Blow molding machine operation condition inference device, method, and program, and learning device

Abstract

A trained model 73 which takes, as input data, measurement values regarding physical property information of a preform, conditions during molding of a bottle, and physical property information of the bottle, and measurement values regarding environment information during manufacturing of the bottle, and which outputs, as output data, operation conditions for a blow molding machine is prepared. By inputting, to the trained model 73, target values 81a regarding the physical property information, which should be satisfied by a molded bottle, and the measurement values 81b regarding the physical property information of the preform, the conditions during molding of the bottle, and the environment information, the operation conditions (heater output, blow pressure, blow timing, rod speed, and cooling water temperature) for the blow molding machine, which should be set to the blow molding machine in order to mold a bottle that satisfies the target values, are inferred.

Description

Apparatus, method and program for inferring operating conditions of a blow molding machine, and learning apparatus

[0001] The present invention relates to an operating condition inference device for a blow molding machine, an operating condition inference method and program for a blow molding machine, and a learning device.

[0002] Many plastic bottles, such as PET bottles, are made by stretch-blow molding a test-tube-shaped preform, which is then stretched using a blow molding machine to form the bottle (see Patent Document 1 for details of blow molding machines).

[0003] Blow molding machines are equipped with a wide variety of settings, each of which is set with setting data (setting values), and the blow molding machine operates according to these settings. In recent years, the variety of bottles has increased, and with the demand for lighter bottles and the use of recycled materials, the variety of preforms has also increased. Naturally, different types of bottles or different types of preforms used to mold bottles require adjustments (changes) to the settings of the blow molding machine. Setting up a blow molding machine often relies on the intuition and experience of a skilled operator.

[0004] International Publication WO2019 / 048419

[0005] An object of the present invention is to support various settings of a blow molding machine.

[0006] An operating condition inference device for a blow molding machine according to at least some embodiments of the present invention is an inference device that infers the operating conditions of a blow molding machine that heats a preform with a heater, moves the heated preform to a mold, stretches the preform axially by inserting a stretch rod into the preform, and stretches the preform outward by supplying pressurized fluid inside, thereby molding a hollow bottle.The inference device is characterized by comprising: a trained model in which parameters are trained so that physical property information or environmental information of the preform is used as input data and the operating conditions of the blow molding machine are used as output data; and an inference means that infers the operating conditions of the blow molding machine that should be set on the blow molding machine in order to mold a bottle that satisfies the target value or target quality, which is inferred by inputting target values ​​of physical property information or environmental information that the molded bottle must satisfy into the trained model.

[0007] According to this invention, the operating conditions of a blow molding machine that should be set in the blow molding machine to mold bottles that satisfy target values ​​or target quality are inferred. The inferred operating conditions are based on a trained model in which parameters are trained so that physical property information or environmental information of the bottle is used as input data and the operating conditions of the blow molding machine are used as output data. The operating conditions of the blow molding machine that will mold bottles with the target physical property information or target quality can be inferred in a short period of time without relying on the experience of the operator.

[0008] Preferably, the input data further includes at least one of physical property information of the preform and molding conditions for the bottle, which increases the possibility of more accurately and precisely estimating the operating conditions of the blow molding machine for molding a bottle having target physical property information or target quality.

[0009] In one embodiment, the inferred operating conditions of the blow molding machine are at least one of the heater output conditions for heating the preform and the pressure conditions of the fluid supplied to the preform. This is because the heater output and fluid pressure (blow pressure) have a significant effect on the physical properties or quality of the bottle molded by the blow molding machine. The trained model infers the heater output and / or fluid pressure that should be set in the blow molding machine to mold a bottle that satisfies the target physical property values ​​or quality. This increases the certainty that a bottle that meets the specified physical property value target or quality target can be molded without relying on the operator's experience.

[0010] In addition to the heater output and fluid pressure (blow pressure) described above, the inferred operating conditions of the blow molding machine may further include the timing conditions for supplying fluid to the preform and / or the speed conditions of the stretch rod inserted into the preform.

[0011] Preferably, the input data includes information regarding the presence or absence of an abnormality in the appearance of the bottle, and the inference means infers the operating conditions of the blow molding machine that should be set in the blow molding machine in order to mold bottles that satisfy the target value or target quality and are free from abnormalities in appearance, thereby making it possible to infer operating conditions for molding bottles of higher quality.

[0012] In one embodiment, the preform physical property information includes data on the material and weight of the preform. The preform physical property information may include data distinguishing whether the preform is made from virgin resin or recycled resin. The preform physical property information may further include an IV (intrinsic viscosity) value of the preform.

[0013] In another embodiment, the bottle forming conditions include data regarding the volume and shape of the bottle.

[0014] In another embodiment, the physical property information of the bottle includes at least one of the full capacity, buckling strength, overall height, wall thickness, and section weight of the bottle.

[0015] In another embodiment, the environmental information includes at least one of the temperature and humidity of the environment surrounding the blow molding machine.

[0016] Preferably, the blow molding machine operating condition inference device includes a setting means for setting the operating conditions of the blow molding machine inferred by the inference means in the blow molding machine, so that the blow molding machine can be operated in accordance with the inferred operating conditions of the blow molding machine.

[0017] The present invention also provides a learning device that generates a trained model suitable for use in a blow operating condition inference device. The learning device according to at least some embodiments of the present invention includes: a training data acquisition means that acquires training data including operating conditions of a blow molding machine that heats a preform with a heater, moves the heated preform to a mold, stretches the preform in its axial direction by inserting a stretch rod into the preform, and stretches the preform outward by supplying a pressurized fluid inside the preform, thereby molding a hollow bottle; and a model generation means that uses the training data to generate a trained model for inferring, from target values ​​or target quality of physical property information that the molded bottle should satisfy, operating conditions of the blow molding machine that should be set in the blow molding machine in order to mold a bottle that satisfies the target value or target quality.

[0018] The present invention also provides a method for inferring the operating conditions of a blow molding machine and a program for causing a computer to execute the method. The program can be stored on a portable recording medium, such as a CD-ROM or semiconductor memory.

[0019] FIG. 1 is a schematic plan view of a blow molding machine. It shows the positional relationship between a preform and a heater zone. It is a vertical cross-sectional view of a mold in which a preform is set. It is a functional block diagram of a blow molding machine. It is a block diagram of a learning device. It is a block diagram of an inference device. It shows the processing results of the inference device. It is a graph showing the relationship between the target value and the actual measured value of the full injection capacity. It is another example showing the processing results of the inference device. It is yet another example showing the processing results of the inference device.

[0020] FIG. 1 is a plan view showing a schematic view of a blow molding machine.

[0021] A large number of preforms P made of resin, for example PET (polyethylene terephthalate), are prepared in a preform supply station 1. The preforms P are supplied along a transport path from the preform supply station 1 to a heating station (oven) 2. The heating station 2 has one or more heater zones 5 (six heater zones 5 are shown in FIG. 1), and the preforms P are heated to a temperature at which they can be blow-molded by passing through the heater zones 5. The number of heater zones 5 can be designed as desired.

[0022] The preforms P heated in the heating station 2 proceed to the molding station 3. The molding station 3 is equipped with a rotating wheel 6 and a plurality of molds 7 that are arranged along the rotating wheel 6 and move around the molding station 3 at a constant speed in accordance with the rotational movement of the rotating wheel 6. The preforms P supplied to the molding station 3 are each set in the molds 7. Note that the molding station 3 does not necessarily need to be equipped with a rotating wheel 6, and the molds 7 may be transported in a linear manner. The number of molds 7 that the molding station 3 is equipped with is also arbitrary.

[0023] The mold 7 (the bottom mold 7c of the mold 7, which will be described later) is provided with a cooling water passage (not shown) through which cooling water passes, and is cooled by cooling water of a predetermined temperature that passes through the cooling water passage. As will be described next, once the bottle B adheres to the inner surface of the mold 7, the bottle B is cooled and solidified.

[0024] A stretch rod is inserted into the preform P through its opening, which stretches the preform P in the axial direction (longitudinal direction). A fluid, typically compressed air, is blown into the preform P through its opening (air blow). The air blow stretches the preform P outward (axially and circumferentially). The air blowing is performed in two stages: first, low-pressure air (hereinafter referred to as "first air") is blown into the preform P to inflate it; then, high-pressure air (hereinafter referred to as "second air") is blown in to adhere the preform P to the inner surface of the mold, thereby producing a bottle B shaped to conform to the shape of the mold 7. After the compressed air and stretch rod are removed from the bottle B, the bottle B is removed from the mold 7. After leaving the molding station 3, the bottle B proceeds to subsequent processes (not shown), such as a sterilization process, beverage filling process, and capping process, resulting in a completed beverage-filled bottle.

[0025] The blow molding machine is provided with a thermometer 61 for measuring the ambient temperature (room temperature) and a hygrometer 62 for measuring the ambient humidity of the blow molding machine. The temperature measured by the thermometer 61 and the humidity measured by the hygrometer 62 can be used to infer the operating conditions of the blow molding machine, as will be described later.

[0026] 2 is a schematic diagram showing the positional relationship between the preform P and the heater zone 5 when the preform P is transported through the heating station 2, and also shows a schematic cross section of the heater zone 5. Also shown in FIG. 2 is a spindle 8 for holding and rotating the preform P. In the heating station 2, the preform P is transported through the heating station 2 while being rotated by the spindle 8 around its axial direction as the rotation axis.

[0027] The heating station 2 of the blow molding machine includes a heater zone 5, as described above, which softens the preform P so that it can be stretched using a stretching rod and air blow. When multiple heater zones 5 are provided, various types of heater zones 5 can be used, including those that heat the entire preform P, those that focus heat on the area around the support ring of the preform P (focus heaters), those that blow air onto the entire preform P without a heater (surface cleaning), and those that blow air onto a portion of the preform P (air knife), allowing for precise control of the heating of the preform P. The heater zone 5 shown in FIG. 2 is a type that heats the entire preform P. Each of the multiple heater zones 5 can be controlled independently of each other.

[0028] The heater zone 5 shown in Fig. 2 includes a plurality of heaters 5a (e.g., halogen lamps) (seven in Fig. 2) arranged at intervals along the axial direction of the preform P. Each of the seven heaters 5a included in the heater zone 5 can be controlled independently of the others. The number of heaters 5a included in the heater zone 5 can also be any number.

[0029] For example, if it is desired to thicken the wall of a bottle B molded by a blow molding machine in a particular area, it is possible to lower the heating temperature of the corresponding portion of the preform P. For example, using the type of heater zone 5 shown in Figure 2, the output of a heater 5a (e.g., one of seven heaters 5a) that heats the portion of the preform P corresponding to the desired thickened wall when molding the bottle B can be lowered. This makes it possible to lower the heating temperature of a portion of the preform P, thereby making it possible to mold a bottle B with thicker walls in some areas.

[0030] 3 is a schematic diagram of the blow molding process, showing a vertical cross section of a mold 7 in which a preform P is set. In FIG. 3, the cooling water passages provided in the bottom mold 7c of the mold 7 described above are omitted.

[0031] The preform P, which has been heated in the heating station 2 and transferred to the molding station 3, is surrounded on its sides by the split dies 7a and 7b that make up the mold 7. The bottom surfaces of the split dies 7a and 7b are closed by a bottom die 7c, and a valve block 13 is joined to the mouth of the preform P. A stretch rod 11 is inserted into the preform P through a hole formed in the valve block 13. The stretch rod 11 stretches the preform P in the axial direction.

[0032] The valve block 13 has two valves V1 and V2 inside. Compressed air (first air, second air) from the air pump 12 is sent through the valve V1 into the preform P. The first air and second air are sequentially blown into the preform P from the valve V1, causing the preform P to expand outward and form the bottle B.

[0033] After the bottle B has been molded, the compressed air inside the bottle B is released to the outside through valve V2 of the valve block 13. The stretch rod 11 is removed from the bottle B, and the bottom mold 7c and split molds 7a and 7b are opened, and the completed bottle B is carried out of the blow molding machine.

[0034] In the forming station 3, it is possible to control the rod speed (advancing speed) of the stretch rod 11, the timing to start blowing the first air, the pressure of the first air, the timing to switch from the first air to the second air, the pressure of the second air, etc.

[0035] For example, if it is desired to reduce the wall thickness near the bottom of bottle B molded by a blow molding machine, one possible solution is to start blowing the first air earlier. By supplying the first air before the stretch rod is fully extended (before the preform P is fully stretched vertically), the wall thickness of the bottom of bottle B will be reduced, and instead the wall thickness from the neck to the shoulder of bottle B can be increased.

[0036] FIG. 4 shows a functional block diagram of the blow molding machine.

[0037] The blow molding machine is broadly divided into a heater function 20 and a blow function 30. The heater function 20 is realized by the plurality of heater zones 5 provided in the heating station 2. The blow function 30 is realized by the stretch rod 11, air pump 12, valve block 13, etc. provided in the molding station 3.

[0038] As described above, each of the plurality of heater zones 5 provided in the heating station 2, and each of the plurality of heaters 5a provided in the heater zone 5, can be controlled individually. In the forming station 3, the pressure of the air supplied from the air pump 12, the timing of the air supply, and the rod speed (advance speed) of the stretch rod 11 can also be controlled. The heater function 20 and the blow function 30 are centrally managed by the control device 40.

[0039] A setting device 50 is connected to the control device 40. The setting device 50 is equipped with a display device (not shown), and a number of setting items are displayed on a setting screen displayed on the display device. The control device 40 controls the heater function 20 and the blow function 30 in accordance with the setting of the number of setting items using the setting screen.

[0040] The setting device 50 can set a number of setting values ​​for various setting items, and the heater function 20 and blow function 30 of the blow molding machine are precisely controlled in accordance with these setting values. By changing the setting values, it is possible to increase or decrease the wall thickness of a specific portion of the bottle B that is finally molded, as described above.

[0041] Specifically, the setting device 50 is set with preform physical property information such as the material and weight of the preform P, bottle molding conditions such as the volume and shape of the bottle B to be molded, heater setting information such as the output of the heater 5a in the heater zone 5, and blow setting information such as the rod speed (advancement (insertion) speed) of the stretch rod 11, the timing to start the first air, the pressure of the first air, the timing to switch from the first air to the second air, and the pressure of the second air. The blow molding machine has many setting items, and skilled operators finely adjust the setting values ​​of many of these setting items on-site. This is because skilled operators have experience in knowing which of the many setting items to adjust and how the results will appear in the bottle B.

[0042] In this embodiment, machine learning is performed as described below to mechanically realize the detailed adjustment of setting items made by a skilled operator, reduce the time required for adjustment work, stabilize the quality of the molded bottle B, and adjust setting items for improving the quality of the bottle B that even a skilled operator may not notice. Target values ​​for physical property information that the molded bottle B must satisfy are given, and the operating conditions to be set in the blow molding machine to mold the bottle B that satisfies those target values ​​(the types of setting data to be set in the blow molding machine and their set values) are inferred as the result of the machine learning process.

[0043] As described below, in this embodiment, the operating conditions of the blow molding machine inferred using machine learning processing are typically the heater output of the blow molding machine. Additionally or instead, the blow pressure may be inferred. Other inferred operating conditions may also include blow timing, rod speed, and the temperature of the cooling water passed through the cooling water passage provided in the bottom mold 7c of the mold 7. In any case, the inferred operating conditions of the blow molding machine, such as heater output, blow pressure, blow timing, rod speed, and cooling water temperature, correlate with the physical properties of the bottle B molded from the preform P by the blow molding machine.

[0044] FIG. 5 shows a block diagram of a learning device 70. The learning device 70 performs machine learning related to blow molding machines and includes a learning data acquisition unit 71 that acquires learning data used for learning, a model generation unit 72 that uses the learning data to generate a learning model used to infer the heater output, blow output, blow timing, rod speed, cooling water temperature, etc. of the blow molding machine, and a generated trained model 73 (its storage unit). The learning device 70 is realized by a computer device equipped with, for example, a processor, memory, storage device, communication device, etc., and operates according to a program that causes the computer device to realize the functions of the learning device 70. The computer device functions as the learning device 70 when the program stored in a portable recording medium 75 is installed on the computer device.

[0045] The learning data acquisition unit 71 acquires (receives) data set or measured when bottle B is actually formed, as learning data, via a network (e.g., the Internet) from multiple bottle molding factories (bottled beverage manufacturing factories) where blow molding machines are installed. The learning data includes the following data:

[0046] (1) Heater Output: This is the output of the multiple heater zones 5 provided in the heating station 2 of the blow molding machine. Data representing the heater output may be the output ratio (total output) (e.g., 80%) of the maximum output of the entire heating station 2, each heater zone 5, or each heater 5a, or may be on / off data for the multiple heaters 5a provided in the heater zone 5. (2) Blow Pressure: This is the pressure value of the first air and the second air blown into the preform P.

[0047] (3) Blow Timing This is the start timing of the first air and the timing when the first air is switched to the second air (switching timing). Instead of the switching timing, a combination of the end timing of the first air and the start timing of the second air may be used. The data representing the blow timing can be the rotation angle of the rotating wheel 6, with the position where the preform P is transferred to the molding station 3 as the reference (0°). For example, if the start timing of the first air is "33°," this means that the first air will start when the preform P is transferred to the molding station 3 and rotated 33° by the rotating wheel 6.

[0048] (4) Rod Speed ​​This is the advancement (insertion) speed of the stretch rod 11 inserted into the preform P.

[0049] (5) Preform Physical Property Information This includes the material, weight, and dimensions of the preform P. The information about the material of the preform P may include data distinguishing whether the preform P is made of virgin resin (new resin) or recycled resin (used resin). It may also include the IV (Intrinsic Viscosity) value of the PET resin used in the preform P. The dimensions generally include the diameter of the opening of the preform P, the diameter of the body, and the length from the support ring to the bottom end (body length).

[0050] (6) Bottle molding conditions The capacity and shape of bottle B. The shape is generally classified into round and square.

[0051] (7) Environmental Information: Measurements of the temperature and humidity around the blow molding machine, as measured by a thermometer 61 and a hygrometer 62.

[0052] (8) Bottle physical property information: Measurement values ​​of the full capacity, buckling strength, overall height, wall thickness, and section weight (weight of each section divided by cutting bottle B transversely (cutting perpendicular to the axial direction) at one or more predetermined locations) of the molded bottle B. For the wall thickness and section weight, values ​​(or more detailed values) for multiple locations of bottle B, such as the neck, shoulder, body, and bottom, are generally used.

[0053] (9) Bottle Appearance Information Bottle B molded from preform P may become cloudy (white). The presence or absence of this whitening is included in the bottle appearance information. The presence or absence of structural defects in bottle B may also be included in the bottle appearance information. There are two types of whitening: whitening due to overstretching and whitening due to crystallization. Whitening due to overstretching is likely to occur when preform P is not heated enough, while whitening due to crystallization occurs when preform P is heated too much and the material of preform P crystallizes.

[0054] (10) Mold Cooling Water Temperature This is the temperature of the cooling water passed through the cooling water passage to cool the bottom mold 7c of the mold 7 described above.

[0055] The model generation unit 72 learns the learning data from the learning data acquisition unit 71 and generates a trained model for inferring the operating conditions that should be set in the blow molding machine in order to mold a bottle B that meets the target, specifically, heater output, blow pressure, blow timing, rod speed, cooling water temperature, etc.

[0056] The learning algorithm used by the model generation unit 72 may be a known algorithm such as supervised learning, unsupervised learning, or reinforcement learning.

[0057] For example, LightGBM, a learning algorithm with high prediction accuracy and short training time, can be used. The trained model 73 represents the relationship between explanatory variables (specifically, information on the physical properties of the bottle, information on the physical properties of the preform, conditions during bottle molding, environmental information, etc.) and objective variables (heater output of the blow molding machine, blow pressure, blow timing, rod speed, cooling water temperature, etc.) and has learned parameters for calculating the values ​​of the objective variables. External setting variables (hyperparameters) for managing the training of the learning model (such as the number of branches in the decision tree) are optimized using, for example, a Bayesian optimization library.

[0058] The output data (objective variables) of the trained model 73 may be selected with priority given to data that significantly affects the physical properties of the bottle B molded by the blow molding machine. Here, an example is described in which five pieces of data (heater output, blow pressure, blow timing, rod speed, and cooling water temperature) of the blow molding machine are treated as output data. However, more types of data may be used as output data, or conversely, fewer types of data may be used as output data, such as heater output data alone, or the heater output and blow pressure (particularly the first air pressure). Since LightGBM calculates the importance (degree of influence) of explanatory variables that affect the objective variable for each explanatory variable, when focusing on a bottle property of interest (e.g., full fill capacity) (one of the explanatory variables), data (objective variables) that significantly affect the bottle property of interest can be selected as the output data of the trained model 73.

[0059] Of course, the input data (explanatory variables) of the trained model 73 can also be limited to, for example, the physical property information of the bottle B to be molded by the blow molding machine (for example, the full filling capacity). Furthermore, if the physical property information of the bottle B to be molded is fixed, the input data of the trained model 73 can be limited to environmental information (either temperature or humidity, or both). However, using multiple types of input data increases the possibility that setting items and setting values ​​that the operator is unaware of will be inferred.

[0060] FIG. 6 is a block diagram of an inference device for inferring the operating conditions to be set in the blow molding machine.

[0061] The inference device 80 comprises an inference data acquisition unit 82 and an inference unit 83. The inference device 80 is also realized by a computer device comprising, for example, a processor, memory, a storage device, a communication device, etc., and operates according to a program that causes the computer device to realize the functions of the inference device 80. The computer device functions as the inference device 80 when the program stored in a portable recording medium 85 is installed on the computer device.

[0062] The inference data acquisition unit 82 is provided with target values ​​(design values) 81a for the physical property information of the bottle B to be molded (at least one of full filling capacity, buckling strength, overall height, wall thickness, and section weight), and preferably with physical property information 81b for the preform P used to mold the bottle B (for example, at least one of the material and weight of the preform P), bottle molding conditions (at least one of the bottle capacity and shape), and environmental information (at least one of the temperature and humidity).

[0063] The inference unit 83 infers the operating conditions of the blow molding machine, in this case the heater output, blow pressure, blow timing, rod speed, and cooling water temperature, using the trained model 73. That is, by inputting the inference data acquired by the inference data acquisition unit 82 into the trained model 73, the inference unit 83 outputs the heater output, blow pressure, blow timing, rod speed, and cooling water temperature such that the bottle to be produced will have the target physical property information.

[0064] The target values ​​of the physical property information of the bottle to be molded may be specific numerical values ​​or may be numerical ranges with a predetermined width. The wall thickness and section weight are generally set at multiple locations on the bottle B, such as the neck, shoulder, body, and bottom.

[0065] By using a trained model 73 that has been trained using a large amount of training data, the heater output, blow pressure, blow timing, rod speed, and cooling water temperature to be set in the setting device 50 of the blow molding machine in order to produce a bottle B having physical properties that meet the target or fall within the target range can be output to the inference device 80. The heater output, blow pressure, blow timing, rod speed, and cooling water temperature data output from the inference device 80 are provided to the setting device 50 of the blow molding machine as setting data, and the blow molding machine operates in accordance with the provided setting data. Bottle B that meets the target physical properties can be produced under operating conditions for the blow molding machine that appear to be set by a skilled operator, and even under operating conditions for the blow molding machine that even a skilled operator would not notice.

[0066] As described above, since the learning data includes bottle appearance information, it is also possible to have the inference device 80 output the heater output, blow pressure, blow timing, rod speed, and cooling water temperature that should be set in the setting device 50 of the blow molding machine in order to produce a bottle B that has the desired physical properties and is free from abnormalities in appearance (for example, no whitening), thereby enabling the molding of a bottle B of higher quality.

[0067] Instead of or in addition to the target values ​​of physical property information that the molded bottle B should satisfy, it is also possible to focus on the quality of bottle B that the molded bottle B should satisfy (target quality of bottle B). In this case, trained model 73 learns the quality of bottle B (whether or not blow molding was successful, whether there is misalignment, whether or not condensation is present, etc.), and inference device 80 infers the operating conditions that should be set in the blow molding machine in order to mold bottle B that satisfies the target quality, based on trained model 73.

[0068] Figure 7 shows the processing results of an inference device 80 using a trained model 73. The bottle molding conditions are "bottle capacity (600 ml)" and "bottle shape (round)," the target value (explanatory variable, input data) for the physical properties of the bottle to be molded is "full capacity" (ml), and the two blow molding machine operating conditions, "heater output" (%) (total output) and "blow pressure" (first air pressure) (bar), are used as inferred values ​​(objective variable, output data). Figure 7 also shows the measured full capacity values ​​of five bottles B actually molded from five preforms P of the same type by setting the heater output and blow pressure inferred for different target full capacity values ​​in the blow molding machine, as well as the error between the target value and the measured value. The "error between the target value and the measured value" also includes the "error (= measured value - target value)," the "error rate (= error ÷ measured value)," the "absolute error rate (= absolute value of the error rate)," and the "mean absolute error rate (= sum of absolute error rates ÷ number of data points)." 8 is a graph with the horizontal axis representing the target value of the full filling capacity and the vertical axis representing the actual measured value of the full filling capacity, and the target value and the actual measured value for each of the five data are plotted with circles. The dashed line in the graph of FIG. 8 is an auxiliary line showing the case where the target value and the actual measured value match.

[0069] The average absolute error rate of the five data points was small at 0.124%, and it was confirmed that by molding Bottle B using the heater output and blow pressure estimated using the trained model 73, it was possible to mold Bottle B with a full filling capacity that was quite close to the target value.

[0070] Figure 9 shows the inference results of an inference device 80 using a trained model 73, with input data being the physical property information of the preform P, such as "weight," "virgin resin / recycled resin," and the IV (Intrinsic Viscosity) value (the inherent viscosity (intrinsic viscosity) of the resin), and the inferred value (output data) being "heater output" (%) (total output), which is one of the operating conditions of the blow molding machine.

[0071] Figure 9 also shows the bottle quality (here, moldability and misalignment) of eight bottles B molded using the blow molding machine with the heater output inferred for different input data (weight, virgin / recycled material, and IV value) ("Inferred Results Reflected"). It also shows the bottle quality of eight bottles B molded using a blow molding machine with the heater output (total output) fixed at 70% without heater output control based on the inferred values ​​("Before Inference"). In the "Moldability" column, "OK" indicates that the blow molding was successful, while "NG" indicates that it was not successful (a bottle B with a distorted shape was molded). "Misalignment" refers to a misalignment between the center of the entire bottle B and the center of its bottom. "OK" indicates that no misalignment occurred, while "NG" indicates that misalignment occurred.

[0072] Comparing No. 1 and No. 3, the weight and IV value of the preform P are the same, but the difference is whether preform P made from virgin resin (new resin) was used (No. 1) or recycled resin (second-hand resin) was used (No. 3). Comparing No. 1 and No. 3, the fact that preform P is made from recycled resin was given as input data (No. 3), leading to the inference that a lower heater output (68%) was used for No. 3 than for No. 1, which was made from virgin resin (70%). As a result, for No. 3, the misalignment that occurred was resolved when the heater output was set to 70%. This inference is consistent with the countermeasure learned from previous experience that recycled resin preform P is more likely to soften when heated than virgin resin preform P, and therefore more likely to cause misalignment. To prevent this, when using recycled resin preforms, the preform temperature after heating should be lowered.

[0073] Comparing No. 1 and No. 2, we see that they have the same weight of preform P and are both made of virgin resin, but differ in the intrinsic viscosity (limiting viscosity) of the PET resin, expressed as an IV value. Comparing No. 1 and No. 2, an IV value of 0.80 was used as input data (No. 2), leading to a lower heater output (67%) for No. 2 than the heater output (70%) for No. 1, which was made from PET resin with an IV value of 0.82. As a result, for No. 2, the misalignment that occurred when the heater output was set to 70% was successfully resolved. This inference is consistent with the approach we have learned from previous experience: preforms made from PET resin with a low IV value tend to soften when heated, making them more susceptible to misalignment. To prevent this, we lower the preform temperature after heating when using preforms made from PET resin with a low IV value.

[0074] Comparing No. 1 and No. 5, both preforms P are made from virgin resin and have the same PET resin IV value, but they differ in weight. By comparing No. 1 and No. 5 and providing weight as one of the input data, a heater output (80%) greater than the heater output (70%) for No. 1's preform P is inferred for No. 5, which is lighter than No. 1. As a result, bottle B can be successfully molded from preform P for No. 5. This inference is consistent with the approach learned from past experience that lighter preforms should have a higher preform heating temperature to eliminate molding abnormalities.

[0075] Figure 10 shows the inference results of an inference device 80 using a trained model 73, with the input data being the environmental information, namely, the "temperature" and "humidity" around the blow molding machine, and the inferred values ​​(output data) being the operating conditions of the blow molding machine, namely, "heater output" (%) (total output) and "mold cooling water temperature" (°C).

[0076] FIG. 10 shows the bottle quality (full capacity and condensation) and preform temperature of each of six bottles B molded by setting the heater output and mold cooling water temperature inferred for different input data (either the ambient temperature or humidity of the blow molding machine) in the blow molding machine, as well as the bottle quality and preform temperature of each of six bottles B molded by a blow molding machine with the heater output (total output) fixed at 65% and the mold cooling water temperature fixed at 8°C.

[0077] Comparing No. 1 and No. 3, their humidity levels were relatively similar (30% and 20%, respectively), but their temperatures were different (25°C for No. 1 and 3 for No. 3). By inputting a temperature of 33°C (No. 3), a lower heater output (62%) was inferred compared to the heater output (65%) required for No. 1 at 25°C. This resulted in the elimination of the deviation (528 ml) from the target full volume (525 ml) for Bottle B, which occurred when the heater output was set to 65%. This inference is consistent with the approach learned from past experience: when the ambient temperature of the blow molding machine is high, the temperature of the heated preform P increases, resulting in a larger full volume for Bottle B. To prevent this, the heater output is reduced when the ambient temperature of the blow molding machine is high, thereby maintaining a constant temperature for Preform P.

[0078] Comparing No. 1 and No. 5, their humidity levels were relatively similar (30% and 35%, respectively), but their temperatures were different (25°C for No. 1 and 20°C for No. 5). By using temperature (=20°C) as one of the input data points (No. 5), a higher heater output (67%) was inferred compared to the heater output (65%) for No. 1 at 25°C. As a result, the deviation (521 ml) from the target full volume (525 ml) for Bottle B, which occurred when heater output was set to 65%, was eliminated. This inference is consistent with the countermeasure learned from past experience: when the ambient temperature of the blow molding machine is low, the temperature of the heated preform P drops, resulting in a smaller full volume for Bottle B. To prevent this, when the ambient temperature of the blow molding machine is low, the heater output is increased to maintain a constant temperature for Preform P.

[0079] Comparing No. 1 and No. 2, they have the same temperature but different humidity (No. 1: 30% humidity, No. 2: 70%). By inputting humidity (70%) as one of the input data (No. 2), a higher mold cooling water temperature (10°C) was inferred compared to the 30% humidity (8°C) for No. 1. As a result, when the mold cooling water temperature was set to 8°C, condensation was prevented from adhering to Bottle B. This inference is consistent with the approach learned from past experience: when the humidity around the blow molding machine is high, condensation easily forms on the mold, causing bottle B to become defective. To prevent this, when the humidity around the blow molding machine is high, the mold cooling water temperature is increased to prevent condensation.

[0080] 1 Preform supply station 2 Heating station 3 Molding station 5 Heater zone 5a Heater 7 Mold 11 Stretching rod 12 Air pump 13 Valve block 20 Heater function 30 Blowing function 40 Control device 50 Setting device 61 Thermometer 62 Hygrometer 70 Learning device 71 Learning data acquisition unit 72 Model generation unit 73 Trained model 75, 85 Recording medium 80 Inference device 82 Inference data acquisition unit 83 Inference unit B Bottle P Preform

Claims

1. An inference device for inferring the operating conditions of a blow molding machine that heats a preform with a heater, moves the heated preform to a mold, stretches the preform in the axial direction by inserting a stretch rod into the preform, and stretches the preform outward by supplying pressurized fluid inside to form a hollow bottle, the device comprising: a trained model in which parameters are trained so that physical property information or environmental information of the bottle is input data and the operating conditions of the blow molding machine are output data; and an inference means for inferring the operating conditions of the blow molding machine that should be set on the blow molding machine in order to mold a bottle that satisfies the target value or target quality, which is inferred by inputting target values ​​of physical property information or environmental information that the molded bottle should satisfy into the trained model.

2. The operating condition inference device for a blow molding machine according to claim 1, wherein the input data further includes at least one of physical property information of the preform and molding conditions of the bottle.

3. The operating condition inference device for a blow molding machine according to claim 1, wherein the operating condition of the blow molding machine is at least one of a heater output condition for heating the preform and a pressure condition of a fluid supplied to the preform.

4. The operating condition inference device for a blow molding machine according to claim 3, wherein the operating conditions of the blow molding machine further include a timing condition for supplying a fluid to the preform, or a speed condition for a stretch rod inserted into the preform.

5. An operating condition inference device for a blow molding machine as described in claim 1, wherein the input data includes information regarding the presence or absence of visual abnormalities in the bottle, and the inference means infers the operating conditions of the blow molding machine that should be set on the blow molding machine in order to mold bottles that satisfy the target value or target quality and are free of visual abnormalities.

6. The operating condition inference device for a blow molding machine according to claim 2, wherein the physical property information of the preform includes data on the material and weight of the preform.

7. The operating condition inference device for a blow molding machine according to claim 2, wherein the physical property information of the preform includes data distinguishing whether the preform is made of virgin resin or recycled resin.

8. The operating condition inference device for a blow molding machine according to claim 2, wherein the physical property information of the preform includes an IV value of the preform.

9. The operating condition inference device for a blow molding machine according to claim 2, wherein the bottle molding conditions include data on the volume and shape of the bottle.

10. The operating condition inference device for a blow molding machine according to claim 1, wherein the physical property information of the bottle includes at least one of the full filling capacity, buckling strength, overall height, wall thickness and section weight of the bottle.

11. The operating condition inference device for a blow molding machine according to claim 1, wherein the environmental information includes at least one of the temperature and humidity around the blow molding machine.

12. The operating condition inference device for a blow molding machine according to claim 1, further comprising a setting means for setting the operating conditions of the blow molding machine inferred by said inference means in said blow molding machine.

13. A learning device comprising: a learning data acquisition means for acquiring learning data including operating conditions of a blow molding machine which heats a preform with a heater, moves the heated preform to a mold, stretches the preform in the axial direction by inserting a stretch rod into the preform, and stretches the preform outward by supplying a pressurized fluid inside, thereby forming a hollow bottle, and physical property information or environmental information of the molded bottle; and a model generation means for using the learning data to generate a trained model for inferring, from a target value or target quality of physical property information that the molded bottle should satisfy, the operating conditions of the blow molding machine that should be set on the blow molding machine in order to mold a bottle that satisfies the target value or target quality.

14. A method for inferring operating conditions of a blow molding machine that heats a preform with a heater, moves the heated preform to a mold, stretches the preform in the axial direction by inserting a stretch rod into the preform, and stretches the preform outward by supplying pressurized fluid into the inside to thereby mold a hollow bottle, comprising: preparing a trained model in which parameters are trained so that physical property information or environmental information of the bottle is input data and the operating conditions of the blow molding machine are output data; and inferring operating conditions of the blow molding machine that should be set on the blow molding machine in order to mold a bottle that satisfies the target value or target quality, which are inferred by inputting target values ​​of physical property information or environmental information that the molded bottle should satisfy, into the trained model.

15. A program for causing a computer to carry out the method according to claim 14.

Images