Method and system for providing operating condition of manufacturing device

Abstract

To provide robust operating conditions resistant to disturbances and external noise.SOLUTION: A method for providing operating conditions of a manufacturing device which manufactures products using an input device, an output device, a processor, and a memory includes: a variation amount input step; a model reading step in which the processor reads a model in which the input is the operating conditions and the output is a characteristic value of the products from the memory; a predicted characteristic value distribution calculation step in which the processor calculates a predicted characteristic value distribution reflecting the variation amount on the predetermined operating conditions using the model; an objective function calculation step in which the processor calculates an objective function based on the predicted characteristic value distribution and the target characteristic value; an operating condition search step in which the processor searches for the predetermined operating conditions which minimize the objective function; and an operating condition output step in which the processor outputs the operating conditions serving as the search results.SELECTED DRAWING: Figure 2

Description

[Technical field]

[0001] The present invention relates to a technology for providing operating conditions of a manufacturing apparatus, for example, to materials informatics. [Background technology]

[0002] As an example of the application of the present invention, the background art of plastic production using an extruder as a production device will be described below.

[0003] In recent years, in order to improve the functionality of plastics, polymer alloys that express functions by adding additives and fillers and mixing multiple polymers are being considered. More advanced mixing and dispersion technology is important for improving these functions, and extruders are used as a means of achieving both continuous production functionality and advanced mixing.

[0004] In the operation of an extruder in kneading, the viscosity of the material inside the extruder is a very important parameter, and one of the important indicators that determines the viscosity is the temperature of the resin. In particular, in the production of plastics, if the resin temperature is too high, it can lead to deterioration of the material properties. Therefore, the spatiotemporal resin temperature distribution inside the extruder must be accurately controlled. However, in reality, it is not easy to accurately control the true resin temperature distribution inside the extruder.

[0005] One of the reasons that it is difficult to control the resin temperature distribution inside the extruder is that it is difficult to directly measure the resin temperature. Although it is possible to directly introduce a temperature sensor inside the extruder, it is difficult to obtain a three-dimensional distribution, and in practice, indirect internal temperature prediction is performed from the standpoint of cost and maintainability. Indirect internal temperature prediction methods include consideration based on knowledge from small-scale tests, or simulation, etc.

[0006] In general, the search for operating conditions of an extruder, including temperature, is conducted on a small-scale test machine before being conducted on a large-scale mass production machine. However, when scaling up using knowledge obtained from small-scale studies, it is difficult to realize the mixing performance of small-scale on a large scale by simply enlarging the equipment dimensions in a similar shape. One of the reasons for this is the problem of heat exchange. For example, the total heat stored by the resin inside the extruder and the amount of heat generated by the reaction are proportional to the volume, so they change as the cube of the dimension ratio when scaled up, but the speed at which the heated resin comes into contact with the inner wall of the extruder and is cooled and controlled by heat exchange is proportional to the area, so they change as the square of the dimension ratio when scaled up. For this reason, the optimal temperature conditions obtained from small-scale studies are judged to be a "process mainly involving heat transfer" or a "process with little temperature dependency" when scaled up, and an appropriate strategy is adopted to suit the process.

[0007] Before actually mass-producing plastics with mass production equipment, in many cases, simulations are used to confirm values ​​that cannot be measured, such as temperature. However, with current simulators, it is still difficult to perform simulations that take into account all operating condition parameters and all factors related to the environment in which the equipment is placed (external disturbances such as temperature and humidity).

[0008] For example, in many simulations, calculations are performed assuming that the filling rate inside the cylinder is 100%, and the difficulty of calculations increases significantly when the filling rate is less than 100%, where a gas-liquid interface exists. Therefore, even if the simulation is performed with a filling rate of 100%, in reality a gas-liquid interface exists, and the temperature differs depending on the phase. Furthermore, even if the phase is the same, it is possible that temperature distributions exist on various scales due to disturbances, and even if a thermometer is used to measure a single point inside the cylinder, temperature fluctuations will always be observed.

[0009] In addition, the cylinder temperature is an operating condition parameter and is controlled to a constant temperature, but depending on the control method, the temperature may constantly fluctuate on a minute scale. Also, uncontrollable disturbances such as weather can cause temperature fluctuations that affect product quality. Conventionally, when these temperature fluctuations affected the production quality and caused it to deviate from the target quality, an experienced worker would adjust the operating condition parameters based on their experience to stabilize and optimize the quality.

[0010] In recent years, advances in machine learning technology have made it possible to construct learning models that predict processing results from operating conditions. It is also possible to use such learning models to search for operating conditions that satisfy the target processing results. However, such operating condition search technology has not yet optimized operating conditions taking into account disturbances and external factors. Furthermore, even for the same type of equipment, there are various external factors such as individual differences in equipment parts and different installation locations, and the effects of these disturbances on processing results are diverse, making it impossible to determine them before installing the equipment. For this reason, optimization that takes into account the time-dependent dispersion and probability distribution of condition values ​​that are disturbed by disturbances and external factors is required.

[0011] Patent Document 1 discloses a control system including a controller that controls the control amount of a control object for machining a workpiece in order to automatically suppress deviation of the machining result from a target state, and a predictor constructed based on a prediction model that receives a disturbance as input and outputs a setting value for setting a quality characteristic value to a quality characteristic target value. [Prior art documents] [Patent documents]

[0012] [Patent Document 1] JP 2022-165143 A Summary of the Invention [Problem to be solved by the invention]

[0013] Disturbances and external disturbances are likely to differ depending on the installation and operation conditions of the equipment. Therefore, it is difficult to accumulate data in advance, and it is difficult to prepare a machine learning model that accepts fluctuations as inputs before introducing the equipment. In addition, it is often difficult to prepare a machine learning model that accepts fluctuations after the handover as inputs not only before and after the introduction of the equipment, but also before and after the handover of operating conditions such as relocating the equipment, replacing equipment parts, and scaling up from development to mass production.

[0014] It is desirable to provide a method for providing an operating condition that is robust against fluctuations and can achieve a target processing result even in the presence of disturbances or external disturbances that cannot be predicted in advance. The present invention aims to provide an operating condition that is robust against disturbances and external disturbances. [Means for solving the problem]

[0015] A preferred embodiment of the present invention is a method for providing operating conditions for a manufacturing apparatus that produces a product, using an input device, an output device, a processor, and a memory, comprising: a variation input step in which the input device inputs a variation amount related to a predetermined operating condition; a model read step in which the processor reads out from the memory a model whose input is the operating condition and whose output is a characteristic value of the product; a target characteristic value input step in which the input device inputs a target characteristic value of the product; a predicted characteristic value distribution calculation step in which the processor uses the model to calculate a predicted characteristic value distribution reflecting the variation amount related to the predetermined operating condition; an objective function calculation step in which the processor calculates an objective function based on the predicted characteristic value distribution and the target characteristic value; an operating condition search step in which the processor searches for the predetermined operating condition that reduces the objective function; and an operating condition output step in which the processor outputs the operating conditions that are the result of the search.

[0016] Another preferred embodiment of the present invention is a system for providing operating conditions for a manufacturing apparatus for producing a product from a raw material, comprising: a variation amount setting unit that receives a variation amount related to a predetermined operating condition from an input device; a predictor that receives the predetermined operating condition and the variation amount using a model in which the input is the operating condition and the output is a characteristic value of the product, and outputs a predicted characteristic value distribution of the product; a target characteristic value setting unit that receives a target characteristic value that is a target characteristic value of the product from the input device; an objective function setting unit that sets an objective function based on the predicted characteristic value distribution and the target characteristic value; a searcher that searches for the predetermined operating condition so as to reduce the objective function; and an output unit that outputs the searched operating condition. Effect of the Invention

[0017] According to the present invention, it is possible to provide robust operating conditions that are resistant to disturbances and external disturbances. [Brief description of the drawings]

[0018] [Figure 1] 4 is a flowchart showing the procedure from start to output of proposed operating conditions in the method of the embodiment. [Diagram 2] FIG. 2 is a functional block diagram for explaining a method for providing operating conditions for the manufacturing apparatus of the embodiment. [Diagram 3] FIG. 2 is a block diagram showing an example of a hardware configuration of a system for providing operating conditions of a manufacturing apparatus according to an embodiment. [Figure 4] FIG. 1 is an explanatory diagram illustrating an example of the configuration of a processing condition derivation system according to a first embodiment; [Diagram 5] 1 is a graph showing the change over time in the value observed by the temperature sensor when the temperature control target value in Example 1 is 220° C. [Figure 6] 1 is a graph showing the change over time in the value observed by the temperature sensor when the temperature control target value in Example 1 is 180° C. [Figure 7] FIG. 2 is a graph showing the relationship between operating temperature and product quality in the subject of Example 1. [Figure 8] FIG. 2 is a functional block diagram for explaining a system for providing operating conditions for the manufacturing apparatus according to the embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0019] Hereinafter, an embodiment of the present invention will be described with reference to the drawings. However, the present invention is not to be interpreted as being limited to the description of the embodiment shown below. It is easily understood by those skilled in the art that the specific configuration can be changed without departing from the idea or intent of the present invention. In addition, the position, size, shape, etc. of each configuration shown in the drawings in this specification may not represent the actual position, size, shape, etc. in order to facilitate understanding of the invention. Therefore, the present invention is not limited to the position, size, shape, etc. disclosed in the drawings, etc.

[0020] In the configuration of the embodiments described below, the same reference numerals are used in common between different drawings for the same parts or parts having similar functions, and duplicated explanations may be omitted. When there are multiple elements having the same or similar functions, they may be described with different subscripts added to the same reference numeral. However, when it is not necessary to distinguish between multiple elements, the subscripts may be omitted.

[0021] The designations "first," "second," "third," and the like in this specification are given to identify components, and do not necessarily limit the number, order, or content. Furthermore, numbers for identifying components are used in different contexts, and a number used in one context does not necessarily indicate the same configuration in another context. Furthermore, a component identified by a certain number does not prevent the component from also having the function of a component identified by another number.

[0022] The publications, patents, and patent applications cited herein are hereby incorporated by reference in their entirety. In this specification, elements referred to in the singular form include the plural form unless the context clearly indicates otherwise.

[0023] A method for providing operating conditions for a manufacturing apparatus for producing a product from raw materials, which is one example of an embodiment described in detail below, is characterized by comprising the steps of: inputting a variation amount for at least one operating condition parameter via an input device; reading from a memory by a processor a model whose input is an operating condition and whose output is a characteristic value of the product; inputting a target characteristic value of a target processed product via the input device; calculating by the processor a predicted characteristic value distribution taking into account the variation amount from the model; calculating by the processor an objective function based on the difference between the predicted characteristic value distribution and the target characteristic value; searching by the processor for operating conditions to be input to the model so as to reduce the objective function based on the difference; and outputting the results of the searched operating conditions.

[0024] (common part) FIG. 1 is a flow chart showing a series of steps from start to output of proposed operating conditions. 2 is a functional block diagram showing a configuration example of the system 1 for providing operating conditions. In the following, the operation principle of the system 1 for providing operating conditions in FIG. 2 will be described with reference to the flowchart in FIG.

[0025] In Fig. 1, first, the fluctuation setting unit 2 receives a fluctuation Vp from a fluctuation observer 3 such as a physical sensor (A1). The fluctuation Vp does not need to be one-dimensional, and it is considered that the fluctuation Vp can be a number of different physical quantities that can observe the state of the manufacturing equipment ME using, for example, sensors for temperature, pressure, etc. Hereinafter, the n-th type of fluctuation component of the fluctuation Vp is expressed as Vpn. The information that the fluctuation observer 3 obtains from the manufacturing equipment ME is mainly data on changes over time, and it is considered that there is basically a dependency on the operating conditions X in many cases.

[0026] Next, the predictor 4 reads out a trained model whose input is the operating condition X and whose output is the predicted characteristic value g(X) of the product from the trained model database DB (A2). Here, the trained model does not necessarily have to be read out from the database, and may be in any form as long as it is simply read out. The operating condition X is generally multidimensional and consists of multiple operating condition parameters Xm.

[0027] Next, the target characteristic value setting unit 6 receives a target characteristic value T (A3). The target characteristic value T does not need to be a one-dimensional quantity, and a specific target value such as a mechanical characteristic value or an optical characteristic value is designated by the user US. Note that the order of execution of (A1) to (A3) is arbitrary.

[0028] Next, the predictor 4 creates a predicted characteristic value distribution 7 that takes into account the fluctuation amount Vp using a trained model (A4). The trained model is assumed to be created in advance using a known machine learning technique. There are three methods for creating the predicted characteristic value distribution 7 using the operating condition parameter Xn that is most related to the fluctuation amount component Vpn. (1-1) is the least computationally intensive, and (1-3) is the most computationally intensive but has high performance.

[0029] For example, the operating condition parameter Xn is the set temperature of a specific location A of the manufacturing equipment ME, and the fluctuation component Vpn is the temperature change at the specific location A of the manufacturing equipment ME. The expression format of the fluctuation component Vpn is arbitrary, but for example, it can be converted into time-series data such as the fluctuation value on the positive side and the fluctuation value on the negative side from the set temperature, which is the operating condition parameter Xn, based on the time-series data of temperature acquired by the fluctuation observer 3. Now, with the set temperature, which is the operating condition parameter Xn, as the base, the absolute value of the maximum fluctuation value on the positive side is Vpnp, and the absolute value of the maximum fluctuation value on the negative side is Vpnm.

[0030] (1-1): Two points, the lower limit fluctuation predicted characteristic value g(Xn-Vpnm) and the upper limit fluctuation predicted characteristic value g(Xn+Vpnp), are calculated, and these two points are used as the predicted characteristic value distribution 7 for the nth type fluctuation amount.

[0031] (1-2): A predicted characteristic value is calculated at a number of search points selected by a design of experiments (DoE) according to the number of searches previously specified by the user US or the number of searches initially set in the system within the range from the lower limit of variation (Xn-Vpnm) to the upper limit of variation (Xn+Vpnp) of the operating condition parameter Xn, and these are regarded as the predicted characteristic value distribution 7 for the n-th type variation amount.

[0032] (1-3): Calculate predicted characteristic values ​​for all points in the range from the lower limit of variation (Xn-Vpnm) to the upper limit of variation (Xn+Vpnp) of the operating condition parameter Xn according to the number of steps previously specified by the user US or the number of steps initially set in the system, and use these as the predicted characteristic value distribution 7 for the nth type variation amount. For example, in the temporal variation of temperature shown in Figures 5 and 6 below, the measured temperature is sampled at a predetermined time interval, the sampled measured temperature is used to calculate predicted characteristic values, and these are used as the predicted characteristic value distribution 7 for the nth type variation amount.

[0033] In a typical predictor using a trained model, the set operating conditions X are used as input and the predicted characteristic value g(X) of the product is used as output. In this embodiment, however, a predicted characteristic value distribution 7 reflecting the fluctuation component Vpn of the operating conditions X is obtained.

[0034] Next, in the target characteristic value setting unit 6, the objective function setting unit 15 calculates the objective function f(X, Vp) based on the difference between the target characteristic value T and the predicted characteristic value distribution 7 (A5). There are two ways to create the objective function f(X, Vp):

[0035] (2-1): The mean squared error (MSE) between each value included in the predicted characteristic value distribution 7 and the target characteristic value T is set as the objective function. That is, f(X, Vp) = Σ(g(X, Vp) - T)^2.

[0036] (2-2): The mean absolute error (MAE: Mean Absolute Error) between each value included in the predicted characteristic value distribution 7 and the target characteristic value T is set as the objective function. That is, f(X, Vp)=Σ|g(X, Vp)-T|.

[0037] Next, the searcher 8 searches for the operating conditions X that minimize the objective function f(X, Vp) (A6). Numerical minimization is performed by an iterative method, and various algorithms can be applied. Algorithms are broadly divided into gradient methods and non-gradient methods, with the BFGS (Broyden-Fletcher-Goldfarb-Shanno) method being used as the gradient method, and the Nelder-Mead method being used as the non-gradient method. The initial operating conditions X0 use values ​​input in advance by the user US or values ​​initially set in the system. During the search, the processes of (A1) to (A5) are repeated.

[0038] Finally, the output device 9 outputs the proposed operating conditions XS, which are the search results, to end the series of processes (A7). The output device 9 may be a user interface such as a display or a speaker.

[0039] (Hardware configuration) An example of the configuration of the system 1 that provides the operating conditions of the control device will be described with reference to Fig. 3. As shown in Fig. 3, the system 1 that provides the operating conditions of the control device according to this embodiment is a computer to which an input device 11 and an output device 12 are connected as an interface unit, and a processor 13 and a memory 14 are connected as a control unit. The system 1 that provides the operating conditions of the control device is connected to the manufacturing equipment ME, and can read the values ​​of various sensors and control the equipment according to the set values ​​of various operating conditions.

[0040] In system 1 in Fig. 2, functions indicated by blocks with bold frames are realized by processor 13 executing calculations based on software stored in memory 14, or by dedicated hardware. In system 1 in Fig. 2, information indicated by blocks with thin frames may be stored as data in memory 14. Here, processor 13 refers to a configuration that executes various calculations, and includes a CPU (Central Processing Unit), a GPU (Graphics Processing Unit), an FPGA (Field Programmable Gate Array), an ASIC (Application Specific Integrated Circuit), etc. EXAMPLES

[0041] In Example 1, a twin-screw extruder is used as a plastic manufacturing device, and a method for providing operating conditions for the manufacturing device of the embodiment is considered to be used to accurately control and manage the resin temperature. In the operation of a twin-screw extruder, it is generally not easy to measure the resin temperature inside the twin-screw extruder. As a direct measurement method, a thermometer is installed on an adapter or an inner wall, or the temperature of the discharged material is directly measured at the extruder outlet, but the measured value varies over time due to disturbance or external disturbance, and even if a time average is obtained, there is no guarantee that it is a spatial average value that represents the resin temperature inside the extruder.

[0042] Figure 4 shows an example of the temperature control status of each cylinder zone of cylinder S of a twin-screw extruder and the predicted internal material temperature value by simulation. Each temperature shown above cylinder S in the extruder in Figure 4 is the control temperature target value of cylinder S. Temperature-related operating conditions X can be set for each cylinder zone. In this case, since there are multiple operating conditions, one method is to prioritize the operating conditions that have the greatest impact on quality characteristics, and then determine the operating conditions in order.

[0043] In the example shown in Figure 4, the raw material M introduced from the hopper HP is first led to the zone intended for melting, where it receives energy from the inner wall of the cylinder, which is controlled at a high temperature, and melts. Since the raw material is in a solid phase here, the solid, liquid, and gas phases all exist inside the cylinder until it melts. The molten material is gradually melted and kneaded in the melting zone, and is then led to the zone intended for kneading continuously as is.

[0044] In this kneading zone, the screw configuration is such that a high pressure state is created using a reverse kneading screw, etc., and the volume ratio of the gas phase is small and the filling rate is high. The material heats up not only due to the cylinder temperature, but also due to viscous heat generated by the rotation of the screw. The material that has been thoroughly kneaded in the kneading zone is finally led to the cooling zone.

[0045] In the cooling zone, the material temperature is higher than the cylinder inner wall temperature, and the material is gradually cooled through the cylinder inner wall. The material temperature inside the cylinder and the cylinder temperature control target are not necessarily the same value.

[0046] Consider inserting one temperature sensor at position T1 in Figure 4. The actual material temperature in the area where the pressure increases, just before the melting zone switches to the kneading zone, is important information.

[0047] Figure 5 shows the time-dependent change in the value observed by the inserted temperature sensor. The temperature observed by the temperature sensor inserted at position T1 was 220°C, which was almost the same as the temperature control target, and the temperature fluctuation Vp was small, and the temperature distribution over time was a symmetric normal distribution centered on 220°C. This observed temperature can be said to be the most direct information on the material temperature inside the cylinder, but it is still not complete in that the value from the temperature sensor is a time interval that captures disturbances on a microscale, and it is only information on one finite volume component inside the cylinder, so it does not provide information on the distribution. Nevertheless, it is very useful for predicting the state inside the cylinder, together with the temperature prediction by simulation. Here, we consider the case where the motion conditions related to the temperature are changed.

[0048] Figure 6 shows the change over time in the value observed by the temperature sensor at position T1 when the operating conditions related to temperature in Figure 4 are changed and the temperature control target value at position T1 is changed to 180°C. Compared to Figure 5, the average observed temperature in Figure 6 is slightly lower than the temperature control target value (180°C), and the temperature fluctuation amount Vp is larger. Also, the temperature distribution over time deviates significantly from a normal distribution.

[0049] Figure 7 is a graph showing the relationship between the operating condition temperature and product quality. For the manufacturing equipment and manufacturing items shown in Figures 4, 5, and 6, the relationship between the T1 operating condition temperature and product quality error as shown in Figure 7 is known. The smaller the product quality error, the better the quality of the product, and we want to prevent products with product quality errors exceeding the allowable threshold from being manufactured by the manufacturing equipment as much as possible. As shown in Figure 7, the product quality error is assumed to be minimal at temperatures of 180°C and 220°C. In addition, the median values ​​of the temperature range where the product quality falls within the allowable threshold are assumed to be 185°C and 223°C. In the following, we consider what degree Celsius is appropriate as the T1 operating condition temperature that is robust against disturbances and external disturbances in such a case.

[0050] Looking only at Fig. 7, it appears that the range of values ​​below the allowable threshold is wider around 185°C than around 223°C, and that setting the T1 operating condition temperature to 185°C increases the possibility that the product quality error will be within the allowable threshold even if the temperature varies slightly. However, as shown in Figs. 5 and 6, when the T1 temperature control target is set to 180°C, the temperature fluctuation amount Vp is larger and the distribution is different compared to when it is set to 220°C, so it cannot be said that 185°C will necessarily minimize the product quality error. In addition, in this embodiment 1, the temperature is one-dimensional, but if it becomes multidimensional, it is even more difficult to derive the optimal proposed operating conditions without numerical optimization calculations.

[0051] The procedure for solving the above situation will be described with reference to the flowchart in Figure 1. First, the fluctuation amount setting unit 2 acquires information on the fluctuation amount Vp according to the operating conditions from the historical data set of the past extruder operating conditions and the T1 sensor acquisition values ​​accumulated in the fluctuation amount observer 3 (A1).

[0052] Then, the predictor reads out from the database DB a trained regression model whose input is the operating conditions of the extruder and whose output is the characteristic values ​​of the product manufactured by the extruder (A2). This trained regression model may be a regression model such as DNN (Deep Neural Network), GBGT (Gradient Boosting Decision Tree), KRR (Kernel Ridge Regression), or others. The trained model to be read out may be a fixed model if the product manufactured by the manufacturing equipment is fixed, or may be selected by the user US each time.

[0053] Thereafter, the user US sets a target characteristic value T of the product to be manufactured by the extruder through an input device, and the value is acquired by the target characteristic value setting unit 6 (A3). At the same time, the user US may also specify and input specific operating conditions as initial operating conditions.

[0054] In this Example 1, the user US sets the initial operating conditions with the T1 temperature control target at 180° C., and wishes to know whether 185° C., 220° C., 223° C., or another temperature is a better condition. Calculation then begins, and the trained model outputs a predicted characteristic value distribution 7 that takes into account the amount of variation (A4).

[0055] Thereafter, the objective function setting unit 15 calculates an objective function f based on the difference between the target characteristic value T and the predicted characteristic value distribution 7 (A5). In this Example 1, since it is considered that the closer the quality of the product manufactured by the extruder is to the target characteristic value T, the better, the mean square error MSE is used as the form of the objective function f.

[0056] Then, the searcher 8 searches for the operating conditions that minimize the objective function f (A6). In this embodiment 1, since it is known that there is a region in which the product quality error becomes large between 185°C and 220°C as the T1 operating condition temperature, the Nelder-Mead method, which is a non-gradient method, is selected as the search algorithm instead of the gradient method that heads toward a minimum value based on a gradient.

[0057] After the search by the searcher 8 is completed, the proposed operating conditions XS are obtained and are output by the output device 9 to be presented to the user US (A7). Through this series of steps, it is possible to provide optimal operating conditions for each manufacturing equipment ME that are resistant to fluctuations such as disturbances and external disturbances.

[0058] In the above, the T1 operating condition temperature is set as the operating condition parameter Xn, and the objective function f is calculated by the above methods (2-1) and (2-2) to obtain the optimal operating condition. In an actual device, the operating condition temperature varies depending on the location and conditions as shown in Figure 4, and there may be m operating condition parameters such as X1 to Xm operating condition temperatures.

[0059] In this way, there may be multiple operating conditions for one manufacturing device, and each operating condition may be the set temperature for a different part of the manufacturing device. In this case, the searcher 8 searches for a set temperature that reduces the objective function for each operating condition. In this case, the operating conditions are determined in order of temperature that has the greatest effect on quality characteristics, for example, by prioritizing optimization of the cooling zone. Alternatively, the operating conditions are optimized by searching for a condition that minimizes the sum of the objective functions of all operating conditions. EXAMPLES

[0060] FIG. 8 is a functional block diagram of a system that provides optimal operating conditions for each piece of manufacturing equipment, the system having actual operating condition data previously performed by the manufacturing equipment and associated measured characteristic value data, and a fluctuation observer that observes the fluctuations of the manufacturing equipment.

[0061] The basic configuration is the same as that shown in Figure 2, but differs in that (8-1) instead of reading out a trained model from a database, this system constructs a self-generated trained model from actual operating condition data and measured characteristic value data using supervised learning, and (8-2) the system includes a fluctuation observer consisting of hardware such as a temperature sensor and a pressure sensor.

[0062] (8-1): The learning device 10 has past actual operating condition data and measured characteristic value data for the manufacturing equipment associated with the system, and uses all or a part of that data to construct a self-generating learned model through supervised learning.

[0063] The actual operating condition data uses data measured by the fluctuation observer 3. The actual measured characteristic value data is obtained by separately sampling the product and measuring the characteristics by a known method. These data can be stored as a database by accumulating past data.

[0064] The learning performed by the learning device 10 at this time may be any machine learning method, such as deep learning, kernel ridge regression, multiple regression, etc. In addition, the training data used for learning is usually filtered by the product to be manufactured, time, etc., and the filtering can be performed automatically or manually. The self-generated trained model constructed by the learning device 10 can be stored in a trained database such as that shown in FIG. 2 and can be used on another occasion.

[0065] (8-2): The fluctuation observer stores the hardware of physical sensors such as temperature measuring devices and the time series data of the physical quantities obtained from the sensors. The fluctuation setting unit calculates the fluctuation amount after filtering the data held by the fluctuation observer with the manufacturing date and time and environmental records based on the user settings or initial settings. Here, environmental records include the date and time of relocation of equipment specific to the equipment that is thought to affect disturbances, and the date and time records of part modifications and repairs specific to the equipment. EXAMPLES

[0066] In the above embodiment, an extruder is taken as an example. However, the present invention can also be applied to technical fields in which other raw materials are denatured to produce products. For example, in a bread machine, the fermentation state of the dough caused by yeast inside is a very important parameter, and the temperature of the dough is one of the important indicators that determine the fermentation. However, in reality, it is not easy to accurately measure or estimate the temperature distribution of the dough inside the machine and control the temperature.

[0067] Using a method basically similar to that of Examples 1 and 2, it is possible to appropriately control the fermentation state of bread dough.

[0068] As described in detail above, this embodiment uses a trained model that predicts the characteristic values ​​that are the processing results of the workpiece from the operating conditions, calculates a predicted characteristic value distribution that takes into account the amount of variation from the model, and optimizes the operating conditions to minimize an objective function related to the difference between the target characteristic value and the predicted characteristic value distribution, thereby making it possible to provide robust operating conditions that are resistant to disturbances and external disturbances.

[0069] That is, according to this embodiment, even if there are machine difference factors such as equipment specifications, different installation environments for each equipment, and individual part differences that may affect the processing results, the user does not need to search for and set operating conditions that take these into account. Also, as an additional effect, it is possible to reduce the probability that stationary points such as maximum values ​​and saddle points are selected as optimal solutions in the operating condition optimization calculation.

[0070] According to the above-described embodiment, an efficient manufacturing technique can be provided, which reduces energy consumption, reduces carbon emissions, prevents global warming, and contributes to the realization of a sustainable society. [Explanation of symbols]

[0071] 1: System for providing operating conditions, 2: Fluctuation amount setting unit, 3: Fluctuation amount observer, 4: Predictor, 5: Input unit, 6: Target characteristic value setting unit, 7: Predicted characteristic value distribution, 8: Searcher, 9: Output unit, 10: Learning unit, 11: Input device, 12: Output device, 13: Processor, 14: Memory, 15: Objective function setting unit

Claims

1. 1. A method for providing operating conditions for a manufacturing apparatus for producing a product using an input device, an output device, a processor, and a memory, comprising: A fluctuation amount input step in which the input device inputs a fluctuation amount related to a predetermined operating condition; a model reading step in which the processor reads from the memory a model whose inputs are the operating conditions and whose outputs are characteristic values ​​of the product; a target characteristic value input step in which the input device inputs a target characteristic value of the product; a predicted characteristic value distribution calculation step in which the processor calculates a predicted characteristic value distribution reflecting a fluctuation amount related to the predetermined operating condition by using the model; an objective function calculation step in which the processor calculates an objective function based on the predicted characteristic value distribution and the target characteristic value; an operating condition search step in which the processor searches for the predetermined operating condition that reduces the objective function; an operating condition output step of outputting the operating conditions resulting from the search; A method for providing operating conditions for a manufacturing apparatus, comprising:

2. The manufacturing apparatus produces the product by modifying a raw material. A method for providing operating conditions for a manufacturing apparatus according to claim 1.

3. The predetermined operating condition is a set value of a predetermined physical quantity, and the fluctuation amount is a fluctuation amount from the set value of the predetermined physical quantity. A method for providing operating conditions for a manufacturing apparatus according to claim 1.

4. The predetermined operating condition is a set temperature of a predetermined location of the manufacturing apparatus, and the fluctuation amount is a fluctuation amount from the set temperature. A method for providing operating conditions for a manufacturing apparatus according to claim 3.

5. The predicted characteristic value distribution calculation step includes: When the input of the model is the operating condition X and the output of the model is the characteristic value g(X) of the product, With respect to the set value Xn of the predetermined physical quantity as a reference, the absolute value of the maximum fluctuation value of the fluctuation quantity on the positive side is Vpnp, and the absolute value of the maximum fluctuation value on the negative side is Vpnm, Two points, a fluctuation lower limit predicted characteristic value g (Xn-Vpnm) and a fluctuation upper limit predicted characteristic value g (Xn+Vpnp), are calculated, and the two points are used to form the predicted characteristic value distribution. A method for providing operating conditions for a manufacturing apparatus according to claim 3.

6. The predicted characteristic value distribution calculation step includes: When the input of the model is the operating condition X and the output of the model is the characteristic value g(X) of the product, With respect to the set value Xn of the predetermined physical quantity as a reference, the absolute value of the maximum fluctuation value of the fluctuation quantity on the positive side is Vpnp, and the absolute value of the maximum fluctuation value on the negative side is Vpnm, Calculate the characteristic values ​​at several search points selected by an experimental design method according to a specified number of searches within a range from a variation lower limit predicted characteristic value g (Xn-Vpnm) to a variation upper limit predicted characteristic value g (Xn+Vpnp), and set these values ​​as the predicted characteristic value distribution. A method for providing operating conditions for a manufacturing apparatus according to claim 3.

7. The predicted characteristic value distribution calculation step includes: When the input of the model is the operating condition X and the output of the model is the characteristic value g(X) of the product, With respect to the set value Xn of the predetermined physical quantity as a reference, the absolute value of the maximum fluctuation value of the fluctuation quantity on the positive side is Vpnp, and the absolute value of the maximum fluctuation value on the negative side is Vpnm, A predetermined number of data points are sampled from the fluctuation amount related to the predetermined operating condition within a range from a fluctuation lower limit predicted characteristic value g (Xn-Vpnm) to a fluctuation upper limit predicted characteristic value g (Xn+Vpnp), the characteristic values ​​are calculated using the sampled data, and these are set as the predicted characteristic value distribution. A method for providing operating conditions for a manufacturing apparatus according to claim 3.

8. The objective function calculation step includes: a mean square error between each value included in the predicted characteristic value distribution and the target characteristic value is set as an objective function; A method for providing operating conditions for a manufacturing apparatus according to claim 3.

9. The objective function calculation step includes: The mean absolute error between each value included in the predicted characteristic value distribution and a target characteristic value is set as an objective function. A method for providing operating conditions for a manufacturing apparatus according to claim 3.

10. There are a plurality of the predetermined operating conditions for one of the manufacturing apparatuses, performing the fluctuation amount input step, the model read step, the predicted characteristic value distribution calculation step, the objective function calculation step, the operating condition search step, and the operating condition output step for each of a plurality of predetermined operating conditions; A method for providing operating conditions for a manufacturing apparatus according to claim 3.

11. A system for providing operating conditions of a manufacturing apparatus for producing a product from a raw material, comprising: A fluctuation amount setting unit that receives a fluctuation amount related to a predetermined operating condition from an input device; a predictor that receives the predetermined operating conditions and the fluctuation amounts and outputs a predicted characteristic value distribution of the product using a model in which the input is an operating condition and the output is a characteristic value of the product; a target characteristic value setting unit that receives a target characteristic value, which is a target characteristic value of the product, from the input device; an objective function setting unit that sets an objective function based on the predicted characteristic value distribution and the target characteristic value; A searcher that searches for the predetermined operating condition so as to reduce the objective function; an output unit that outputs the searched operating conditions; A system for providing operating conditions for a manufacturing device, comprising:

12. A learning machine that generates a self-generated trained model by performing supervised learning using actual operating condition data and actual measured characteristic value data actually acquired from the manufacturing apparatus, The predictor uses the self-generated trained model as the model. The system for providing operating conditions of a manufacturing device according to claim 11.

13. The predetermined operating condition is a set temperature of a predetermined location of the manufacturing apparatus, and the fluctuation amount is a fluctuation amount from the set temperature. The system for providing operating conditions of a manufacturing device according to claim 11.

14. The searcher searches for a set temperature of a predetermined location of the manufacturing equipment that reduces the objective function. The system for providing operating conditions of a manufacturing device according to claim 13.

15. There are a plurality of the predetermined operating conditions for one of the manufacturing apparatuses, Each of the predetermined operating conditions is a set temperature of a different portion of the manufacturing apparatus, The searcher searches for a set temperature that reduces the objective function for each of the predetermined operating conditions. The system for providing operating conditions of a manufacturing device according to claim 14.

Images