610) is to receive the measurement signal of the thermal analysis, Step (620) is to determine a plurality of sliding windows based on the measurement signal, wherein each sliding window is assigned to a corresponding section in the measurement curve having a plurality of measurement points, Step (630) involves using an artificial intelligence module configured for classification to determine whether the thermal effect of the sample that forms the basis of the thermal analysis is present in each of a plurality of sliding windows, and the artificial intelligence module determines a continuous section of the thermal effect based on the measurement curve, Methods that include...

2. A method according to claim 1, comprising supplying a specific section of the thermal effect to a production planning and / or control system and / or quality control system.

3. A method according to claim 1 or 2, wherein the section of the thermal effect is determined based on a temperature series by a lower effect limit and an upper effect limit, the upper and lower effect limits dividing the section of the thermal effect from different thermal effects in the measurement curve or from sections without thermal effects in the measurement curve.

4. A method according to any one of claims 1 to 3, wherein the determination of the thermal effect by the artificial intelligence module is Extracting at least one measurement point from each of the aforementioned multiple sliding windows, The process involves supplying each of the plurality of sliding windows to the artificial intelligence module, wherein the artificial intelligence module determines whether or not the thermal effect exists for each of the extracted measurement points, and, based on the determination of the thermal effect, determines the section of the thermal effect for each of the extracted measurement points across the plurality of sliding windows. Methods that further include the above.

5. A method according to claim 4, comprising determining the section of the thermal effect for each of the adjacent and extracted measurement points across the plurality of sliding windows based on whether the same thermal effect was predicted, and grouping the adjacent and extracted measurement points having the same thermal effect.

6. A method according to claim 4 or 5, wherein the determination for each of the at least one extracted measurement point is performed for the central region or center point of each of the plurality of sliding windows.

7. A method according to any one of claims 1 to 6, wherein the artificial intelligence module determines whether the thermal effect exists for each of the plurality of sliding windows with respect to different sliding window sizes, and determines the section of the thermal effect based on whether the same thermal effect has been determined for different sliding window sizes.

8. A method according to claim 7, comprising counting the determinations for each measurement point obtained with the different sliding window sizes, and determining the section of the thermal effect based on the counts.

9. A method according to any one of claims 1 to 8, wherein the thermal effect is assigned to a glass transition, melting, or crystallization process of the sample material for thermal analysis.

10. An apparatus (100) for evaluating the measurement signal of thermal analysis, wherein the apparatus is A data interface (110) configured to receive the measurement signal of the thermal analysis, wherein the measurement signal shows a measurement curve based on a temperature series, and the data interface (110) Processing logic (120), The processing logic (120) is provided with, Based on the measurement signal, it is configured to determine a plurality of sliding windows, and each sliding window is assigned to a corresponding section of the measurement curve having a plurality of measurement points. An apparatus comprising: an artificial intelligence module configured for classification, which is executed by the processing logic described above, and which is configured to determine whether or not the thermal effect of the sample material that forms the basis of the thermal analysis exists for each of a plurality of sliding windows, and which is configured to determine a continuous section of the thermal effect based on the measurement curve.

11. The apparatus according to claim 10, wherein the artificial intelligence module includes at least one of a support vector machine and a random forest method.

12. A method (700) for generating training data for an artificial intelligence module to be trained to evaluate measurement signals of thermal analysis, wherein the method is Step (710) is to receive a training dataset, wherein the training dataset has a plurality of samples, each including a measurement signal of a thermal analysis and at least one thermal effect assigned to each of the measurement signals of a sample that forms the basis of the thermal analysis, and each of the measurement signals shows a measurement curve based on a temperature series or a time series, The steps include applying multiple sliding windows to the multiple samples (720), Step (730) is to assign a label to each sliding window, where each label indicates the corresponding thermal effect as a continuous section of the thermal effect based on each measurement curve, A step (740) of generating training data based on the training dataset and each of the assigned labels, Methods that include...

13. A method according to claim 12, further comprising supplying the generated training data to an artificial intelligence module.

14. A computer-readable medium storing learning data generated according to the method of claim 12, or a data carrier signal for transferring learning data generated according to the method of claim 12 or 13.

15. A device (200) for generating training data for an artificial intelligence module to be trained to evaluate measurement signals of thermal analysis, wherein the device (200) A data interface (210) configured to receive a training dataset, wherein the training dataset has a plurality of samples, each including a measurement signal of a thermal analysis and at least one thermal effect assigned to each measurement signal of a sample underlying the thermal analysis, and each measurement signal shows a measurement curve based on a temperature series, and the data interface (210) Processing logic (220), The processing logic (220) is provided with, It is configured to apply multiple sliding windows to the multiple samples mentioned above. It is configured to assign a label to each sliding window, and each label indicates the corresponding thermal effect as a continuous section of the thermal effect based on each measurement curve. A device configured to generate training data based on the aforementioned training dataset and each assigned label.

16. The apparatus according to claim 15, wherein the apparatus is further configured to provide the generated learning data to an artificial intelligence module and / or to supply the generated learning data to the artificial intelligence module.

17. A computer program, which, when the computer program is executed by a computer, includes an instruction causing the computer to execute the method according to any one of claims 1 to 9 and / or the method according to claim 12 or 13.

18. A computer-readable medium comprising, when a computer program is executed by the computer, an instruction causing the computer to execute the method according to any one of claims 1 to 9 and / or the method according to claim 12 or 13.

Images