State estimation device, state estimation method, and state estimation program
The state estimation system uses machine learning to convert production data into feature vectors, enabling real-time estimation of chemical substance states during production, addressing the challenge of quality control without process disruption.
Patent Information
- Application Number
- JP2023508241
- Authority / Receiving Office
- JP · JP
- Patent Type
- Patents
- Current Assignee / Owner
- Filing Date
- 2021-03-24
- Publication Date
- 2024-11-26
- Estimated Expiration
- 2041-03-24
AI Technical Summary
Existing methods, such as those described in Patent Documents 1 and 2, struggle to accurately estimate the state of a chemical substance during its production process without disrupting the production flow.
A state estimation system that includes an acquisition unit, an extraction unit, and an estimation unit, utilizing machine learning to convert time-series data from production sensors into feature vectors, allowing for real-time estimation of the chemical substance's state by comparing current data with pre-learned models.
Enables real-time estimation of the chemical substance's state during production, ensuring quality control without halting the process, and providing timely feedback for process adjustments.
Smart Images

Figure 0007589798000001 
Figure 0007589798000002 
Figure 0007589798000003
Abstract
Description
[Technical field]
[0001] The present invention relates to a technique for estimating the state of a chemical substance in its production process. [Background technology]
[0002] It is important to confirm that normal processing is being performed in a generating device in a chemical plant, or that a product is being generated in accordance with standards. However, it is often difficult to directly confirm the internal state of the generating device during the generating process. Also, for example, when a product is generated using a chemical reaction, if the generating process is stopped midway to confirm the state of the target chemical substance to be generated, sufficient characteristics are often not obtained even if the process is restarted. Patent Documents 1 and 2 disclose technologies for determining the state of a target object.
[0003] Patent Document 1 relates to an anomaly detection method for detecting anomalies in a plant. In the monitoring method of Patent Document 1, measurement data measured in the plant is used as input data, and anomalies are detected using a machine-learned learning model. Patent Document 2 discloses an example of a data processing method for detecting anomalies using similarities in time-series data. [Prior art documents] [Patent documents]
[0004] [Patent Document 1] International Publication No. 2011 / 086805 [Patent Document 2] International Publication No. 2020 / 049666 Summary of the Invention [Problem to be solved by the invention]
[0005] However, in Patent Documents 1 and 2, it is difficult to estimate the state of the product during its production process.
[0006] An object of the present invention is to provide a state estimation system etc. that solves the above-mentioned problems. [Means for solving the problem]
[0007] In order to solve the above problems, the state estimation device of the present invention includes an acquisition unit, an extraction unit, an estimation unit, and an output unit. The acquisition unit acquires first time series data related to the production environment of the target chemical substance. The extraction unit extracts features of the first time series data. The estimation unit estimates the state of the target chemical substance using an estimation model that has been machine-learned to learn the relationship between the state of the target chemical substance in the production process and the features of second time series data related to the production environment, based on the features of the first time series data. The output unit outputs the state estimated by the estimation unit.
[0008] The state estimation method of the present invention acquires first time series data related to the production environment of a target chemical substance. The state estimation method of the present invention extracts feature quantities of the first time series data. The state estimation method of the present invention estimates the state of the target chemical substance using an estimation model that has been machine-learned to determine the relationship between the state of the target chemical substance in the production process and the feature quantities of second time series data related to the production environment, based on the feature quantities of the first time series data. The state estimation method of the present invention outputs an estimated state.
[0009] The program recording medium of the present invention records a state estimation program. The state estimation program causes a computer to execute a process of acquiring first time series data related to the production environment of a target chemical substance. The state estimation program causes a computer to execute a process of extracting a feature amount of the first time series data. The state estimation program causes a computer to execute a process of estimating a state of the target chemical substance using an estimation model that has machine-learned the relationship between the state of the target chemical substance in the production process and the feature amount of second time series data related to the production environment, based on the feature amount of the first time series data. The state estimation program outputs an estimated state. Effect of the Invention
[0010] According to the present invention, the state of a chemical substance in the production process can be estimated even when the production is in progress. [Brief description of the drawings]
[0011] [Figure 1] 1 is a diagram showing an outline of a configuration of a first exemplary embodiment of the present invention. [Diagram 2] 1 is a diagram illustrating an example of a configuration of a state estimation device according to a first embodiment of the present invention. [Diagram 3] FIG. 2 is a diagram illustrating an example of an operation flow of the state estimating device of the first exemplary embodiment of the present invention. [Figure 4] FIG. 2 is a diagram illustrating an example of an operation flow of the state estimating device of the first exemplary embodiment of the present invention. [Diagram 5] FIG. 2 is a diagram illustrating an example of time-series data of a measurement result in the first embodiment of the present invention. [Figure 6] FIG. 2 is a diagram showing an example of a display screen in the first embodiment of the present invention. [Figure 7] FIG. 2 is a diagram showing an example of a display screen in the first embodiment of the present invention. [Figure 8] FIG. 4 is a diagram illustrating an example of a configuration of a state estimation device according to a second embodiment of the present invention. [Figure 9] FIG. 11 is a diagram illustrating an example of an operation flow of a state estimation device according to a second exemplary embodiment of the present invention. [Figure 10] FIG. 13 is a diagram illustrating an example of another configuration of the embodiment of the present invention. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS
[0012] A first embodiment of the present invention will be described in detail with reference to the drawings. FIG. 1 is a diagram showing an outline of the configuration of a state estimation system of this embodiment. The state estimation system of this embodiment includes a state estimation device 10, a sensor 20, and a terminal device 30. A plurality of sensors 20 are included. The state estimation device 10 and each sensor 20 are connected via a network. Furthermore, the state estimation device 10 and the terminal device 30 are connected via a network.
[0013] The state estimation system of this embodiment is a system that estimates a characteristic value of a product in a production process of a chemical substance by using measurement data of a sensor 20. The state estimation system of this embodiment acquires time-series measurement data for a predetermined period of time from a sensor 20 attached inside or outside a production device in the production process of the product, and estimates a characteristic value in the production process of the product by using the acquired measurement data. In the following, a product whose state is to be estimated in the production process is also referred to as a target chemical substance.
[0014] The state estimation system of this embodiment generates reference data in advance using time-series data of the measurement results by the sensor 20 when a product was generated in the past and characteristic values of the product at the stage when the generation of the product is completed. The state estimation system estimates the characteristic values of the product by using the similarity between the time-series data measured in the generation process of the product and the reference data.
[0015] The product is, for example, a granular object (hereinafter referred to as "granular material"). In the case of a granular material, the characteristic value is, for example, the particle size. The granular material is, for example, an oxygen absorber, a desiccant, an abrasive, a resin, a medicine, or a powdered food product. The product may be an object with other properties, such as a liquid. The characteristic value may be viscosity, light transmittance, or chromaticity. The characteristic value may indicate a distribution of a physical quantity that indicates a characteristic. The following explanation will be given taking as an example a case in which the particle size of a granular material is estimated using the measurement results of the sensor 20.
[0016] The data items of the generation environment measured by the sensor 20 are set using physical quantities that change depending on the state and characteristics of the granular material during the generation process of the granular material. The data items of the generation environment measured by the sensor 20 are set by selecting one or more physical quantities from, for example, vibration, pressure, temperature, load of the stirring device, light transmittance inside the generation device, and sound inside the device. The data items measured by the sensor 20 may be set using physical quantities other than those mentioned above. Furthermore, sensors 20 measuring the same physical quantity may be installed at multiple locations.
[0017] The following describes the configuration of the state estimation device 10. Fig. 2 is a diagram showing an example of the configuration of the state estimation device 10. The state estimation device 10 includes an acquisition unit 11, an extraction unit 12, an estimation unit 13, a data management unit 14, a model generation unit 15, a storage unit 16, an input unit 17, and an output unit 18.
[0018] The acquisition unit 11 acquires, when generating the reference data, time series data of the measurement results measured by the multiple sensors 20 when the granular material was generated in the past, and data on the particle size at the final stage. That is, the acquisition unit 11 acquires multidimensional time series data. The final stage refers to a predetermined time period including the time when the particle size of the granular material reaches a design value, that is, the target value, and the generation is terminated. The predetermined time period is set in advance as a length at which the time series data of the measurement results of the sensors 20 reflects the characteristics of the properties. The acquisition unit 11 associates the acquired time series data of the measurement results with the data on the particle size at the final stage and stores them in the storage unit 16. The acquisition unit 11 also acquires time series data of the measurement results of the multiple sensors 20 during the generation process. The acquisition unit 11 stores the acquired time series data of the measurement results in the storage unit 16.
[0019] When generating the reference data, the extraction unit 12 extracts feature quantities of the time series data from the time series data at a predetermined time in the initial stage and the final stage. Here, the initial stage refers to, for example, the period from the start of generation until a predetermined time has elapsed. The predetermined time is set in advance as a length suitable for detecting a variation in the characteristic value of the granular matter, or is set arbitrarily by the operator.
[0020] The extraction unit 12 extracts data from the time series data of the measurement results during past manufacturing from the data at the time when generation started until a predetermined time has passed as time series data of the initial stage. If the measurement data fluctuates greatly immediately after the start, the start time of the initial stage may be set to a time when a preset time has elapsed from the start of generation. Furthermore, the extraction unit 12 extracts the final data from the time series data of the measurement results during past generation, that is, the data from a time a predetermined time before the data when generation was completed to the final data, as time series data of the final stage.
[0021] In the generation process, the extraction unit 12 extracts, as time series data at a target time point for particle size estimation, data from a predetermined time before the last acquired data to the last acquired data from the time series data stored in the storage unit 16. The extraction unit 12 also extracts feature quantities from the time series data extracted in the generation process.
[0022] The estimation unit 13 converts the extracted time series data of the measurement results for the predetermined time into a feature vector indicating the features of the time series data for the predetermined time, using an estimation model generated by machine learning.
[0023] The estimation unit 13 converts time series data for a predetermined time into a feature vector by using an estimation model previously generated by machine learning to convert the time series data into a real vector and then converting the real vector into a binary vector. The estimation unit 13 converts the time series data for the predetermined time into a feature vector using the estimation model to extract a feature amount of the time series data for the predetermined time. The generation of the estimation model will be described later.
[0024] A real vector is a vector in which the values of each dimension are real numbers. A binary vector representing a feature vector is a vector in which the values of each dimension are either one of two values, such as 1 and -1, or 0 and 1.
[0025] The estimation model used by the estimation unit 13 to convert data is configured to convert S×T pieces of numerical data into an n-dimensional binary feature vector, where S is the number of sensors, T is the number of time points, and n is the number of dimensions of the binary vector. The number of time points is the number of time points at which data is used for conversion by the estimation model among the times at which time-series data is measured within a specified time. When the number of data within a specified time is greater than the number of time points set in the estimation model, the extraction unit 12 extracts data for the number of time points set in the estimation model from the measurement data, and then performs conversion by the estimation model.
[0026] The estimation unit 13 uses the feature vector of the reference data read out from the memory unit 16 and the feature vector converted from the time series data of the measurement results during the generation process to estimate the particle size at the current time, i.e., at the time when the time series data was measured during the generation process.
[0027] The estimation unit 13 estimates the particle size at the current time using the similarity between the feature vectors of the measurement result at the initial stage of the reference data and the current time, and the similarity between the measurement result at the current time and the feature vector at the final stage of the reference data. The estimation unit 13 calculates the particle size at the current time from, for example, the Euclidean distance between the feature vectors at the initial stage of the reference data and the current time, the Euclidean distance between the feature vectors at the current time and the final stage of the reference data, and the particle size at the final stage of the reference data. The estimation unit 13 may calculate the distance between the feature vectors by a method other than the Euclidean distance as long as it can calculate the distance between the feature vectors in the feature amount space. The estimation unit 13 may calculate the distance between the feature vectors by, for example, the Hamming distance.
[0028] When generating the reference data, the data management unit 14 stores the data of the feature vector at the initial stage, the data of the feature vector at the final stage, and the data of the particle size at the final stage in the storage unit 16 in association with each other. The data of the feature vector at the initial stage and the data of the feature vector at the final stage may be further stored in association with the time series data before conversion. The reference data is generated, for example, for each set value of the manufacturing conditions and the particle size. The reference data is generated for each set value of the manufacturing conditions and the particle size, using the time series data measured during manufacturing and the particle size at the final stage. The reference data may be set for each manufacturing condition.
[0029] When estimating the particle size in the generation process, the data management unit 14 reads out the data of the feature vector in the initial stage, the data of the feature vector in the final stage, and the data of the particle size in the final stage, which are used for estimating the particle size, from the storage unit 16. The data management unit 14 identifies reference data that meets the conditions inputted, for example, by the operator through the terminal device 30, and reads out the data from the storage unit 16. The data management unit 14 may also read out from the storage unit 16 the reference data in which the time-series data measured in the generation process is similar to the data in the initial stage.
[0030] The model generation unit 15 generates an estimation model by machine learning, which is used when the estimation unit 13 converts time series data at a predetermined time into a feature vector. The model generation unit 15 generates the estimation model by machine learning using a recurrent neural network, for example. The model generation unit 15 generates the estimation model by the method disclosed in International Publication No. 2020 / 049666, for example.
[0031] The model generation unit 15 performs machine learning using multiple time-series measurement data of a predetermined time as learning data, and generates a data estimation model that is a learned model. The model generation unit 15 performs machine learning so that multiple learning data are converted into multiple real vectors that maintain the relative similarity between the multiple learning data. That is, the model generation unit 15 performs machine learning so that mutually similar learning data are converted into mutually similar real vectors, and mutually dissimilar learning data are converted into mutually dissimilar real vectors. The model generation unit 15 stores the generated data of the estimation model in the storage unit 16.
[0032] For example, an estimation model is generated for each generation device, and reference data is generated for each set value of generation conditions and particle size. The estimation model can be used even if the set values of generation conditions and particle size are different, as long as the number of sensors 20, the items to be measured, and the number of time points to be extracted from the time-series data are the same. Therefore, by generating an estimation model in advance for each type of generation device and the installation form of the sensor 20, and generating reference data for each product to be generated, it becomes possible to estimate the particle size of granular matter during the generation process.
[0033] The memory unit 16 stores data of the machine-learned estimation model generated by the model generation unit 15. The memory unit 16 stores time-series data of the measurement results of the sensor 20 acquired by the acquisition unit 11. The memory unit 16 also stores a feature vector converted from the time-series data of the measurement results in the initial stage, a feature vector converted from the time-series data of the measurement results in the final stage, and the particle size in the final stage as reference data in association with each other. The reference data also stores information on the generation conditions and the target value of the particle size.
[0034] The input unit 17 acquires input data input to the terminal device 30 by an operation of an operator from the terminal device 30. The input unit 17 may acquire input data input by an operation of an operator from an input device connected to the state estimation device 10.
[0035] The output unit 18 outputs the particle size estimation result to the terminal device 30. The output unit 18 may output the particle size estimation result to a display device (not shown) connected to the state estimation device 10.
[0036] The processes in the acquisition unit 11, the extraction unit 12, the estimation unit 13, the data management unit 14, the model generation unit 15, the input unit 17, and the output unit 18 are implemented by, for example, a CPU (Central Processing Unit). The processing in the acquisition unit 11, the extraction unit 12, the estimation unit 13, the data management unit 14, the model generation unit 15, the input unit 17, and the output unit 18 may be performed in another information processing device connected via a network.
[0037] The storage unit 16 is configured using, for example, a hard disk drive. The storage unit 16 may be configured with other types of storage devices such as a non-volatile semiconductor storage device or a combination of multiple types of storage devices. The storage unit 16 may be provided on a storage device connected to the state estimation device 10. The storage unit 16 may also be provided on a storage device controlled by an information processing device connected via a network.
[0038] The sensor 20 is of a type that corresponds to the physical quantity to be measured. The sensor 20 measures the corresponding physical quantity inside or outside the production device, and sends the measurement result to the state estimation device 10. The sensor 20 is installed, for example, inside the production chamber of the production device or the flow path of the product, so as to measure the physical quantity of the production environment. The production environment refers to, for example, the atmosphere in the production chamber. Also, the physical quantity of the production environment refers to, for example, the temperature in the production chamber. The physical quantity of the production environment may include the physical quantity of the product. The physical quantity of the production environment may be, for example, an item whose value changes as the chemical reaction of the product progresses, such as the torque of a stirring device or the flow rate in a pipe.
[0039] The terminal device 30 displays on a display device (not shown) display data of the particle size estimation result acquired from the state estimating device 10. In addition, the terminal device 30 transmits the manufacturing conditions and the target particle size value input by an operator's operation to the state estimating device 10 as input data.
[0040] The operation of the state estimation system of this embodiment will be described below.
[0041] 3, the acquisition unit 11 of the state estimation device 10 acquires time-series data of measurement data of the sensor 20 when a granular material was previously manufactured and data on the particle size at the final stage (step S11). The acquisition unit 11 also acquires the generation conditions when the granular material was generated.
[0042] The acquiring unit 11 acquires, for example, via a network, time series data of measurement data from the sensor 20 when the granular material was previously produced, data on the particle size at the final stage, and production conditions, which are stored in a production management server (not shown). The time series data of measurement data from the sensor 20 when the granular material was previously produced, data on the particle size at the final stage, and production conditions may be input to the terminal device 30 by an operator's operation, and acquired from the terminal device 30. The acquiring unit 11 stores in the memory unit 16 the acquired time series data of measurement data from the sensor 20 when the granular material was previously produced, data on the particle size at the final stage, and production conditions.
[0043] When the time series data of the measurement data and the data of the particle size at the final stage are stored in the storage unit 16, the extraction unit 12 extracts time series data of the measurement results at a predetermined time in the initial stage and the final stage from the stored data (step S12). The extraction unit 12 extracts feature values from the time series data at the predetermined time. The extraction unit 12 extracts data for a predetermined number of time points for a predetermined item of the measurement data obtained by measuring the generation environment as feature values. When the number of data is greater than the predetermined number of time points, the extraction unit 12 extracts data for the predetermined number of time points. For example, the extraction unit 12 extracts data for the predetermined number of time points from the time series data for the predetermined time so that the time intervals of the extracted data are uniform.
[0044] Fig. 5 is a diagram showing a schematic example of time series data measured by four sensors. Fig. 5 shows the time series data of the measurement results of sensor A, sensor B, sensor C, and sensor D. The extraction unit 12 extracts data within the dotted lines indicated as the start stage and the end stage in Fig. 5 as time series data for a predetermined time. The horizontal axis of Fig. 5 shows time, and the vertical axis shows a schematic change in the measurement value.
[0045] In Fig. 3, when the feature amount is extracted, the estimation unit 13 converts the time series data of the measurement results in the initial stage and the final stage into a real vector using an estimation model, and further converts the real vector into a binary vector to convert it into a feature vector (step S13). After converting into a feature vector, the estimation unit 13 associates the data of the feature vector in the initial stage and the feature vector in the final stage with the data of the particle size in the final stage and the generation conditions, and stores them in the storage unit 16 as reference data (step S14). The data of the particle size in the final stage is also used as information on the target value of the particle size when selecting the reference data when the particle size is estimated. The information on the target value of the particle size in the reference data may be a set value that was targeted when the measurement data that is the basis of the reference data was measured, or a value inputted via the terminal device 30 by the operator.
[0046] When the feature vector data and the final stage particle size data are stored in the memory unit 16, if there is unconverted measurement result time series data (Yes in step S15), the state estimation device 10 returns to step S12 and performs the process of converting the unconverted measurement result time series data into a feature vector. When the conversion process has been completed for all the acquired measurement data (No in step S15), the state estimation device 10 ends the operation of generating the reference data.
[0047] Next, the operation of the state estimating device 10 when estimating a particle size during the generation process of a granular object will be described.
[0048] When starting the generation of granular material, the input unit 17 acquires input data of the selection result of reference data according to the generation conditions and the target particle size value input to the terminal device 30 by the operator's operation from the terminal device 30. When the input data of the selection result of reference data is acquired, the data management unit 14 reads out the corresponding reference data from the storage unit 16.
[0049] 4, the acquisition unit 11 acquires time series data of measurement results during the generation process of granular matter from the sensor 20 (step S21). When the time series data is acquired, the extraction unit 12 extracts feature quantities from the time series data for a predetermined time period, from data at past times to the most recently acquired data, and the time series data for a predetermined time at the present time. When the feature quantities are extracted, the estimation unit 13 uses the time series data for the predetermined time at the present time as input data, and converts the time series data into a real vector using an estimation model, and further converts it into a binary vector to convert it into a feature vector (step S22).
[0050] When the feature of the time series data at the current time is converted into a feature vector, the estimation unit 13 calculates the distance in the feature space between the current time and the initial stage, and between the current time and the final stage, using the feature vector converted from the time series data at the current time and the feature vectors at the initial stage and the final stage (step S23). After calculating the distance in the feature space, the estimation unit 13 estimates the particle size at the current time using each calculated distance and the particle size data at the final stage included in the reference data (step S24).
[0051] The estimation unit 13, for example, calculates the ratio of the distance between the initial stage and the current time and between the current time and the final stage, and estimates the particle size at the current stage using the particle size at the final stage by using this ratio as the degree of progress of granular material generation. For example, when the distance between the initial stage and the current time is A, the distance between the current time and the final stage is B, the particle size at the initial stage is 0, and the particle size at the final stage is R, the estimation unit 13 calculates the particle size at the current stage using the formula (A / (A+B))×R. When the particle size at the initial stage is R, I , the final particle size is R F And as the story progresses When the particle size increases, R = R F -R I In addition, if the particle size in the initial stage is R I , the final particle size is R F When the particle size decreases with the progress, R = R I -R F R I It becomes.
[0052] When the particle size is estimated, the estimation unit 13 determines whether the particle size has reached a reference value. The estimation unit 13 determines that the particle size has reached the final stage when the particle size at the current time is, for example, equal to or greater than the reference value, and determines that the particle size is in the intermediate stage when it is less than the reference value. When the granular material is a product generated by breaking down a large mass, the estimation unit 13 determines that the particle size at the current time is equal to or less than the reference value, and determines that the particle size is in the intermediate stage when the distance is greater than the reference value. The estimation unit 13 may also calculate the distance between the feature vector converted from the current time series data and the feature vector of the final stage, and determine that the particle size has reached the final stage when the distance is within the reference value, and determine that the particle size is in the intermediate stage when the distance is greater than the reference value.
[0053] If the particle size estimated in step S24 does not reach the standard and is identified as being in an intermediate stage (No in step S25), the output unit 18 outputs data on the particle size estimation result to the terminal device 30 (step S27). Upon receiving the particle size estimation result, the terminal device 30 displays the particle size estimation result on a display device (not shown). When the data on the particle size estimation result is output, the state estimation device 10 executes again the process of obtaining the time series data at the current time point in step S21, and continues estimating the particle size in the generation process.
[0054] FIG. 6 is a diagram showing a schematic example of a display screen of the particle size estimation result. In the example of FIG. 6, the target value of the particle size is shown as the set particle size, and the estimated value of the particle size is shown as the current value. The right side of FIG. 6 shows an example in which the current measurement values of the sensors A, B, C, and D are displayed. The sensor measurement values are added to the estimation result by the state estimation device 10 and output, for example. The sensor measurement values may also be displayed as time series data of the measurement results used when estimating the particle size. When the sensor measurement values are displayed as time series data, the final stage time series data used to generate the reference data may also be displayed together, so that the difference from the current measurement data can be visually recognized.
[0055] Fig. 7 is a schematic diagram showing an example of a display screen in which a diagram of the particle size estimation result is further displayed on the display screen of Fig. 6. The upper left corner of Fig. 7 shows an estimated state of the granular material generated using the current particle size estimation result as the current estimated state. The lower left corner of Fig. 7 shows the initial and final stage states, and further shows where the current estimated state is located between the initial and final stages so that it can be visually recognized. By displaying the current estimated state in this way, the operator managing the generation process can more easily recognize the current state.
[0056] In step S25 of Figure 4, when the particle size has reached the standard and is determined to be in the final stage (Yes in step S25), the output unit 18 sends information indicating that the final stage has been reached and the estimated particle size as estimation result data to the terminal device 30 (step S26).
[0057] When the terminal device 30 receives the estimation result data including the information indicating that the final stage has been reached, it displays the information indicating that the final stage has been reached and the particle size estimation result on a display device (not shown). The worker can end the production of the granular product by checking the information indicating that the final stage has been reached. The information indicating that the final stage has been reached may also be output to a control device of the device, which then ends the granular material production process. In step S26, when the information indicating that the final stage has been reached and the estimated particle size as the estimation result data are output, the state estimation device 10 ends the operation related to the particle size estimation process.
[0058] In the above description, the data management unit 14 reads out the reference data according to the input result from the storage unit 16, but the reference data having similar time series data at the early stage of the generation process may be read out from the storage unit 16. In such a configuration, the time series data at the early stage of the generation process is acquired by the acquisition unit 11 and converted into a binary feature vector by the extraction unit 12. The estimation unit 13 calculates the distance between the feature vector obtained by converting the time series data at the early stage of the generation process and the feature vector at the early stage stored as the reference data, and identifies the reference data having a similar early stage. The estimation unit 13 estimates the particle size at the current time point using the identified reference data and the feature vector converted from the measurement data acquired in the generation process.
[0059] The model generating unit 15 may use the generation conditions of the product as input data when generating an estimation model by machine learning. When the generation conditions are used when generating an estimation model, the estimation unit 13 can estimate the particle size from the generation conditions and the feature quantities of the time-series data of the generation process, even without acquiring the selection result of the generation conditions input by the operation of the operator in the generation process. As the generation conditions of the product, for example, one or more items among the pressure in the generation device, the temperature in the generation device, the input amount of the raw material, the input speed of the raw material, the input pressure of the raw material, the stirring speed, and the stirring torque are used. The generation conditions of the product may be items other than those mentioned above.
[0060] In the above description, the particle size at the current time point is estimated based on the feature vector converted from the time series data of two sections, the initial stage and the final stage, during past generation, as a reference, but a reference stage may be set between the initial stage and the final stage. In such a configuration, the extraction unit 12 extracts time series data for a predetermined time between the initial stage and the final stage and converts it into a feature vector. The data management unit 14 associates the converted feature vector with the feature vectors of the initial stage and the final stage as a feature vector of an intermediate stage and stores it in the storage unit 16 as reference data. In addition, the data management unit 14 further associates and stores data on the particle size of the granular material measured by sampling the product when the time series data of the intermediate stage is measured.
[0061] In the generation process, the estimation unit 13 calculates the distance between each stage using the feature vector converted from the measurement data and the feature vector of each stage of the reference data, and identifies whether the current time is between the initial stage and the intermediate stage, or between the intermediate stage and the final stage. When the current time is between the initial stage and the intermediate stage, the estimation unit 13 estimates the current particle size using the ratio of the particle size and the distance at the intermediate stage. When the current time is between the intermediate stage and the final stage, the estimation unit 13 estimates the current particle size using the ratio of the difference between the particle size at the intermediate stage and the particle size at the final stage and the distance.
[0062] Here, it is assumed that the measurement data of the time series at the final stage is the second time series data, the measurement data of the time series at the initial stage is the third time series data, the measurement data of the time series at the intermediate stage is the fourth time series data, and the measurement data of the time series at the current time is the first time series data. It is also assumed that the extraction unit 12 converts the second time series data into a second feature vector, the third time series data into a third feature vector, the fourth time series data into a fourth feature vector, and the first time series data into a first feature vector. At this time, the estimation unit 13 calculates the distance between the first feature vector at the current time and the third feature vector at the initial stage, the second feature vector at the final stage, and the fourth feature vector at the intermediate stage of the reference data, respectively.
[0063] When there is no intermediate reference data, the estimation unit 13 estimates the particle size at the current time, i.e., when the first time series data is obtained, using the distance between the first feature vector and the second feature vector, the distance between the first feature vector and the third feature vector, and the particle size at the time of measuring the second time series data.
[0064] When there is reference data for the intermediate stage, the estimation unit 13 calculates the distance between the first feature vector and the second feature vector, the distance between the first feature vector and the third feature vector, and the distance between the first feature vector and the fourth feature vector. When the first time series data at the current time is before the intermediate stage, the distance between the first feature vector and the third feature vector, the distance between the first feature vector and the fourth feature vector, and the particle size at the time of measurement of the fourth time series data are used to estimate the particle size at the current time, i.e., when the first time series data is acquired. When the measurement data at the current time is after the intermediate stage, the distance between the first feature vector and the fourth feature vector, the distance between the first feature vector and the second feature vector, the particle size at the time of measurement of the fourth time series data, and the particle size at the time of measurement of the second time series data are used to estimate the particle size at the current time, i.e., when the first time series data is acquired. In addition, when the particle size at the initial stage is other than 0, the particle size data at the initial stage is associated with the reference data, and the estimation unit 13 estimates the particle size at the current time using the particle size at the initial stage as well. Increasing the number of standards in this way brings the measurement times of the two reference data closer to the current time, improving the accuracy of particle size estimation by the state estimation device 10. There may also be multiple intermediate stages. The estimation unit 13 may estimate the particle size at the current time using the distance between the feature vectors of the current time and the final stage, and the distance between the feature vectors of the initial stage and the final stage.
[0065] The output unit 18 may output the progress of the generation process and information on advice for the worker to the terminal device 30 using the estimation result of the estimation unit 13. The estimation unit 13, for example, compares the speed of grain size expansion with the actual data. When the comparison shows that the grain size expansion is fast, for example, the estimation unit 13 estimates that there is a risk of quality deterioration. The estimation unit 13 identifies the speed of progress using, for example, the ratio of the time from the initial stage to the final stage to the elapsed time up to the present, and the ratio of the current grain size to the grain size at the final stage. In addition, when there is reference data for intermediate stages, the estimation unit 13 may identify the speed of grain size expansion using the time from the start of generation to the arrival of each stage and the grain size at each stage, using the time to reach each stage stored in the memory unit 16 in association with the feature vector of each stage.
[0066] The output unit 18 outputs to the terminal device 30, according to the estimation by the estimation unit 13, a sentence such as "The production speed is fast" as the progress status and a sentence such as "Please lower the temperature as the particle size is increasing too fast and quality deterioration is likely to occur" as the advice. Also, for example, when the estimation unit 13 estimates that the particle size is increasing slowly, the output unit 18 outputs to the terminal device 30 a sentence such as "The production speed is slow" as the progress status and a sentence such as "Please add catalyst A as the particle size is increasing too slow and quality deterioration is likely to occur" as the advice. The correspondence between the estimation result of the estimation unit 13 and the sentence to be output is stored in advance in the storage unit 16.
[0067] The state estimation device 10 of this embodiment extracts time series data for a predetermined time in the initial stage and the final stage from multidimensional time series data measured by the multiple sensors 20 when generating granular materials, converts the data into a feature vector using an estimation model, and stores the data as reference data. In addition, the state estimation device 10 extracts data for a predetermined time from the time series measurement data of the multiple sensors 20 during the generation process of the granular materials, and converts the data into a feature vector using an estimation model. The state estimation device 10 estimates the current particle size by calculating the distance between the feature vector converted from the current time series data during the generation process of the granular materials and the feature vector of the reference data generated in advance. In this way, the state estimation device 10 generates the reference data in advance, and during the generation process, converts the measurement data at the current time into a feature vector and estimates the particle size using the distance between the feature vectors, thereby making it possible to estimate the state of the product even when the internal state cannot be confirmed. Furthermore, the state estimation device 10 of this embodiment generates reference data in advance, and in the generation process, the state estimation device 10 only performs the process of converting the current measurement data into a feature vector and estimating the particle size using the distance between the feature vectors, thereby reducing the amount of data processing. By reducing the amount of necessary data processing, the state estimation device 10 reduces the time required to estimate the particle size in the generation process, and can estimate the state of the product in real time. As a result, the state estimation system of this embodiment can estimate the state of a chemical substance in the generation process even during generation.
[0068] Second Embodiment The second embodiment of the present invention will be described in detail with reference to the drawings. Fig. 8 is a diagram showing an example of the configuration of a state estimation device 100 according to the present embodiment. The state estimation device 100 includes an acquisition unit 101, an extraction unit 102, an estimation unit 103, and an output unit 104.
[0069] The acquisition unit 101 acquires first time series data related to the production environment of the target chemical substance. The target chemical substance is a product whose state in the production process by a chemical reaction is to be estimated. The extraction unit 102 extracts a feature amount of the first time series data. The extraction unit 102 extracts the feature amount of the first time series data. The estimation unit 103 estimates the state of the target chemical substance using an estimation model that machine-learns the relationship between the state of the target chemical substance in the production process and the feature amount of the second time series data related to the production environment based on the feature amount of the first time series data. The output unit outputs the state estimated by the estimation unit. The output unit 104 outputs the state estimated by the estimation unit 103. Here, the acquisition unit 11 of the first embodiment is an example of the acquisition unit 101. Also, the acquisition unit 101 is one aspect of the acquisition means. The extraction unit 12 is an example of the extraction unit 102. Also, the extraction unit 102 is one aspect of the extraction means. The estimation unit 13 and the data management unit 14 are an example of the estimation unit 103. Moreover, the estimation unit 103 is one aspect of the estimation means. The output unit 18 is an example of the output unit 104. Moreover, the output unit 104 is one aspect of the output means.
[0070] The operation of the state estimation device 100 of this embodiment will be described. FIG. 9 is a diagram showing an example of the operation flow of the state estimation device 100. The acquisition unit 101 acquires first time series data related to the production environment of the target chemical substance (step S101). The extraction unit 102 extracts feature amounts of the first time series data (step S102). The estimation unit 103 estimates the state of the target chemical substance using an estimation model that has been machine-learned to learn the relationship between the state of the target chemical substance in the production process and the feature amounts of second time series data related to the production environment based on the feature amounts of the first time series data (step S103). The output unit 104 outputs the estimated state (step S104).
[0071] In the state estimation device 100 of this embodiment, the acquisition unit 101 acquires time series data of the target chemical substance, and the extraction unit 102 extracts feature quantities of the time series data. Furthermore, the estimation unit 103 estimates the state of the target chemical substance from the feature quantities of the first time series data, using second time series data of the production environment when the target chemical substance was produced and an estimation model generated from the state of the target chemical substance. By estimating the state in this manner, the state estimation device 10 of this embodiment can estimate the state of the chemical substance in the production process, even during production.
[0072] Each process in the state estimation device 10 of the first embodiment and the state estimation device 100 of the second embodiment can be performed by executing a computer program on a computer. FIG. 10 shows an example of the configuration of a computer 200 that executes a computer program that performs each process in the state estimation device 10 of the first embodiment and the state estimation device 100 of the second embodiment. The computer 200 includes a CPU 201, a memory 202, a storage device 203, an input / output I / F (Interface) 204, and a communication I / F 205. is.
[0073] The CPU 201 reads out and executes a computer program for performing each process from the storage device 203. The CPU 201 may be configured by a combination of a CPU and a GPU (Graphics Processing Unit). The memory 202 is configured by a DRAM (Dynamic Random Access Memory) or the like, and temporarily stores the computer program executed by the CPU 201 and data being processed. The storage device 203 stores the computer program executed by the CPU 201. The storage device 203 is configured by, for example, a non-volatile semiconductor storage device. Other storage devices such as a hard disk drive may be used for the storage device 203. The input / output I / F 204 is an interface that receives input from an operator and outputs display data and the like. The communication I / F 205 is an interface that transmits and receives data between the sensor 20 and the terminal device 30. The terminal device 30 may also be configured in a similar manner.
[0074] The computer program used to execute each process may be stored in a recording medium and distributed. As the recording medium, for example, a magnetic disk such as a magnetic tape for recording data or a hard disk may be used. Also, an optical disk such as a CD-ROM (Compact Disc Read Only Memory) may be used as the recording medium. A non-volatile semiconductor memory device may also be used as the recording medium.
[0075] A part or all of the above-described embodiments can be described as, but is not limited to, the following supplementary notes.
[0076] [Appendix 1] An acquisition means for acquiring first time-series data regarding the generation environment of a target chemical substance; An extraction means for extracting a feature amount of the first time series data; an estimation means for estimating a state of the target chemical substance by using an estimation model that is machine-learned to understand the relationship between a state of the target chemical substance in a production process and a feature amount of second time-series data related to the production environment, based on the feature amount of the first time-series data; an output means for outputting the state estimated by the estimation means; A state estimation device comprising:
[0077] [Appendix 2] The state of the target chemical substance estimated by the estimation means is at least one of the size of the target chemical substance in a production process, the progress of the production process, and whether the state of the target chemical substance is normal or not. 2. A state estimation device as recited in claim 1.
[0078] [Appendix 3] The output means outputs a diagram or an image relating to the state of the target chemical substance estimated by the estimation means. 3. A state estimation device according to claim 1 or 2. [Appendix 4] The first time series data is at least one of time series data of temperature in a production process of the target chemical substance, time series data of sound emitted by a production device for producing the target chemical substance, and time series data of vibration of the production device. 4. A state estimation device according to claim 1.
[0079] [Appendix 5] The estimation means estimates a time until the production of the target chemical substance is completed, The output means outputs the time estimated by the estimation means. 5. A state estimation device according to claim 1.
[0080] [Appendix 6] When the characteristic value indicating the state of the target chemical substance estimated by the estimation means satisfies a predetermined condition criterion, The output means outputs information indicating that the generation is completed. 6. A state estimation device according to claim 1.
[0081] [Appendix 7] The estimation means estimates a progress status and advice of a production process based on the estimated state and a production time from a start of production of the target chemical substance until the state is reached, The output means outputs the progress status and advice. 7. A state estimation device according to any one of appendix 1 to 6.
[0082] [Appendix 8] The estimation model is generated by further machine learning the production conditions of the target chemical substance. 8. A state estimation device according to any one of appendix 1 to 7.
[0083] [Appendix 9] The estimation model may further include a generating unit. 9. A state estimation device according to any one of appendix 1 to 8.
[0084] [Appendix 10] Obtain first time-series data on the generation environment of a target chemical substance; Extracting features of the first time series data; estimating a state of the target chemical substance using an estimation model that has been machine-learned to determine a relationship between a state of the target chemical substance in a production process and a feature amount of second time-series data related to the production environment, based on the feature amount of the first time-series data; Print the estimated state A state estimation method comprising:
[0085] [Appendix 11] The estimated state of the target chemical substance is at least one of the size of the target chemical substance in the production process, the progress of the production process, and whether the state of the target chemical substance is normal or not. 11. The state estimation method according to claim 10.
[0086] [Appendix 12] Output a diagram or image regarding the estimated state of the target chemical substance 12. The state estimation method according to claim 10 or 11. [Appendix 13] The first time series data is at least one of time series data of temperature in a production process of the target chemical substance, time series data of sound emitted by a production device for producing the target chemical substance, and time series data of vibration of the production device. 13. A state estimation method according to any one of appendix 10 to 12.
[0087] [Appendix 14] Estimating the time until completion of production of the target chemical substance; Print the estimated time 14. A state estimation method according to any one of claims 10 to 13.
[0088] [Appendix 15] When the characteristic value indicating the estimated state of the target chemical substance satisfies a predetermined condition criterion, Prints information indicating that generation is complete 15. A state estimation method according to any one of claims 10 to 14.
[0089] [Appendix 16] estimating a progress status and advice of a production process based on the estimated state and a production time from the start of production of the target chemical substance to reaching the state; Outputting the progress and advice 16. A state estimation method according to any one of claims 10 to 15.
[0090] [Appendix 17] The estimation model is generated by further machine learning the production conditions of the target chemical substance. 17. A state estimation method according to any one of claims 10 to 16.
[0091] [Appendix 18] A process of acquiring first time-series data regarding the generation environment of a target chemical substance; A process of extracting a feature amount of the first time series data; a process of estimating a state of the target chemical substance using an estimation model that is machine-learned to understand a relationship between a state of the target chemical substance in a production process and a feature amount of second time-series data related to the production environment, based on the feature amount of the first time-series data; A process to output the estimated state A program recording medium having a state estimation program recorded thereon for causing a computer to execute the above.
[0092] The present invention has been described above by taking the above-mentioned embodiment as an exemplary example. However, the present invention is not limited to the above-mentioned embodiment. That is, the present invention can be applied in various aspects that can be understood by a person skilled in the art within the scope of the present invention. [Explanation of symbols]
[0093] 10 State Estimation Device 11 Acquisition Department 12 Extraction part 13 Estimation part 14 Data Management Department 15 Model Generation Unit 16 Memory section 17 Input section 18 Output section 20 Sensors 30 Terminal Equipment 100 State Estimation Device 101 Acquisition Department 102 Extraction part 103 Estimation part 200 Computers 201 CPU 202 Memory 203 Storage device 204 Input / Output Interface 205 Communication I / F
Claims
1. An acquisition means for acquiring first time-series data regarding the production environment of a target chemical substance; An extraction means for extracting a feature amount of the first time series data; an estimation means for estimating a state of the target chemical substance by using an estimation model that is machine-learned to understand a relationship between a state of the target chemical substance in a production process and a feature amount of second time-series data related to the production environment, based on a feature amount of the first time-series data; an output means for outputting the state estimated by the estimation means; A state estimation device comprising:
2. The state of the target chemical substance estimated by the estimation means is at least one of the size of the target chemical substance in the production process, the progress of the production process, and whether the state of the target chemical substance is normal or not. The state estimation device according to claim 1 .
3. The output means outputs a diagram or an image relating to the state of the target chemical substance estimated by the estimation means. The state estimation device according to claim 1 .
4. The first time series data is at least one of time series data of temperature in a production process of the target chemical substance, time series data of sound emitted by a production device for producing the target chemical substance, and time series data of vibration of the production device. The state estimation device according to claim 1 .
5. The estimation means estimates a time until the production of the target chemical substance is completed, The output means outputs the time estimated by the estimation means. The state estimation device according to claim 1 .
6. When the characteristic value indicating the state of the target chemical substance estimated by the estimation means satisfies a predetermined condition criterion, The output means outputs information indicating that the generation is completed. The state estimation device according to claim 1 .
7. The estimation means estimates a progress status and advice of the production process based on the estimated state and a production time from the start of production of the target chemical substance to reaching the state, The output means outputs the progress status and advice. The state estimation device according to claim 1 .
8. The estimation model is generated by further machine learning the production conditions of the target chemical substance. The state estimation device according to claim 1 .
9. Obtaining first time-series data on the generation environment of a target chemical substance; Extracting features of the first time series data; estimating a state of the target chemical substance using an estimation model that has been machine-learned to determine a relationship between a state of the target chemical substance in a production process and a feature amount of second time-series data related to the production environment, based on the feature amount of the first time-series data; Print the estimated state A state estimation method comprising:
10. A process of acquiring first time-series data regarding the generation environment of a target chemical substance; A process of extracting a feature amount of the first time series data; a process of estimating a state of the target chemical substance using an estimation model that is machine-learned to understand a relationship between a state of the target chemical substance in a production process and a feature amount of second time-series data related to the production environment, based on a feature amount of the first time-series data; A process to output the estimated state A state estimation program that causes a computer to execute the above.
Citation Information
Patent Citations
Productive relay system maving sampling quantity circuit
JP1978065945A
Pharmaceutical manufacturing control device, pharmaceutical manufacturing control method, pharmaceutical manufacturing control program and pharmaceutical manufacturing system
JP2013029323A
Tablet quality determination device, tablet manufacturing device, and tablet quality determination method
JP2020173162A
Quality control device for solid, pharmaceutical products
US20060135055A1
Anomaly detection method and anomaly detection system
WO2011086805A1
