Information processing device, information processing method, and computer program

Abstract

Provided are an information processing device, an information processing method, and a computer program that make it possible to appropriately acquire a desired physical quantity through machine learning from a portion of waveform data pertaining to thermal characteristics. The information processing device 1, which performs data analysis based on waveform data pertaining to thermal characteristics through machine learning, comprises: a processor for executing a computer program; and a storage device for storing the computer program. The storage device further stores parameters used for the data analysis derived on the basis of teaching data including a plurality of pieces of waveform data to which annotations indicating peaks are respectively given. The processor receives the waveform data pertaining to the thermal characteristics, identifies the peaks of a waveform on the basis of the curvature or the variation value of the waveform data and a threshold included in the parameter, and updates the threshold on the basis of the peaks of the identified waveform data and the peaks of the teaching data.

Description

Information processing device, information processing method, and computer program

[0001] The present invention relates to an information processing device, an information processing method, and a computer program.

[0002] In materials development or product testing, it is important to know how a material behaves with temperature changes. To achieve this, a technique called thermal analysis is used, in which a material is subjected to controlled temperature changes and the physical quantities of the material are evaluated. Generally, values ​​such as heat flow, mass, and elastic modulus are measured as a function of temperature and time, which can be considered as essentially continuous values, and recorded in the form of waveform data, a type of unstructured data.

[0003] To obtain physical quantities related to the thermal properties of a material from the waveform data obtained in this way, signal processing of the waveform data is usually required. For example, one type of signal processing involves determining the temperature at which a material undergoes a phase transition when the first or higher derivatives of the waveform data take discontinuous values. Another type of signal processing involves determining the area enclosed by a peak in the heat flow near the transition temperature as the enthalpy change of the material before and after the phase transition. However, this signal processing often relies on the user's knowledge of materials and thermal analysis, and its highly personal nature has often made technology transfer difficult.

[0004] There are attempts to automate this signal processing using machine learning to reduce dependency on individual skills and simultaneously improve throughput. However, this signal processing can differ depending on the purpose of the thermal analysis. For example, should a single peak be the subject of processing to obtain enthalpy change? It is usually difficult to collect the necessary training data for machine learning to achieve sufficient accuracy for each use case and purpose.

[0005] Patent Literature 1 discloses a thermal analysis data processing device capable of accurately detecting inflection point temperatures from thermal analysis data of various substances. The device processes thermal analysis data of a sample and includes a condition setting unit that sets detection conditions for the inflection point temperature by machine learning training data including known data, which is known thermal analysis data, and correct temperatures, which are inflection point temperatures specified in advance in the known data, and a temperature detection unit that detects the inflection point temperature from the thermal analysis data of the sample in accordance with the detection conditions.

[0006] JP 2023-80556 A

[0007] The technology disclosed in Patent Document 1 uses machine learning to automate the detection of inflection point temperature, one of the physical quantities to be obtained in thermal analysis. However, Patent Document 1 does not allow for the acquisition of physical quantities other than inflection point temperature. In particular, inflection point temperature can be determined in principle based on the first-order derivative of the waveform, regardless of the use case or analytical purpose, and is relatively easy to acquire as a physical quantity. Therefore, it is applicable to physical quantities that are more difficult to obtain, such as the enthalpy change near the transition point. There is a need for an apparatus that can accurately acquire physical quantities according to the use case or analytical purpose through machine learning using a small amount of training data.

[0008] One object of the present disclosure is to provide an information processing device, an information processing method, and a computer program that enable a desired physical quantity to be appropriately acquired from a portion of waveform data related to thermal characteristics by machine learning.

[0009] In order to solve the above problem, an information processing device according to one aspect of the present invention is an information processing device that performs data analysis based on waveform data related to thermal properties using machine learning, and includes a processor that executes a computer program and a storage device that stores the computer program.The storage device further stores parameters used in the data analysis that are derived based on teacher data including multiple waveform data each of which has an annotation indicating a peak.The processor receives the waveform data related to thermal properties, identifies a peak in the waveform based on the curvature or variation value of the waveform data and a threshold value included in the parameter, and updates the threshold based on the identified peak in the waveform data and the peak in the teacher data.

[0010] According to the present invention, a waveform peak can be identified based on the curvature or variation value of the waveform data related to thermal characteristics and a threshold value included in the parameter, and the threshold value can be updated based on the identified peak of the waveform data and the peak of the teacher data.

[0011] FIG. 1 is a block diagram of an information processing apparatus according to a first embodiment. FIG. 2 is a block diagram illustrating an example of a hardware configuration of a computer. FIG. 3 is an explanatory diagram illustrating an example of input waveform data according to the first embodiment. FIG. 4 is an explanatory diagram illustrating an example of peak data according to the first embodiment. FIG. 5 is an explanatory diagram illustrating an example of a learning information setting screen according to the first embodiment. FIG. 6 is a flowchart illustrating an example of a procedure for calculating curvatures, etc., performed by a curvature, etc., calculation unit according to the first embodiment. FIG. 7 is a flowchart illustrating an example of a peak determination procedure performed by a peak determination unit according to the first embodiment. FIG. 8 is a flowchart illustrating an example of a parameter update procedure performed by a parameter update unit according to the first embodiment. FIG. 9 is an explanatory diagram illustrating an example of parameter data. FIG. 10 is an explanatory diagram illustrating an example of difference data. FIG. 11 is a block diagram of an information processing apparatus according to a second embodiment. FIG. 12 is an explanatory diagram illustrating an example of peak boundary data according to the second embodiment. FIG. 13 is a flowchart illustrating an example of a boundary determination procedure performed by a boundary determination unit according to the second embodiment. FIG. 14 is a flowchart illustrating an example of a parameter update procedure performed by a parameter update unit according to the second embodiment. FIG. 15 is a block diagram of an information processing apparatus according to a third embodiment. FIG. 16 is an explanatory diagram illustrating an example of material physical quantity data. FIG. 17 is a flowchart illustrating an example of a parameter update procedure performed by a parameter update unit according to a third embodiment. FIG. 18 is a block diagram of an information processing device according to a fourth embodiment. FIG. 19 is a flowchart illustrating an example of a robustness information calculation procedure performed by a robustness information calculation unit according to a fourth embodiment. FIG. 20A is an explanatory diagram illustrating an example of a robustness information display screen according to a fourth embodiment. FIG. 20B is an explanatory diagram illustrating another example of a robustness information display screen according to a fourth embodiment. FIG. 20C is an explanatory diagram illustrating another example of a robustness information display screen according to a fourth embodiment. FIG. 21 is a block diagram of an information processing device according to a fifth embodiment. FIG. 22 is a flowchart illustrating an example of a robustness information calculation procedure performed by a robustness information calculation unit according to a fifth embodiment. FIG. 23A is an explanatory diagram illustrating an example of a robustness information display screen according to a fifth embodiment. FIG. 23B is an explanatory diagram illustrating another example of a robustness information display screen according to a fifth embodiment. FIG. 23C is an explanatory diagram illustrating another example of a robustness information display screen according to a fifth embodiment. FIG. 24 is a block diagram of an information processing device according to a sixth embodiment. FIG. 25 is an explanatory diagram illustrating an example of input waveform data according to a sixth embodiment.FIG. 26 is a flowchart illustrating an example of a parameter update procedure performed by the parameter update unit according to the sixth embodiment.

[0012] Hereinafter, an embodiment of the present invention will be described with reference to the drawings. In this embodiment, parameters used in machine learning for thermal analysis are optimized using a small amount of waveform data. As a result, in this embodiment, machine learning is realized to obtain physical quantities corresponding to differences in use cases or analytical purposes from waveform data related to thermal properties. In this embodiment, as described below, by defining a basic algorithm and adjusting only important parameters, physical quantities corresponding to differences in use cases or analytical purposes can be obtained from a small amount of data.

[0013] In this embodiment, for example, in data analysis in thermal analysis, points where the curvature (or variation value) exceeds a threshold are identified as waveform peaks, and the identified threshold is updated based on a comparison between the peaks of the waveform data to be analyzed and the peaks of the training data (waveform). As a result, in this embodiment, for materials that undergo slow phase transitions, such as polymeric materials, parameters (thresholds) to be used in thermal analysis can be identified from a small amount of training data, enabling accurate thermal analysis.

[0014] The information processing device disclosed in this embodiment has a processor that executes a computer program and a storage device that stores the computer program, and is an information processing device that performs data analysis based on waveform data related to thermal characteristics.The storage device stores parameters used for data analysis derived based on teacher data including multiple waveform data, each of which has an annotation indicating a peak (the apex of the peak waveform).The processor receives the waveform data related to thermal characteristics, identifies the peak of the waveform based on the curvature or variation value of the waveform data and a threshold value included in the parameter, and updates the threshold based on the identified peak of the waveform data and the peak of the teacher data.

[0015] A first embodiment will be described with reference to Figures 1 to 10. Figure 1 is a block diagram of an information processing apparatus according to the first embodiment. Figure 2 is a block diagram showing the hardware configuration of a computer included in the information processing apparatus.

[0016] 1, an information processing device 1 includes, for example, computers 100-1 and 100-2 and a terminal 101. The computers 100-1 and 100-2 and the control terminal 101 are connected to a communication network 102. The communication network 102 is a facility that enables data to be sent and received via wired or wireless lines.

[0017] The computer 100-1 is a computer that can access teacher waveform data 110. The teacher waveform data 110 stores, for example, input waveform data 111 relating to thermal characteristics and peak data 112 that specifies a corresponding peak from which a physical quantity is to be acquired.

[0018] The computer 100-2 includes, for example, a learning information input receiving unit 120, a curvature etc. calculation unit 121, a peak determination unit 122, and a parameter update unit 123. The computer 100-2 is a computer that outputs the results of machine learning using teacher waveform data 110 as input, and is a computer that can access parameter data 124 and difference data 125. Hereinafter, when there is no need to distinguish between the computers 100-1 and 100-2, they may be collectively referred to as the computer 100. The terminal 101 is a control computer terminal that can access the computers 100-1 and 100-2.

[0019] Referring to FIG. 2, the computer 100 includes, for example, a processor 201, a main memory device 202, a secondary memory device 203, a network interface 204, an input device 205, and an output device 206.

[0020] The secondary storage device 203 is a device for storing data in a writable and readable manner. Data to be used in calculation processing and the results of calculation processing are stored in the secondary storage device 203.

[0021] The processor 201 is a device that reads data stored in the secondary storage device 203 into the primary storage device 202 and executes a software computer program (not shown). The curvature calculation unit 121, the peak determination unit 122, and the parameter update unit 123 in the computer 100-2 are realized by the processor 201 executing the software computer program.

[0022] The network interface 204 is a device that transmits and receives data to and from other devices via the communication network 102. The network interface 204 can transmit data resulting from processing by the processor 201 to other devices via the communication network 102. The network interface 204 can receive data to be processed by the processor 201 from other devices via the communication network 102.

[0023] The input device 205 is a device such as a keyboard or mouse that accepts user operations and acquires information input by the user. The output device 206 is a device that outputs information to a monitor display or a printer, and presents information to the user by displaying it on a screen, for example.

[0024] Returning to Fig. 1, the computer 100-1 can access teacher waveform data 110. The teacher waveform data 110 stores input waveform data 111 and peak data 112.

[0025] 3 is an explanatory diagram illustrating an example of input waveform data 111 according to Example 1. The input waveform data 111 includes, for the input waveform data 111 related to thermal characteristics, a feature 310 that describes the feature of the waveform, and a waveform descriptor 320 that describes the measurement value for each time and temperature.

[0026] The feature 310 is assigned an input waveform data ID 311, which is an individual identification number for the input waveform data 111. The feature 310 can include information used in analyzing the input waveform data, such as a temperature change computer program for performing unit conversion for the numerical values ​​stored in the waveform descriptor 320, limiting the measurement target, and the amount of sample.

[0027] The waveform descriptor 320 stores time 330-1, temperature 330-2, and measurement values ​​340, such as heat flow, at each time point. Time 330-1 and temperature 330-2 are recorded at intervals that can be considered substantially continuous values, and plotting the corresponding measurement values ​​340 results in a continuous waveform graph. Hereinafter, time 330-1 and temperature 330-2 may both be referred to as "time." The following description will use an example of a heating program that performs analysis while applying heat flow to a substance, and will explain the correspondence between high and low temperatures before and after the time. However, the present invention can also be applied to observing the thermal characteristics of a substance during cooling by appropriately interpreting high and low temperatures. The first-order and higher-order derivatives of measurement values ​​340 with respect to time 330-1 and temperature 330-2 can be essentially obtained from the difference values ​​with respect to time 330-1 and temperature 330-2. Therefore, the terms "derivative" and "difference" will be used interchangeably below. The point where the absolute values ​​of the waveform, first derivative, and higher derivatives are extremely large corresponds to the phase transition point of the substance. Calculating the peak area of ​​the waveform around the phase transition point corresponds to the "calculation of physical quantities" as the purpose of thermal analysis.

[0028] FIG. 4 is an explanatory diagram illustrating an example of peak data 112 according to the first embodiment. For input waveform data 111 identified by an input waveform data ID 311, measured values ​​340 corresponding to time 330-1 and temperature 330-2 are plotted to obtain a waveform 400. In this waveform 400, a peak location 410 to be extracted is identified by values ​​such as time 330-1 and temperature 330-2. It is not limited to peaks, and the identification of a phase transition point 420 to be extracted may also be included. Even if a peak-shaped location 430 in the waveform is not targeted for extraction depending on knowledge of the target material or the purpose of the analysis, and is not explicitly included in the peak data 112, the purpose of the analysis can be reflected.

[0029] 1, the computer 100-2 has a learning information input accepting unit 120 that accepts input of learning information. The learning information input accepting unit 120 displays necessary information on the terminal 101 to accept input from a user regarding machine learning setting information. The learning information input accepting unit 120 applies the machine learning setting information input from the terminal 101 to processing in the curvature etc. calculation unit 121 and the like.

[0030] 5 shows an example of a screen 500 for accepting input of machine learning settings. This screen 500 is generated by the learning information input accepting unit 120 and provided to the user from the terminal 101. The screen 500 may be called a learning information input screen or a settings screen. The screen 500 is displayed on the terminal 101, accepts input from the user regarding machine learning settings, and transmits the input to the curvature etc. calculation unit 121, etc.

[0031] As an example, the screen 500 may include a learning data specification field 510, a learning parameter search method specification field 520, an initial parameter determination method setting field 530, a learning parameter search range setting field 540, a loss function specification field 550, a behavior setting field when peaks are overdetected 560, a behavior setting field when peaks are underdetected 570, and a preprocessing information setting field 580.

[0032] The learning data specification field 510 is a field for selecting the teacher waveform data 110. The extraction target according to the purpose of each analysis is specified by the peak data 112, and by performing machine learning using the peak data 112, the purpose of the analysis can be appropriately reflected.

[0033] The learning parameter search method specification field 520 is a field for selecting a parameter search method for parameter updating by machine learning. The parameter update unit 123 calculates the difference between the peak candidate information calculated using the parameters specified based on this method and the peak data 112, and machine learning is performed based on this value to select optimal parameters for peak extraction. The user can specify that the learning parameter search range set in the learning parameter search range setting field 540 be searched by grid search, or can specify that parameters be searched by Bayesian optimization such as GP-UCB (Gaussian Process Upper Confidence Bound).

[0034] The initial parameter determination method setting field 530 is a field for selecting a method for determining the initial values ​​of the learning parameters. The parameter update unit 123 calculates the difference between the peak candidate information calculated using the parameters specified based on this method and the peak data 112, and machine learning is performed based on this value so that optimal parameters for peak extraction can be selected. The user can specify that the initial values ​​be determined randomly from the learning parameter search range set in the learning parameter search range setting field 540, or can specify a file containing past machine learning results to adopt parameters that were determined to be optimal in the past machine learning results as the initial values.

[0035] The learning parameter search range setting field 540 is a field for setting the upper and lower limits of the search range when searching for learning parameters. By appropriately specifying the range, the user can set it so that efficient machine learning is performed without unnecessary parameter searches. By setting different search ranges for positive and negative values ​​for variation values ​​and curvature, it is also possible to set different sensitivities for upward and downward peaks (not shown).

[0036] The loss function specification field 550 is a field for selecting a loss function to be used as an evaluation method when the parameter update unit 123 evaluates the difference between the calculated peak candidate information and the peak data 112. By specifying the loss function, the user can easily reflect the purpose of analysis by controlling the weights of the multiple differences, particularly when multiple peaks are included in one input waveform data.

[0037] The peak overdetection behavior setting field 560 and the peak underdetection behavior setting field 570 are fields for setting behavior when the number of peaks included in the peak candidate information calculated using specified parameters differs from the number of peaks included in the peak data 112, making it impossible to simply calculate the difference information between them. By specifying these setting fields, the user can assign a large penalty value to overdetection or underdetection and search for parameters that prevent overdetection or underdetection. Furthermore, the user can control the detection sensitivity and easily reflect the analysis objective by setting the value of the loss function related to overdetection or underdetection to zero.

[0038] The preprocessing information setting field 580 is a field for setting preprocessing that is performed before the curvature etc. calculation unit 121 calculates the curvature etc. of the input waveform data 111. For example, the user can set smoothing of the waveform 400. The user can set information such as the range of time 330-1 and temperature 330-2 that the smoothing refers to, or can set the information to be the subject of machine learning in this information processing device and input setting information related to the machine learning.

[0039] The user sets some or all of the following: learning data specification field 510, learning parameter search method specification field 520, initial parameter determination method setting field 530, learning parameter search range setting field 540, loss function specification field 550, behavior setting field for excessive peak detection 560, behavior setting field for insufficient peak detection 570, and preprocessing information setting field 580, and then presses the save button 501. As a result, the information set in each of the units 510-580 is applied to processing by the curvature etc. calculation unit 121 and the like. This setting information can also be stored in the secondary storage device 203.

[0040] The GUI (Graphical User Interface) shown in the screen 500 of Fig. 5 is an example and is not limited to this. The user does not necessarily need to set all setting items, and there is no prohibition on the settings being determined automatically in an appropriate manner.

[0041] 1, the computer 100-2 has a curvature etc. calculation unit 121. The curvature etc. calculation unit 121 calculates the curvature etc. of the waveform 400 related to the input waveform data 111 that has been input.

[0042] 6 is a flowchart showing an example of a procedure for calculating curvatures, etc., by the curvature, etc., calculation unit 121. The curvature, etc., calculation unit 121 acquires setting information I specified by a user operating the terminal 101, and further acquires input waveform data 111 specified by the setting information I (S601).

[0043] The curvature etc. calculation unit 121 performs preprocessing on the input waveform data 111 based on the acquired setting information I (S602). The preprocessing is smoothing of the waveform 400, etc., as specified in the preprocessing information setting field 580.

[0044] The curvature calculation unit 121 calculates the variation and curvature of the waveform 400 at each time specified by the time 330-1 and temperature 330-2 of the input waveform data 111 as curvature information (S603). As described above, the variation and curvature can be calculated using the difference between values ​​at neighboring points at each time. The curvature information calculated here does not preclude values ​​related to third-order or higher derivatives. The curvature information may be calculated from values ​​at a certain instant in time, or may be information based on changes in values ​​within a certain time range (a range of values ​​over multiple steps).

[0045] The curvature etc. calculation unit 121 transmits the curvature etc. information calculated in step S603 to the peak determination unit 122, and ends the process (S604).

[0046] Returning to Fig. 1, the computer 100-2 has a peak determination unit 122. The peak determination unit 122 calculates peak candidates based on specific parameters from the received information such as curvature.

[0047] 7 is a flowchart showing an example of a procedure for determining peak candidates by the peak determining unit 122. The peak determining unit 122 acquires setting information I specified by the user operating the terminal 101, curvature information etc. transmitted from the curvature etc. calculating unit 121, and parameter data 124 and difference data 125 (described later) (S701). The parameter data 124 is information on a small number of parameter sets that control the behavior of information processing for determining peaks, and is data related to the history of parameter sets that were targeted in past trials. The difference data 125 is data storing the difference value evaluated by the parameter updating unit 123 (described later) regarding the difference between peak candidate information calculated for parameter sets that were targeted in past trials and the peak data 112.

[0048] The peak determination unit 122 creates a learning parameter set P to be tried this time from the history contained in the parameter data 124 and the difference data 125 based on the setting information I set in the learning parameter search method specification field 520, the initial parameter determination method setting field 530, the learning parameter search range setting field 540, etc. (S702).

[0049] The peak determination unit 122 calculates threshold information for the variation and curvature of the waveform 400, which is used to identify peak candidates, from the setting information I and the created training parameter set P (S703). The parameter data 124 and the training parameter set P may include the variation and curvature thresholds themselves. The variation and curvature thresholds may be ratios of the variation and curvature thresholds to the maximum values ​​of the variation and curvature of the waveform 400 for each input waveform data 111.

[0050] The peak determination unit 122 compares the curvature information with the curvature threshold information, and extracts the time 330-1 or temperature 330-2 of the input waveform data 111 corresponding to a time having a variation value or curvature exceeding the threshold as peak candidate information (S704). When multiple nearby times are simultaneously extracted as peak candidates, the peak determination unit 122 may extract only the time with the largest value as a representative value as a peak candidate, and delete the other times from the peak candidates.

[0051] The peak determining unit 122 transmits the extracted peak candidate information to the parameter updating unit 123 and ends the process (S705).

[0052] Returning to Fig. 1, the computer 100-2 has a parameter update unit 123. The parameter update unit 123 compares the received peak candidate information with the peak data 112 to calculate a difference, stores the difference in difference data 125 as an evaluation of the learning parameter set P, and updates the parameter data 124 based on the evaluation.

[0053] 8 is a flowchart showing an example of a parameter update procedure performed by the parameter update unit 123. The parameter update unit 123 acquires setting information I specified by a user operating the terminal 101, peak candidate information transmitted from the peak determination unit 122, input waveform data 111, peak data 112, and parameter data 124 (S801).

[0054] The parameter update unit 123 evaluates the difference between the time at which a peak included in the peak candidate information is designated and the time at which a peak stored in the peak data 112 is designated, based on the setting information I designated in the loss function designation field 550, the behavior setting field for peak overdetection 560, the behavior setting field for peak underdetection 570, and the preprocessing information setting field 580, and calculates difference information (S802).

[0055] The parameter update unit 123 stores the currently tried learning parameter set P in the parameter data 124, and stores the corresponding difference information together with the input waveform data ID 311 related to the input waveform data 111 in the difference data 125 (S803).

[0056] The parameter update unit 123 refers to the difference data 125 and updates the parameter data 124 as necessary (S804). That is, the parameter update unit 123 refers to the difference data 125 related to the parameter data 124 that was previously tried for the same or similar input waveform, and if it is determined that the difference data related to the learned parameter set P is suitable for actual peak candidate extraction, for example, if the difference data related to the learned parameter set P has a smaller value than the difference data 125, it performs an update process, such as setting a flag to that effect, for the learned parameter set P stored in the parameter data 124, and then terminates the process (S804). The parameter update unit 123 can refer to the feature 310 when determining the identity or similarity of the input waveforms.

[0057] Returning to FIG. 1, the computer 100 - 2 can access the parameter data 124 and the difference data 125 .

[0058] 9 is an explanatory diagram illustrating an example of the parameter data 124 according to Example 1. Each parameter set stored in the parameter data 124 is assigned a parameter set ID 910, which is an individual identification number.

[0059] For the parameter sets stored in the parameter data 124, the parameter update unit 123 refers to the difference data 125, and for those parameter sets that are determined to be suitable for actual peak candidate extraction based on the past trial history, this is recorded in the optimum flag 920 together with information on what waveform or substance, or what experimental conditions the parameter set is suitable for.

[0060] Here, assuming that extraction should be performed using a single parameter set for all substances and all experimental conditions, it is not prohibited to use a data format in which only a single parameter set ID is recorded separately and the optimal flag 920 is not attached to each record of the parameter data 124.

[0061] The parameter data 124 stores the values ​​of parameters 930, which are a small number of parameters that control the behavior of peak candidates and are the subject of search. As described above for the learning parameter P, the parameters 930 can be a threshold value 930-1 for the variation value and a threshold value 930-2 for the curvature. They can also be the ratios of the threshold values ​​for the variation value and curvature to the maximum values ​​of the variation value and curvature for the waveform 400 associated with each piece of input waveform data 111. If set in the preprocessing information setting field 580, a parameter 930-3 related to preprocessing can be included in the parameter data 124.

[0062] It is possible that an optimal parameter set cannot be uniquely determined for one waveform or one substance. In this case, an optimal flag may be set for each of multiple parameter sets, and the results calculated using the multiple parameter sets may be provided to the user, who may then select one of the parameter sets.

[0063] 10 is an explanatory diagram showing an example of the difference data 125 according to Example 1. The difference data 125 associates the parameter set ID 910 assigned when the learning parameter P was stored in the parameter data 124 with the input waveform data ID 311 assigned to the input waveform data that was the target when the parameter update unit 123 evaluated the difference information, and stores the value of the difference information as a difference value 1010.

[0064] As described above, in the first embodiment, when extracting peaks from waveform data related to thermal properties obtained by thermal analysis, the learning target is limited to only a small number of parameters, such as curvature thresholds, that control the behavior of the extraction. Therefore, according to the first embodiment, a user can perform machine learning that enables extraction of peaks that reflect the analysis objectives of each use case by simply creating a relatively small number of peak data 112. The waveform data related to thermal properties may be one or more of thermal analysis DSC (Differential Scanning Calorimetry) data, thermogravimetry (TG) data, thermomechanical analysis (TMA) data, dynamic mechanical analysis (DMA) data, and differential thermal analysis (DTA) data.

[0065] Example 2 will be described with reference to Figures 11 to 14. In each of the following examples, including this example, differences from the examples described earlier will be mainly described. Example 2 shows an example of an information processing device 1A for performing machine learning that enables the extraction of peaks that reflect the analysis objective for each use case when defining peak boundaries related to the identification of a baseline for further calculating the area swept by an extracted peak candidate.

[0066] 11 is a block diagram of an information processing apparatus 1A according to the second embodiment. In the second embodiment, the teacher waveform data 110 stored in the computer 100-1 includes peak boundary data 1102 in addition to the input waveform data 111. The computer 100-2 includes a learning information input receiving unit 120, a curvature etc. calculating unit 121, a peak determining unit 122, parameter data 124, and difference data 125, as well as a boundary determining unit 1122 and a parameter updating unit 1123.

[0067] 12 is an explanatory diagram showing an example of peak boundary data 1102 according to Example 2. The peak boundary data 1102 can record, for each peak location 410 to be extracted from the waveform 400 according to the peak data 112, a peak start time 1211 indicating the point at which the peak should be analyzed, and a peak end time 1212 indicating the point at which the peak should be analyzed. Physical quantities related to phase transition, such as the solidification onset and end point, can be calculated from the waveform 400, the peak location 410, the peak start time 1211, and the peak end time 1212. For example, the area enclosed by the waveform 400 and a baseline 1221 connecting the peak start time 1211 and the peak end time 1212 with a line segment or a smooth curve is calculated as the enthalpy change related to the phase transition.

[0068] 11, the computer 100-2 includes a boundary determining unit 1122. Based on the peak candidate information determined by the peak determining unit 122, the boundary determining unit 1122 calculates a peak boundary candidate for the peak candidate.

[0069] 13 is a flowchart showing an example of a boundary determination procedure by the boundary determination unit 1122. The boundary determination unit 1122 acquires setting information I specified by a user operating the terminal 101, a learning parameter set P, peak candidate information, and input waveform data 111 (S1301). The peak determination unit 122 of the second embodiment transmits the learning parameter set P together with the peak candidate information. The boundary determination unit 1122 acquires the peak candidate information and the learning parameter set P transmitted from the peak determination unit 122.

[0070] The boundary determination unit 1122 divides the waveform 400 related to the input waveform data 111 into segments for each peak candidate time included in the peak candidate information, and creates segmented waveform data (S1302). The boundary determination unit 1122 may create multiple types of segmented waveform data by dividing the data for each order of differentiation that the peak determination unit 122 refers to when calculating the peak candidates.

[0071] The boundary determination unit 1122 calculates a section feature characterizing each section for each of the divided section waveform data (S1303). The boundary determination unit 1122 includes parameters required for calculating the section feature in a training parameter set P, and does not prevent the parameters from being subject to optimization by machine learning implemented by the information processing system under machine learning settings included in the setting information I. In one embodiment, the section feature is the slope of the waveform in the section. In another embodiment, the section feature is the slope of the waveform within a range further limited within the section to an area where the curvature is equal to or less than a certain value. In yet another embodiment, the curvature threshold referenced when limiting the range within the section is included in the training parameter set P and is subject to optimization by machine learning implemented by the information processing system 1A.

[0072] The boundary determination unit 1122 calculates threshold information for the variation or curvature for each piece of segment waveform data from the calculated segment feature, the learning parameter set P, and the setting information I (S1304). In one embodiment, the threshold information for the segment curvature, etc., uses a range obtained by increasing or decreasing a certain value with respect to the waveform slope within a range further limited to an area within the segment where the curvature is equal to or less than a certain value as a variation threshold. In another embodiment, the certain value to be increased or decreased is included in the learning parameter set P and is subject to optimization by machine learning realized by the information processing system 1A.

[0073] The boundary determination unit 1122 extracts peak boundary candidates based on the results of comparing the calculated segmental curvature etc. threshold information with the variation value or curvature value at each time of the segmented waveform data, and creates peak boundary candidate information (S1305). The peak boundary candidate times are compared, and the later one is set as peak start time 1211 for the peak that divides the waveform at the end time of the segmented waveform data, and the earlier one is set as peak end time 1212 for the peak that divides the waveform at the start time of the segmented waveform data. The boundary determination unit 1122 may include processing such as extracting only the time furthest or closest to the peak candidate time, or the average time of those times, from among the times at which the variation value or curvature value exceeds the segmental curvature etc. threshold value, as the peak boundary.

[0074] The boundary determining unit 1122 determines whether the peak start time and peak end time have been identified for all peak candidates (S1306). If it is determined that the peak start time and peak end time have been identified for all peak candidates, the boundary determining unit 1122 proceeds to step S1308.

[0075] If it is determined that the peak start time and peak end time could not be identified for all peak candidates, the boundary determination unit 1122 modifies the threshold information for the partition curvatures, etc. (S1307). The modification includes changing the values ​​of the parameters referenced when calculating the threshold information for the partition curvatures, etc., and then recalculating the threshold information for the partition curvatures, etc. If it is determined that the peak start time and peak end time could not be identified for all peak candidates, the boundary determination unit 1122 may assign a large value as the difference information 1010, send the information to the parameter update unit 1123, and end the process (not shown).

[0076] The boundary determining unit 1122 transmits the created peak boundary candidate information to the parameter updating unit 1123, and ends the process (S1308).

[0077] 11, the computer 100-2 has a parameter update unit 1123. The parameter update unit 1123 compares the peak boundary candidate information with the peak boundary data 1102, calculates the difference, stores the difference in difference data 125 as an evaluation of the learning parameter set P, and updates the parameter data 124 based on the evaluation.

[0078] 14 is a flowchart showing an example of a parameter update procedure performed by the parameter update unit 1123. The parameter update unit 1123 acquires setting information I specified by a user operating the terminal 101, input waveform data 111, peak boundary data 1102, parameter data 124, difference data 125, and peak boundary candidate information (S1401).

[0079] The parameter updating unit 1123 uses the acquired peak boundary candidate information as the peak start time 1211 and peak end time 1212 related to the peak. The parameter updating unit 1123 calculates a physical quantity related to the waveform 400 and compares the calculated physical quantity with a physical quantity related to the waveform 400 calculated using the start time 1211 and end time 1212 stored in the peak boundary data 1102. The parameter updating unit 1123 evaluates the difference obtained as a result of the comparison based on setting information I specified in the loss function specification field 550, the behavior setting field 560 when peaks are excessively detected, the behavior setting field 570 when peaks are underdetected, and the preprocessing information setting field 580, and calculates difference information (S1402).

[0080] In one embodiment, the comparison of the physical quantities is performed by directly comparing the start time 1211 or the end time 1212. In another embodiment, the comparison of the physical quantities is performed by comparing the area enclosed by the waveform 400 and a baseline 1221 connecting the time of the coagulation start point or the coagulation end point, or the peak start time 1211 and the peak end time 1212 with a line segment or a smooth curve. In yet another embodiment, the comparison is performed by comparing a weighted linear sum of one or more physical quantities.

[0081] The parameter update unit 1123 updates the parameter data 124 and the difference information 125 based on the calculated difference information, similarly to steps S803 and S804 in the parameter update unit 123 of the first embodiment.

[0082] In this way, in Example 2, when physical quantities related to phase transitions are extracted from waveform data related to thermal properties obtained by thermal analysis, the learning target is limited to only a small number of parameters that control the behavior of extraction, such as the curvature threshold and the section curvature threshold. As a result, Example 2 enables a user to extract peak boundaries that reflect the analysis objective for each use case simply by creating a relatively small number of peak boundary data 1102, and enables machine learning to calculate physical quantities based on the extracted peak boundary data.

[0083] 15 to 17 , a third embodiment will be described. In the third embodiment, when a user is unable to create peak data 112 or peak boundary data 1102 and is given only external knowledge about the physical quantities of a material to be analyzed, machine learning is performed to determine peak boundary candidates by comparison with the physical quantities, and parameters suitable for the peak boundary candidates are obtained.

[0084] 15 is a block diagram of an information processing apparatus 1B according to a third embodiment. In the third embodiment, the teacher waveform data 110 included in the computer 100-1 stores only input waveform data 111, and does not include peak data 112 or peak boundary data 1102 corresponding to all of the input waveform data 111. The computer 100-1 according to the third embodiment includes material data 1510. The material data 1510 stores material physical quantity data 1502 that records physical quantities related to the substance corresponding to the input waveform data 111. In the third embodiment, the computer 100-2 includes a learning information input receiving unit 120, a curvature etc. calculation unit 121, a peak determination unit 122, a boundary determination unit 1122, parameter data 124, and difference data 125, as well as a parameter update unit 1523.

[0085] FIG. 16 is an explanatory diagram showing an example of material physical quantity data 1502 according to Example 3. The material physical quantity data is assigned an input waveform data ID 311 that identifies the corresponding material and the experimental conditions under which the measurement was performed. Furthermore, the material physical quantity data 1502 is assigned a peak ID 1610, which is an individual identification number that specifies which phase transition point the physical quantity is associated with. Furthermore, the material physical quantity data 1502 stores physical quantity values ​​1621 for the corresponding peaks for each type of physical quantity. Each record of the material physical quantity data 1502 does not necessarily have to record values ​​for all physical quantities 1621 for each identified peak. The peak ID 1610 is not limited to a single value, and does not prevent peaks from being identified by a time range of the waveform 400 for identifying the calculated peak candidate.

[0086] 15, the computer 100-2 has a parameter update unit 1523. The parameter update unit 1523 optimizes a parameter set for extracting peak boundary candidates so as to reproduce the physical quantities recorded in the material physical quantity data 1502.

[0087] 17 is a flowchart showing an example of a parameter update procedure by the parameter update unit 1523. The parameter update unit 1523 acquires setting information I specified by a user operating the terminal 101, input waveform data 111, material physical quantity data 1502, parameter data 124, difference data 125, peak candidate information, and peak boundary candidate information (S1701).

[0088] The parameter update unit 1523 uses the acquired peak boundary candidate information as the peak start time 1211 and peak end time 1212 related to the peak, calculates a physical quantity related to the waveform 400, compares it with the physical quantity value 1621 recorded in the material physical quantity data 1502 identified by the corresponding input waveform data ID 311 and peak ID 1610, evaluates the difference based on the setting information I specified in the loss function specification field 550, the behavior setting field 560 when peaks are excessively detected, the behavior setting field 570 when peaks are insufficiently detected, and the preprocessing information setting field 580, and calculates difference information (S1702).

[0089] Thereafter, the parameter update unit 1523 updates the parameter data 124 and the difference data 125 based on the calculated difference information in the same process as the parameter update unit 1123 according to the second embodiment (S803, S804).

[0090] Thus, according to Example 3, even if peak boundary data is not obtained and only external knowledge regarding the physical quantities of the material to be analyzed is given, machine learning can be performed to determine peak boundary candidates by comparing with the physical quantities, and parameters suitable for the peak boundary candidates can be obtained.

[0091] 18 to 20C, a fourth embodiment will be described. In the fourth embodiment, an example of an information processing device 1C is shown, which, when extracting peak candidates from new waveform data using the parameter set for peak candidate extraction optimized in the first embodiment, simultaneously displays the result of extracting peak candidates as a robustness evaluation result even when a small perturbation is applied to the value of the parameter set.

[0092] 18 is a block diagram of an information processing apparatus 1C according to Example 4. In Example 4, waveform data 1810 provided in a computer 100-1 stores input waveform data 111. A computer 100-2 includes a learning information input receiving unit 120, a curvature etc. calculating unit 121, a peak determining unit 122, parameter data 124, and a robustness information calculating unit 1823.

[0093] 19 is a flowchart illustrating an example of a robustness information calculation procedure performed by the robustness information calculation unit 1823 according to Example 4. The robustness information calculation unit 1823 acquires setting information I, curvature information, input waveform data 111, and parameter data 124 (S1901).

[0094] The robustness information calculation unit 1823 references the feature 310 of the input waveform data 111 and the optimization flag 920 of the parameter data 124 to obtain a parameter set P to be used for calculating peak candidates, and applies perturbations to the values ​​of various parameters in the parameter set P to create one or more parameter sets P' for robustness evaluation (S1902). Perturbations do not necessarily have to be applied to all parameters in the parameter set P, but can be applied to only a small number of parameters selected by any method. There are also no limitations on the magnitude and distribution of the perturbations.

[0095] The robustness information calculation unit 1823 creates a curvature etc. threshold information set by calculating and combining curvature etc. threshold information based on each of the robustness evaluation parameter sets P' in a manner similar to that in which the peak determination unit 122 in Example 1 calculates curvature etc. threshold information based on the setting information I and the learning target parameter set P (S1903).

[0096] The robustness information calculation unit 1823 extracts peak candidate information by comparing each of the curvature etc. threshold information included in the created curvature etc. threshold information set with the received curvature etc. information, in a manner similar to that in which the peak determination unit 122 according to Example 1 compares the curvature etc. information with the curvature etc. threshold information to extract peak candidate information, and calculates a peak candidate information set by combining the extracted peak candidate information (S1904).

[0097] The robustness information calculation unit 1823 associates each record of the calculated peak candidate information set with each robustness evaluation parameter set P'. The robustness information calculation unit 1823 references the setting information I and creates robustness information necessary for displaying a robustness information display screen (described later) on the terminal (S1905). The robustness information is information that associates the difference between the values ​​of each record of the parameter set P that is optimal for analyzing the target input waveform data 111 and the robustness evaluation parameter set P' with the time of the peak candidate recorded in the peak candidate information. In one embodiment, the robustness information is created using only records of the peak candidate information set whose difference from the peak candidate based on the parameter set P exceeds a threshold set in the setting information I, rather than using all records of the robustness evaluation parameter set P'.

[0098] The robustness information calculation unit 1823 transmits the generated robustness information and ends the process (S1906). The transmitted robustness information is output to the terminal 101 as a robustness information display screen and is perceived by the user.

[0099] 20A is an explanatory diagram showing an example of a robustness information display screen according to Example 4. A screen 2000A displaying robustness information may include, for example, an input waveform identification field 2001 and a perturbation variation display field 2011. When one piece of input waveform data 111 includes multiple peak candidates, it is also possible to display multiple sets of this information for each peak candidate on the same screen or on multiple screens that can be switched by operation.

[0100] The input waveform identification field 2001 displays information related to the characteristics 310 of the target input waveform data 111 and information identifying the target peak candidate. These can be identifiers assigned in the same manner as the input waveform data ID 311 and peak ID 1610.

[0101] The perturbation variation display field 2011 displays robustness information. In one embodiment, for each robustness evaluation parameter set P′ obtained by perturbing one or two parameters selected by any method from the parameter set P, the perturbation variation display field 2011 visually displays the difference between the calculated peak candidate time and the peak candidate time associated with the parameter set P. In one embodiment, the difference is displayed using a shade of color or the like, allowing the magnitude of the change in the peak candidate time in response to the change in the parameter to be visually recognized from the degree of change in the shade of color.

[0102] 20B is an explanatory diagram showing another example of the robustness information display screen according to Example 4. As an example, a screen 2000B displaying robustness information displays, on a command line, an input waveform identification field 2001, parameter set variation information 2021, and peak candidate variation information 2022. Furthermore, when one piece of input waveform data 111 includes multiple peak candidates, it is also possible to display multiple sets of this information for each peak candidate on the same screen or multiple screens that can be switched by operation.

[0103] The parameter set variation information 2021 is information relating to the robustness evaluation parameter set P' among the robustness information. The peak candidate variation information 2022 is information relating to the difference with the corresponding peak candidate information or peak candidate relating to the parameter set P. These do not necessarily have to be displayed for all records of the robustness evaluation parameter set P', and may be displayed only for representative records extracted by an appropriate method.

[0104] 20C is an explanatory diagram showing another example of the robustness information display screen according to Example 4. As an example, a screen 2000C displaying robustness information displays an input waveform identification field 2001, a set safe range display field 2031, a parameter acceptable range display field 2032, and the like on the terminal 101 and is perceived by the user. When one piece of input waveform data 111 includes multiple peak candidates, it is also possible to display multiple sets of this information for each peak candidate on the same screen or multiple screens that can be switched by operation.

[0105] The set safe range display field 2031 is information relating to the range of allowable error for the peak candidate. In one embodiment, this range is determined by the set information I.

[0106] The parameter acceptable range display field 2032 displays boundary value information indicating whether the difference between the peak candidate time calculated for the robustness evaluation parameter set P′ and the peak candidate time associated with the parameter set P deviates from the acceptable error range for the peak candidate, along with the value of the robustness evaluation parameter set P′. In one embodiment, the robustness evaluation parameter set P′ is generated by perturbing only one parameter from the parameter set P, and the parameter acceptable range display field 2032 displays information about the boundary value indicating whether the deviation occurs. In one embodiment, the parameter acceptable range display field 2032 also displays an evaluation of the magnitude of the boundary value displayed in this field. The evaluation may be the result of classification into a finite number of classes based on the magnitude of the boundary value, which is determined based on the setting information I. The evaluation may be the result of classification into a finite number of classes based on the magnitude of the boundary value, which is determined based on the distribution of boundary values ​​obtained by performing the same process on similar input waveform data extracted based on the feature 310 of the input waveform data 111.

[0107] As described above, according to the fourth embodiment, when peak candidates are extracted from new waveform data using a parameter set for peak candidate extraction, the result of extracting peak candidates can be obtained as a robustness evaluation result even when a small perturbation is applied to the values ​​of the parameter set. In particular, by employing the perturbation variation display field 2011 and the parameter acceptable range display field 2032 as a display method for calling attention to perturbations that have low robustness and may cause problems in terms of experimental reproducibility, the user can limit the experiments to be considered to those that may contain problems, without making a judgment on the majority of experiments with few problems.

[0108] 21 to 26C , a fifth embodiment will be described. In the fifth embodiment, an example of an information processing device is shown for extracting peak boundary candidates from new waveform data using the parameter set for peak boundary candidate extraction optimized in the second embodiment, and simultaneously extracting peak boundary candidates and calculating physical quantities even when a small perturbation is applied to the values ​​of the parameter set, and displaying the results as robustness evaluation results.

[0109] 21 is a block diagram of an information processing apparatus 1D according to Example 5. In Example 5, a computer 100-2 includes a robustness information calculation unit 2123 in addition to a learning information input receiving unit 120, a curvature etc. calculation unit 121, a boundary determination unit 1122, a peak determination unit 122, and parameter data 124.

[0110] 22 is a flowchart illustrating an example of a robustness information calculation procedure performed by the robustness information calculation unit 2123 according to the fifth embodiment. The robustness information calculation unit 2123 acquires information through processing similar to that of the robustness information calculation unit 1823 according to the fourth embodiment (S1901), creates a robustness evaluation parameter set P′ (S1902), creates a curvature etc. threshold information set (S1903), and calculates a peak candidate information set (S1904). However, in the fifth embodiment, the robustness evaluation parameter set P′ to which perturbations are applied is not limited to only parameters included in the parameter set P, and does not preclude, for example, directly applying perturbations to the peak start time and peak end time calculated using the parameter set P.

[0111] The robustness information calculation unit 2123 associates each record of the robustness evaluation parameter set P' with each record of the peak candidate information set created based on it, and transmits them to the boundary determination unit 1122 together with the setting information I. The boundary determination unit 1122 transmits peak boundary candidate information for the set of records in the same manner as in Example 2, and then ends the process. The robustness information calculation unit 2123 combines the peak boundary candidate information to create a peak boundary candidate information set (S2205).

[0112] The robustness information calculation unit 2123 uses each record of the peak boundary candidate information set as the peak start time 1211 and peak end time 1212 related to the peak, calculates the physical quantities related to the waveform 400, and creates a physical quantity set (S2206).

[0113] The robustness information calculation unit 2123 associates the robustness evaluation parameter set P' with records in the physical quantity set calculated based on each record, and generates robustness information from them (S2207). The robustness information is information that associates the difference between the values ​​of each record of the parameter set P that is optimal for analyzing the target input waveform data 111 and the robustness evaluation parameter set P' with the difference in the values ​​of the physical quantities calculated based on them. In one embodiment, the robustness information is generated using only records of the robustness evaluation parameter set P' for which the difference between the physical quantities exceeds the safety range threshold included in the setting information I, rather than using all records of the robustness evaluation parameter set P'.

[0114] The robustness information calculation unit 2123 transmits the generated robustness information and ends the process (S2208). The transmitted robustness information is output to the terminal 101 as a robustness information display screen and is perceived by the user.

[0115] FIG. 23A is an explanatory diagram showing an example of a robustness information display screen according to Example 5. The robustness information display screen 2300A may include, for example, an input waveform identification field 2001 and a perturbation-variation value display field 2311. When one input waveform data 111 includes multiple peak candidates, it is possible to display multiple sets of information for each peak candidate on the same screen or multiple screens that can be switched by operation. It is also possible to simultaneously display multiple sets of perturbation-variation value display fields 2311 for multiple physical quantities on the same screen or multiple screens that can be switched by operation. In Example 5, the input waveform identification field 2001 does not prevent the display of the value of the physical quantity associated with the parameter set P, along with information related to the feature 310 of the target input waveform data 111 and information identifying the target peak candidate.

[0116] The perturbation variation display field 2311 displays robustness information. In one embodiment, for each robustness evaluation parameter set P′ in which one or two parameters selected by an arbitrary method from the parameter set P are perturbed, the field visually displays the difference between the time of the calculated physical quantity and the time of the physical quantity related to the parameter set P. In another embodiment, the difference is displayed using a shade of color or the like, so that the magnitude of the change in the time of the peak candidate corresponding to the change in the parameter can be visually recognized from the degree of change in the shade of color. In yet another embodiment, the field visually displays only parameter regions in which the difference in physical quantity deviates from a range determined based on the setting information I, or only parameter regions in which the difference does not deviate from the range determined based on the setting information I.

[0117] 23B is an explanatory diagram showing another example of the robustness information display screen according to Example 5. As an example, a screen 2300B displaying robustness information displays an input waveform identification field 2001, parameter set variation information 2321, and physical quantity variation information 2322 on a command line. When one piece of input waveform data 111 includes multiple peak candidates, it is also possible to display multiple sets of this information for each peak candidate on the same screen or multiple screens that can be switched by operation.

[0118] The parameter set variation information 2321 is information related to the robustness evaluation parameter set P' among the robustness information. The physical quantity variation information 2322 is information related to the corresponding physical quantity or information related to the difference with the physical quantity related to the parameter set P. These do not necessarily have to be displayed for all records of the robustness evaluation parameter set P', and may be displayed only for representative records extracted by an appropriate method.

[0119] 23C is an explanatory diagram showing another example of the robustness information display screen according to Example 5. As an example, the robustness information display screen 2300C displays an input waveform identification field 2001, a set safe range display field 2331, a parameter acceptable range display field 2332, and the like on the terminal 101 and is perceived by the user. In a case where one piece of input waveform data 111 includes multiple peak candidates, it is also possible to display multiple sets of this information for each peak candidate on the same screen or multiple screens that can be switched by operation.

[0120] The set safe range display field 2331 is information relating to the range of allowable error for the physical quantity. In one embodiment, this range is determined by the set information I.

[0121] The physical quantity allowable range display field 2332 displays boundary value information indicating whether a difference between a physical quantity calculated for a robustness evaluation parameter set P′ and a physical quantity calculated based on the parameter set P deviates from an allowable error range for the physical quantity, along with the value of the robustness evaluation parameter set P′. In one embodiment, the robustness evaluation parameter set P′ is generated by perturbing only one parameter from the parameter set P, and the physical quantity allowable range display field 2332 displays information on the boundary value indicating whether the deviation occurs. In one embodiment, the physical quantity allowable range display field 2332 also displays an evaluation of the magnitude of the boundary value displayed in this field. The evaluation may be a result of classification into a finite number of classes based on the magnitude of the boundary value, which is determined based on the setting information I. The evaluation may be a result of classification into a finite number of classes based on the magnitude of the boundary value, which is determined based on the distribution of boundary values ​​obtained by performing the same process on similar input waveform data extracted based on the feature 310 of the input waveform data 111.

[0122] As described above, according to the fifth embodiment, when peak boundary candidates are extracted from new waveform data using a parameter set for peak boundary candidate extraction, the robustness is directly evaluated based on the physical quantity that is the ultimate target of the thermal analysis when a small perturbation is applied to the values ​​of the parameter set, and the result is obtained as a robustness evaluation result. Furthermore, by directly perturbing the values ​​of the peak boundary candidates, it is possible to evaluate the robustness of the extracted peak boundary candidates themselves with respect to the measurement of the physical quantity. In particular, by employing the perturbation variation display field 2311 and the physical quantity acceptable range display field 2332 as display methods for drawing attention to perturbations that are low in robustness and may cause problems in terms of, for example, experimental reproducibility, the user can limit the experiments to be considered to those that may contain problems, without making judgments based on the majority of experiments with few problems.

[0123] 24 to 26, a sixth embodiment will be described. In the sixth embodiment, when a user is unable to create peak data and only input waveform data relating to similar materials or similar experimental conditions is obtained, machine learning related to determining peak candidates is performed based on the assumption that these data have similar peak candidates, and parameters suitable for the peak candidates are obtained.

[0124] 24 is a block diagram of an information processing apparatus 1E according to Example 6. In Example 6, the teacher waveform data 110 included in the computer 100-1 stores a plurality of pieces of input waveform data 2411. The computer 100-2 includes a learning information input receiving unit 120, a curvature etc. calculating unit 121, a peak determining unit 122, parameter data 124, and difference data 125, as well as a parameter updating unit 2423.

[0125] 25 is an explanatory diagram showing an example of input waveform data 2411 according to Example 6. The input waveform data includes a feature 310 that describes the feature of the waveform, and a waveform descriptor 320 that describes measurement values ​​for each time and temperature. Of these, the waveform descriptor 320 is configured with the same content as the input waveform data 111 according to Example 1.

[0126] The feature 310 according to Example 6 is assigned an input waveform data ID 311, which is an individual identification number for the input waveform data 2411. The feature 310 may include information used in analyzing the input waveform data, such as a temperature change computer program for performing unit conversions for the numerical values ​​stored in the waveform descriptor 320, limiting the measurement target, and sample amounts. The feature 310 according to Example 6 includes main material group information 2512, which is information necessary for characterizing the relationship between the input waveform data 2411 to determine whether peak candidates exist at similar times. In one embodiment, the main material group information 2512 includes information related to the composition of the main material of the target substance associated with the feature 310 for the input waveform data 2411, as information sufficient to determine whether peak candidates exist at similar times. In another embodiment, the main material group information 2512 includes thermal history information related to thermal analysis.

[0127] 26 is a flowchart showing an example of a parameter updating procedure by the parameter updating unit 2423 according to the sixth embodiment. The parameter updating unit 2423 acquires setting information I, peak candidate information, and parameter data 124 for each piece of input waveform data 2411 (S2601). The parameter updating unit 2423 does not necessarily need to acquire peak candidate information for all pieces of input waveform data 2411. For a specific waveform to be learned among the input waveform data 2411, the parameter updating unit 2423 may refer to main material group information 2512 and acquire information limited to waveforms having peak candidates at similar times.

[0128] The parameter update unit 2423 refers to the main material group information 2512 for a specific waveform to be learned from the input waveform data 2411, and calculates difference information based on the setting information I from the difference between the times of the peak candidates related to the peak candidate information for waveforms that are estimated to have peak candidates at similar times (S2602).

[0129] The parameter update unit 2423 stores the currently tried learning parameter set P in the parameter data 124, and stores the corresponding difference information in the difference data 125 together with the input waveform data ID 311 relating to the specific waveform to be learned from the input waveform data 2411 (S2603).

[0130] The parameter update unit 2423 performs the same processing as the parameter update unit 123 in Example 1 by referring to the differential data 125 and, if necessary, performing an update process on the learning parameter set P stored in the parameter data 124, and then terminates the processing (S804).

[0131] Thus, according to Example 6, even if the user is unable to create peak data and only input waveform data relating to similar materials or similar experimental conditions is available, machine learning for determining peak candidates can be performed based on the assumption that these data have similar peak candidates, and parameters suitable for the peak candidates can be obtained.

[0132] The present invention is not limited to the above-described embodiments, and includes various modifications and equivalent configurations within the spirit and scope of the appended claims. For example, the above-described embodiments have been described in detail to clearly explain the present invention, and the present invention is not necessarily limited to configurations including all of the described configurations. Furthermore, part of the configuration of one embodiment may be replaced with the configuration of another embodiment. Furthermore, the configuration of another embodiment may be added to the configuration of one embodiment. Furthermore, part of the configuration of each embodiment may be added to, deleted from, or replaced with other configurations.

[0133] Furthermore, each of the aforementioned configurations, functions, processing units, processing means, etc. may be realized in hardware, for example by designing some or all of them as integrated circuits, or may be realized in software, by a processor interpreting and executing a computer program that realizes each function.

[0134] Information such as computer programs, tables, files, etc. that realize each function can be stored in a storage device such as a memory, hard disk, or SSD (Solid State Drive), or in a recording medium such as an IC (Integrated Circuit) card, SD card, or DVD (Digital Versatile Disc).

[0135] In addition, the control lines and information lines shown are those that are considered necessary for the explanation, and do not necessarily show all the control lines and information lines that are necessary for implementation. In reality, it can be assumed that almost all components are interconnected.

[0136] The above-described embodiment discloses a configuration that can be expressed as follows.

[0137] (Representation 1) An information processing device that performs data analysis based on waveform data related to thermal characteristics using machine learning, comprising a processor that executes a computer program and a storage device that stores the computer program, wherein the storage device further stores parameters used in the data analysis derived based on teacher data including multiple waveform data each of which has an annotation indicating a peak, and wherein the processor receives waveform data related to thermal characteristics, identifies a peak of the waveform based on a curvature or variation value of the waveform data and a threshold value included in the parameter, and updates the threshold value based on the identified peak of the waveform data and the peak of the teacher data.

[0138] (Representation 2) An information processing device according to Representation 1, wherein each of the plurality of waveform data of the teacher data is further provided with an annotation indicating a peak boundary, and the processor divides the waveform data into a plurality of segment waveform data based on the identified peak, calculates segment features characterizing the segment waveform data using the parameters, identifies peak boundaries of the waveform data based on the calculated segment features, and updates the parameters based on the peak boundaries of the identified waveform and the peak boundaries of the teacher data.

[0139] (Representation 3) An information processing device according to Representation 1 or 2, wherein the processor identifies multiple peaks in waveform data based on multiple types of parameters set to different values, and performs analytical robustness evaluation from the identified multiple peaks.

[0140] (Representation 4) An information processing device described in any one of Representations 1-3, wherein the processor identifies multiple peak boundaries in waveform data based on multiple types of parameters set to different values, calculates multiple peak physical quantities based on the identified multiple peak boundaries, and evaluates the robustness of the analysis from the calculated multiple peak physical quantities.

[0141] (Representation 5) An information processing device that performs data analysis based on waveform data related to thermal characteristics using machine learning, comprising a processor that executes a computer program and a storage device that stores the computer program, wherein the storage device further stores parameters used in the data analysis derived based on teacher data including multiple waveform data, and wherein the processor receives multiple waveform data related to thermal characteristics, identifies peaks of the multiple waveforms based on the curvature or variation value of the multiple waveform data and a threshold value included in the parameter, and updates the threshold value based on the relationship between the peaks of the identified multiple waveform data.

[0142] (Representation 6) An information processing device that performs data analysis based on waveform data related to thermal properties using machine learning, comprising: a processor that executes a computer program; and a storage device that stores the computer program. The storage device further stores parameters used in the data analysis derived based on teacher data including a plurality of waveform data, each of which is annotated to indicate a physical quantity corresponding to a peak. The processor receives a plurality of waveform data related to thermal properties, identifies peaks in the plurality of waveforms based on curvatures or variation values ​​of the plurality of waveform data and a threshold value included in the parameters, divides the plurality of waveform data into a plurality of segmented waveform data based on the identified peaks, calculates segment features that characterize the segmented waveform data using the parameters, identifies peak boundaries of the waveform data based on the calculated segment features, calculates peak physical quantities based on the identified plurality of peak boundaries, and updates the parameters based on the calculated peak physical quantities, the peak boundaries of the identified waveforms, and the peak boundaries of the teacher data.

[0143] (Representation 7) An information processing device according to any one of Representations 1 to 6, wherein the data analysis based on waveform data relating to the thermal characteristics is thermal analysis of a polymer material.

[0144] (Representation 8) In the information processing device according to Representation 8, the waveform data relating to the thermal characteristics may be one or more of DSC (Differential Scanning Calorimetry) data of thermal analysis, thermogravimetry (TG) data, thermomechanical analysis (TMA) data, dynamic mechanical analysis (DMA) data, and differential thermal analysis (DTA) data.

[0145] (Representation 9) An information processing method for performing data analysis based on waveform data relating to thermal properties using machine learning with an information processing device, wherein the information processing device includes a processor that executes a computer program and a storage device used by the processor, and the storage device stores the computer program and parameters used in the data analysis derived based on teacher data including a plurality of waveform data each annotated to indicate a peak, and the processor identifies a peak of the waveform based on the curvature or variation value of the waveform data and a threshold value included in the parameter, and updates the threshold value based on the identified peak of the waveform data and the peak of the teacher data.

[0146] (Representation 10) A computer program that causes a processor executing the computer program to perform a determination process of identifying a peak of the waveform based on the curvature or variation value of waveform data related to thermal characteristics and a threshold value included in the parameter, and an update process of updating the threshold value based on the identified peak of the waveform data and the peak of the teacher data.

[0147] (Representation 11) An information processing device that performs data analysis based on waveform data related to thermal characteristics using machine learning, comprising a processor that executes a computer program and a storage device that stores the computer program, wherein the storage device further stores parameters used in the data analysis derived based on training data including multiple waveform data each annotated to indicate a peak, and wherein the processor receives the waveform data related to thermal characteristics and performs predetermined processing based on some of the information contained in the waveform data, thereby optimizing settings used in machine learning to analyze the waveform data.

[0148] (Representation 12) An information processing device in which the specified processing is a processing described in any one of Representations 1-9.

[0149] 1, 1A, 1B, 1C, 1D, 1E: information processing device, 100: computer, 101: terminal, 102: communication network, 110: teacher waveform data, 111: input waveform data, 112: peak data, 120: learning information input receiving unit, 121: curvature etc. calculation unit, 122: peak determination unit, 123: parameter update unit, 124: parameter data, 125: difference data, 201: processor, 202: main memory device, 203: Sub-storage device, 204: network interface, 205: input device, 206: output device, 1102: peak boundary data, 1122: boundary determination unit, 1123: parameter update unit, 1502: material physical quantity data, 1510: material data, 1523: parameter update unit, 1810: waveform data, 1823: robustness information calculation unit, 2123: robustness information calculation unit, 2411: input waveform data, 2423: parameter update unit

Claims

1. An information processing device that performs data analysis based on waveform data related to thermal characteristics through machine learning, comprising: a processor that executes a computer program; and a storage device that stores the computer program, wherein the storage device further stores parameters used in the data analysis derived based on teacher data including a plurality of waveform data, each of which is annotated to indicate a peak, wherein the processor receives waveform data related to thermal characteristics, identifies a peak of the waveform based on a curvature or variation value of the waveform data and a threshold value included in the parameters, and updates the threshold value based on the identified peak of the waveform data and the peak of the teacher data.

2. An information processing device as described in claim 1, wherein each of the multiple waveform data of the teacher data is further provided with an annotation indicating a peak boundary, and the processor divides the waveform data into multiple segment waveform data based on the identified peaks, calculates segment features characterizing the segment waveform data using the parameters, identifies peak boundaries of the waveform data based on the calculated segment features, and updates the parameters based on the peak boundaries of the identified waveform and the peak boundaries of the teacher data.

3. An information processing device according to claim 1, wherein the processor identifies a plurality of peaks in the waveform data based on a plurality of types of the parameters set to different values, and performs an analysis robustness evaluation from the identified plurality of peaks.

4. An information processing device according to claim 2, wherein the processor: identifies a plurality of peak boundaries in the waveform data based on a plurality of types of the parameters set to different values; calculates a plurality of peak physical quantities based on the identified plurality of peak boundaries; and evaluates the robustness of the analysis from the calculated plurality of peak physical quantities.

5. An information processing device that performs data analysis based on waveform data relating to thermal characteristics through machine learning, comprising: a processor that executes a computer program; and a storage device that stores the computer program, wherein the storage device further stores parameters used in the data analysis derived based on teacher data including a plurality of waveform data, wherein the processor receives multiple waveform data relating to thermal characteristics, identifies peaks of the multiple waveforms based on a curvature or variation value of the multiple waveform data and a threshold value included in the parameters, and updates the threshold value based on the relationship between the peaks of the identified multiple waveform data.

6. An information processing device that performs data analysis based on waveform data relating to thermal characteristics through machine learning, comprising: a processor that executes a computer program; and a storage device that stores the computer program, wherein the storage device further stores parameters used in the data analysis derived based on teacher data including a plurality of waveform data, each of which is annotated to indicate a physical quantity corresponding to a peak, wherein the processor receives a plurality of waveform data relating to thermal characteristics, identifies peaks of the plurality of waveforms based on a curvature or variation value of the plurality of waveform data and a threshold value included in the parameters, divides the plurality of waveform data into a plurality of segmented waveform data based on the identified peaks, calculates segment features characterizing the segmented waveform data using the parameters, identifies peak boundaries of the waveform data based on the calculated segment features, calculates a peak physical quantity based on the identified plurality of peak boundaries, and updates the parameters based on the calculated peak physical quantity, the peak boundaries of the identified waveform, and the peak boundaries of the teacher data.

7. An information processing device according to any one of claims 1 to 6, wherein the data analysis based on the waveform data relating to the thermal characteristics is a thermal analysis of a polymer material.

8. An information processing device according to claim 7, wherein the waveform data relating to the thermal characteristics may be any one or more of thermal analysis DSC (Differential Scanning Calorimetry) data, thermogravimetry (TG) data, thermomechanical analysis (TMA) data, dynamic mechanical analysis (DMA) data, and differential thermal analysis (DTA) data.

9. An information processing method for performing data analysis based on waveform data relating to thermal properties through machine learning using an information processing device, wherein the information processing device includes a processor that executes a computer program and a storage device used by the processor, and the storage device stores the computer program and parameters used in the data analysis derived based on teacher data including a plurality of waveform data each of which is annotated to indicate a peak, and the processor identifies a peak of the waveform based on a curvature or variation value of the waveform data and a threshold value included in the parameters, and updates the threshold value based on the identified peak of the waveform data and the peak of the teacher data.

10. A computer program that causes a processor that executes the computer program to execute a determination process for determining a peak of the waveform based on a curvature or variation value of waveform data related to thermal characteristics and a threshold value included in the parameters, and an update process for updating the threshold value based on the determined peak of the waveform data and a peak of the teacher data.

Images