COMPUTER IMPLEMENTED METHOD, SYSTEM, AND COMPUTER PROGRAM FOR THERMAL ANALYSIS OF A SAMPLE OF MATERIAL - Patent application

Abstract

The present invention relates to a computer-implemented method for the thermal analysis of a sample of material. [Solution] The method includes providing first data to a first software module (6), the first data representing an observable response signal of the sample exposed to an excitation that generates an observable response and the excitation as a function of time, the response signal representing a thermal effect due to the sample, the first software module (6) being configured to receive the first data as input, calculate a thermal analysis measurement curve from the first data, and output second data suitable for representing the measurement curve, the thermal analysis measurement curve enabling identification of the thermal effect, the method further includes providing the second data to a second software module, the second software module including an artificial intelligence engine (9) configured for automatic identification of a thermal effect, the second software module being configured to receive the second data as input and output third data representative of the thermal effect automatically identified by the artificial intelligence engine (9).

Description

[Technical field]

[0001] The invention relates to a computer implemented method for thermal analysis of a sample of material, said method comprising providing first data to a first software module, said first data representing an observable response signal of said sample exposed to an excitation generating an observable response and said excitation as a function of time, said response signal representing a thermal effect due to said sample, said first software module being configured to receive said first data as input, to calculate a thermal analysis measurement curve from said first data, and to output second data suitable for representing said measurement curve, said thermal analysis measurement curve allowing identification of said thermal effect.Furthermore, the invention relates to a system for thermal analysis of a sample of material, a computer program for thermal analysis of a sample of material, a training software module of a computer program for thermal analysis, and a computer readable data carrier storing said computer program. [Background technology]

[0002] Thermal analysis studies the physical and chemical properties of a substance as they change with temperature. According to the International Confederation for Thermal Analysis and Calorimetry, thermal analysis is a group of techniques in which the physical properties of a substance are measured as a function of temperature while the substance is exposed to a controlled temperature program. That is, in thermal analysis, a sample of said substance is exposed to an excitation that generates an observable response. A response signal corresponding to said observable response is measured using a measuring means. First data (i.e. a time series of excitation and response signal) containing both the response signal and the excitation as a function of time is used to calculate a thermal analysis measurement curve, thereby allowing the identification of said thermal effect. In particular, the response signal, and possibly the excitation, contained in the first data, if a thermal effect exists, are converted into a unit(s) in which the thermal effect is observable. The unit into which the response signal is converted depends on the thermal analysis technique used and / or the thermal effect to be detected. This unit is usually one of physical units (including size) and chemical units. The units to which the excitation is converted, if not already in this unit, are units representing time, such as seconds, or temperature, such as °C, °F, or K. In many thermal analysis techniques, and therefore in many embodiments of the present invention, the units to which the excitation is converted, if not already in this unit, are units representing temperature.

[0003] The excitation may be considered to provide an independent variable of the thermoanalytical measurement curve and an independent variable of data representing the thermoanalytical measurement curve, such as the second data or the fourth data discussed further below, and the response signal may be considered to provide a dependent variable of the thermoanalytical measurement curve and a dependent variable of data representing the thermoanalytical measurement curve, such as the second data or the fourth data, meaning a variable that is dependent on the independent variable.

[0004] As mentioned, in many thermal analysis techniques, and therefore in many embodiments of the present invention, the independent variable has units that represent temperature.

[0005] Examples of thermal analysis techniques, responses, and response quantities, as well as excitations and excitation quantities, are given further below.

[0006] In a computer-implemented method for thermal analysis, first data may be provided to a first software module, said first software module being configured to calculate said measurement curve from the first data. The first software module may be further configured to output second data suitable for representing the measurement curve, i.e. the second data comprises data allowing to represent the curve shape and / or curve progression of the measurement curve.

[0007] In particular, the second data make it possible to represent the measurement curve in such a way that the curve shape can be reproduced in detail, which means that the curve shape can be reproduced in such a way that irregularities such as peaks, changes in slope, etc. can be made visible by use of the second data. In other words, the second data comprises data that allows a direct depiction of the measurement curve, in particular a direct graphical depiction of its curve shape and curve progression.

[0008] For example, the second data may include tuples of data points. In embodiments where the independent variable has units representing temperature, the values ​​of the tuples representing the independent variable are given in units representing temperature.

[0009] A graphical representation of the second data, respectively the measurement curve, may be displayed on the display means.The above mentioned first software module is known in the prior art, for example as STARe Software by Mettler-Toledo.

[0010] An example of thermal analysis is Differential Scanning Calorimetry (DSC). In a DSC experiment, a sample of a substance and a known reference, typically an empty crucible, are exposed to an excitation comprising a temperature program. The excitation may be in the form of a linear temperature slope superimposed with a periodic temperature modulation with a predefined frequency and amplitude. This generates an observable response comprising a response corresponding to the heat flow caused by the difference between the heat flow to the sample and the heat flow to the reference. The heat flow may be measured by using a measuring means, for example via a thermoelectric voltage, which is a response signal. The first data, which comprises the response signal as a function of time and the temperature of the temperature program as a function of time, allows the calculation of a DSC curve, i.e. a measurement curve. More details on DSC can be found, for example, in the standard textbook "Differential Scanning Calorimetry - An Introduction for Practitioners" by G. Hoehne, W. Hemminger, H.-J. Flammersheim.

[0011] Once the measurement curve is calculated, it may be displayed, for example, on a display means and examined for thermal effects by a human expert. In the case of DSC, such thermal effects may include glass transition, cold crystallization, or melting. Thermal effects are reflected in the shape of the corresponding measurement curve, for example as peaks, dips, steps, etc. However, the interpretation of measurement curves and the identification of thermal effects in such curves is a difficult task. The analysis and interpretation of the curves requires not only a significant amount of experience in thermal analysis, but also knowledge of possible reactions that a particular sample may undergo. Overlapping effects and artifacts make this process even more difficult. Thermal analysis is therefore difficult, cumbersome, and prone to errors.

[0012] In the past, electronic data processing has proved to be a useful tool for the analysis of thermoanalytical measurement curves. DE102013011730B3 discloses a method for evaluating thermoanalytical measurement curves by using electronic data processing means. The measurement curves are classified according to the changes in the characteristics present in the measurement curves and features associated with the thermal effects reflected by these changes are calculated, thereby obtaining a feature vector. This feature vector is compared by using an algorithm with feature vectors of known substances stored in a database, thereby obtaining a match probability. US2011 / 0301860A1 discloses a computer-implemented method for classifying DSC plasma thermograms using a similarity metric to identify pre-characterized inflammatory diseases.

[0013] Furthermore, there are computer-implemented methods for thermal analysis in the prior art that utilize artificial intelligence. In CN105823863A, the thermogravimetric curve of coal is first analyzed to extract values ​​of feature points associated with the curve. These values ​​are then used as input for a neural network that outputs data related to the quality of the coal. BRPI0604102A uses an expert system to relate molecular fragments to mass losses determined by thermogravimetric analysis. As these methods are based on models, there is room for improvement.

[0014] Interest in computer-implemented methods that utilize artificial intelligence has grown tremendously in recent years. Some related documents are not related to thermal analysis and thermal analysis measurements, but rather to the temperature behavior of samples. For example, CN113760660A provides a three-dimensional multi-core chip temperature prediction method in which historical temperature information is input into a neural network to predict the expected future temperature distribution of a three-dimensional multi-core chip, in particular to predict the location of potential hot spots. Summary of the Invention [Problem to be solved by the invention]

[0015] It is therefore an object of the present invention to provide a computer-implemented method for thermal analysis that is efficient and reliable. [Means for solving the problem]

[0016] According to a first aspect of the present invention, this object is achieved in that the above-mentioned method further comprises providing said second data to a second software module, said second software module comprising an artificial intelligence engine configured for automatic identification of thermal effects, said second software module being configured to receive said second data as input and to output third data representative of said thermal effects automatically identified by said artificial intelligence engine.

[0017] The third data typically includes a type, such as a name or identifier, of the thermal effect. Melting, vaporization, crystallization, recrystallization, and phase transitions such as glass transitions, decomposition, pyrolysis, depolymerization, degradation, polymerization, cross-linking, sublimation, and desorption are examples of thermal effects.

[0018] The third data may further include at least one of information about artifacts and / or confidence values ​​associated with the identified thermal effect and information such as an excitation range of said thermal effect, in particular an excitation range in the thermal analysis measurement curve or in the second data. The excitation range may be defined by a lower and an upper limit value of the excitation range, and the thermal effect may occur between said lower and upper values.

[0019] The identification of thermal effects in a measurement curve by the artificial intelligence engine according to the first aspect of the present invention is quicker and more prone to errors than an identification performed by a human expert. Moreover, it does not require the extraction of feature point values ​​from the measurement curve, as disclosed, for example, in CN105823863A, and is therefore also quicker and more prone to errors than known computer-implemented methods using artificial intelligence.

[0020] The method according to the first aspect of the invention uses a modular approach: a first software module calculates the thermoanalytical measurement curve from the experimental data, i.e. the observable response signal and the excitation, and a second software module identifies the thermal effects reflected in the second data by using artificial intelligence. Such a modular approach is very efficient, since the two software modules can be optimized for their respective tasks. Furthermore, known software modules configured to calculate thermoanalytical measurement curves, such as STARe software from Mettler-Toledo, may be used as the first software module.

[0021] Moreover, the first and second software modules communicate with each other via a clearly defined data interface, and the second data, which is the output of the first software module, is accepted as input by the second software module. Thus, the components, subroutines, etc. of the first and second software modules can be changed without affecting the functioning of the method, as long as the data structure of the second data is kept unchanged. In this way, the method of the invention is very flexible. For example, the first software module may be understood as a base software for thermal analysis. This base software may include several main programs, for example an installation program for setting up the equipment, creating users and registering data on reference materials, a control program representing the measuring means or measuring balance and making it possible to create methods and experiments for routine operation, and a calculation program for calculating said thermal analysis measurement curves. Reference materials may be used for calibration and adjustment. The calculation program may be configured to accept said first data as input and to output said second data. Updating the different programs of the first software module does not affect the method according to the first aspect of the invention, as long as the data structure of the second data is kept unchanged.

[0022] The material analyzed by the method according to the first aspect of the invention may be any material, including a pure material or a mixture. The thermal effect may be any thermal effect and may include an artefact. In the case of an artefact, the third data may include information that the thermal effect is an artefact.

[0023] The first data may include data obtained by subjecting the sample to a thermal analysis measurement, including, but not limited to, Differential Scanning Calorimetry (DSC), Differential Thermal Analysis (DTA), Thermogravimetric Analysis (TGA), Evolved Gas Analysis (EGA), Thermomechanical Analysis (TMA), or Dynamic Mechanical Analysis (DMA). The first data may be obtained by applying only one of these measurement methods, or may be obtained by applying two or more of these measurement methods simultaneously.

[0024] The excitation, particularly the excitation as a function of time (i.e. the excitation may be time dependent), may be known in advance, so that a measurement of its value is not necessary. In another embodiment, the excitation may not be known in advance. Thus, a measurement means may be provided for measuring the value of said excitation, particularly as a function of time. For thermoanalytical measurements, the excitation may include an excitation corresponding to a variable temperature.

[0025] The response signal represents the thermal effect. There is a functional relationship between the response signal and the observable response. For example, in the case of DSC, the response may include a response quantity corresponding to the heat flow caused by the difference between the heat flow to the sample and the heat flow to the reference, and the response signal may include a thermoelectric voltage corresponding to the heat flow. Thus, the heat flow may be calculated from the thermoelectric voltage.

[0026] The first data may include data tuples, in particular time-ordered data tuples including an excitation and a corresponding response signal at a particular instance in time and a respective time instant.

[0027] The first and second software modules are executed by a data processing means. The first and second software modules may be executed by the same data processing means or by different data processing means that may be located at different locations. The first software module may be installed on a cloud computer. The first software module may be a multi-user application. The first software module may include a plurality of first software modules. Each of the plurality of first software modules may be executed by a separate data processing means, for example by a personal computer of one of the plurality of users. The second software module may be executed by a server or a cloud computer, for example, that may be located at the location of the software provider or the user or elsewhere. The first software module may be integrated in a Laboratory Information Management System (LIMS). The first software module may be executed by a data processing means that is part of the measurement means for thermal analysis.

[0028] The measurements performed to obtain the first data may be performed by a user of the method according to the first aspect of the invention. In this case, the method may further comprise exposing the sample to an excitation generating an observable response and measuring a corresponding observable response signal of the sample as a function of time. Alternatively, the measurements may be performed by a third party and the first data may be provided to a user of the method according to the first aspect of the invention for thermal analysis. The first data, representing both the observable response signal and the excitation as a function of time, is provided to a first software module. The first software module is configured to calculate a thermal analysis measurement curve from the first data.

[0029] The measurement curve calculated by the first software module may be a two-dimensional curve. Such a measurement curve establishes a relationship between a certain excitation or time and a corresponding observable response. For example, in the case of DSC, the measurement curve may establish a relationship between temperature or time and heat flow.

[0030] However, the measurement curve may also be multidimensional. The measurement curve may include only one multidimensional curve. Alternatively, the measurement curve may include two or more two-dimensional curves. In both cases, the measurement curve establishes a relationship between a particular excitation or time and each of a number of corresponding observable responses. For example, in the case of DSC, the measurement curve may establish a relationship between time, heat flow, reference temperature, and sample temperature.

[0031] The second data, which is the output of the first software module, may include or be any data representative of the shape of the measurement curve. For example, the measurement curve may be i and / or time t i And some physical quantity Q i , for example, to establish relationships between observable responses. The second data is then a set of tuples of values ​​of the data points (t i ,T i ,Q iAdditionally or alternatively, the second data may include pixel data of a graphical depiction of the measurement curve. In the case of a DSC, the second data may include a tuple of data points (t i ,Φ i ,T S , i ,T r , i ), where t i is the time, and Φ i is the heat flow and T S , i is the sample temperature, T r , i is the temperature of the reference. The reference can be an empty crucible. i ,Φ i ) can represent a DSC curve.

[0032] The second data is provided to a second software module. The second software module includes an artificial intelligence engine configured for automatic identification of thermal effects. The artificial intelligence engine can be operative to detect characteristic features of thermal effects reflected in the second data and / or the measurement curve. The artificial intelligence engine can include at least one artificial intelligence algorithm, in particular a machine learning algorithm. The machine learning algorithm can be trained using training data. That is, the machine learning algorithm can be first trained using training data and then used for identification of thermal effects. The machine learning algorithm can include a supervised learning algorithm. That is, the machine learning algorithm can accept training input data and make predictions based on the underlying model / function, and when those predictions are wrong in light of the training output data, the machine learning algorithm is corrected. In particular, the supervised learning algorithm can include a classification algorithm.

[0033] The training data may be created in the following manner: a known training material may be exposed to an excitation that generates an observable response, and first data is created, including both the corresponding response signal and the excitation as a function of time. The first data is used to calculate a thermal analysis measurement curve. Second training data suitable for representing the shape of the measurement curve is then created, e.g., a tuple of data points of the curve, or pixel data of a graphical depiction of the measurement curve. Furthermore, the measurement curve and / or the second data are analyzed, e.g., by a human expert, for thermal effects present in the second data and / or the measurement curve. In this way, third training data may be determined, which is representative of the thermal training effect reflected in the measurement curve. The third training data may, for example, include the type of effect and the location of the effect in the measurement curve.

[0034] The second and third training data are then used to train an artificial intelligence algorithm, e.g., a machine learning algorithm, of the artificial intelligence engine. A template of the artificial intelligence algorithm to be trained with the training data may be provided. The artificial intelligence algorithm template accepts the second data as input and outputs the third data in a manner dictated by the engine parameters. The second training data is provided as input to the artificial intelligence algorithm template, and the output of the algorithm is compared to the third training data. The engine parameters are then modified until the output of the algorithm is substantially equal to the third training data associated with the second training data. By "substantially equal" it is meant that the distance between the output and the third training data is less than some threshold value within some standard. In this manner, the artificial intelligence algorithm used in the artificial intelligence engine may be trained and then deployed to the second software module.

[0035] The third data may be any data representative of the thermal effect automatically identified by the artificial intelligence engine. The third data may include a confidence value associated with the identified effect.

[0036] In one embodiment of the method according to the first aspect of the invention, the artificial intelligence engine may include at least one neural network for automatic identification of the thermal effect, the at least one neural network including an input layer of input neurons for receiving the second data and an output layer of output neurons for outputting the third data, the second data being provided to the input neurons of the input layer. The third data representative of the thermal effect may then be obtained as an output in an output layer of the at least one neural network. Furthermore, the neural network may include one or more hidden layers. Each layer of the neural network includes a plurality of artificial neurons. The neurons in successive layers may be connected by weighted links with adjustable weights. Each neuron receives an input value and generates an output value by applying a nonlinear function to the input value. The output value is transmitted to neurons in successive hidden layers or presented as an output of the neural network in the output layer.

[0037] The neurons of the neural network may be S-shaped neurons. S-shaped neurons have an associated weight w i with N inputs x i Accept

number

[0038] An artificial intelligence engine according to the invention may include one, two, three or more neural networks. If there is more than one neural network, the neural networks may differ in their network architecture, for example in the number of input and / or output neurons, and / or in the number of hidden layers, and / or in the weights, and / or in the biases, and / or in the topology. The method of the invention works for any network architecture.

[0039] In one embodiment, the artificial intelligence engine may include at least two sub-engines, preferably neural networks, for the automatic identification of the thermal effect, and the second software module may be further configured to receive selection data to select one of the sub-engines for the automatic identification of the thermal effect and provide the second data to an input of the selected sub-engine, which in a preferred embodiment where the sub-engine is a neural network, is done by providing the second data to an input layer of the selected neural network, and the selection data may be provided to the second software module. The at least two sub-engines may be configured, for example, for identification of thermal effects of materials in different material classes (e.g., metals, polymers, etc.), and the selection data may include data associated with the material classes. However, there may also be other specific characteristics associated with each of the sub-engines, and the selection data may include data that allows for selecting the desired sub-engine. For example, if the sub-engines are neural networks, the sub-engines may differ in type of neurons, number of layers, topology, version number, type of training data used for training the respective neural networks, etc. The second data may include the selection data. A user can input the selection data into the first and / or second software module. By enabling the selection of one of the sub-engines for automatic identification of thermal effects, the efficiency and accuracy of the method can be improved since the sub-engine most suitable for identifying thermal effects present in a particular material can be pre-selected.

[0040] In one example of the method according to the first aspect of the invention, the method may further comprise deploying at least one sub-engine in the second software module and / or removing at least one sub-engine from the second software module and / or deactivating at least one sub-engine in the second software module, preferably a sub-engine being a neural network. In this way, the artificial intelligence engine may be made user-specific. By deploying at least one sub-engine in the second software module, a user-specific sub-engine and / or an updated sub-engine etc. may be provided to the second software module and may be used for the automatic evaluation of thermal effects. By removing at least one of the sub-engines and then deploying a sub-engine or vice versa, the sub-engine may be replaced in the artificial intelligence engine. Since there is a clearly defined interface between the first and second software modules, as described above, it is feasible to deploy and / or remove at least one sub-engine and / or to stop the operation of at least one sub-engine, without affecting the functional principle of the method according to the first aspect of the invention. Furthermore, the first software module can be modified, for example by updating the calculation routine of the thermal measurement curve, or by updating any other routine in the first software module, as long as the data structure of the second data is not changed. Thus, the method according to the first aspect of the invention is very flexible, since both the sub-engine and the first software module can be updated or modified without affecting the functional principle of the method, as long as the data interface between the first and second software modules is not changed.

[0041] In another embodiment of the method according to the first aspect of the invention, said method may further comprise the use of a training software module for creating a trained sub-engine, preferably a trained neural network, and deploying the trained sub-engine in said second software module, whereby the training software module is configured to create the trained sub-engine by using expert training data associated with at least one training sample of a training material, said expert training data comprising second and third data sets of training data, said second training data being suitable for representing a training measurement curve attributable to said training sample, said training measurement curve enabling the identification of at least one thermal training effect attributable to said training sample, said third training data being representative of said thermal training effect, and the training software module further comprises the step of: providing a sub-engine template linking input data to output data in a content defined by a set of engine parameters, and deploying said trained sub-engine in said second software module; and creating a trained sub-engine by determining engine parameters such that, for a majority of the set, when second training data of said data set is input to said trained sub-engine, third data output by the sub-engine trained with the engine parameters is essentially equal to the third training data of one of the data sets, whereby the sub-engine template is preferably a blank neural network including an input layer for receiving said second data, an output layer for outputting said third data, and preferably one or more intermediate layers, as well as weights representing connections between the input layer, the output layer and preferably the intermediate layers, and the engine parameters are preferably weights determined such that, for a majority of the data sets, when second training data of said data set is received by the input layer, third data output by the output layer is essentially equal to the third training data of one of the data sets, whereby the determined weights define a trained neural network.

[0042] A trained sub-engine is created by starting with a sub-engine template. Such a template links input data to output data in a manner defined by a set of engine parameters, i.e., there is a functional relationship between the input data and the output data. When the trained sub-engine is a trained neural network, the sub-engine template is a blank neural network that includes an input layer of input neurons for receiving second data and an output layer of output neurons for outputting third data. Furthermore, the blank neural network may include one or more intermediate layers of neurons arranged in series between the input layer and the output layer. The neurons in the successive layers are connected by links associated with weights that are engine parameters.

[0043] The engine parameters are selected such that for a majority of the data sets, when second training data of said data sets is input to said trained sub-engine, third data output by the trained sub-engine is essentially equal to third training data of one of the data sets, where "essentially equal" may mean that the difference between the third data and the third training data within some standard is less than a threshold value.

[0044] An optimization routine may be used to select the engine parameters accordingly. For example, at the start of training, the engine parameters may be set to some initial values. These values ​​may be selected, for example, randomly. The second training data of one of the data sets may then be provided as input data to the sub-engine templates, and the output of each of the sub-engine templates may be compared to the third training data, for example, via some distance within a standard. The engine parameters may then be modified so that the output of the sub-engine templates approaches the third training data to a desired accuracy within the said standard. Here, a gradient optimization method may be used.

[0045] If the sub-engine is a neural network, the engine parameters may be weights. These weights may then be selected such that for a majority of the data sets, when a second training data of said data set is received by the input layer, the third data output by the output layer is essentially equal to the third training data of one of the data sets. The determined weights then define a trained neural network.

[0046] The expert training data may include a data set of one or more training samples of one particular training material. Alternatively, the expert training data may include a data set of one or more training samples of several different training materials. The training materials may belong to one material class, e.g. polymer, PET, metal, etc. The trained sub-engine may then be well suited to identifying thermal effects due to materials belonging to said material class. Alternatively, the training materials may belong to different material classes.

[0047] A trained sub-engine may be created from a sub-engine template, preferably a blank neural network, by using expert training data provided by a user. In this manner, a user-specific sub-engine, preferably a user-specific neural network, may be created. A manner of generating a neural network is disclosed, for example, in US2021 / 0012206A1.

[0048] In case the artificial intelligence engine includes two or more sub-engines, it may be beneficial if said second software module is further configured to receive selection data for selecting one of the sub-engines for the automatic identification of thermal effects and to provide second data to the selected sub-engine, said selection data may be provided to said second software module as described above. Thus, a user may, for example, select a trained sub-engine by providing corresponding selection data to the second software module.

[0049] According to one example, the training measurement curve may be obtained by applying thermal analysis measurements to the training samples, and / or the training measurement curve may be a theoretical measurement curve corresponding to the training samples, and the third training data may be obtained by human identification of thermal training effects present in the training measurement curve. That is, the measurement curve may be analyzed by a human expert for possible thermal effects, and / or a theoretical measurement curve including thermal training effects may be created and analyzed by a human expert, and these measurement curves may be used for training the trained sub-engine. In one embodiment, both the measurement curve obtained by thermal analysis measurements and the theoretical measurement curve corresponding to the training samples may be used to train the trained sub-engine.

[0050] The third data may be any data representative of the thermal effect automatically identified by the artificial intelligence engine. In one example, the third data may include an excitation range and a type of the thermal effect. The excitation range may be defined by a lower and an upper limit of an excitation range, and the thermal effect may occur between the lower and upper values. The type of effect may include, for example, a glass transition, a melting point, and a cold crystallization peak. If the thermal effect is an artifact, the third data may include this information.

[0051] In one embodiment, the method further comprises providing said third data and fourth data suitable for representing said measurement curve to an evaluation software module configured to receive said third and fourth data as input and to calculate a value of at least one characteristic mass associated with said at least one thermal effect by means of said fourth data. The fourth data may be provided by the first software module. The fourth data may be equal to or comprise the second data. However, the fourth data may also be different from the second data. As explained above, for example when the second data comprises pixel data representing a measurement curve, the fourth data may be provided at a time t i , temperature T i , and physical quantity Q i A tuple of data point values ​​(t i ,Ti,Q i ).

[0052] The evaluation software module may further be configured to output the characteristic quantity. The characteristic quantity may be any physical quantity associated with the thermal effect. For example, the third data may include the type of effect and the corresponding excitation range. For example, in the case of DSC, the characteristic quantity may include the glass transition temperature, the onset temperature of the thermal effect, the melting point temperature, the peak value, the midpoint temperature, the integral, etc., depending on the type of effect.

[0053] In other words, the evaluation software module may be configured to determine a value from the third data and the fourth data provided to the evaluation software module, the value being a characteristic of a thermal effect observable in the fourth data and enumerated in the third data, and the evaluation software module may be further configured to output said value.

[0054] The first, second and evaluation software modules communicate with each other via well-defined data interfaces - the second data, which is the output of the first software module, is accepted as input by the second software module, the third data, which is the output of the second software module, and the fourth data, which is the output of the second software module, are accepted as input by the evaluation software module. Thus, components, subroutines, etc. of the first, second and evaluation software modules can be modified without affecting the function of the method, as long as the data structures of the second, third and fourth data are kept unchanged.

[0055] The evaluation software module may be a software module independent of the first and second software modules, however, the evaluation module may also be implemented as a routine within the first software module, which may further be configured to receive the third data as an input.

[0056] The method according to the first aspect of the invention may further comprise displaying on a display means a graphical representation of the thermal analysis measurement curve and / or data associated with the excitation range of the thermal effect and / or data associated with the type of said thermal effect and / or the value of the at least one characteristic quantity, thereby allowing a graphical representation of the results of the thermal analysis.

[0057] The excitation may include one or more excitation doses. The excitation may include an excitation dose corresponding to a variable temperature. The variable temperature may be a temperature program. Additionally or alternatively, the excitation may include an excitation dose corresponding to a variable power, and / or an excitation dose corresponding to a variable pressure, and / or an excitation dose corresponding to a variable radiation dose, and / or an excitation dose corresponding to a variable stress or strain, and / or an excitation dose corresponding to a variable atmosphere of gas, and / or an excitation dose corresponding to a variable magnetic field. In this manner, thermal analysis measurements may be performed, including differential scanning calorimetry (DSC), differential thermal analysis (DTA), thermogravimetric analysis (TGA), evolved gas analysis (EGA), thermomechanical analysis (TMA), or dynamic mechanical analysis (DMA). Only one of these measurements may be performed, or two or more of these measurements may be performed simultaneously. When the excitation includes more than one excitation dose, two or more thermal analysis methods are applied to the sample simultaneously.

[0058] The response may include one or more response quantities, such as a response quantity corresponding to a temperature difference in a dynamic thermal analysis method, and / or a response quantity corresponding to a heat flow in a dynamic thermal analysis method, in particular a heat flow caused by a difference between a heat flow to a sample of a substance and a heat flow to a known reference, and / or a response quantity corresponding to a difference in heating power in a dynamic power compensation thermal analysis method, and / or a response quantity corresponding to a change in length in a dynamic thermomechanical analysis method, and / or a response quantity corresponding to a change in weight in a dynamic thermogravimetric analysis method, and / or a response quantity corresponding to a force in a dynamic mechanical analysis method, and / or a response quantity corresponding to a change in length in a dynamic mechanical analysis method, and / or a response quantity corresponding to a change in voltage in a dynamic dielectric analysis method.

[0059] According to a second aspect of the present invention there is provided a system for thermal analysis of a sample of a substance, said system comprising: a measurement means operative to measure a response signal of a sample exposed to an excitation producing an observable response, and to output first data representative of the response signal and the excitation as a function of time, the response signal being indicative of a thermal effect attributable to the sample; and a data processing means having a first software module configured to receive first data as input, to calculate a thermoanalytical measurement curve from the first data, and to output second data suitable to represent said measurement curve, said thermoanalytical measurement curve allowing for the identification of said thermal effect, said data processing means further comprising a second software module including an artificial intelligence engine configured for automatic identification of a thermal effect, said second software module configured to receive the second data as input, and to output third data representative of said thermal effect identified by the artificial intelligence engine.

[0060] The method according to the second aspect of the invention is adapted to implement the method according to the first aspect of the invention. Everything said above with respect to the method according to the first aspect of the invention also applies to the system according to the second aspect of the invention.

[0061] The system may include one measurement means or multiple measurement means. Preferably, the measurement means is a sensor, such as a heat flow sensor or a temperature sensor. Each of the multiple measurement means may be operative to measure a specific type of response signal of a sample exposed to a specific excitation generating a specific observable response, said specific response signal being representative of a thermal effect due to said sample. For example, one of the measurement means may be operative to perform a DSC measurement, another measurement means may be operative to perform a TGA measurement, etc. However, the system may further include several measurement means operative to measure the same type of response signal. Each measurement means of the multiple measurement means may be in communication with a data processing means and operative to provide first data to the data processing means. The measurement means may include a transmitting unit for transmitting the first data to the data processing means, and the data processing means may include a receiving unit for receiving the first data.

[0062] The data processing means may be a single data processing means or may comprise multiple data processing means. For example, a first data processing means may be operative to execute a first software module and a second data processing means may be operative to execute a second software module. Furthermore, the first data processing means may comprise a server or a cloud computer and may be accessible by several users. In this way, a multi-user environment may be created. In an embodiment, the measurement means may comprise the first data processing means and / or the second data processing means.

[0063] According to a third aspect of the present invention, there is provided a computer program for thermal analysis of a sample of a material, the computer program comprising a first software module configured to receive as input first data representative of a response signal of the sample exposed to an excitation generating an observable response and the excitation as a function of time, to calculate from the first data a thermal analysis measurement curve enabling identification of the thermal effect, and to output second data suitable to represent the measurement curve, the response signal being representative of a thermal effect attributable to the sample, the computer program further comprising a second software module including an artificial intelligence engine configured for automatic identification of a thermal effect, the second software module configured to receive as input the second data and to output third data representative of the thermal effect identified by the artificial intelligence engine.

[0064] A computer program according to the third aspect of the invention is adapted to implement the method according to the first aspect of the invention and to be used in the system according to the second aspect of the invention. Everything said above with respect to the method according to the first aspect of the invention and the system according to the second aspect of the invention also applies to the computer program according to the third aspect of the invention.

[0065] According to a fourth aspect of the present invention there is provided a training software module of a computer program for thermal analysis of a sample of a substance, said training software module being configured to receive expert training data comprising second and third data sets of training data, said second training data being suitable for representing a training measurement curve attributable to a training sample, said training measurement curve enabling identification of at least one thermal training effect attributable to said training sample, and said third training data being representative of said thermal training effect, and / or said training software module is further adapted to receive second data suitable for representing a thermoanalytical measurement curve enabling the identification of at least one thermal training effect from the first software module according to the third aspect of the invention, and to generate associated third data from a human input representative of a human identification of the at least one thermal training effect, thereby generating expert training data; whereby the training software module creates a trained sub-engine, preferably a trained neural network, based on the training data, the trained sub-engine being suitable for being deployed in a second software module according to the third aspect of the invention; Thereby, the training software module is preferably part of the computer program according to the third aspect of the invention.

[0066] The training software module according to the fourth aspect of the invention is configured to create a trained sub-engine by using expert training data. As explained above with respect to the method according to the first aspect of the invention, this trained sub-engine may be created by providing a sub-engine template linking input data to output data in a manner defined by a set of engine parameters, and by determining engine parameters such that, for a majority of the data sets, when a second training data of said data set is input to said trained sub-engine, a third data output by the sub-engine trained with the engine parameters is essentially equal to the third training data of one of the data sets. As explained above, the trained sub-engine may be a user-specific sub-engine, in particular a user-specific neural network.

[0067] In one example, the sub-engine template may be a blank neural network including an input layer of input neurons for receiving the second data, an output layer of output neurons for outputting the third data, and preferably one or more intermediate layers of neurons. Neurons in different layers may be linked by connections with associated weights, the weights being engine parameters. The weights are determined such that for a majority of the data sets, when the second training data of the data set is received by the input layer, the third data output by the output layer is essentially equal to the third training data of one of the data sets, whereby the determined weights define a trained neural network.

[0068] The engine parameters may be determined as described above in relation to the method according to the first aspect of the invention.

[0069] According to a fifth aspect of the present invention there is provided a computer readable data carrier having stored thereon a computer program according to the third aspect of the present invention.

[0070] In the following description, the invention will be explained in more detail by way of example with reference to the drawings, in which: [Brief description of the drawings]

[0071] [Figure 1] FIG. 2 shows a schematic depiction of a system according to a second embodiment of the present invention. [Diagram 2] FIG. 2 shows a flow chart of an embodiment of a method according to a first aspect of the present invention. [Figure 3a] FIG. 2 is a diagram showing an example of a DSC measurement curve calculated by a first software module. [Figure 3b] FIG. 3b shows the measurement curve shown in FIG. 3a, in which the thermal effects identified by the second software module and the associated characteristic quantities calculated by the evaluation module are displayed. [Figure 4] FIG. 2 illustrates a flow chart for training a neural network for use in an artificial intelligence engine in accordance with the present invention. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0072] FIG. 1 is a schematic depiction of a system 1 according to a second embodiment of the invention. The system comprises three measuring means 2a, 2b, 2c, each of which is operable to measure a response signal of a sample exposed to an excitation generating an observable response, said response signal being representative of a thermal effect due to said sample. For example, the measuring means 2a can be operable to perform a DSC measurement. The excitation can then comprise a variable temperature in the form of a temperature program. The observable response can comprise a heat flow caused by a difference between the heat flow into the sample of the substance and the heat flow into a known reference, the known reference being typically an empty crucible. The corresponding response signal can be a thermoelectric voltage measured by the measuring means 2a. The measuring means 2b can be operable to perform, for example, a TGA measurement and the measuring means 2c can be operable to perform, for example, a TMA measurement. Alternatively, all three measuring means 2a, 2b, 2c can be operable to perform a DSC measurement. In figure 1, each of the measurement means 2a, 2b, 2c performs a thermal analytical measurement on a respective sample 3a, 3b, 3c, which may be samples of the same material or of different materials.

[0073] Each of the measuring means 2a, 2b, 2c is operable to output first data representative of both the response signal and said excitation as a function of time. For example, in the case of the measuring means 2a configured to perform DSC measurements, the first data may include thermoelectric voltages representative of the heat flow to the sample 3a and to a reference (not shown), the temperature of the sample 3a and the temperature of the reference, all as a function of time. Preferably, the first data is time-sequenced.

[0074] Each of the measuring means 2a, 2b, 2c can communicate with a data processing means. The data processing means may be accessible by several user PCs 11a, 11b. The data processing means may include a first data processing means 4a and a second data processing means 4b that communicate with each other. The first data processing means 4a may include a CPU 5 and may be operable to execute a first software module 6, an evaluation software module 7 and, optionally, one or more other software modules 8. One of these other software modules may include an installation program for setting up the equipment, creating users and registering data on reference materials. Another one of these software modules may include a control program that represents the measuring means or measuring scale and allows creating methods and experiments for routine operation. That is, the first data processing means 4a may send commands to the measuring means 2a, 2b, 2c (see arrows A1a, A2a, A3a from the first data processing means 4a to the measuring means 2a, 2b, 2c). Alternatively, although not shown in FIG. 1, each measurement means may include a first data processing means executing a first software module.

[0075] The measuring means 2a, 2b, 2c are operable to transmit respective first data to the first data processing means 4a (see arrows A1b, A2b, A3b). Each of the measuring means 2a, 2b, 2c may comprise a transmitting unit for transmitting the first data to the first data processing means 4a. The first data processing means 4a may comprise a receiving unit for receiving the first data. The first software module 6 is configured to receive said first data as input, to calculate a thermoanalytical measurement curve enabling the identification of said thermal effects, and to output second data suitable to represent said measurement curve. For example, in the case of DSC, the measurement curve may comprise a relationship between the temperature of the probe and the heat flow. The second data may comprise data representative of this measurement curve. For example, the second data may comprise a relationship between the sample temperature T S (ti ) and heat flow Φ S (T S (t i )) pairs of data points (T S (t i ),Φ S (T S (t i ))) is the time-dependent temperature T S (t i ) as a function of t i is not only a point in time but also a reference temperature, time t i , and the material class.

[0076] The first data processing means 4a is operable to transmit second data to the second data processing means 4b (see arrow A4). The first data processing means 4a may comprise a transmitting unit for transmitting the second data to the second data processing means 4b. The second data processing means 4b may comprise a receiving unit for receiving said second data. The second data processing means 4b comprises a second software module. The second software module comprises an artificial intelligence engine 9 configured for automatic identification of thermal effects, said second software module being configured to receive as input said second data and to output third data representative of said thermal effects identified by said artificial intelligence engine 9.

[0077] The artificial intelligence engine 9 may include at least one neural network. In Fig. 1, the artificial intelligence engine 9 includes four neural networks 10a, 10b, 10c, 10d for identifying thermal effects in DSC measurement curves and, optionally, further neural networks (not shown) for identifying thermal effects in measurement curves obtained by the same or other thermal analysis measurement methods, e.g., TGA and TMA. The four neural networks 10a, 10b, 10c, 10d may be configured for identifying thermal effects of materials belonging to different material classes. For example, the first neural network 10a may be configured for identifying thermal effects of polymers, the second neural network 10b may be configured for identifying thermal effects of metals, the third neural network 10c may be a neural network for identifying general thermal effects, and the fourth neural network 10d may be a user-specific neural network as further specified below. The second software module is operative to receive the second data as input and to select one of the neural networks for automatic identification of thermal effects depending on selection data defining a material class included in the second data. For example, when the selection data is "polymeric", the first neural network 10a is selected for identification of thermal effects. Furthermore, possible selection data for the example artificial intelligence engine described above can specify "metallic", "generic" and "user specific".

[0078] Each of the neural networks 10a, 10b, 10c, 10d is configured to accept as input the second data and output third data representative of the thermal effect identified by the neural network. The third data may include an excitation range and a type of the thermal effect. For example, in the case of DSC, the third data may include a lower and upper temperature value of a temperature range, and a type of the thermal effect, such as a glass transition.

[0079] The third data is then sent to an evaluation software module 7 executed by the first data processing means 4a (see arrow A5). The second software module may include a sending unit for sending the third data to the evaluation software module 7. The first software module may include a receiving unit for receiving the third data. The evaluation software module 7 is configured to receive as input the third and fourth data suitable for representing the measurement curve and to calculate the value of at least one characteristic mass associated with the at least one thermal effect by means of the fourth data. The evaluation software module 7 may be implemented as a routine in the first software module, which may further be configured to receive the third data as input. For example, in the case of DSC, the third data may include the identified thermal effect defined by the "glass transition" as well as the lower and upper temperature values ​​of the range in which this effect occurs, and the evaluation software module 7 may be configured to calculate the onset and midpoint temperatures of the glass transition as characteristic masses.

[0080] Figure 2 shows a flow chart of an embodiment of the method according to the first aspect of the invention. In step S1, a sample of a substance is exposed to an excitation which generates an observable response. A corresponding response signal represents the thermal effect due to said sample. In the case of DSC, the sample of a substance and a reference are subjected to a temperature program. This generates an observable response, i.e. a heat flow caused by the difference between the heat flow to the sample and the heat flow to the reference. In a DSC measurement, the heat flow is measured via the thermoelectric voltage, which is the response signal. The heat flow is proportional to said thermoelectric voltage. In this way, first data is obtained, which includes both the excitation and the response signal as a function of time. That is, for a DSC measurement, the first data is obtained at a time t i , temperature T i , and the thermoelectric voltage V i The data may include time-ordered data in which

[0081] In step S2, the first data is provided to a first software module, which is executed by a data processing means, for example the first data processing means 4a shown in FIG.

[0082] In step S3, a first software module receives as input said first data and calculates from said first data a thermoelectric measurement curve that allows the identification of said thermal effect. For example, in the case of DSC, the measurement curve represents the relationship between temperature or time and heat flow. In FIG. 3, a graphical representation of a DSC curve (temperature on the x-axis, heat flow on the y-axis) calculated from first data obtained for a PET sample is depicted. The sample is heated from 30° C. to 290° C. In this temperature range, three characteristic changes are visible: a step, a peak and a dip. Each of these characteristic changes corresponds to a thermal effect. The first software module outputs second data suitable to represent the measurement curve. For example, the second data may represent pairs of data points of x and y values ​​of the measurement curve (x i ,y i ). Alternatively, the second data may include pixel data of a graphical depiction of the measurement curve.

[0083] In step S4, the second software module receives the second data. The second software module includes an artificial intelligence engine configured for automatic identification of thermal effects. As described above with reference to FIG. 1, the artificial intelligence engine may include at least one neural network. The second software module is configured to receive the second data as input and to output third data representative of the thermal effects automatically identified by the artificial intelligence engine. The third data may include an excitation range and a type of associated thermal effect. For example, for the measurement curve depicted in FIG. 3a, the third data may include the following information: glass transition in the temperature range [70° C., 100° C.], crystallization in the temperature range [120° C., 180° C.], melting in the temperature range [210° C., 270° C.].

[0084] Figures 3a and 3b show, as an example, the measured curves and the depiction of the identified thermal effects together with the associated calculated properties for the example of PET annealed at 65°C for 10 hours. The small arrows indicate the direction along the vertical axis representing the exothermic behavior of the sample. In Figures 3a and 3b, the arrows point upwards, whereby the rising curves illustrate processes in which energy is provided from the sample to its surroundings. Such exothermic processes of the sample are many chemical reactions or, in this case, crystallization. Endothermic processes in which the sample consumes energy from the surroundings, such as melting or glass transition in the illustrated cases, look like sloping curves.

[0085] In step S5, the evaluation software module receives the third data and fourth data suitable for representing the measurement curve. The fourth data is represented by pairs of x and y data points of the measurement curve (x i ,y i). The evaluation software module is configured to receive the third and fourth data as input and to calculate the value of at least one characteristic mass associated with the at least one thermal effect by means of the fourth data. The result of this calculation may be displayed on a display means, for example in combination with a measurement curve. An example of such a representation is shown in FIG. 3b, where three characteristic changes are identified in the measurement curve as "glass transition", "crystallization" and "melting" and are labeled "glass transition", "crystallization" and "melting". For the glass transition, the following characteristic masses are calculated: onset temperature ("onset"), midpoint temperature according to ISO 11357-2, 3rd Edition, published in 2020 ("Midpoint ISO"), midpoint temperature according to Richardson ("Midpoint Richardson"), glass transition step height according to Richardson ("Delta cp Richardson"). For crystallization, an integral is calculated that represents the peak area between the measured curve and the baseline in the range of crystallization. For melting, an integral is calculated that represents the peak area between the measured curve and the baseline in the range of melting.

[0086] Fig. 4 depicts a flow chart for training a neural network according to the invention. In a first step S11, a thermoanalytical measurement is performed on a training sample, i.e. the training sample is exposed to an excitation that generates an observable response. For example, in the case of DSC, a sample of the material and a reference, such as an empty crucible, are exposed to a temperature program. This results in an observable response in the form of a heat flow caused by the difference between the heat flow to the sample and the heat flow to the reference. In a DSC measurement, the heat flow may be measured via a thermoelectric voltage, which is the response signal. In this way, a first data set may be obtained, which includes both the excitation and the response signal as a function of time.

[0087] In step S12, first data are input to a first software module, which calculates a thermoanalytical measurement curve, for example a DSC curve, from said first data and outputs second training data suitable for representing said measurement curve. The thermoanalytical measurement curve allows to identify thermal effects present in a sample. The second data may for example be pairs of data points of x and y values ​​of the measurement curve (x i ,y i ).

[0088] In step S13, a human expert identifies thermal effects present in the thermal analysis measurement curve. For example, the thermal analysis measurement curve obtained by means of the first software module may be displayed on a display means. The human expert can identify the location and type of thermal effects reflected in the measurement curve. In this way, third training data may be obtained which are representative of said thermal effects and which include for example the type and location of thermal effects reflected in the measurement curve.

[0089] In step 14, a blank neural network is provided that includes at least an input layer of input neurons and an output layer of output neurons connected by weighted links. The neural network may further include one or more intermediate layers of neurons arranged in series between the input layer and the output layer. The input layer accepts the second data as input and outputs the third data.

[0090] In step S15, the blank neural network is trained using expert training data including the second and third training data. For this purpose, the weights of the neural network are set to some initial values. Then, the second training data is provided to the input layer of the neural network, and the third data is obtained at the output layer of the neural network. The third data is compared with the third training data, and the weights are adjusted until the third data is essentially equal to the third training data. As known in the art, a gradient optimization method may be used to adjust the weights. In this way, a trained neural network is generated.

[0091] In step S16, the trained neural network is deployed in a second software module including an artificial intelligence engine, which may be used for the methods, systems and computer programs according to the invention.

Claims

**Claim 1** A computer-implemented method for thermal analysis of a sample of a substance, the method comprising the step of providing first data to a first software module (6), the first data representing, as a function of time, an observable response signal of the sample exposed to an excitation that generates an observable response and the excitation, the response signal representing a thermal effect due to the sample, the first software module (6) being configured to receive the first data as an input, calculate a thermal analysis measurement curve from the first data, and output second data suitable for representing the measurement curve, the thermal analysis measurement curve enabling identification of the thermal effect, the method further comprising the step of providing the second data to a second software module, the second software module having an artificial intelligence engine (9) configured for automatic identification of the thermal effect, the second software module being configured to receive the second data as an input and output third data representing the thermal effect automatically identified by the artificial intelligence engine (9). A computer-implemented method, characterized in that it is as described above. **Claim 2** The method according to claim 1, wherein the artificial intelligence engine (9) comprises at least one neural network (10a, 10b, 10c, 10d) for automatic identification of the thermal effect, the at least one neural network (10a, 10b, 10c, 10d) comprising an input layer of input neurons for receiving the second data and an output layer of output neurons for outputting the third data, the second data being provided to the input layer. **Claim 3** The artificial intelligence engine (9) includes at least two sub-engines (10a, 10b, 10c, 10d), preferably neural networks (10a, 10b, 10c, 10d), for the automatic identification of the thermal effect. The second software module is further configured to receive selection data for selecting one of the sub-engines for the automatic identification of the thermal effect, and to provide the second data as input to the selected sub-engine (10a, 10b, 10c, 10d). This is done, in a preferred embodiment where the sub-engines (10a, 10b, 10c, 10d) are neural networks (10a, 10b, 10c, 10d), by providing the second data to the input layer of the selected neural network (10a, 10b, 10c, 10d). The selection data is provided to the second software module, according to the method of claim 1.

4. The method further includes the step of deploying at least one sub-engine to the second software module, and / or the step of removing at least one sub-engine from the second software module, and / or the step of stopping the operation of at least one sub-engine within the second software module. Preferably, the sub-engine is a neural network, according to the method of claim 2.

5. The method further includes the use of a training software module for creating a trained sub-engine, preferably a trained neural network, and the step of deploying the trained sub-engine to the second software module, whereby The training software module is configured to create the trained sub-engine by using expert training data associated with at least one training sample of a training substance. The expert training data includes a data set of second and third training data. The second training data is suitable for representing a training measurement curve resulting from the training sample. The training measurement curve enables the identification of at least one thermal training effect resulting from the training sample. The third training data represents the thermal training effect. The training software module further provides a sub - engine template that links input data to output data with content defined by a set of engine parameters, and, for most of the data set, when the second training data of the data set is input into the trained sub - engine, the engine parameters are determined such that the third data output by the trained sub - engine using the engine parameters is essentially equal to one of the third training data of the data set, thereby configuring the trained sub - engine to be created, whereby the sub - engine template is preferably a blank neural network including an input layer for receiving the second data, an output layer for outputting the third data, and preferably one or more intermediate layers, and weights representing connections between the input layer, the output layer, and preferably the intermediate layers, the engine parameters are preferably weights determined such that for most of the data set, when the second training data of the data set is received by the input layer, the third data output by the output layer is essentially equal to one of the third training data of the data set, whereby the determined weights define the trained neural network, the method according to any one of claims 1 to 4. Claim 6 The training measurement curve is obtained by applying a thermal analysis measurement result to the training sample, and / or the training measurement curve is a theoretical measurement curve corresponding to the training sample, and the third training data is obtained by human identification of the thermal training effect present in the training measurement curve, the method according to claim 5. Claim 7 The third data includes an excitation range and a type of the thermal effect, the method according to any one of claims 1 to 4. Claim 8 The method further includes providing the third data and the fourth data suitable for representing the measurement curve to an evaluation software module, and the evaluation software module is configured to receive the third data and the fourth data as inputs and calculate, based on the fourth data, a value of at least one characteristic quantity associated with the at least one thermal effect. The method according to any one of claims 1 to 4.

9. The method further includes displaying, on a display means, the thermal analysis measurement curve, and / or data associated with the excitation range of the thermal effect, and / or data associated with the type of the thermal effect, and / or a graphical depiction of the value of at least one characteristic quantity. The method according to any one of claims 1 to 4.

10. The excitation includes an excitation amount corresponding to a variable temperature, and / or an excitation amount corresponding to a variable power, and / or an excitation amount corresponding to a variable pressure, and / or an excitation amount corresponding to a variable radiation dose, and / or an excitation amount corresponding to a variable stress or strain, and / or an excitation amount corresponding to a variable gas atmosphere, and / or an excitation amount corresponding to a variable magnetic field. The method according to any one of claims 1 to 4.

11. The response includes a response amount corresponding to a temperature difference in a dynamic thermal analysis method, and / or a response amount corresponding to a heat flow in a dynamic thermal analysis method, in particular, a heat flow caused by a difference between the heat flow to a sample of the substance and the heat flow to a known reference, and / or a response amount corresponding to a difference in heating power in a dynamic power compensation thermal analysis method, and / or a response amount corresponding to a change in length in a dynamic thermomechanical analysis method, and / or a response amount corresponding to a change in weight in a dynamic thermogravimetric analysis method, and / or a response amount corresponding to a force in a dynamic mechanical analysis method, and / or a response amount corresponding to a change in length in a dynamic mechanical analysis method, and / or a response amount corresponding to a change in voltage in a dynamic dielectric analysis method. The method according to any one of claims 1 to 4.

12. A system (1) for thermal analysis of a sample (3a, 3b, 3c) of a substance, wherein the system (1) is Measuring means (2a, 2b, 2c), wherein the measuring means (2a, 2b, 2c) is adapted to measure a response signal of a sample exposed to an excitation that generates an observable response, and to output first data representing the response signal and the excitation as a function of time, the response signal representing a thermal effect due to the sample, the measuring means (2a, 2b, 2c); Data processing means (4a, 4b) having a first software module (6), the first software module (6) being configured to receive the first data as an input, calculate a thermal analysis measurement curve from the first data, and output second data suitable for representing the measurement curve, the thermal analysis measurement curve enabling the identification of the thermal effect, the data processing means (4a, 4b), the data processing means (4a, 4b) further having a second software module including an artificial intelligence engine (9) configured for automatic identification of thermal effects, the second software module being configured to receive the second data as an input and output third data representing the thermal effect identified by the artificial intelligence engine (9), a system (1) characterized in that it is so configured. Claim 13 A computer program for thermal analysis of a sample of a substance, the computer program comprising a first software module (6), the first software module (6) receiving as input first data representing a response signal of the sample exposed to an excitation that generates an observable response and the excitation as a function of time, calculating a thermal analysis measurement curve that enables identification of the thermal effect, and outputting second data suitable for representing the measurement curve, the response signal representing a thermal effect due to the sample, the computer program further comprising a second software module including an artificial intelligence engine (9) configured for automatic identification of the thermal effect, the second software module being configured to receive the second data as input and output third data representing the thermal effect identified by the artificial intelligence engine (9). A training software module of a computer program for thermal analysis of a sample of a substance, the training software module being configured to receive expert training data including a data set of second and third training data, the second training data being suitable for representing a training measurement curve resulting from a training sample, the training measurement curve enabling identification of at least one thermal training effect resulting from the training sample, the third training data representing the thermal training effect, and / or the training software module is further configured to receive second data suitable for representing a thermal analysis measurement curve enabling identification of at least one thermal training effect from the first software module according to claim 13, and to create the expert training data by being configured to create third data related from a human input representing a human identification of the at least one thermal training effect. ​ Thereby, the training software module creates a trained sub - engine, preferably a trained neural network, based on the training data, and the trained sub - engine is suitable for being deployed to the second software module according to claim 13. Thereby, the training software module is preferably a training software module that is part of the computer program according to claim 13.

15. A computer - readable data carrier storing the computer program according to claim 13.

Images