Machine learning device, design support device, inference device, machine learning method, design support method, and inference method

Abstract

We provide a machine learning device that enables visualization of the fluid flowing through a pump. [Solution] The machine learning device 3 includes a learning data acquisition unit 300 that acquires multiple sets of learning data 11 consisting of input data and output data, a machine learning unit 301 that uses the multiple sets of learning data 11 acquired by the learning data acquisition unit 300 to train a learning model 11 on the correlation between input data and output data using machine learning, and a trained model storage unit 33 that stores the learning model 11 on which the correlation has been trained by the machine learning unit 301. The input data includes pump design information indicating design requirement values ​​for design items when designing a pump, and the output data includes fluid feature information indicating fluid feature values ​​of the fluid at each element position of the fluid flowing through the pump.

Description

Technical Field

[0001] The present invention relates to a machine learning device, a design support device, an inference device, a machine learning method, a design support method, and an inference method.

Background Art

[0002] "Specific speed Ns" is a parameter that characterizes the fluid flow inside a pump, and is the most important similarity law that governs the performance characteristics of the pump. For this reason, various pump types and performance characteristics are organized by the specific speed Ns, and in pump design, a series development method based on the specific speed Ns is adopted. Note that the specific speed Ns when the units of rotational speed N, flow rate Q, and head H are [min-1], [m3 / min], and [m] respectively is calculated by the following formula. Ns=N·Q1 / 2 / H3 / 4 [min-1,m3 / min,m]

[0003] In pump design, a baseline pump type with a specific speed Ns close to a value that satisfies the design requirements of each design item such as pressure and flow rate is selected, and the shape of the impeller of that pump is adjusted (trimming) to optimize the pump performance. For example, Patent Document 1 discloses a design support system for improving the efficiency of centrifugal pumps with a relatively low specific speed Ns.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] The design support system disclosed in Patent Document 1 is primarily for centrifugal pumps, and a new design support system is required for pumps of other types. Therefore, a general-purpose system for designing the shape of an impeller without predetermining the type of pump is desirable. In such a general-purpose system, a means for visually confirming the condition of the fluid flowing through the pump when the impeller rotates is useful.

[0006] In view of the above problems, the present invention aims to provide a machine learning device, a design support device, an inference device, a machine learning method, a design support method, and an inference method that enable visualization of the condition of the fluid flowing through a pump. [Means for solving the problem]

[0007] To achieve the above objective, a machine learning apparatus according to one aspect of the present invention is: A training data acquisition unit that acquires multiple sets of training data consisting of input data and output data, A machine learning unit uses multiple sets of the training data acquired by the training data acquisition unit to train a learning model on the correlation between the input data and the output data using machine learning. The machine learning unit then stores the learned model in which the correlation has been learned, and the machine learning unit stores the learned model in a learned model. The aforementioned input data is This includes pump design information that shows the design requirements for the design items when designing a pump. The output data mentioned above is: This includes fluid feature information indicating the fluid feature values ​​of the fluid at each element position of the fluid flowing through the pump.

[0008] Furthermore, a design support device according to one aspect of the present invention is A data acquisition unit that acquires input data, The system includes a data generation unit that inputs the input data acquired by the data acquisition unit into a learning model and generates output data for said input data, The aforementioned learning model, This is a trained model that has learned the correlation between the input data and the output data using machine learning. The aforementioned input data is This includes pump design information that shows the design requirements for the design items when designing a pump. The output data mentioned above is: This includes fluid feature information indicating the fluid feature values ​​of the fluid at each element position of the fluid flowing through the pump. [Effects of the Invention]

[0009] According to one aspect of the present invention, a machine learning device is provided that generates a learning model 11 that, in response to input data including pump design information, generates output data including fluid feature information that shows the fluid feature values ​​of the fluid at each element position of the fluid flowing through the pump. Furthermore, according to one aspect of the present invention, a design support device generates output data including fluid feature information that shows the fluid feature values ​​of the fluid at each element position of the fluid flowing through the pump, in response to input data including pump design information. This makes it possible to visualize the condition of the fluid flowing through the pump based on the fluid feature values ​​at each element position shown by the fluid feature information.

[0010] Other issues, configurations, and effects will be clarified in the embodiments for carrying out the invention described later. [Brief explanation of the drawing]

[0011] [Figure 1] This is an overall diagram showing an example of pump design system 1. [Figure 2] This is a data configuration diagram showing an example of pump design information 12, impeller feature information 13, and fluid feature information 14. [Figure 3] This is a block diagram showing an example of machine learning device 3. [Figure 4A] This figure shows an example of the first training data 10A and the first training model 11A. [Figure 4B] This figure shows an example of the second training data 10B and the second training model 11B. [Figure 5]This is a block diagram showing an example of the design support device 4. [Figure 6] This is a function explanatory diagram showing an example of the design support device 4. [Figure 7] This is a hardware configuration diagram showing an example of the computer 900. [Figure 8] This is a flowchart showing an example of a machine learning method by the machine learning device 3. [Figure 9A] This is a flowchart showing an example of a design support method by the design support device 4. [Figure 9B] This is a flowchart (continuation of Fig. 9A) showing an example of the design support method by the design support device 4. [Figure 10] This is a screen configuration diagram showing an example of the feature visualization screen 15A. [Figure 11] This is a screen configuration diagram showing an example of the modified feature visualization screen 15B. [Detailed Description of the Invention]

[0012] Hereinafter, embodiments for carrying out the present invention will be described with reference to the drawings. In the following description, The scope necessary for explanation to achieve the object of the present invention is schematically shown, and the scope necessary for explanation of the relevant parts of the present invention will be mainly described, and parts whose description is omitted shall be based on known techniques.

[0013] Figure 1 is an overall diagram showing an example of the pump design system 1. The pump design system 1 functions as a system that supports a designer who designs the shape of the pump 2. The pump 2 is roughly classified into types such as centrifugal pumps (including volute pumps, volute pumps with guide vanes, etc.), mixed flow pumps (including volute mixed flow pumps, mixed flow pumps with guide vanes), and axial flow pumps, depending on the magnitude of the specific speed Ns calculated from the flow rate Q, head H, and rotational speed N. The pump design system 1 is not limited to the above types, and is a system that can be used when designing any type of turbopump.

[0014] The pump design system 1 comprises, as its main components, a machine learning device 3, a design support device 4, a design database device 5, a fluid analysis device 6, and a designer terminal device 7. Each of the devices 3 to 7 is, for example, composed of a general-purpose or dedicated computer (see Figure 7 below) and is connected to a wired or wireless network 8, enabling the mutual transmission and reception of various data and information (some information is shown in Figure 1).

[0015] The machine learning device 3 operates as the main component of the machine learning learning phase. For example, it acquires training data 10 from the design database device 5 and the fluid analysis device 6, and generates a learning model 11 used by the design support device 4 through machine learning. The trained learning model 11 is provided to the design support device 4 via the network 8, recording media, etc. The machine learning device 3 employs, for example, supervised learning as its machine learning method.

[0016] The design support device 4 operates as the main component of the machine learning inference phase and assists in the design of the pump 2 using the learning model 11 generated by the machine learning device 3. For example, the design support device 4 receives pump design information 12 from the designer terminal device 7, which indicates the design requirements for the design items when designing the pump 2, and generates impeller feature information 13 and fluid feature information 14 as feature information that can be used as reference when designing the shape of the pump 2, based on the pump design information 12.

[0017] The design database device 5 stores previously designed data 50, which includes design requirements, design results, and evaluation results for pump 2, from when a designer (or another designer) designed pump 2 through trial and error in the past. The previously designed data 50 includes, for example, pump design information 12 as design requirements, pump design information 12 when pump 2 was designed, shape parameters of pump 2 and impeller feature information 13 generated from the shape parameters of pump 2 as design results, and fluid feature information 14 generated by experiments using the actual pump 2 or a model, or by simulations using the fluid analysis device 6, as evaluation results for pump 2. The previously designed data 50 is used as training data 10 by the machine learning device 3.

[0018] The fluid analysis device 6 generates fluid characteristic information 14 of the pump 2, which is identified by the pump's shape parameters and pump design information 12, by performing simulations based on computational fluid dynamics (CFD). The simulation results from the fluid analysis device 6 are used as training data 10 by the machine learning device 3.

[0019] The designer terminal device 7 is a terminal device used by the designer. The designer terminal device 7 accepts various input operations (e.g., pump design information 12, etc.) via a display screen such as an application or browser, and displays various information (e.g., impeller characteristic information 13 and fluid characteristic information 14, etc.) via the display screen. Although Figure 1 shows only one designer terminal device 7, multiple designer terminal devices 7 may be connected to the pump design system 1. The designer terminal device 7 may be used by any user other than the designer.

[0020] Figure 2 is a data configuration diagram showing an example of pump design information 12, impeller feature information 13, and fluid feature information 14.

[0021] Pump design information 12 is information that shows the design requirements for the design items when designing pump 2. Pump design information 12 includes design requirements for at least one of the design items: suction pressure, discharge pressure, suction temperature, flow rate, fluid type, and fluid characteristics. The fluid type is specified by, for example, purified water, wastewater, seawater, liquefied petroleum, etc. Fluid characteristics are defined by, for example, temperature, density, compressibility, viscosity, etc. In the example in Figure 2, the pump design information 12 is illustrated in which the design requirements include design requirements for suction pressure, discharge pressure, suction temperature, and flow rate. Note that the design items are not limited to the example in Figure 2, and other design items may be used.

[0022] Impeller feature information 13 is information that indicates the shape feature value of the impeller equipped in pump 2. The impeller shape feature value represents the characteristics when the shape of pump 2 is classified by, for example, the type of pump 2 or the number of impeller stages. In the example in Figure 2, the impeller shape feature value is classified into, for example, impeller feature 1 to impeller feature N, and impeller feature 1 to impeller feature 6 are assigned to a 2-stage centrifugal pump, a 1-stage centrifugal pump, a 2-stage mixed-flow pump, a 1-stage mixed-flow pump, a 2-stage axial-flow pump, and a 1-stage axial-flow pump, respectively. In this case, impeller feature information 13 shows "Impeller feature 2 (1-stage centrifugal pump)" as the shape feature value of the impeller.

[0023] The definition of impeller shape feature values ​​is not limited to the example in Figure 2, and the classification method and number of classifications can be changed as appropriate. For example, impeller shape feature values ​​may be classified based on the specific speed Ns. Also, impeller shape feature values ​​may represent the shape of the impeller in terms of size or numerical values. For example, impeller feature information 13 may include at least one of the following as impeller shape design values: blade width, number of blades, blade exit angle, inlet diameter, and outlet diameter.

[0024] The fluid feature information 14 is information that shows the fluid feature values ​​of the fluid at each element position of the fluid flowing through pump 2. The fluid element positions are represented as positions in a two-dimensional coordinate system or a three-dimensional coordinate system, and are, for example, arranged at predetermined intervals. The fluid feature values ​​of the fluid represent the characteristics of the fluid flowing around the impeller and inside the case of pump 2 when pump 2 is operating. The fluid feature value at each element position (pressure in the example in Figure 2) can be visualized by representing it as a grayscale or color contour plot. The fluid feature information 14 includes at least one of the following as fluid feature values ​​for each element position: pressure, velocity, vorticity, and density. In this case, the fluid feature value may be a scalar value, a vector value, or a value along the X, Y, and Z axes, respectively. Also, the fluid feature value may be a relative value or an absolute value. In the example shown in Figure 2, the fluid feature information 14 includes the pressure and velocity at each position in the 3D coordinate system as fluid feature values ​​of the fluid at each element position of the fluid.

[0025] Furthermore, the fluid feature values ​​are not limited to the example in Figure 2, and other fluid feature values ​​may be used. Also, the fluid feature information 14 may further include impeller feature values ​​relating to the impeller when the pump 2 is operating, and the impeller feature values ​​may include at least one of the axial thrust value, radial thrust value, deformation amount at each position of the impeller, and stress at each position of the impeller. Furthermore, the fluid feature information 14 may further include pump performance feature values ​​relating to the pump performance when the pump 2 is operating, and the pump performance feature values ​​may include at least one of the performance curve (QH curve) based on the relationship between the flow rate Q and head H of the pump 2, and the performance curve (Q-η curve) based on the relationship between the flow rate Q and efficiency η of the pump 2. good.

[0026] (Machine learning device 3) Figure 3 is a block diagram showing an example of a machine learning device 3. The machine learning device 3 comprises a control unit 30, a communication unit 31, a training data storage unit 32, and a trained model storage unit 33. The machine learning device 3 is composed of, for example, the computer 900 shown in Figure 7, which will be described later.

[0027] The control unit 30 functions as a learning data acquisition unit 300 and a machine learning unit 301. The communication unit 31 is connected to external devices (e.g., design support device 4, design database device 5, fluid analysis device 6, etc.) via the network 8 and functions as a communication interface for sending and receiving various types of data.

[0028] The training data acquisition unit 300 is connected to an external device, for example, via a network 8, and acquires training data 10, which consists of input data and output data. The training data 10 is used as training data, validation data, and test data in supervised learning. The output data that makes up the training data 10 is used as correct labels in supervised learning.

[0029] Two methods are given below as examples of how the learning data acquisition unit 300 acquires the learning data 10. In the first method, the learning data acquisition unit 300 receives multiple pre-designed data 50 from the design database device 5 and acquires multiple sets of learning data 10 based on the pump design information 12, impeller feature information 13, and fluid feature information 14 contained in each pre-designed data 50. In the second method, the learning data acquisition unit 300 works in cooperation with the fluid analysis device 6 and performs a simulation while appropriately varying the simulation conditions, thereby acquiring multiple sets of learning data 10 based on the simulation results. For example, the learning data acquisition unit 300 generates multiple simulation conditions by varying the shape parameters of the pump 2 and the pump design information 12 within a predetermined range, and acquires multiple sets of learning data 10 based on the fluid feature information 14 generated as a simulation result for each simulation condition.

[0030] The training data acquisition unit 300 acquires multiple sets of training data 10 by repeatedly executing the above method or by combining it as appropriate. At that time, multiple sets of training data 10 are acquired so that the specific speed Ns of the pump 2 is distributed within a predetermined range (for example, 50 to 4000). The training data acquisition unit 300 may employ a method different from the above. In this embodiment, the training data acquisition unit 300 acquires a first set of training data 10A and a second set of training data 10B as training data 10, but the details of the data configuration will be described later.

[0031] The learning data storage unit 32 is a database that stores multiple sets of learning data 10 acquired by the learning data acquisition unit 300. The specific configuration of the database constituting the learning data storage unit 32 can be designed as appropriate.

[0032] The machine learning unit 301 performs machine learning using multiple sets of training data 10 stored in the training data storage unit 32. Specifically, the machine learning unit 301 inputs multiple sets of the first training data 10A into the first learning model 11A and trains the first learning model 11A to learn the correlation between the first input data and the first output data contained in the first training data 10A, thereby generating a trained first learning model 11A. The machine learning unit 301 also inputs multiple sets of the second training data 10B into the second learning model 11B and trains the second learning model 11B to learn the correlation between the second input data and the second output data contained in the second training data 10B, thereby generating a trained second learning model 11B.

[0033] The machine learning unit 301 can employ any method when performing machine learning, such as online learning, batch learning, or mini-batch learning. The machine learning unit 301 may also perform predetermined preprocessing on the input data to be input to the learning model 11, or perform predetermined postprocessing on the output data output from the learning model 11.

[0034] The trained model storage unit 33 is a database that stores the trained model 11 (specifically, the adjusted weight parameter set) generated by the machine learning unit 301. The trained model 11 stores a first trained model 11A and multiple second trained models 11B, each corresponding to a shape feature value of the impeller. In the example in Figure 3, the case where multiple second trained models 11B are stored corresponds to impeller features 1 to N.

[0035] The trained model 11 stored in the trained model storage unit 33 is provided to the actual system (for example, the design support device 4) via the network 8 or a recording medium. In Figure 3, the training data storage unit 32 and the trained model storage unit 33 are shown as separate storage units, but they may be composed of a single storage unit. Furthermore, the training data storage unit 32 or the trained model storage unit 33 may be replaced by the storage unit of an external computer (for example, a server-type computer or a cloud-type computer), in which case the machine learning unit 301 only needs to access the external computer.

[0036] Figure 4A shows an example of the first training data 10A and the first learning model 11A. The first training data 10A consists of first input data and first output data and is used for machine learning of the first learning model 11A.

[0037] The first input data includes pump design information 12, which shows the design requirements for the design items when designing pump 2. The details of pump design information 12 are the same as those shown in Figure 2, so the explanation is omitted.

[0038] The first output data includes impeller feature information 13, which shows the shape feature values ​​of the impeller of the pump 2, and fluid feature information 14, which shows the fluid feature values ​​of the fluid at each element position of the fluid flowing through the pump 2. The details of the impeller feature information 13 and fluid feature information 14 are the same as those shown in Figure 2, so the explanation is omitted.

[0039] The first learning model 11A employs, for example, a neural network structure and comprises an input layer 110, a hidden layer 111, and an output layer 112. Synapses (not shown) connect each neuron between each layer, and each synapse is associated with a weight. The weight parameter set, consisting of the weights of each synapse, is adjusted by machine learning.

[0040] The input layer 110 has a number of neurons corresponding to the pump design information 12 as the first input data, and each value of the pump design information 12 is input to each neuron. The output layer 112 has a number of neurons corresponding to the impeller feature information 13 and fluid feature information 14 as the first output data, and the prediction results (inference results) of the impeller feature information 13 and fluid feature information 14 for the pump design information 12 are output as the first output data.

[0041] Figure 4B shows an example of the second training data 10B and the second training model 11B. The second training data 10B consists of second input data and second output data and is used for machine learning of the second training model 11B.

[0042] The second input data includes pump design information 12, which shows the design requirements for the design items when designing pump 2. The details of pump design information 12 are the same as those shown in Figure 2, so the explanation is omitted.

[0043] The second output data includes fluid feature information 14, which shows the fluid feature values ​​of the fluid at each element position of the fluid flowing through pump 2. The details of the fluid feature information 14 are the same as those shown in Figure 2, so the explanation is omitted.

[0044] The second learning model 11B corresponds to impeller feature 1 to impeller feature N, respectively. Therefore, the second learning model 11B corresponding to impeller feature 1 is trained using the second training data 10B, which includes pump design information 12 and fluid feature information 14 for pump 2 having an impeller classified as impeller feature 1 as the second input data and second output data. Similarly, the second learning model 11B corresponding to impeller feature N is trained using the second training data 10B, which includes pump design information 12 and fluid feature information 14 for pump 2 having an impeller classified as impeller feature N as the second input data and second output data. The other second learning models 11B are also trained using their respective second training data 10B.

[0045] The second learning model 11B, like the first learning model 11A, employs, for example, a neural network structure and comprises an input layer 110, a hidden layer 111, and an output layer 112. Synapses (not shown) connect each neuron between each layer, and each synapse is associated with a weight. The weight parameter set, consisting of the weights of each synapse, is adjusted by machine learning.

[0046] The input layer 110 has a number of neurons corresponding to the pump design information 12 as second input data, and each value of the pump design information 12 is input to each neuron. The output layer 112 has a number of neurons corresponding to the fluid feature information 14 as second output data, and the prediction result (inference result) of the fluid feature information 14 for the pump design information 12 is output as second output data.

[0047] In this embodiment, the data configuration of the training data 10 and the training model 11 was described as being as shown in Figures 4A and 4B. However, multiple data configurations with different conditions, such as machine learning methods and data types, may be adopted. In that case, the training data acquisition unit 300 acquires multiple types of training data 10 corresponding to the multiple data configurations with different conditions, and the machine learning unit 301 performs machine learning using each of these training data 10.

[0048] (Design support device 4) Figure 5 is a block diagram showing an example of the design support device 4. Figure 6 is a functional diagram showing an example of the design support device 4. The design support device 4 comprises a control unit 40, a communication unit 41, and a learned model storage unit 42. The design support device 4 is composed of, for example, the computer 900 shown in Figure 7, which will be described later.

[0049] The control unit 40 functions as a data acquisition unit 400, a data generation unit 401, and an output processing unit 402. The data acquisition unit 400 functions as a first data acquisition unit 400A and a second data acquisition unit 400B. The data generation unit 401 functions as a first data generation unit 401A and a second data generation unit 401B. The communication unit 41 is connected to external devices (e.g., machine learning device 3, design database device 5, etc.) via the network 8 and functions as a communication interface for sending and receiving various types of data.

[0050] The trained model storage unit 42 is a database that stores a first trained model 11A and a plurality of second trained models 11B corresponding to impeller features 1 to N, respectively, as trained trained trained models 11. The number of trained trained models 11 stored in the trained model storage unit 42 is not limited to the above example; for example, multiple trained trained models with different conditions, such as machine learning methods and data types, may be stored and used selectively or in parallel. The trained model storage unit 42 may be replaced by the storage unit of an external computer (for example, a server-type computer or a cloud-type computer), in which case the design support device 4 only needs to access the external computer.

[0051] The first data acquisition unit 400A is connected to an external device via the communication unit 41 and the network 8 to acquire first input data. For example, the first data acquisition unit 400A acquires first input data by receiving the pump design information 12 of the pump 2 to be designed, which has been entered by the designer on the design requirement input screen displayed on the designer terminal device 7, from the designer terminal device 7. Alternatively, the first data acquisition unit 400A may acquire first input data by referring to the pump design information 12 included in the existing design data 50.

[0052] The first data generation unit 401A generates first output data for the first input data acquired by the first data acquisition unit 400A by inputting the first input data into the first learning model 11A.

[0053] The second data acquisition unit 400B acquires second input data based on the first input data acquired by the first data acquisition unit 400A. In this case, the second data acquisition unit 400B acquires second input data based on the modified pump design information 12 in which the design requirements have been changed for the pump design information 12 included in the first input data. For example, the second data acquisition unit 400B acquires second input data by receiving the modified pump design information 12 in which the design requirements have been changed by the designer for the feature visualization screen displayed on the designer terminal device 7 from the designer terminal device 7.

[0054] The second data generation unit 401B generates second output data for the second input data acquired by the second data acquisition unit 400B by inputting the second input data into one of the multiple second learning models 11B. In this case, the second data generation unit 401B generates the second output data by inputting the second input data into the second learning model 11B that corresponds to the shape feature value indicated by the impeller feature information 13 included in the first output data, among the multiple second learning models 11B. In the example in Figure 6, since the impeller feature information 13 included in the first output data indicates the impeller feature N as a shape feature value, the second data generation unit 401B is shown inputting the second input data into the second learning model 11B corresponding to the impeller feature N.

[0055] Furthermore, the first data generation unit 401A may perform predetermined preprocessing on the first input data to be input to the first learning model 11A, and the second data generation unit 401B may perform predetermined preprocessing on the second input data to be input to the second learning model 11B. In addition, the first data generation unit 401A may perform predetermined postprocessing on the first output data output from the first learning model 11A, and the second data generation unit 401B may perform predetermined postprocessing on the second output data output from the second learning model 11B.

[0056] The output processing unit 402 generates visualization information for visualizing the fluid feature values ​​of each element position indicated by the fluid feature information 14, based on the first output data generated by the first data generation unit 401A or the second output data generated by the second data generation unit 401B, and provides this information to, for example, the designer terminal device 7. The visualization information is then visualized by the designer terminal device 7. By displaying it on the screen, the designer can visually confirm the fluid characteristic values ​​at each element's location.

[0057] (Hardware configuration of each device) Figure 7 is a hardware configuration diagram showing an example of computer 900. Each of the devices 3 to 7 and the pump 2 that make up the pump design system 1 are configured with a general-purpose or dedicated computer 900.

[0058] As shown in Figure 7, the computer 900 comprises, as its main components, a bus 910, a processor 912, memory 914, an input device 916, an output device 917, a display device 918, a storage device 920, a communication interface unit 922, an external device interface unit 924, an I / O device interface unit 926, and a media input / output unit 928. Note that the above components may be omitted as appropriate depending on the intended use of the computer 900.

[0059] The processor 912 consists of one or more arithmetic processing units (CPU (Central Processing Unit), MPU (Micro-Processing Unit), DSP (Digital Signal Processor), GPU (Graphics Processing Unit), NPU (Neural Processing Unit), etc.) and operates as a control unit that oversees the entire computer 900. The memory 914 stores various data and programs 930 and consists of volatile memory (DRAM, SRAM, etc.) that functions as main memory, and non-volatile memory (ROM), flash memory, etc.

[0060] The input device 916 consists of, for example, a keyboard, mouse, numeric keypad, electronic pen, joystick, microphone, etc., and functions as an input unit. The output device 917 consists of, for example, a sound (voice) output device, a vibration device, etc., and functions as an output unit. The display device 918 consists of, for example, a liquid crystal display, an organic EL display, electronic paper, a projector, etc., and functions as an output unit. The input device 916 and the display device 918 may be configured as an integrated unit, such as a touch panel display. The storage device 920 consists of, for example, an HDD, an SSD, etc., and functions as a storage unit. The storage device 920 stores various data necessary for the execution of the operating system and program 930.

[0061] The communication I / F unit 922 is connected by wire or wireless to a network 940 such as the Internet or an intranet (which may be the same as network 8 in Figure 1) and functions as a communication unit that sends and receives data with other computers according to a predetermined communication standard. The external device I / F unit 924 is connected by wire or wireless to external devices 950 such as cameras, printers, scanners, and reader / writers and functions as a communication unit that sends and receives data with external devices 950 according to a predetermined communication standard. The I / O device I / F unit 926 is connected to I / O devices 960 such as various sensors and actuators and functions as a communication unit that sends and receives various signals and data with the I / O devices 960, such as detection signals from sensors and control signals to actuators. The media input / output unit 928 consists of, for example, a drive device such as a DVD drive or CD drive, a memory card slot, and a USB connector, and reads and writes data to media (non-temporary storage media) 970 such as DVDs, CDs, memory cards, and USB memory.

[0062] In the computer 900 having the above configuration, the processor 912 calls and executes the program 930 stored in the storage device 920 in the memory 914, and controls various parts of the computer 900 via the bus 910. The program 930 may also be stored in the memory 914 instead of the storage device 920. The program 930 is an insulator. The program 930 may be recorded on media 970 in a tollable or executable file format and provided to computer 900 via media input / output unit 928. The program 930 may also be provided to computer 900 by downloading it via network 940 through communication interface unit 922. Alternatively, computer 900 may implement various functions realized by the processor 912 executing program 930 using hardware such as FPGA (Field-Programmable Gate Array) or ASIC (Application Specific Integrated Circuit).

[0063] Computer 900 is an electronic device of any form, consisting of, for example, a stationary computer or a portable computer. Computer 900 may be a client computer, a server computer, a cloud computer, or an embedded computer such as a control panel or controller (including microcontrollers, programmable logic controllers, and sequencers).

[0064] (Machine learning methods) Figure 8 is a flowchart illustrating an example of a machine learning method using the machine learning device 3. Below, we will describe the case in which a learning model 11 (either the first or second learning model 11A or 11B) is generated using multiple sets of training data 10 (for example, either the first or second training data 10A or 10B shown in Figures 4A and 4B).

[0065] First, in step S100, the training data acquisition unit 300 acquires a desired number of training data 10 as preparation for starting machine learning, and stores the acquired training data 10 in the training data storage unit 32. The number of training data 10 prepared here should be set considering the inference accuracy required for the final learning model 11.

[0066] Next, in step S110, the machine learning unit 301 prepares a pre-training model 11 in order to start machine learning. The pre-training model 11 prepared here consists of the neural network model exemplified in Figures 4A and 4B, and the weights of each synapse are set to their initial values.

[0067] Next, in step S120, the machine learning unit 301 randomly selects, for example, one set of training data 10 from the multiple sets of training data 10 stored in the training data storage unit 32.

[0068] Next, in step S130, the machine learning unit 301 inputs the pump design information 12 (input data) contained in a set of training data 10 to the input layer 110 of the prepared pre-training (or training) learning model 11. As a result, fluid feature information 14, etc. (output data) is output from the output layer 112 of the learning model 11 as an inference result, but this output data is generated by the pre-training (or training) learning model 11. Therefore, in the pre-training (or training) state, the output data output as an inference result shows information different from the fluid feature information 14, etc. (ground truth labels) contained in the training data 10.

[0069] Next, in step S140, the machine learning unit 301 compares the fluid feature information 14 etc. (ground truth labels) included in the set of training data 10 acquired in step S120 with the fluid feature information 14 etc. (output data) output as an inference result from the output layer 112 in step S130, and performs machine learning by adjusting the weight of each synapse (backpropagation). In this way, the machine learning unit 301 allows the learning model 11 to learn the correlation between the input data and the output data.

[0070] Next, in step S150, the machine learning unit 301 determines whether predetermined learning termination conditions have been met, for example, based on the evaluation value of the error function, which is based on the fluid feature information 14 etc. (ground truth labels) included in the training data 10 and the fluid feature information 14 etc. output as inference results (output data), or based on the remaining number of untrained training data 10 stored in the training data storage unit 32.

[0071] In step S150, if the machine learning unit 301 determines that the learning termination condition has not been met and that machine learning should continue (No in step S150), it returns to step S120 and repeats steps S120 to S140 multiple times using the untrained training data 10 on the learning model 11 that is currently being trained. On the other hand, in step S150, if the machine learning unit 301 determines that the learning termination condition has been met and that machine learning should be terminated (Yes in step S150), it proceeds to step S160.

[0072] Then, in step S160, the machine learning unit 301 stores the trained model 11 (set weight parameter group) generated by adjusting the weights associated with each synapse in the trained model storage unit 33, and the series of machine learning methods shown in Figure 8 is completed. In the machine learning method, step S100 corresponds to the training data acquisition step, steps S110 to S150 are the machine learning steps, and step S160 is the trained model storage step.

[0073] As described above, the machine learning device 3 and machine learning method according to this embodiment provide a learning model 11 that generates output data including fluid feature information 14 showing the fluid feature values ​​of the fluid at each element position of the fluid flowing through the pump 2, based on input data including pump design information 12. This makes it possible to visualize the condition of the fluid flowing through the pump 2 based on the fluid feature values ​​at each element position shown by the fluid feature information 14.

[0074] (Design support method) Figures 9A and 9B are flowcharts illustrating an example of a design support method using the design support device 4. In the following description, it is assumed that the learned model storage unit 42 stores a first learned model 11A and a plurality of second learned models 11B corresponding to impeller features 1 to N, respectively, as learned learned models 11.

[0075] First, in step S200, when the designer terminal device 7 receives an input operation from the designer to start design support for pump 2, it displays the design requirement input screen. When the designer performs the operation to input pump design information 12 for pump performance on the design requirement input screen, in step S201, the design support device 4 transmits the input pump design information 12 to the design support device 4.

[0076] Next, in step S210, the first data acquisition unit 400A of the design support device 4 acquires the first input data by receiving the pump design information 12 input by the designer from the designer terminal device 7.

[0077] Next, in step S211, the first data generation unit 401A inputs the first input data acquired in step S210 into the first learning model 11A, thereby generating first output data for the first input data.

[0078] Next, in step S212, the output processing unit 402 generates visualization information based on the first output data generated in step S211 and transmits the visualization information to the designer terminal device 7.

[0079] Next, in step S220, when the designer terminal device 7 receives visualization information from the design support device 4, it displays the feature visualization screen 15A based on that visualization information.

[0080] Figure 10 is a screen configuration diagram showing an example of the feature visualization screen 15A. The feature visualization screen 15A includes a pump design information display field 150A that displays pump design information 12 included in the first input data, an impeller feature information display field 151 that displays impeller feature information 13 included in the first output data, a fluid feature information display field 152A that displays fluid feature information 14 included in the first output data, a fluid feature value selection field 153 for selecting the fluid feature value to be displayed, a change button 154 that instructs to change the design requirement value, a save button 155 that instructs to save the visualization information, and an exit button 156 that instructs to end the design support.

[0081] The pump design information display area 150A displays the design requirements for each design item indicated in the pump design information 12. The design requirements displayed in the pump design information display area 150A are shown to be editable by the user. In the example in Figure 10, the design requirements for suction pressure, discharge pressure, suction temperature, and flow rate are shown to be editable.

[0082] The impeller feature information display area 151 displays the shape feature values ​​of the impeller indicated by the impeller feature information 13. In the example in Figure 10, "1-stage centrifugal pump" is displayed as impeller feature 2. If the impeller feature information 13 further includes at least one of the following as the shape design values ​​of the impeller: blade width, number of blades, blade exit angle, inlet diameter, and outlet diameter, these may also be displayed in the impeller feature information display area 151.

[0083] The fluid feature information display area 152A displays the fluid feature values ​​for each element location indicated by the fluid feature information 14, visualized in two or three dimensions. The display can be zoomed in and out, and the viewpoint position can be changed as desired. In the example in Figure 10, "pressure" is selected in the fluid feature value selection area 153, so the fluid feature information display area 152A displays a contour plot showing the pressure at each element location. The contour plot may display multiple fluid feature values ​​in combination; for example, a contour plot showing both "pressure" and "flow velocity" may be displayed in the fluid feature information display area 152A. In that case, for example, a "pressure / flow velocity" option can be added to the fluid feature value selection area 153. Furthermore, the fluid feature values ​​for each element location may be displayed as a dynamic animation in the fluid feature information display area 152A. Additionally, when displaying the fluid feature values ​​for each element location in the fluid feature information display area 152A, the fluid flow may be displayed as streamlines.

[0084] Furthermore, if the fluid feature information 14 further includes at least one of the following as impeller feature values: axial thrust value, radial thrust value, deformation amount at each position of the impeller, and stress at each position of the impeller, these may be displayed on the feature visualization screen 15A. In that case, the deformation amount at each position of the impeller and the stress at each position of the impeller may be displayed superimposed on the contour plot shown in Figure 10, for example. Also, if the fluid feature information 14 further includes at least one of the following as pump performance feature values: QH curve and Q-η curve, graphs representing these may be displayed in the fluid feature information display section 152A.

[0085] Then, when the designer presses the save button on the feature visualization screen 15A, the design support device 4 saves the visualization information as a still image or video in any file format. Also, when the designer changes any of the design requirements in the pump design information display field 150A on the feature visualization screen 15A and presses the change button 154, in step S221, the design support device 4 transmits the modified pump design information 12 to the design support device 4 based on the modified design requirements on the feature visualization screen 15A. In the example in Figure 9B, the changes to the pump design information 12 include a change in discharge pressure from "0.50" to "0.60" and a change in flow rate to " The diagram illustrates the case where the value is changed from "1200" to "1100".

[0086] Next, in step S230, the second data acquisition unit 400B acquires the second input data by receiving the modified pump design information 12, which has been changed by the designer, from the designer terminal device 7.

[0087] Next, in step S231, the second data generation unit 401B generates second output data by inputting the second input data acquired in step S230 into the second learning model 11B among the multiple second learning models 11B that corresponds to the shape feature value indicated by the impeller feature information 13 included in the first output data generated in step S211. In the example of Figure 10, since the impeller feature information 13 included in the first output data indicates impeller feature 2 ("centrifugal pump stage 1") as a shape feature value, the second input data is input into the second learning model 11B corresponding to impeller feature 2 ("centrifugal pump stage 1").

[0088] Next, in step S232, the output processing unit 402 generates visualization information based on the second output data generated in step S231 and transmits the visualization information to the designer terminal device 7.

[0089] Next, in step S240, when the designer terminal device 7 receives visualization information from the design support device 4, it updates the feature visualization screen 15A and displays the feature visualization screen 15B based on that visualization information.

[0090] Figure 11 is a screen configuration diagram showing an example of the modified feature visualization screen 15B. The feature visualization screen 15B has the same screen configuration as the feature visualization screen 15A shown in Figure 10, and includes a pump design information display field 150B, an impeller feature information display field 151, a fluid feature information display field 152B, a fluid feature value selection field 153, a change button 154, a save button 155, and an exit button 156.

[0091] The pump design information display area 150B displays the design requirement values ​​for each design item indicated by the modified pump design information 12 included in the second input data. At that time, the design requirement values ​​displayed in the pump design information display area 150B are displayed in a way that allows the user to further modify them.

[0092] The fluid feature information display area 152B displays the fluid feature values ​​for each element position indicated by the fluid feature information 14 included in the second output data, visualized in two or three dimensions.

[0093] Then, when the designer determines that the design requirements for pump 2 are met with the fluid characteristic values ​​of each element position visualized in the fluid characteristic information display section 152B of the feature visualization screen 15B, and presses the finish button 156, the series of design support methods shown in Figures 10A and 10B are terminated. In the design support method, step S210 corresponds to the first data acquisition step, step S211 to the first data generation step, step S230 to the second data acquisition step, step S231 to the second data generation step, and steps S212 and S232 to the output processing steps.

[0094] Furthermore, if the designer changes any of the design requirements in the pump design information display section 150B of the feature visualization screen 15B and presses the change button 154, the process returns to step S221, and steps S221 to S240 are performed in the same manner as above, thereby further updating the feature visualization screen 15B.

[0095] Furthermore, various types of information generated or transmitted by the design support device 4 (pump design information 12, impeller feature information 13, fluid feature information 14, and visualization information) are used by the design support device 4 and the design data The data may be stored in the base device 5.

[0096] Furthermore, in step S230, depending on the changes made to the modified pump design information 12, it may be determined whether the first data acquisition unit 400A acquires the first input data based on the modified pump design information 12, or whether the second data acquisition unit 400B acquires the second input data based on the modified pump design information 12. For example, if it is predicted that the shape feature values ​​of the impeller will change due to the changes made to the pump design information 12, the first data acquisition unit 400A should determine to acquire the first input data based on the modified pump design information 12. In that case, the process should return to step S210, and the processing from step S210 onward should be carried out in the same manner as described above.

[0097] As described above, the design support device 4 and design support method according to this embodiment generate output data including fluid feature information 14 that shows the fluid feature values ​​of the fluid at each element position of the fluid flowing through the pump 2, based on input data including pump design information 12. This makes it possible to visualize the condition of the fluid flowing through the pump 2 based on the fluid feature values ​​at each element position shown by the fluid feature information 14.

[0098] Furthermore, the first data acquisition unit 400A and the first data generation unit 401A generate impeller feature information 13 and fluid feature information 14 for the initial pump design information 12 using the first learning model 11A, while the second data acquisition unit 400B and the second data generation unit 401B generate fluid feature information 14 for the modified pump design information 12 using the second learning model 11B selected based on the impeller feature information 13. This allows for the design of the basic shape of the impeller by identifying the shape feature values ​​of the impeller of the pump 2 to be designed, and by visualizing the condition of the fluid flowing into the pump 2 based on the fluid feature values ​​at each element position indicated by the fluid feature information 14, the design of the basic shape of the impeller can be supported. Then, by visualizing the condition of the fluid flowing into the pump 2 based on the fluid feature values ​​at each element position indicated by the fluid feature information 14, the impact of changes in the design requirements can be confirmed, thereby supporting the adjustment work by the designer. In other words, designers can fine-tune the required design values ​​while checking the visualized fluid conditions based on the initial design requirements.

[0099] (Other embodiments) The present invention is not limited to the embodiments described above, and can be implemented with various modifications without departing from the spirit of the invention. All such modifications are included in the technical concept of the present invention.

[0100] In the above embodiment, the machine learning device 3 and the design support device 4 were described as being composed of separate devices, but they may be composed of a single device. Furthermore, the machine learning device 3 and the design support device 4 may function as at least one of the design database device 5, the fluid analysis device 6, and the designer terminal device 7.

[0101] In the above embodiment, a case in which a neural network is used as the learning model 11 that realizes machine learning by the machine learning device 3 was described, but other machine learning models may also be used. Other machine learning models include, for example, tree-type models such as decision trees and regression trees, ensemble learning such as bagging and boosting, recurrent neural networks, convolutional neural networks, and neural network types such as LSTM (deep learning). (including), hierarchical clustering, non-hierarchical clustering, k-nearest neighbors, k-means clustering, etc. Examples include multivariate analyses such as Sterling's method, principal component analysis, factor analysis, logistic regression, and Gaussian process regression, as well as support vector machines and regression kriging.

[0102] (Machine learning programs and design support programs) The present invention can also be provided in the form of a program (machine learning program) 930 that causes a computer 900 to function as each part of the machine learning device 3 according to the above embodiment. Alternatively, the present invention can also be provided in the form of a program (design support program) 930 that causes a computer 900 to execute each part of the design support device 4 according to the above embodiment.

[0103] (Inference device, inference method, and inference program) The present invention can be provided not only in the form of the design support device 4 (design support method or design support program) according to the above embodiment, but also in the form of an inference device (inference method or inference program) used to support the design of the pump 2. In that case, the inference device (inference method or inference program) may include a memory and a processor, the processor of which may execute a series of processes. The series of processes includes a data acquisition process (data acquisition step) that acquires input data including pump design information 12 that indicates design requirement values ​​for design items when designing the pump 2, and an inference process (inference step) that, once the input data has been acquired by the data acquisition process, infers output data including fluid characteristic information 14 that indicates fluid characteristic values ​​of the fluid at each element position of the fluid flowing through the pump 2 from the input data.

[0104] By providing the inference device (inference method or inference program) in the form of an inference device, it becomes easier to apply to various devices compared to implementing a design support device. It will be obvious to those skilled in the art that when the inference device (inference method or inference program) infers pump performance, it may apply the inference method performed by the inference unit of the design support device using the trained learning model 11 generated by the machine learning device 3 and machine learning method according to the above embodiment. [Explanation of Symbols]

[0105] 1... Pump design system, 2... Pump, 3... Machine learning device, 4... Design support device 5…Design database device, 6…Fluid analysis device, 7…Designer terminal device, 10...Training data, 10A...First training data, 10B...Second training data, 11...Learning model, 11A...Second learning model, 11B...Second learning model, 12...Pump design information, 13...Impeller feature information, 14...Fluid feature information, 30...Control unit, 31...Communication unit, 32...Learning data storage unit, 33...Trained model memory unit, 40...Control unit, 41...Communication unit, 42...Learned model memory unit, 300...Training data acquisition unit, 301...Machine learning unit, 400...Data acquisition unit, 400A...First data acquisition unit, 400B...Second data acquisition unit, 401...Data generation unit, 401A...First data generation unit, 401B...Second data generation unit, 402...Output processing unit

Claims

1. A training data acquisition unit that acquires multiple sets of training data consisting of input data and output data, A machine learning unit uses multiple sets of the training data acquired by the training data acquisition unit to train a learning model on the correlation between the input data and the output data using machine learning. The machine learning unit then stores the learned model in which the correlation has been learned, and the machine learning unit stores the learned model in a learned model. The aforementioned input data is This includes pump design information that shows the design requirements for the design items when designing a pump. The output data mentioned above is: The fluid feature information includes fluid feature values ​​of the fluid at each element position of the fluid flowing through the pump, Machine learning device.

2. The aforementioned pump design information includes the following design items: Inhalation pressure, Discharge pressure, suction temperature, flow rate, The type of the aforementioned fluid, and This indicates the design requirement value for at least one of the characteristics of the fluid. The machine learning apparatus according to claim 1.

3. The fluid characteristic information is expressed as the fluid characteristic value at each element position, The pressure at each of the aforementioned element positions, The flow velocity at each of the aforementioned element positions, The vorticity at each of the aforementioned element positions, and This indicates at least one density at each of the aforementioned element locations. The machine learning apparatus according to claim 1.

4. The output data mentioned above is: The pump further includes impeller feature information indicating the shape feature values ​​of the impeller, A machine learning apparatus according to any one of claims 1 to 3.

5. A data acquisition unit that acquires input data, The system includes a data generation unit that inputs the input data acquired by the data acquisition unit into a learning model and generates output data for said input data, The aforementioned learning model, This is a trained model that has learned the correlation between the input data and the output data using machine learning. The aforementioned input data is This includes pump design information that shows the design requirements for the design items when designing a pump. The output data mentioned above is: This includes fluid feature information that shows the fluid feature values ​​of the fluid at each element position of the fluid flowing through the pump, Design support equipment.

6. The output data mentioned above is: The pump further includes impeller feature information indicating the shape feature values ​​of the impeller, The design support device according to claim 5.

7. The data acquisition unit, This consists of a first data acquisition unit and a second data acquisition unit. The data generation unit, This consists of a first data generation unit and a second data generation unit. The aforementioned learning model, A first trained model is a pre-trained model that has learned the correlation between the first input data and the first output data using machine learning, These are multiple second learning models, each corresponding to the aforementioned shape feature values, and which are pre-trained models that have learned the correlation between the second input data and the second output data through machine learning. The first input data includes the pump design information, The first output data includes the fluid feature information and the impeller feature information, The second input data includes the pump design information, The second output data includes the fluid characteristic information, The first data acquisition unit is, The first input data is obtained, The first data generation unit is: The first input data acquired by the first data acquisition unit is input to the first learning model to generate the first output data for the first input data. The second data acquisition unit described above is: Based on the modified pump design information, in which the design requirement value has been changed with respect to the pump design information included in the first input data acquired by the first data acquisition unit, the second input data is acquired. The second data generation unit is: The second input data acquired by the second data acquisition unit is input to the second learning model among a plurality of second learning models that corresponds to the shape feature value indicated by the impeller feature information included in the first output data generated by the first data generation unit, thereby generating the second output data. The design support device according to claim 6.

8. An inference device comprising memory and a processor, The aforementioned processor, A data acquisition process that acquires input data including pump design information that shows the design requirements for the design items when designing a pump, When the input data is acquired by the data acquisition process described above, an inference process is performed on the input data to infer output data that includes fluid feature information indicating the fluid feature values ​​of the fluid at each element position of the fluid flowing through the pump. Reasoning device.

9. The training data acquisition process involves acquiring multiple sets of training data consisting of input data and output data, A machine learning step in which a learning model learns the correlation between the input data and the output data using machine learning, using multiple sets of the training data acquired in the training data acquisition step, The system includes a trained model storage step, which stores the trained model, which has learned the correlation relationship through the machine learning step, in a trained model storage unit. The aforementioned input data is This includes pump design information that shows the design requirements for the design items when designing a pump. The output data mentioned above is: This includes fluid feature information that shows the fluid feature values ​​of the fluid at each element position of the fluid flowing through the pump, Machine learning methods.

10. The data acquisition process involves obtaining input data, The system includes a data generation step which generates output data for the input data obtained in the data acquisition step by inputting the input data into a learning model, The aforementioned learning model, This is a trained model that has learned the correlation between the input data and the output data using machine learning. The aforementioned input data is This includes pump design information that shows the design requirements for the design items when designing a pump. The output data mentioned above is: This includes fluid feature information that shows the fluid feature values ​​of the fluid at each element position of the fluid flowing through the pump, Design support method.

11. The output data mentioned above is: The pump further includes impeller feature information indicating the shape feature values ​​of the impeller, The design support method according to claim 10.

12. The aforementioned data acquisition process is as follows: This consists of a first data acquisition process and a second data acquisition process. The aforementioned data generation process is: This consists of a first data generation process and a second data generation process. The aforementioned learning model, A first trained model is a pre-trained model that has learned the correlation between the first input data and the first output data using machine learning, These are multiple second learning models, each corresponding to the aforementioned shape feature values, and which are pre-trained models that have learned the correlation between the second input data and the second output data through machine learning. The first input data includes the pump design information, The first output data includes the fluid feature information and the impeller feature information, The second input data includes the pump design information, The second output data includes the fluid characteristic information, The first data acquisition step is: The first input data is obtained, The first data generation step is: By inputting the first input data acquired in the first data acquisition step into the first learning model, the first output data is generated for the first input data. The second data acquisition step described above is: Based on the modified pump design information obtained by the first data acquisition step, in which the design requirement values ​​have been changed from the pump design information included in the first input data obtained by the first data acquisition step, the second input data is acquired. The second data generation step is as follows: The second input data acquired in the second data acquisition step is selected from among a plurality of second learning models, and the second learning model corresponds to the shape feature value indicated by the impeller feature information included in the first output data generated in the first data generation step. By inputting this, the second output data is generated. The design support method according to claim 11.

13. An inference method performed by an inference device comprising memory and a processor, The aforementioned processor, A data acquisition process that acquires input data including pump design information that shows the design requirements for the design items when designing a pump, When the input data is acquired by the data acquisition process described above, an inference process is performed on the input data to infer output data that includes fluid feature information indicating the fluid feature values ​​of the fluid at each element position of the fluid flowing through the pump. Reasoning method.

Images