Method for adaptation and information processing apparatus

Abstract

To obtain a machine learning model useful for measurement by field equipment with ease.SOLUTION: An information processing apparatus 4 includes an acquisition unit 411 and an update unit 412. The acquisition unit 411 acquires process data indicating information related to a measurement target of field equipment and product specification data indicating information related to the field equipment. The update unit 412 updates parameters or the structure of a pre-trained model configured to calculate estimation data on the basis of sensor values obtained from sensors provided in the field equipment, using the process data and the product specification data.SELECTED DRAWING: Figure 5

Description

[Technical Field]

[0001] The present invention relates to an adaptation method and an information processing device. [Background technology]

[0002] Conventionally, there is known a technology related to a field device that measures an output value indicating a state of a fluid, including the flow rate, density, etc. For example, Patent Document 1 discloses a field device that measures a state quantity of a fluid, and that is capable of obtaining a more accurate measurement value even when a slug flow occurs. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Patent No. 6608396 Summary of the Invention [Problem to be solved by the invention]

[0004] However, with conventional technology, if a contaminant such as an air bubble is mixed into the fluid, it becomes difficult to accurately measure the output value, and an error may occur in the measured output value. The field device described in Patent Document 1 is intended to solve this problem. However, although this field device can be applied to fluids whose output value remains approximately constant when no contaminants are mixed in, it has been difficult to apply it to fluids whose output value changes over time due to operation, such as in a distillation process.

[0005] To accurately measure output values ​​related to fluids that change over time, machine learning models can be used. However, training a machine learning model to adapt to various situations requires domain knowledge and experience, and may also require a large amount of training data. For these reasons, obtaining a useful machine learning model is not easy.

[0006] The present disclosure aims to provide an adaptation method and an information processing device that can easily obtain a machine learning model that is useful for measurements using field devices. [Means for solving the problem]

[0007] An adaptation method according to one aspect is characterized in that a computer acquires first data indicating information about an object to be measured by a measuring device and second data indicating information about the measuring device, and performs a process of updating, based on the first data and the second data, parameters or structure of a trained first model that calculates estimated data based on sensor values ​​of a sensor provided in the measuring device.

[0008] An information processing device according to one aspect is characterized by having an acquisition unit that acquires first data indicating information about an object to be measured by a measuring device and second data indicating information about the measuring device, and an update unit that updates, based on the first data and the second data, parameters or structure of a trained first model that calculates estimated data based on sensor values ​​of a sensor provided in the measuring device.

[0009] An information processing device according to one aspect is characterized by having a sensor for measuring the state of a fluid, a memory unit for storing information about a second model obtained by updating the parameters or structure of a learned first model that calculates estimated data based on the sensor value of the sensor based on first data indicating information about the fluid and second data indicating information about the device itself, and a calculation unit for calculating estimated data based on the sensor value using the second model. [Effects of the Invention]

[0010] According to one embodiment, a machine learning model useful for measurements using field devices can be easily obtained. [Brief explanation of the drawings]

[0011] [Figure 1]1 is a block diagram showing a schematic configuration of a field device according to a first embodiment of the present disclosure. [Figure 2] FIG. 10 is a schematic diagram for explaining an overview of regression processing and discrimination processing by a calculation unit. [Figure 3] FIG. 10 is a schematic diagram illustrating an example of a regression process executed using a trained regression model. [Figure 4] FIG. 10 is a schematic diagram illustrating an example of a discrimination process executed using a trained discrimination model. [Figure 5] FIG. 1 is a diagram illustrating a configuration of an information processing device and adaptive processing. [Figure 6] FIG. 10 is a diagram illustrating an example of an adapted model. [Figure 7] FIG. 10 is a diagram illustrating an example of an adapted model. [Figure 8] 10 is a flowchart illustrating an example of an operation of the information processing device. [Figure 9] 10 is a flowchart illustrating a first example of an operation of a field device. [Figure 10] 10 is a flowchart illustrating a second example of the operation of the field device. [Figure 11] FIG. 10 is a schematic diagram for explaining a third example of the operation of the field device. [Figure 12] FIG. 10 is a schematic diagram for explaining a third example of the operation of the field device. [Figure 13] 10 is a flowchart illustrating a fourth example of the operation of the field device. [Figure 14] FIG. 10 is a block diagram showing a schematic configuration of a field device according to a second embodiment of the present disclosure. [Figure 15] FIG. 2 is a diagram illustrating an example of a hardware configuration. DETAILED DESCRIPTION OF THE INVENTION

[0012] The background and problems of the prior art will now be described in more detail.

[0013] In a measuring device such as a field device, if a fluid to be measured contains contaminants such as air bubbles, it becomes difficult to accurately measure output values ​​that indicate the state of the fluid, including the flow rate, density, etc. For example, errors may occur in the measured output values.

[0014] For example, Patent Document 1 discloses a Coriolis flowmeter as an example of a field device. In a Coriolis flowmeter, for example, when air bubbles are mixed into a fluid as contaminants, the measured output value, including the density of the fluid, is affected by the mixed air bubbles. In this case, as shown in FIG. 4 of Patent Document 1, the measured output value obtained is lower than the actual value when no air bubbles are mixed in. This is expected to result in a large error between the measured value and the actual value.

[0015] To solve this problem, the field device described in Patent Document 1 uses the drive current value of the Coriolis flowmeter, called the drive current, as a threshold value for determining whether a slug flow or the like is occurring. In this case, a hysteresis is generally provided for the threshold value. As described above, the field device determines whether air bubbles are mixed in the fluid, and corrects the measurement value when air bubbles are mixed in based on the normal output value when no air bubbles are mixed in.

[0016] However, while the field device described in Patent Document 1 can be applied to fluids whose output value remains approximately constant when no contaminants are present, it has been difficult to apply it to fluids whose output value changes over time during operation, such as in a distillation process.In addition, accurate correction is difficult depending on conditions including the type and amount of small bubbles such as microbubbles, and the duration of time when bubbles remain present.

[0017] In order to solve the above problems, the present disclosure aims to provide a field device that can more accurately measure an output value even when contaminants are mixed into the fluid as the output value of the fluid changes over time.

[0018] Hereinafter, one embodiment of the present disclosure will be mainly described with reference to the accompanying drawings.

[0019] (First embodiment) 1 is a block diagram showing a schematic configuration of a field device 1 according to a first embodiment of the present disclosure. The configuration and functions of the field device 1 according to the first embodiment will be mainly described with reference to FIG.

[0020] The field device 1 may be referred to as, for example, a measuring device. In this disclosure, the term "field device" includes any device that performs measurement processing on a physical quantity to be measured and acquires a measurement value. In this disclosure, the term "physical quantity" includes, for example, the temperature, pressure, flow rate, and pH of fluids, including gases and liquids, generated in a plant facility where the field device is located, as well as the corrosion level and vibration level of the plant facility. Without being limited thereto, the physical quantity also includes status output values, including temperature and pressure, associated with actuators, including valves, motors, and relays.

[0021] In the present disclosure, "plant facilities" include, for example, chemical and other industrial plants, as well as plants that manage and control wellheads including gas fields and oil fields, etc. In addition, plant facilities may include plants that manage and control power generation such as hydroelectric power, thermal power, and nuclear power, plants that manage and control environmental power generation such as solar power and wind power, and plants that manage and control water supply and sewage systems, dams, etc.

[0022] For example, the field device 1 includes any device that measures at least one output value that indicates the state of a fluid. In this disclosure, the term "fluid" includes, for example, gas, liquid, and solid slug. The "output value" includes, for example, at least one of the density, volumetric flow rate, mass flow rate, and bubble volume fraction of the fluid. The "bubble volume fraction" refers to, for example, the proportion of gas components contained in the fluid. The proportion of the gas components is based on a weight or volume basis. The field device 1 measures, for example, the flow rate of a fluid flowing through a pipe. The field device 1 includes, for example, a Coriolis flowmeter. The Coriolis flowmeter is an example of a field device. The field device 1 includes a detection unit 10, a processing unit 20, and a calculation unit 30 included in the processing unit 20.

[0023] The detection unit 10 vibrates a measuring pipe 11 through which a fluid to be measured flows, and detects the vibrations upstream and downstream and the temperature of the measuring pipe 11. In addition to the measuring pipe 11, the detection unit 10 has a vibrator 12, an upstream sensor 13, a downstream sensor 14, and a temperature sensor 15.

[0024] The measuring pipe 11 includes, for example, a straight pipe type having both ends fixedly supported by support members. The measuring pipe 11 is not limited to this, and may include, for example, a U-shaped pipe or other shapes.

[0025] The vibrator 12 includes any vibration module that mechanically vibrates the measuring tube 11 up and down. The vibrator 12 is provided around the measuring tube 11 through which the fluid flows. For example, the vibrator 12 is installed near the center of the measuring tube 11. The vibrator 12 is electrically connected to the processing unit 20.

[0026] The upstream sensor 13 includes any sensor capable of detecting vibrations of the measuring pipe 11 caused by the vibrator 12. The upstream sensor 13 is fixed on the side where the fluid flows into the measuring pipe 11, that is, on the upstream side of the vicinity of the center of the measuring pipe 11 where the vibrator 12 is installed.

[0027] The downstream sensor 14 includes any sensor capable of detecting vibrations of the measuring pipe 11 caused by the vibrator 12. The downstream sensor 14 is fixed on the side where the fluid flows out of the measuring pipe 11, that is, on the downstream side of the vicinity of the center of the measuring pipe 11 where the vibrator 12 is installed.

[0028] Each of the upstream sensor 13 and the downstream sensor 14 is electrically connected to the calculation unit 30 of the processing unit 20 .

[0029] The temperature sensor 15 includes any sensor capable of measuring the surface temperature of the measuring pipe 11. The temperature sensor 15 is fixed, for example, on the surface of the measuring pipe 11 downstream of the vicinity of the center where the vibrator 12 is installed. The temperature sensor 15 is fixed near the downstream sensor 14. The temperature sensor 15 is electrically connected to the calculation unit 30 of the processing unit 20. The temperature sensor 15 is used to reduce measurement errors of the output value due to temperature fluctuations.

[0030] The operation of the detection unit 10 configured as above will be mainly described.

[0031] The vibrator 12 vibrates the measuring tube 11 in a predetermined vibration mode in response to the driving current IR output from the processing unit 20. For example, the vibrator 12 vibrates the measuring tube 11 in a primary vibration mode in which vibration nodes appear only at both ends of the measuring tube 11, which is fixed and supported by a support member.

[0032] When the fluid to be measured flows through the measuring pipe 11 while the vibrator 12 is vibrating the measuring pipe 11 in the primary vibration mode, the measuring pipe 11 vibrates in a secondary vibration mode in which vibration nodes appear at both ends of the measuring pipe 11, which is fixed and supported by the support members, and at a position between them. In reality, the measuring pipe 11 may vibrate in a vibration mode in which two types of vibration modes, the primary vibration mode and the secondary vibration mode, are superimposed.

[0033] The upstream sensor 13 measures the amount of displacement on the upstream side of the measuring pipe 11 vibrating in the above vibration mode. The upstream sensor 13 outputs the measured amount of displacement as a displacement signal SA to the calculation unit 30 of the processing unit 20. The downstream sensor 14 measures the amount of displacement on the downstream side of the measuring pipe 11 vibrating in the above vibration mode. The downstream sensor 14 outputs the measured amount of displacement as a displacement signal SB to the calculation unit 30 of the processing unit 20.

[0034] The temperature sensor 15 measures the surface temperature of the measuring pipe 11 on the surface located on the downstream side of the measuring pipe 11. The temperature sensor 15 outputs the measured surface temperature of the measuring pipe 11 to the calculation unit 30 of the processing unit 20 as a temperature signal ST.

[0035] The processing unit 20 has an excitation circuit 21, an output unit 22, and a memory unit 23 in addition to a calculation unit 30. The calculation unit 30 has a density calculation unit 31, a mass flow rate calculation unit 32, a volumetric flow rate calculation unit 33, and an analysis unit 34. The analysis unit 34 further has a discrimination unit 35 and a regression unit 36. The discrimination unit 35 has a bubble discrimination unit 351, a stop discrimination unit 352, and a phase state discrimination unit 353. The regression unit 36 ​​has a density calculation unit 361, a mass flow rate calculation unit 362, a volumetric flow rate calculation unit 363, and a bubble volume fraction calculation unit 364.

[0036] The excitation circuit 21 is connected to the vibrator 12. The excitation circuit 21 drives the vibrator 12, for example, in a sinusoidal manner, by outputting a drive current IR corresponding to the displacement signal SA to the vibrator 12. The excitation circuit 21 may also drive the vibrator 12, for example, in a sinusoidal manner, by outputting a drive current IR corresponding to a displacement signal SB instead of the displacement signal SA to the vibrator 12.

[0037] The output unit 22 includes any output interface that outputs information to notify the user of the field device 1. The output interface includes, for example, a display that outputs information as a video and a speaker that outputs information as a sound. The display includes, for example, a liquid crystal display and an organic EL (Electro Luminescent) display.

[0038] The storage unit 23 includes any storage module including a hard disk drive (HDD), a solid state drive (SSD), an electrically erasable programmable read-only memory (EEPROM), a read-only memory (ROM), and a random access memory (RAM). The storage unit 23 functions as, for example, a main storage device, an auxiliary storage device, or a cache memory. The storage unit 23 is not limited to being built into the field device 1, and may include an external storage module connected via a digital input / output port such as a universal serial bus (USB). The storage unit 23 stores any information used in the operation of the field device 1.

[0039] The calculation unit 30 includes one or more processors. In the present disclosure, a "processor" refers to, for example, but is not limited to, a general-purpose processor or a dedicated processor specialized for a specific process. The calculation unit 30 executes various calculation processes required for the operation of the field device 1.

[0040] The calculation unit 30 measures the vibration frequency of the measuring tube 11 based on at least one of the displacement signals SA and SB output from the detection unit 10. The density calculation unit 31 of the calculation unit 30 calculates the mass of the fluid to be measured, i.e., the density, based on the measured vibration frequency. The mass flow rate calculation unit 32 of the calculation unit 30 calculates the mass flow rate of the fluid flowing through the measuring tube 11 based on the phase difference between the displacement signals SA and SB output from the detection unit 10. The volumetric flow rate calculation unit 33 of the calculation unit 30 calculates the volumetric flow rate of the fluid flowing through the measuring tube 11 by dividing the mass flow rate calculated by the mass flow rate calculation unit 32 by the density calculated by the density calculation unit 31.

[0041] The detection unit 10 is used to acquire, as data, at least one type of sensor value required for the calculation of the output value as described above by the calculation unit 30. The calculation unit 30 calculates the output value based on the data acquired using the detection unit 10. In the present disclosure, "data" includes, for example, at least one type of data: phase difference data, vibration frequency data, fluid temperature data, drive current data, fluid pressure data, total mass data during operation, and total volume data during operation.

[0042] The phase difference data is data in which the phase difference between the displacement signals SA and SB outputted respectively from the upstream sensor 13 and the downstream sensor 14 of the detection unit 10 is used as a sensor value. The vibration frequency data is data in which the vibration frequency of the measuring pipe 11 measured based on at least one of the displacement signals SA and SB outputted respectively from the upstream sensor 13 and the downstream sensor 14 of the detection unit 10 is used as a sensor value.

[0043] The fluid temperature data is data in which the temperature of the measuring tube 11 is measured based on the temperature signal ST output from the temperature sensor 15 of the detection unit 10 and used as a sensor value. The drive current data is data in which the drive current IR output from the excitation circuit 21 to operate the vibrator 12 of the detection unit 10 is used as a sensor value.

[0044] The fluid pressure data is data obtained by inputting measurements from a pressure meter or the like attached separately from the field device 1, or data in which the pressure set by a fixed value in the calculation unit 30 is used as a sensor value. The total mass data during operation is data in which the sensor value is the total mass of the fluid flowing through the measuring pipe 11 during operation of the field device 1, measured based on the displacement signals SA and SB output from the upstream sensor 13 and downstream sensor 14 of the detection unit 10, respectively. The total volume data during operation is data in which the sensor value is the total volume of the fluid flowing through the measuring pipe 11 during operation of the field device 1, measured based on the displacement signals SA and SB output from the upstream sensor 13 and downstream sensor 14 of the detection unit 10, respectively.

[0045] The analysis unit 34 of the calculation unit 30 obtains the results of at least one of a regression process and a discrimination process related to the output value. The regression process and the discrimination process are performed using a machine learning model previously constructed based on data when a contaminant is mixed into a fluid and data acquired using the detection unit 10. In the following description, the machine learning model will be simply referred to as a model. The calculation unit 30 obtains the results by performing at least one of a regression process and a discrimination process. When the analysis unit 34 detects that a contaminant has been mixed into a fluid, it regresses a value required as a measurement value of the output value based on the previously constructed trained model and outputs the output value.

[0046] In this disclosure, the term "contaminants" includes, for example, air bubbles. In the following, the contaminants will be described as air bubbles, but are not limited to this. The contaminants may include, for example, any foreign matter that affects the measured value of the output value obtained by the field device 1. The same description as below also applies to foreign matter.

[0047] In the present disclosure, the "regression process" for the output value includes, for example, a process of regressing a value required as a measured value of the output value based on a pre-constructed trained model when the measured value of the output value indicates an abnormality due to the inclusion of air bubbles in the fluid. The "discrimination process" for the output value includes, for example, a process of determining whether air bubbles are present in the fluid, a process of determining whether the operation of the field device 1 has stopped, and a process of determining the phase state of the fluid to be measured.

[0048] In this disclosure, the term "phase state" refers to the total number of gas and liquid phases in a multiphase flow of a fluid containing two or more mixed components, such as a gas and a liquid. For example, when gas bubbles are mixed into a two-component fluid separated by a liquid phase in a multiphase flow, the phase state becomes three.

[0049] Fig. 2 is a schematic diagram for explaining an outline of the regression processing and discrimination processing by the calculation unit 30 in Fig. 1. Fig. 2 shows an outline of the processing executed using a trained model.

[0050] In the regression process, the model is constructed using a model that outputs a regression result, such as multiple regression analysis, a neural network, a support vector regression, a Gaussian process regression, a regression tree, a logistic regression, or an autoregressive model. In the discrimination process, the above model is constructed using a model that outputs a classification result, such as a logistic regression, a neural network, a support vector machine, a classification tree, a change point detection, a k-nearest neighbor method, or a k-means method. In the present disclosure, "neural network" includes, for example, a self-organizing map, a convolutional neural network, a recurrent neural network, and a long-short-term memory neural network.

[0051] The input layer 51 uses at least one type of sensor value obtained inside the field device 1. The field device 1 is, for example, a Coriolis flowmeter. For example, the sensor values ​​used in the input layer 51 are included in the above data. More specifically, the sensor values ​​are included in the phase difference data, vibration frequency data, fluid temperature data, drive current data, fluid pressure data, total mass data during operation, and total volume data during operation.

[0052] The output layer 53 uses at least one type of estimated data related to the output value output from the field device 1 and estimated data related to the discrimination process. For example, the estimated data used in the output layer 53 relates to the volumetric flow rate, fluid density, mass flow rate, and bubble volume fraction. For example, the estimated data used in the output layer 53 relates to determining whether or not bubbles are mixed into the fluid, determining the phase state of the fluid to be measured, and determining whether or not the operation of the field device 1 has stopped.

[0053] The analysis unit 34 of the calculation unit 30 constructs the pre-trained model 52 by learning using data and labels obtained when air bubbles are mixed into the fluid. Learning includes supervised learning, unsupervised learning, semi-supervised learning, and the like. In this embodiment, the pre-trained model 52 is constructed by supervised learning. The pre-trained model 52 is an example of a first model. The teacher data serving as labels used to construct the pre-trained model 52 includes, for example, time series data obtained for at least one type of output value when air bubbles are mixed into the fluid. The time series data is continuous, including the time when air bubbles are mixed in. For example, the teacher data is time series data obtained by recording the state of the fluid as photographs or videos using an imaging device such as a camera, and the time series data includes continuous labels indicating the time when air bubbles are confirmed to be mixed in. The analysis unit 34 constructs the pre-trained model 52 by using the data acquired using the detection unit 10 of the field device 1 and the cumulative amount over time obtained from the feature values ​​as feature values.

[0054] The training data for the regression model is obtained using measuring instruments such as level gauges, weight scales, and density meters, as well as visual inspection. Generally, the relationship volumetric flow rate = mass flow rate / density holds. Therefore, if training data is prepared for at least two of the volumetric flow rate, mass flow rate, and density, it is possible to indirectly obtain training data for the remaining one. The analysis unit 34 evaluates the regression from data related to various sensor values ​​based on the obtained training data.

[0055] FIG. 3 is a schematic diagram illustrating an example of regression processing executed using a trained regression model L1. FIG. 3 shows, as an example, the regression model L1 of a neural network during the regression processing. FIG. 4 is a schematic diagram illustrating an example of discrimination processing executed using a trained discrimination model L2. FIG. 4 shows, as an example, the discrimination model L2 of a neural network during discrimination processing in which the number of classes is 2. Both the regression model L1 and the discrimination model L2 are examples of the pre-trained model 52 or an adapted model described below. The adapted model is an example of a second model.

[0056] The analysis unit 34 of the calculation unit 30 obtains results when the regression process and the discrimination process are performed using different models. That is, the analysis unit 34 uses independent models for the regression process and the discrimination process. As an example, the analysis unit 34 uses the same regression model L1 for different types of output values ​​in the regression process. As an example, the analysis unit 34 uses the same discrimination model L2 for different types of discrimination in the discrimination process.

[0057] The analysis unit 34 can set any number of layers and any number of neurons for the hidden layers. The analysis unit 34 sets general nonlinear functions such as a ramp function, a sigmoid function, and a hyperbolic tangent function as activation functions. The analysis unit 34 sets a softmax function for the output layer when performing multi-value classification based on the perspective of the backpropagation method.

[0058] The regression unit 36 ​​of the analysis unit 34 executes regression processing of the fluid density, mass flow rate, volumetric flow rate, bubble volume fraction, etc. from the obtained sensor values ​​based on the regression model L1. In Fig. 3, as an example, regression processing of the volumetric flow rate by the volumetric flow rate calculation unit 363 is shown.

[0059] The density calculation unit 361 executes a regression process of the fluid density. For example, in the regression process of the fluid density, the density calculation unit 361 uses a logistic regression model using one or more of the total mass during operation, the total volume during operation, and the fluid temperature among the sensor values.

[0060] The mass flow rate calculation unit 362 executes regression processing of the mass flow rate. For example, in the regression processing of the mass flow rate, the mass flow rate calculation unit 362 uses a linear regression model that uses, among the sensor values, the phase difference, the fluid temperature, the drive current, the density of the fluid before the regression processing, and the total volume during operation.

[0061] The volumetric flow rate calculation unit 363 executes regression processing of the volumetric flow rate. For example, in the regression processing of the volumetric flow rate, the volumetric flow rate calculation unit 363 uses a linear regression model that uses, among the sensor values, the phase difference, the fluid temperature, the drive current, the density of the fluid before the regression processing, and the total mass during operation.

[0062] The bubble volume fraction calculation unit 364 executes regression processing of the bubble volume fraction. For example, in the regression processing of the bubble volume fraction, the bubble volume fraction calculation unit 364 uses a linear regression model using the phase difference, fluid temperature, drive current, fluid density before regression processing, total mass during operation, and total volume during operation among the sensor values.

[0063] The discrimination unit 35 of the analysis unit 34 executes discrimination processes such as discriminating whether or not air bubbles are mixed in, discriminating whether operation has stopped, and discriminating the phase state from the obtained sensor values ​​based on the discrimination model L2. In Fig. 4, as an example, the discrimination process of whether or not air bubbles are mixed in by the air bubble discriminator 351 is shown.

[0064] The air bubble detector 351 executes a process of determining whether or not air bubbles have been mixed in. For example, the air bubble detector 351 uses change point detection using a phase difference among sensor values ​​in the process of determining whether or not air bubbles have been mixed in.

[0065] The stoppage determination unit 352 executes a process for determining whether or not the operation is stopped. For example, in the process for determining whether or not the operation is stopped, the stoppage determination unit 352 uses, among the sensor values, the phase difference, the fluid temperature, the drive current, the density of the fluid before the regression process, the total mass during operation, and the total volume during operation.

[0066] The phase state determination unit 353 executes a phase state determination process. For example, in the phase state determination process, the phase state determination unit 353 uses the phase difference, fluid temperature, drive current, fluid density before regression processing, total mass during operation, and total volume during operation, which are sensor values.

[0067] (Model adaptation) For example, if the field device 1 is a Coriolis flowmeter, a stable output value can be obtained even when air bubbles are present by using the pre-trained model 52 described above. However, there may be cases where the conditions under which the pre-trained model 52 is used exceed the range of conditions assumed for training. In such cases, training the pre-trained model 52 to suit all conditions would be difficult because a large amount of true values ​​(teaching data) would be required.

[0068] Therefore, in this embodiment, the model is adapted to various conditions using limited information, and as a result, according to this embodiment, the field device 1 can accurately measure the output value.

[0069] For example, differences in conditions may be affected by differences in the characteristics of the object to be measured by the field device 1 and differences in the field device 1 as a product. Therefore, the information processing device of this embodiment adapts the model to the conditions using process data, which is data related to the object to be measured by the field device 1, and product specification data, which is data related to the field device 1. The process data is an example of first data, and the product specification data is an example of second data.

[0070] 5 is a diagram illustrating the configuration and adaptation processing of an information processing device. Here, the information processing device 4 executes a software application for training a pre-trained model 52 and adapting the pre-trained model 52 to conditions. The information processing device 4 is a computer such as a server, PC, or smartphone connected to the field device 1. Furthermore, for example, communication is performed between the information processing device 4 and the field device 1 in accordance with a protocol such as HART, ModBus, or Brain.

[0071] The information processing device 4 may be a server on a cloud, or may be a server that functions as a host that manages multiple field devices in a plant. In this embodiment, the information processing device 4 and the field device 1 are different devices. However, the information processing device 4 and the field device 1 may be the same device.

[0072] 5, the information processing device 4 includes an acquisition unit 411, an update unit 412, a provision unit 413, and a storage unit 42. The information processing device 4 also includes an interface for inputting and outputting data between the field device 1 and other devices.

[0073] The acquiring unit 411, the updating unit 412, and the providing unit 413 are realized by a processor that executes various types of arithmetic processing required for the operation of the field device 1.

[0074] The acquisition unit 411 acquires process data, product specification data, and model information. The model information acquired by the acquisition unit 411 is information for constructing a pre-adaptation model. For example, the model information acquired by the acquisition unit 411 is parameters for constructing the pre-trained model 52.

[0075] For example, if the model is a neural network, the parameters are vectors or tensors whose elements are the weights and biases of each layer. Also, for example, the parameters may be coefficients and constant terms in a regression model and a support vector machine.

[0076] The update unit 412 adapts the model constructed based on the model information acquired by the acquisition unit 411 to the conditions using the process data and product specification data. For example, the update unit 412 updates the parameters using a machine learning technique. A specific example of the adaptation process will be described later.

[0077] The providing unit 413 provides the field device 1 or another device with model information for constructing an adapted model. For example, the providing unit 413 provides updated parameters. The device that has received the parameters inputs the sensor values ​​into the adapted model to obtain estimated data. The obtained estimated data is the output result of the field device 1.

[0078] The storage unit 42 includes any storage module including an HDD, SSD, EEPROM, ROM, RAM, etc. The storage unit 42 functions as, for example, a main storage device, an auxiliary storage device, or a cache memory.

[0079] The storage unit 42 is not limited to being built into the information processing device 4, and may include an external storage module connected via a digital input / output port such as a USB. The storage unit 42 stores any information used in the operation of the information processing device 4. The storage unit 42 stores first model information 421 and second model information 422.

[0080] The first model information 421 is information about the model before updating by the update unit 412. For example, the first model information 421 is parameters of the pre-trained model 52. The second model information 422 is information about the model after updating by the update unit 412. The second model information 422 is provided to another device by the providing unit 413.

[0081] Here, the manufacturer is a person who manufactures the field device 1. The user is a person who uses the field device 1 purchased from the manufacturer in a plant or the like. The process data may be provided by the user. The product specification data may be provided by the manufacturer.

[0082] For example, since the field device 1 can be used to measure a variety of objects, the manufacturer cannot uniquely identify what object the field device 1 will be used to measure at the time of shipping the field device 1. For this reason, the manufacturer prepares a pre-trained model 52 that has a certain degree of versatility or a pre-trained model 52 that is adapted to specific conditions.

[0083] The user inputs process data, product specification data, and model information of the pre-trained model 52 into the information processing device 4 to obtain model information adapted to desired conditions. At this time, the user may receive product specification data from the manufacturer.

[0084] In addition, when a user orders a field device 1, the manufacturer receives process data from the user, inputs the process data, product specification data, and model information of the pre-trained model 52 into the information processing device 4, and can prepare model information adapted to the desired conditions.

[0085] For example, the process data is one or more of the composition, composition ratio, physical property constant, thermal conductivity, heat transfer coefficient, concentration, etc. of the object to be measured. The process data may also include the temperature, pressure, viscosity, etc. of the object to be measured. Note that when the field device 1 is a Coriolis flowmeter, the object to be measured is a fluid.

[0086] The product specification data includes data representing the physical characteristics of the field device 1. For example, the product specification data is one or more of the outer shape, shape, size (volume), length, weight, etc. of the field device 1. The product specification data may also include calibration values ​​resulting from the structure of the sensor of the field device 1.

[0087] The product specification data may also be information for identifying the field device 1. For example, the product specification data may be a pre-shipment lot number or a product number of the field device 1. In this case, the information processing device 4 can refer to a table that associates information for identifying the field device 1 with physical features, and acquire data representing the physical features based on the product specification data.

[0088] Examples of adaptation processing by the update unit 412 include additional learning and output conversion. Additional learning is processing in which process data and product specification data are added to explanatory variables input to the pre-trained model 52 (the model before adaptation), and then additional learning (or re-learning from the initial state) is performed. Output conversion is processing in which values ​​output from the pre-trained model 52 are converted according to the process data and product specification data.

[0089] (Additional training of the model) The additional training will be described using FIG. 6. FIG. 6 is a diagram showing an example of an adapted model. As shown in FIG. 6, an input layer 61 in an additionally trained model 62 (adapted model) has process data and product specification data added thereto, compared to the input layer 51 in the pre-trained model 52 of FIG. 2. For example, adding data to the input layer is achieved by updating the model structure. For example, updating the model structure includes adding nodes to a neural network and adding terms to a regression equation. On the other hand, the output layer 63 in the additionally trained model 62 is the same as the output layer 53 in the pre-trained model 52 of FIG. 2.

[0090] (Model output transformation) The output transformation will be described using Fig. 7. Fig. 7 is a diagram showing an example of an adapted model. The input layer 71, pre-trained model 72, and output layer 73 in Fig. 7 are the same as the input layer 51, pre-trained model 52, and output layer 53 in Fig. 2. The estimated data in the output layer 73 and the process data and product specification data in the input layer 75 are input to the transformation function 74.

[0091] The conversion function 74 converts the estimated data based on the process data and product specification data, and outputs the converted estimated data. For example, the conversion function 74 may be an arithmetic operation, a linear sum, or a kernel. Furthermore, for example, the conversion function 74 may use various nonlinear functions to map an output value in an N-dimensional space spanning the dimensions of N pieces of input information. The coefficients, parameters, etc. in the conversion function 74 are determined based on the process data and product specification data. By adding such a conversion function 74, the structure of the entire model is updated.

[0092] The transformation function 74 may be a machine learning model such as a neural network using supervised learning, in which case the parameters of the transformation function 74 are determined based on the process data and the product specification data.

[0093] Fig. 8 is a flowchart illustrating an example of the operation of the information processing device. The series of processes shown in Fig. 8 are realized by the information processing device 4 executing a software application (hereinafter, referred to as an app). For example, a UI screen is displayed on the display of the information processing device 4. Then, the series of processes is started by pressing a button (including a virtual button on the display) for operating the information processing device 4.

[0094] The information processing device 4 first selects a mode by receiving a command to select either the learning mode or the parameter read / write mode. For example, it is assumed that buttons corresponding to two processes, the learning mode and the parameter read / write mode, are initially displayed on the UI screen. In this case, the mode is selected by pressing the button. The information processing device 4 performs processing according to the selected mode. Note that the mode selection is not limited to button operation, and may also be performed by command line input, etc.

[0095] In step S101, when the learning mode is selected (step S101, learning mode), the information processing device 4 checks whether or not process data and product specification data are input (step S102).

[0096] When the process data and the product specification data are input (Step S102, input), the information processing device 4 accepts the input of the process data and the product specification data (Step S103), and then starts learning (Step S104).

[0097] If the learning process has ended normally (step S105, Yes), the information processing device 4 calculates and saves parameters (step S106) and proceeds to step S108. The saved parameters are provided to field devices, etc. If the conditions for terminating the app are met (step S108, Yes), the information processing device 4 ends the process. The conditions for terminating the app include, for example, pressing the end button or the passage of a certain period of time. If the conditions for terminating the app are not met (step S108, No), the information processing device 4 returns to step S101.

[0098] If the learning process has not ended normally (step S105, No), the information processing device 4 outputs an error (step S107). At this time, a message indicating that an error has occurred is displayed on a display provided in, for example, the field device 1 or the information processing device 4. The message indicating that an error has occurred may also be displayed on a portable terminal or the like connected to the information processing device 4.

[0099] If the parameter read / write mode is selected in step S101 (step S101, parameter read / write mode), the information processing device 4 checks whether or not parameters (learned parameters) are saved (step S109). The parameters are saved in step S106. The model parameters may be saved in the information processing device 4 or the field device 1, or in a server on a cloud different from the information processing device 4.

[0100] For example, the information processing device 4 stores information for constructing an adapted model obtained by updating the pre-trained model in a server on the cloud that is accessible to the field device 1 and is different from the information processing device 4. This allows the field device 1 to omit learning and easily use the adapted model, as will be described below.

[0101] Next, the information processing device 4 selects which of the saved parameters to read (step S110). If the reading of the parameters is canceled (step S110, Cancel), the information processing device 4 proceeds to step S111.

[0102] When the information processing device 4 has selected which parameters to read (step S110, selection of saved parameters), it reads the selected parameters (step S112).

[0103] Here, the information processing device 4 checks whether or not a device (field device 1) is connected (step S113). If a device is not connected (step S113, No), the information processing device 4 proceeds to step S111. If a device is connected (step S113, Yes), the information processing device 4 writes parameters to the device (step S114).

[0104] If the writing has been completed successfully (Yes in step S115), the information processing device 4 outputs a message indicating that the writing has been completed successfully (step S116). At this time, a message indicating that the processing has been completed successfully is displayed on a display provided in the field device 1 or the information processing device 4, for example.

[0105] If the writing did not end normally (step S115, No), the information processing device 4 proceeds to step S111.

[0106] In this way, the information processing device 4 acquires process data indicating information about the object to be measured by the field device 1 and product specification data indicating information about the field device 1, and updates the parameters or structure of the trained pre-trained model that calculates estimated data based on the sensor values ​​of the sensors provided in the field device 1 based on the process data and product specification data.

[0107] This allows the model to be adapted to various situations without requiring domain knowledge, experience, or a large amount of training data, making it possible to easily obtain a model that is useful for measurements by the field device 1.

[0108] The product specification data reflects in the model information about physical and mechanical differences such as the shape and diameter of the field device 1. The process data also reflects in the model information about the physical properties of the measurement object and changes over time.

[0109] The information processing device 4 can perform additional learning. That is, the information processing device 4 updates the parameters of the adapted model in which process data and product specification data are added to the variables input to the pre-trained model, as described in FIG.

[0110] Furthermore, as described with reference to FIG. 7, the information processing device 4 adds to the pre-trained model a function that converts the estimation data and whose calculation method is determined by the process data and the product specification data.

[0111] The information processing device 4 may also be a field device 1. In this case, the detection unit 10 includes a sensor that measures the state of the fluid. The storage unit 23 stores information about the adapted model. The calculation unit 30 uses the adapted model to calculate estimated data based on the sensor value.

[0112] The model in the following description may be the pre-trained model 52 or the pre-trained model 52 that has been adapted by the information processing device 4.

[0113] Fig. 9 is a flowchart for explaining a first example of the operation of the field device 1 of Fig. 1. Fig. 9 summarizes the outline of the main operations realized by the field device 1.

[0114] In step S200, the calculation unit 30 stores at least one type of sensor value, acquired using the detection unit 10, necessary for calculating the output value in the storage unit 23 as data.

[0115] In step S201, the calculation unit 30 determines whether or not a sufficient number of samples has been obtained to calculate the average value of a moving window for a predetermined time interval set for the sensor values ​​stored in step S200. If the calculation unit 30 determines that a sufficient number of samples has been obtained, it executes the process of step S202. If the calculation unit 30 determines that a sufficient number of samples has not been obtained, it executes the process of step S205.

[0116] In step S202, when it is determined that the number of samples has been obtained in step S201, the calculation unit 30 calculates the average value of a moving window for a predetermined time interval set for the sensor values ​​stored in step S200.

[0117] In step S203, the calculation unit 30 determines whether or not air bubbles have been detected in the fluid based on the average value calculated in step S202. If the calculation unit 30 determines that air bubbles have been detected, it executes the process of step S204. If the calculation unit 30 determines that air bubbles have not been detected, it executes the process of step S205.

[0118] The calculation unit 30 determines whether or not air bubbles have been detected by comparing the average value of a moving window for a predetermined time interval set for at least one of the output value and the sensor value (for example, the sensor value in the flowchart of FIG. 9) with a first threshold value. The calculation unit 30 determines that air bubbles have been detected when the average value exceeds the first threshold value. The calculation unit 30 determines the first threshold value to be compared with the average value and used to determine whether or not air bubbles have been detected using the above-described discrimination model L2.

[0119] In step S204, if the calculation unit 30 determines that the intrusion of air bubbles has been detected in step S203, it determines whether or not it has detected a stop of operation of the field device 1. If the calculation unit 30 determines that it has detected a stop of operation of the field device 1, it executes the process of step S205. If the calculation unit 30 determines that it has not detected a stop of operation of the field device 1, it executes the process of step S206.

[0120] In step S205, if the calculation unit 30 determines that the stop of operation has been detected in step S204, it executes normal processing when the operation of the field device 1 is stopped. If the calculation unit 30 determines that the number of samples has not been obtained in step S201, it executes normal processing related to the measurement of the output value of the field device 1. If the calculation unit 30 determines that the intrusion of air bubbles has not been detected in step S203, it executes normal processing related to the measurement of the output value of the field device 1.

[0121] In step S206, if the calculation unit 30 determines that the operation has not been stopped in step S204, but has detected the inclusion of air bubbles in step S203, the calculation unit 30 executes regression processing related to the output value using the pre-constructed model and the data acquired in step S200. In this way, when the calculation unit 30 determines that air bubbles have been mixed into the fluid, it executes regression processing using the regression model L1 and outputs an output value.

[0122] As described above, the calculation unit 30 uses change-point detection using the average value of a moving window over an arbitrary time interval. When detecting the presence of air bubbles, the calculation unit 30 constantly monitors at least one type of sensor value used for the regression process and the discrimination process. The calculation unit 30 determines that the presence of air bubbles has been detected when the monitored sensor value exceeds a set first threshold. To reduce false detections due to factors other than the presence of air bubbles, such as noise, the calculation unit 30 calculates the average value of a moving window over a predetermined time interval and compares it with the first threshold. The first threshold is determined based on a discriminant model L2 previously constructed by supervised learning using training data.

[0123] Fig. 10 is a flowchart for explaining a second example of the operation of the field device 1 of Fig. 1. Fig. 10 shows more specifically the determination process for determining whether or not a stop is performed, which is executed in step S204 of Fig. 9.

[0124] In step S300, the calculation unit 30 stores the calculated at least one type of output value in the storage unit .

[0125] In step S301, the calculation unit 30 determines whether or not a sufficient number of samples has been obtained to calculate the average value of a moving window for a predetermined time interval set for the output values ​​stored in step S300. If the calculation unit 30 determines that a sufficient number of samples has been obtained, it executes the process of step S302. If the calculation unit 30 determines that a sufficient number of samples has not been obtained, it executes the process of step S309.

[0126] In step S302, when it is determined that the number of samples has been obtained in step S301, the calculation unit 30 calculates the average value of a moving window for a predetermined time interval set for the output values ​​stored in step S300.

[0127] In step S303, the calculation unit 30 determines whether the average value calculated in step S302 is equal to or less than a second threshold. If the calculation unit 30 determines that the average value is equal to or less than the second threshold, it executes the process of step S304. If the calculation unit 30 determines that the average value is greater than the second threshold, it executes the process of step S305. The calculation unit 30 determines the second threshold value, which is set for at least one type of output value and is used to set a flag indicating that the vehicle is running or stopped, using the above-mentioned discrimination model L2.

[0128] In step S304, if the calculation unit 30 determines in step S303 that the difference is equal to or less than the second threshold value, it sets flag 1 and stores it in the storage unit 23 as information.

[0129] In step S305, if the calculation unit 30 determines in step S303 that the value is greater than the second threshold value, it sets flag 0 and stores the flag in the storage unit 23 as information.

[0130] In step S306, the calculation unit 30 determines whether or not a sufficient number of samples has been obtained to calculate the average value of the moving window for the predetermined time interval set for the flag values ​​stored in steps S304 and S305. If the calculation unit 30 determines that a sufficient number of samples has been obtained, it executes the process of step S307. If the calculation unit 30 determines that a sufficient number of samples has not been obtained, it executes the process of step S309.

[0131] In step S307, if it is determined that the number of samples has been obtained in step S306, the calculation unit 30 calculates the average value of the flag values. The average value is calculated, for example, by setting a moving window of a predetermined time interval for the flag values ​​stored in steps S304 and S305. The calculation unit 30 determines whether the calculated average value is 0.5 or greater. If the calculation unit 30 determines that the average value is 0.5 or greater, it executes the process of step S308. If the calculation unit 30 determines that the average value is less than 0.5, it executes the process of step S309.

[0132] In step S308, if the calculation unit 30 determines in step S307 that the average value is equal to or greater than 0.5, it determines that the operation of the field device 1 is stopped.

[0133] In step S309, if the calculation unit 30 determines in step S307 that the average value is less than 0.5, it determines that the field device 1 is in operation and executes normal processing for measuring the output value of the field device 1. If the calculation unit 30 determines in step S301 that the number of samples has not been obtained, it executes normal processing for measuring the output value of the field device 1. If the calculation unit 30 determines in step S306 that the number of samples has not been obtained, it executes normal processing for measuring the output value of the field device 1.

[0134] 10 is used to determine between the first and second runs and between the second and third runs when, for example, a distillation application is performed two or three times a day. Generally, a field device may not have a sufficient hardware memory configuration due to power consumption and cost limitations.

[0135] It is easy to reduce false positives related to shutdowns by setting a wide range of average values ​​for output values, but there is a limit to the memory of the arrays used for averaging calculations. Field devices often calculate output value measurements in milliseconds. For example, to calculate the average value for 1 second with a sampling period of 10 ms, an array of 100 values ​​is required. In reality, with a 1 second interval, there is a very high risk of false positives if the sensor value or output value changes suddenly due to noise or other factors. Therefore, an array of about 60 seconds is required.

[0136] Therefore, we prepared an array for 60 seconds, using the value for every 100 cycles of 10 ms as the representative value. As a result, while it would normally be necessary to prepare an array of 6,000 values ​​to calculate the average value for 60 seconds, it is now sufficient to prepare an array containing only 60 representative values ​​for 1 second.

[0137] In determining whether the field device 1 is stopped, the calculation unit 30 does not immediately determine that the device is stopped when the average output value over 60 seconds falls below the second threshold, but instead sets a flag of 1 when the value falls below the second threshold and stores the flag in an array for 60 seconds. The calculation unit 30 similarly sets a flag of 0 when the device is operating, and calculates the average value for these flags to determine that the value is stopped when the value is above an arbitrary threshold, for example, 0.5. This type of stop determination significantly reduces the risk of false detection of a stopped state while the field device 1 is operating.

[0138] The method of calculating the average value, the process of determining whether the field device 1 is stopped, and the process of determining whether the field device 1 is running vary depending on the capabilities and application of the field device 1, and may be changed as appropriate in accordance with the specifications and application of the field device 1.

[0139] 11 and 12 are schematic diagrams for explaining a third example of the operation of the field device 1 of FIG. 1. One set of graphs showing the time changes in volumetric flow rate and density included in FIG. 11 shows the state when the stop determination process described with reference to FIG. 10 is not executed. Another set of graphs showing the time changes in volumetric flow rate and density included in FIG. 12 shows the state when the stop determination process described with reference to FIG. 10 is executed. In FIGS. 11 and 12, solid lines indicate output values ​​output by normal processing related to output value measurement of the field device 1. Dashed lines indicate output values ​​output by regression processing of the field device 1.

[0140] For example, in the graph in FIG. 11 where the vertical axis represents the volumetric flow rate, the stoppage determination process described with reference to FIG. 10 is not performed, and the field device 1 is determined to have stopped when the volumetric flow rate value falls below the second threshold and temporarily drops into the shaded area. The model is constructed by learning only when the field device 1 is operating. Therefore, when the field device 1 is not operating, it is difficult to calculate an accurate output value through regression processing. Note that the boundary line above the shaded area (positive direction of the vertical axis) in the volumetric flow rate graph in FIG. 11 corresponds to the second threshold.

[0141] If the output value drops instantaneously and the field device 1 is erroneously detected as stopped, as in the graph in Figure 11 where the vertical axis is the volumetric flow rate, the total volume during operation used in the regression process of density is reset to zero. As a result, the density output value from the regression process also returns to its initial value, as in the graph in Figure 11 where the vertical axis is the density.

[0142] To reduce such inaccurate output, the stoppage determination process described with reference to FIG. 10 is executed. According to this determination process, even when the output value drops due to momentary noise, the calculation unit 30 determines that the noise is the cause and executes a process that is less susceptible to the effects of noise. As shown in FIG. 12, the calculation unit 30 does not determine that the field device 1 has stopped when the volumetric flow rate temporarily drops below the second threshold and falls into the shaded area, but determines that the field device 1 is stopped when the volumetric flow rate continues to fall below the second threshold. The time variation of the mass flow rate also exhibits a similar trend to that of the volumetric flow rate. Therefore, the calculation unit 30 can perform the determination process using the mass flow rate instead of the volumetric flow rate in the same manner as for the volumetric flow rate.

[0143] Fig. 13 is a flowchart for explaining a fourth example of the operation of the field device 1 of Fig. 1. Fig. 13 shows an outline of the correction process for the output value output by the regression process of the field device 1.

[0144] It is expected that output values ​​such as the density of a fluid may change over time during operation, as in a distillation process. When regression processing is performed on such a fluid, the output value is highly dependent on the output value of the fluid at the beginning of the process. Therefore, if the output value at the start of the process differs for each operation due to differences in the user's operating method, the output value output by the regression processing may differ.

[0145] The calculation unit 30 corrects the output value based on the difference between the output value output by the regression processing and the output value when no air bubbles are mixed into the fluid. More specifically, the calculation unit 30 stores in the memory unit 23 the difference between the output value under normal conditions when no air bubbles are mixed into the fluid and the output value based on the regression processing when an air bubble is detected. The calculation unit 30 performs correction processing for the output value by adding a correction value to the output value when an air bubble is detected. The calculation unit 30 stores the differences used in the correction processing in an array in the memory unit 23, removes outliers, and then calculates the average of the differences to use as the correction value.

[0146] In step S400, the calculation unit 30 determines whether or not the inclusion of air bubbles in the fluid has been detected, for example, using a method similar to that of step S203 in Fig. 9. If the calculation unit 30 determines that the inclusion of air bubbles has been detected, it executes the process of step S401. If the calculation unit 30 determines that the inclusion of air bubbles has not been detected, it executes the process of step S407.

[0147] In step S401, when the calculation unit 30 determines that air bubbles have been detected in step S400, it stores in the memory unit 23 the difference between the output value under normal circumstances when no air bubbles have been detected in the fluid and the output value based on the regression processing when an air bubble has been detected.

[0148] In step S402, the calculation unit 30 determines whether or not a sufficient number of samples has been obtained to calculate the average value of the differences stored in step S401. If the calculation unit 30 determines that a sufficient number of samples has been obtained, it executes the process of step S403. If the calculation unit 30 determines that a sufficient number of samples has not been obtained, it executes the process of step S406.

[0149] In step S403, when it is determined that the number of samples has been obtained in step S402, the calculation unit 30 excludes outliers from the differences stored in step S401.

[0150] In step S404, the calculation unit 30 calculates the average value of the differences stored in step S401 as a correction value after removing the outliers in step S403.

[0151] In step S405, the calculation unit 30 adds the correction value calculated in step S404 to the output value obtained by the regression processing when an air bubble is detected, and outputs the corrected output value to, for example, the output unit 22.

[0152] In step S406, if it is determined that the number of samples has not been obtained in step S402, the calculation unit 30 outputs the output value obtained by the regression process to the output unit 22, for example.

[0153] In step S407, if the calculation unit 30 determines that no air bubbles have been detected in step S400, it outputs to the output unit 22, for example, an output value obtained during normal processing in which neither regression processing nor correction processing has been performed.

[0154] The algorithm used in the field device 1 will be described in detail below as an example.

[0155] At the same time as the start of operation of the field device 1, an internal timer count starts. Every time the timer count reaches 100 cycles, the calculation unit 30 of the field device 1 stores in the memory unit 23 at least one of the output value and the sensor value output by normal processing, which are used to determine whether operation has stopped.

[0156] To determine whether or not air bubbles are present, the calculation unit 30 must acquire sample values ​​equal to or greater than the number of arrays prepared. For example, if the calculation unit 30 prepares an array for 60 seconds, it will calculate the average value after 60 seconds have elapsed. Therefore, even if air bubbles are present in the fluid, the determination process for determining whether or not air bubbles are present will not function until the average value is calculated. Therefore, there is a trade-off between the size of the array, which is a measure to reduce false detection, and the loading time for determining whether or not air bubbles are present. However, the timer count starts after the field device 1 is powered on, so it is executed only once. Therefore, the above does not usually pose a major problem.

[0157] When the timer count reaches or exceeds the prepared array, the calculation unit 30 always simultaneously executes the determination process regarding the presence or absence of air bubbles and the determination process regarding the operation stop. The first threshold value and the second threshold value obtained by machine learning using the above-mentioned discrimination model L2 are implemented in the field device 1.

[0158] In the process of determining whether or not air bubbles are present, the calculation unit 30 compares the average value of at least one type of sensor value with the first threshold value, as described above. In the process of determining whether or not operation should be stopped, the calculation unit 30 compares the output value, which is the application purpose, with the second threshold value, as described above.

[0159] If the calculation unit 30 determines that air bubbles are mixed in the fluid, it executes regression processing using a pre-constructed, trained regression model L1 and outputs an output value. If the calculation unit 30 determines that the field device 1 is stopped, it determines that operation of an application such as distillation has stopped, and sets the timer count, the correction value of the output value, and the corresponding difference array to 0.

[0160] As a result, the timer count is reset, and when the application process such as distillation is restarted, the output values ​​are stored in the array again. As a result, the calculation unit 30 repeatedly executes the process of determining whether or not air bubbles are mixed in and the process of determining whether or not operation has been stopped.

[0161] The field device 1 according to the first embodiment described above can measure the output value more accurately even when contaminants are mixed into the fluid as the output value of the fluid changes over time. The field device 1 acquires the results of at least one of a regression process and a discrimination process related to the output value, which are executed using a model previously constructed based on data obtained when contaminants are mixed into the fluid and the acquired data. This makes the field device 1 easy to apply to fluids whose output value changes over time, such as in a distillation process. The field device 1 can output an accurate measurement value of the output value through the regression process, even when contaminants such as air bubbles are mixed into such a fluid.

[0162] The field device 1 obtains a result by executing at least one of a regression process and a discrimination process. The field device 1 executes at least one of the regression process and the discrimination process by itself. This allows the field device 1 to internally complete various processes from the learning phase for building a model to the estimation phase using the trained model.

[0163] The field device 1 can perform more accurate learning processing based on the labels by constructing a model in advance through supervised learning using data and labels from when a contaminant is mixed into a fluid. The field device 1 can perform at least one of a regression processing and a discrimination processing with higher accuracy using the model constructed through such a learning processing.

[0164] The field device 1 acquires the results when the regression process and the discrimination process are executed using different models. The field device 1 uses independent models for the regression process and the discrimination process. By using multiple models in this way, the field device 1 can achieve more accurate output in the regression process and the discrimination process.

[0165] The field device 1 can determine the first threshold value for determining whether or not a contaminant is present by using a model, thereby more appropriately determining the first threshold value. The field device 1 can objectively and accurately determine the first threshold value, which has conventionally been arbitrarily set by a user based on the user's experience, etc. This allows the field device 1 to more accurately perform the process of determining whether or not a contaminant is present using the first threshold value.

[0166] When the field device 1 determines that a contaminant has been mixed into the fluid, it executes regression processing using the model and outputs an output value. As a result, even if an abnormal value would be generated in the output value under normal processing due to the influence of a contaminant that has been mixed into the fluid, the field device 1 can regress the output value and output a more accurate measurement value.

[0167] The field device 1 corrects the output value based on the difference between the output value output by the regression process and the output value when no contaminants are present in the fluid. As a result, even if there are variations in the operating method and output value at the start of the process between operations, the field device 1 can acquire and correct errors in the output value after the regression process that occur based on such variations. The field device 1 can accurately correct the output value after the regression process even if conditions, including the type and amount of small bubbles such as microbubbles and the duration for which bubbles remain present, vary between operations.

[0168] The field device 1 can determine the second threshold for setting a flag indicating that the device is operating or stopped by using a model, thereby more appropriately determining the second threshold. The field device 1 can objectively and accurately determine the second threshold, which has conventionally been arbitrarily set by a user based on the user's experience, etc. This allows the field device 1 to more accurately perform the determination process for determining whether the device is stopped, using the second threshold, to reduce false detections of operation stoppages.

[0169] Since the output value includes at least one of the density, volume flow rate, mass flow rate, and bubble volume fraction of the fluid, the field device 1 can perform measurement and regression processing of the output value required to function as, for example, a Coriolis flowmeter. The field device 1 can function as a Coriolis flowmeter.

[0170] In the first embodiment, the field device 1 is described as constructing a model in advance by supervised learning using data and labels when a contaminant is mixed into a fluid, but this is not limited to this. The field device 1 may construct a model used in regression processing in advance by unsupervised learning or semi-supervised learning. The field device 1 may construct a model used in discrimination processing in advance by unsupervised learning or semi-supervised learning.

[0171] In the first embodiment, the field device 1 acquires results when the regression process and the discrimination process are performed using different models, but this is not limiting. The field device 1 may use the same common model for the regression process and the discrimination process instead of using independent models for each process. Conversely, the field device 1 may further use different regression models L1 for different types of output values ​​in the regression process. For example, the regression models L1 for mass flow rate, volumetric flow rate, and density do not necessarily need to be identical to each other. The field device 1 may further use different discrimination models L2 for different types of discrimination in the discrimination process.

[0172] In the first embodiment, the field device 1 determines the first threshold value for determining whether or not a contaminant is present using a model, but this is not limiting. The field device 1 does not have to perform such a determination process using a model. The field device 1 may determine the first threshold value to a value appropriately set by a user.

[0173] In the first embodiment, when the field device 1 determines that a contaminant has been mixed into the fluid, it executes regression processing using a model and outputs an output value. However, the present invention is not limited to this. The field device 1 does not necessarily have to execute such regression processing.

[0174] In the first embodiment, the field device 1 corrects the output value output by the regression process, but the present invention is not limited to this. The field device 1 does not necessarily have to perform such correction processing.

[0175] In the first embodiment, the field device 1 determines the second threshold value for setting the operating or stopped flag using a model, but this is not limiting. The field device 1 does not need to perform such a determination process using a model. The field device 1 may determine the second threshold value to a value appropriately set by the user.

[0176] In the first embodiment, the output value includes at least one of the density, volumetric flow rate, mass flow rate, and bubble volume fraction of the fluid, but is not limited to this. The output value may include any other physical quantity measurable by the field device. Accordingly, the field device 1 is not limited to a Coriolis flowmeter. The field device 1 may include other field devices.

[0177] In the first embodiment, the fluid is described as including a liquid, but is not limited to this and may include a gas.

[0178] (Second embodiment) Fig. 14 is a block diagram showing a schematic configuration of a field device 1 according to a second embodiment of the present disclosure. The configuration and functions of the field device 1 according to the second embodiment will be mainly described with reference to Fig. 14. The field device 1 according to the second embodiment differs from the first embodiment in that it does not have an analysis unit 34 by itself.

[0179] Other configurations, functions, effects, modifications, etc. are the same as those of the first embodiment, and the corresponding explanations also apply to the field device 1 according to the second embodiment. In the following, components that are the same as those of the first embodiment are given the same reference numerals, and their explanations will be omitted. Differences from the first embodiment will be mainly explained.

[0180] In the first embodiment, the calculation unit 30 acquires a result by executing at least one of the regression processing and the discrimination processing. However, this is not limited to this. The calculation unit 30 does not have to execute at least one of the regression processing and the discrimination processing by itself. The calculation unit 30 may acquire the result of at least one of the regression processing and the discrimination processing executed outside the field device 1 via communication. For example, the calculation unit 30 may acquire the result of at least one of the regression processing and the discrimination processing executed by an external device from an external device via communication. In other words, the analysis unit 34 including both the discrimination unit 35 and the regression unit 36 ​​may be provided outside the field device 1.

[0181] The external device having the analysis unit 34 includes, for example, one or more server devices capable of communicating with each other, used in edge computing, cloud systems, etc. The external device is not limited to these, and may include any general-purpose electronic device such as a mobile device, a mobile phone, a smartphone, a tablet, etc., a PC (Personal Computer), or another electronic device dedicated to the field device 1.

[0182] The calculation unit 30 of the field device 1 further includes a data communication unit 37. The data communication unit 37 includes a communication module conforming to any communication standard based on wireless or wired communication. The communication standard includes a wireless LAN (Local Area Network) standard, a short-range wireless communication standard, mobile communication standards such as 4G (4th Generation) and 5G (5th Generation), and Internet standards. The field device 1 is connected to the above-mentioned external devices via the data communication unit 37 so as to be able to communicate with them.

[0183] The field device 1 does not need to perform calculations related to at least one of the regression processing and the discrimination processing by acquiring, via communication, the results of at least one of the regression processing and the discrimination processing executed outside the field device 1. Therefore, the field device 1 can reduce the calculation load related to its operation.

[0184] In the second embodiment, both the discrimination unit 35 and the regression unit 36 ​​are described as being provided outside the field device 1, but this is not limiting. Either the discrimination unit 35 or the regression unit 36 ​​may be provided inside the field device 1, and the other may be provided outside the field device 1.

[0185] In the second embodiment, various processes from the learning phase for constructing a model to the estimation phase using the learned model are executed outside the field device 1, but this is not limiting. Either one of the various processes in the learning phase or the various processes in the estimation phase may be executed by the field device 1, and the other may be executed by an external device.

[0186] [system] The information including the processing procedures, control procedures, specific names, various data, output values ​​and parameters shown in the above documents and drawings can be changed arbitrarily unless otherwise specified.

[0187] Furthermore, the components of each device shown in the figure are conceptual functional components and do not necessarily have to be physically configured as shown in the figure. In other words, the specific form of distribution and integration of each device is not limited to that shown in the figure. In other words, all or part of them can be functionally or physically distributed and integrated in any unit depending on various loads, usage conditions, etc.

[0188] Furthermore, all or any part of the processing functions performed by each device may be realized by a CPU and a program analyzed and executed by the CPU, or may be realized as hardware using wired logic.

[0189] [Hardware] Next, an example of the hardware configuration of the information processing device 4 will be described. Fig. 15 is a diagram illustrating an example of the hardware configuration. As shown in Fig. 15, the information processing device 4 has a communication device 400a, an HDD (Hard Disk Drive) 400b, a memory 400c, and a processor 400d. The components shown in Fig. 15 are connected to each other via a bus or the like.

[0190] The communication device 400a is a network interface card or the like, and communicates with other servers. The HDD 400b stores programs and DBs that operate the functions shown in FIG.

[0191] The processor 400d reads out a program that executes the same processes as those of the processing units shown in Fig. 5 from the HDD 400b etc. and loads it into the memory 400c, thereby operating a process that executes the functions described in Fig. 5 etc. For example, this process executes the same functions as those of the processing units included in the information processing device 4.

[0192] In this way, the information processing device 4 executes the learning method by reading and executing the program. The information processing device 4 can also realize functions similar to those of the above-described embodiment by reading the program from a recording medium using a medium reading device and executing the read program. Note that the program in these other embodiments is not limited to being executed by the information processing device 4. For example, the present invention can be similarly applied to cases where another computer or server executes the program, or where these execute the program in cooperation with each other.

[0193] This program can be distributed via a network such as the Internet. In addition, this program can be recorded on a computer-readable recording medium such as a hard disk, a flexible disk (FD), a CD-ROM, a magneto-optical disk (MO), or a digital versatile disk (DVD), and can be executed by being read from the recording medium by a computer.

[0194] Some examples of combinations of the disclosed technical features are set out below.

[0195] (1) The computer acquiring first data indicating information about a measurement object of a measurement device and second data indicating information about the measurement device; The parameters or structure of a trained first model that calculates estimated data based on a sensor value of a sensor provided in the measuring device are updated based on the first data and the second data. 2. An adaptive method comprising: (2) The updating process updates parameters of a second model by adding the first data and the second data to variables input to the first model. The adaptation method according to (1) above. (3) The updating process adds, to the first model, a function that transforms the estimation data, and whose calculation method is determined by the first data and the second data. The adaptation method according to (1) or (2) above. (4) The acquiring process acquires, as the first data, one or more of a composition, a composition ratio, a physical constant, a thermal conductivity, a heat transfer coefficient, and a concentration of the measurement object. The adaptation method according to any one of (1) to (3), characterized in that: (5) The acquiring process acquires data indicating physical characteristics of the measurement device as the second data. The adaptation method according to any one of (1) to (4), characterized in that: (6) The acquiring process acquires, as the second data, one or more of the outer shape, shape, size, length, and weight of the measuring device. The adaptation method according to any one of (1) to (5), (7) The acquiring process acquires data indicating information for identifying the measurement device as the second data. The adaptation method according to any one of (1) to (6), (8) Information for constructing a second model obtained by updating the first model through the updating process is stored in a server on a cloud that is accessible by the measurement device. The adaptation method according to any one of (1) to (7), (9) On the computer, acquiring first data indicating information about a measurement object of a measurement device and second data indicating information about the measurement device; The parameters or structure of a trained first model that calculates estimated data based on a sensor value of a sensor provided in the measuring device are updated based on the first data and the second data. An adaptive program that causes a process to be executed. (10) A computer-readable non-transitory storage medium, On the computer, acquiring first data indicating information about a measurement object of a measurement device and second data indicating information about the measurement device; The parameters or structure of a trained first model that calculates estimated data based on a sensor value of a sensor provided in the measuring device are updated based on the first data and the second data. A storage medium that stores an adaptive program for executing processing. [Explanation of symbols]

[0196] 1 Field devices 4. Information processing equipment 10. Detection unit 11 Measuring tube 12 Shaker 13 Upstream sensor 14 Downstream sensor 15 Temperature Sensor 20 Processing section 21 Excitation circuit 22 Output section 23 Memory section 30 Arithmetic section 31 Density calculation section 32 Mass flow rate calculation section 33 Volumetric flow rate calculation unit 34 Analysis Units 35 Discrimination Unit 351 Bubble detector 352 Stop determination section 353 Phase state determination unit 36 Recurrence Unit 361 Density calculation section 362 Mass flow rate calculation section 363 Volumetric flow calculation unit 364 Bubble volume fraction calculation unit 37 Data Communications Department 411 Acquisition Department 412 Update Department 413 Provision Department 42 Storage section 421 First Model Information 422 Second Model Information IR drive current L1 regression model L2 discrimination model SA Displacement Signal SB Displacement signal ST temperature signal

Claims

1. The computer acquiring first data indicating information about a measurement object of a measurement device and second data indicating information about the measurement device; The parameters or structure of a trained first model that calculates estimated data based on a sensor value of a sensor provided in the measuring device are updated based on the first data and the second data.

2. An adaptive method comprising:

2. The updating process updates parameters of a second model by adding the first data and the second data to variables input to the first model.

2. The method of claim 1 .

3. The updating process adds, to the first model, a function that transforms the estimation data, the calculation method of which is determined by the first data and the second data.

2. The method of claim 1 .

4. The acquiring process acquires, as the first data, one or more of a composition, a composition ratio, a physical constant, a thermal conductivity, a heat transfer coefficient, and a concentration of the measurement object.

2. The method of claim 1 .

5. The acquiring process acquires data indicating physical characteristics of the measuring device as the second data.

2. The method of claim 1 .

6. The acquiring process acquires, as the second data, one or more of an outer shape, a shape, a size, a length, and a weight of the measuring device.

6. The method of claim 5.

7. The acquiring process acquires data indicating information for identifying the measuring device as the second data.

2. The method of claim 1 .

8. Information for constructing a second model obtained by updating the first model through the updating process is stored in a server on a cloud that is accessible by the measurement device.

2. The method of claim 1 .

9. an acquisition unit that acquires first data indicating information about a measurement target of a measurement device and second data indicating information about the measurement device; an update unit that updates parameters or a structure of a trained first model that calculates estimated data based on a sensor value of a sensor provided in the measurement device, based on the first data and the second data; An information processing device comprising:

10. a sensor for measuring a state of the fluid; a storage unit that stores information about a second model obtained by updating parameters or a structure of a learned first model that calculates estimated data based on a sensor value of the sensor based on first data that indicates information about the fluid and second data that indicates information about the device itself; and a calculation unit that calculates estimated data based on the sensor value using the second model; An information processing device comprising:

Images