Device for predicting the rate of defective moldings

Abstract

To provide a prediction device that can accurately predict a defect incidence rate in a molded product molded from resin powder by powder slush molding.SOLUTION: A prediction device 20 includes: a learning part 22 that machine-learns a method for calculating a defect incidence rate of a molded product F molded according to manufacturing conditions, as training data of a plurality of past manufacturing conditions consisting of a softening point of a synthetic resin of resin particles, an average particle size of the resin particles, a percentage of moisture contained in resin powder P, and a heating temperature of a mold 90, and the defect incidence rate of the molded product F for each of the plurality of past manufacturing conditions; and a prediction part 23 that causes a learning model to calculate the defect incidence rate of the molded product F to be molded, and predicts this as the defect incidence rate of the molded product F to be molded, by inputting the softening point of the synthetic resin under the manufacturing conditions for planned molding, the average particle size of the resin particles, the percentage of moisture contained in the resin powder P, and the heating temperature of the mold 90 into the learning model, which is machine-learned by the learning part 22.SELECTED DRAWING: Figure 4

Description

[Technical field]

[0001] The present invention relates to a prediction device for the defect occurrence rate of molded products. [Background technology]

[0002] Molded products, such as vehicle parts, such as instrument panels, are manufactured by powder slush molding. In powder slush molding, synthetic resin powder is poured into a powder box, and a heated mold is rotated together with the powder box to form a powder layer along the mold. The resin powder gels on contact with the mold and adheres to the mold as a powder layer. The powder layer is solidified by cooling together with the mold, and the molded product is removed from the mold. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] JP 2007-203581 A Summary of the Invention [Problem to be solved by the invention]

[0004] However, powder slush molding can produce defective products with poor wall thickness, scratches, and the like, and it is difficult to accurately predict the rate at which such defective products will occur.

[0005] The present invention has been made in consideration of these problems, and an object of the present invention is to provide a prediction device that can accurately predict the defect occurrence rate of molded products molded by powder slush molding from resin powder. [Means for solving the problem]

[0006] As a result of extensive research, the inventors have discovered a new finding that the proportion of moisture contained in the resin powder greatly affects the quality of the molded product when molding by powder slush molding. The present invention is based on this new finding, and the prediction device for the defect occurrence rate of molded products according to the present invention is a prediction device for predicting the defect occurrence rate of molded products to be molded by powder slush molding from resin powder, and uses as training data a plurality of manufacturing conditions used in past molding, which are composed of a softening point of the synthetic resin constituting the resin powder, an average particle size of the resin particles constituting the resin powder, the proportion of moisture contained in the resin powder when performing the powder slush molding, and a heating temperature of the mold when performing the powder slush molding, and the defect occurrence rate of molded products molded by powder slush molding for each of the plurality of manufacturing conditions. and a prediction unit which has a learning model machine-learned by the learning unit, and causes the learning model to calculate a defect rate of the molded product by inputting, as the manufacturing conditions to be molded, the softening point of the synthetic resin constituting the resin powder, the average particle size of the resin particles constituting the resin powder, the proportion of moisture contained in the resin powder when performing the powder slush molding, and the heating temperature of the molding mold when performing the powder slush molding, and predicts the calculated defect rate as the defect rate of the molded product to be molded.

[0007] The manufacturing conditions, which are comprised of the softening point of the synthetic resin constituting the resin powder, the average particle size of the resin particles constituting the resin powder, the proportion of moisture contained in the resin powder when performing powder slush molding, and the heating temperature of the mold when performing powder slush molding, are parameters that contribute to the occurrence of defects in the molded product.

[0008] Specifically, if the resin powder contains too much water when powder slush molding is performed, the resin powder is likely to form lumps, making it difficult to soften. As a result, the synthetic resin adhering to the mold becomes non-uniform, making it easy to produce defective products. However, if the softening point of the synthetic resin constituting the resin powder and the heating temperature of the mold when powder slush molding are changed, it is difficult to specify the softened state of the synthetic resin. Furthermore, if the average particle size of the resin particles constituting the resin powder is changed, the amount of water adhering to the resin particles also changes, and the degree to which the resin powder is likely to form lumps also changes. Therefore, it is difficult to predict the defective rate of molded products based only on the percentage of water contained in the resin powder when powder slush molding is performed.

[0009] Therefore, according to the present invention, a combination of past manufacturing conditions, which are composed of the softening point of the synthetic resin, the average particle size of the resin particles, the proportion of moisture contained in the resin powder, and the heating temperature of the mold, and the defect occurrence rate of molded products molded according to the past manufacturing conditions is used as training data. Specifically, using this training data, the learning unit can machine-learn a calculation method (algorithm) for the defect occurrence rate of molded products molded under the past manufacturing conditions, and construct a machine-learned learning model.

[0010] The prediction unit has a learning model constructed by the learning unit through machine learning. Thus, by inputting the softening point of the synthetic resin constituting the resin powder, the average particle size of the resin particles constituting the resin powder, the proportion of moisture contained in the resin powder when performing powder slush molding, and the heating temperature of the mold when performing powder slush molding as any manufacturing conditions to be molded, into the learning model of the prediction unit, the defect occurrence rate of the molded product to be molded can be predicted with high accuracy.

[0011] In a more preferred embodiment, the prediction device further includes a moisture estimation unit that estimates the percentage of moisture contained in the resin powder when the powder slush molding is performed from the humidity of the storage location for the resin powder during its storage period, and the prediction unit predicts the defect rate using the moisture percentage estimated by the moisture estimation unit as the moisture percentage contained in the resin powder when the powder slush molding is performed.

[0012] According to this aspect, the moisture percentage is accurately estimated by the moisture estimation unit from the humidity of the storage location during the storage period of the resin powder. Therefore, the moisture percentage contained in the resin powder when performing powder slush molding can be accurately predicted without actually measuring it immediately before performing powder slush molding. As a result, the estimated moisture percentage can be used by the prediction unit to improve the accuracy of prediction of the defect rate of molded products. Effect of the Invention

[0013] According to the present invention, it is possible to accurately predict the defect occurrence rate of molded products molded by powder slush molding from resin powder. [Brief description of the drawings]

[0014] [Figure 1] 1(a) to 1(d) are schematic diagrams for explaining a manufacturing method for a molded product in which a defect rate is predicted, and are diagrams for explaining powder slush molding. [Diagram 2] 1 is a schematic diagram of a prediction system including a prediction device for predicting a defect occurrence rate of a molded product according to a first embodiment. FIG. [Diagram 3] FIG. 3 is a block diagram of the prediction device shown in FIG. 2. [Figure 4] 3A is a diagram for explaining machine learning by the learning unit shown in FIG. 2, and FIG. 3B is a diagram for explaining utilization of a learning model by the prediction unit shown in FIG. 2. [Diagram 5] FIG. 3 is a flow diagram illustrating a prediction method using the prediction device shown in FIG. 2. [Figure 6]FIG. 11 is a block diagram of a prediction device according to a second embodiment. [Figure 7] 7(a) is a diagram for explaining machine learning by the second learning unit of the moisture estimation unit shown in FIG. 6, and FIG. 7(b) is a diagram for explaining utilization of the second learning model of the moisture calculation unit shown in FIG. 2. [Figure 8] FIG. 7 is a flow diagram for explaining a prediction method using the prediction device shown in FIG. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0015] Hereinafter, a prediction device for predicting a defect rate of a molded product according to a first and second embodiments of the present invention will be described with reference to FIGS. 1 to 8. FIG.

[0016] First, a molding method for molding a molded product F by powder slush molding from a resin powder P will be described with reference to Fig. 1. Figs. 1(a) to (d) are schematic diagrams for explaining a manufacturing method for a molded product with a predicted defect rate, and are diagrams for explaining powder slush molding.

[0017] As shown in Fig. 1(a), a molding die 90 for molding a molded product F is heated to a predetermined heating temperature by a heating device 91. While maintaining the temperature of the heated molding die 90, as shown in Fig. 1(b), the molding die 90 is connected to a powder box 93 containing a resin powder P. In this connected state, the molding die 90 is rotated together with the powder box 93.

[0018] As a result, the resin powder P contained in the powder box 93 moves to and melts on the molding surface of the heated molding die 90. As a result, a powder layer F1 is formed along the molding surface of the molding die 90.

[0019] Next, as shown in Fig. 1(c), the powder layer F1 is solidified by cooling the powder layer F1 together with the mold 90 in a cooling device 95. Finally, as shown in Fig. 1(d), the powder layer F1 is peeled off from the mold 90, thereby releasing the molded product F from the mold 90.

[0020] However, when the above-mentioned molded product F is molded by powder slush molding, defective products may be generated whose shape and thickness do not meet the specifications. Specifically, if the resin powder P contains too much moisture when powder slush molding is performed, the resin powder P tends to form lumps and is difficult to soften. As a result, the synthetic resin (resin particles) adhering to the molding die 90 becomes non-uniform, and defective products are likely to occur.

[0021] However, when the softening point of the synthetic resin constituting the resin powder P and the heating temperature of the molding die 90 during powder slush molding change, it is difficult to determine the softened state of the synthetic resin. Furthermore, when the average particle size of the resin particles constituting the resin powder changes, the amount of moisture adhering to the resin particles also changes, and this changes the degree to which the resin powder P tends to form lumps. Therefore, it is difficult to predict the defect rate of the molded product F based only on the percentage of moisture contained in the resin powder P during powder slush molding.

[0022] [First embodiment] From this viewpoint, in this embodiment, a prediction system 1 including a prediction device 20 for predicting the defect occurrence rate of a molded product F shown in Fig. 2 is used. Fig. 2 is a schematic diagram of the prediction system 1 including the prediction device 20 for predicting the defect occurrence rate of a molded product F according to the first embodiment.

[0023] The prediction system 1 according to this embodiment includes an input device 10, a prediction device 20, an output device 30, and a moisture content measuring instrument 40. The input device 10 is configured with a keyboard or the like, and is a device for inputting teacher data, to be described later, manufacturing conditions for planned molding, and the like, to the prediction device 20, and is connected to the prediction device 20.

[0024] The prediction device 20 is composed of a calculation device (not shown) such as a CPU, and a storage device (not shown) such as a RAM, a ROM, etc. The storage device stores teacher data and a learning model, which will be described later, and the calculation device is a device that reads data from the storage device, constructs a learning model by machine learning, and performs processing to utilize the learned learning model. The prediction device 20 is connected to an output device 30.

[0025] The output device 30 is configured with a display or the like, and is a device that displays the result of the defect occurrence rate predicted by the prediction device 20. The input device 10 and the output device 30 may form a touch panel display, and the input device 10, the prediction device 20, and the output device 30 may form a tablet terminal, a mobile terminal, or the like.

[0026] Furthermore, the moisture content measuring device 40 is a generally known device that measures the percentage of moisture contained in a resin powder made of a synthetic resin by feeding the resin powder into the device. The measurement principle of this device is not particularly limited as long as it can measure the percentage of moisture (moisture rate) contained in the resin powder.

[0027] For example, the moisture content measuring device 40 may be a device that measures the moisture content (moisture ratio) of the resin powder P using a loss on drying method, or may be a device that measures the moisture content (moisture ratio) of P using a moisture meter. The moisture content (measurement result) of the resin powder P measured by the moisture content measuring device 40 is input to the prediction device 20 via the input device 10.

[0028] 2, the resin powder P is stored in storage bags R1, R2, R3, etc., and therefore even if the resin powder P is the same type, the percentage of moisture in the resin powder P contained in each storage bag R1, R2, R3, etc. will differ if the arrival dates and times of the storage bags R1, R2, R3, etc., the outside air conditions during the storage period, etc. are different. Therefore, in this embodiment, the percentage of moisture contained in the resin powder P is actually measured for each storage bag R1, R2, R3, etc.

[0029] The software of the prediction device 20 will be described below with reference to Fig. 3 and Fig. 4. Fig. 3 is a block diagram of the prediction device 20 shown in Fig. 2. Fig. 4(a) is a diagram for explaining machine learning by the learning unit shown in Fig. 2, and Fig. 4(b) is a diagram for explaining utilization of a learning model by the prediction unit shown in Fig. 2.

[0030] The prediction device 20 includes, as software, a storage unit 21, a learning unit 22, and a prediction unit 23. As described above, the inventors have found that the defect rate of the molded product F molded by powder slush molding is greatly affected by the manufacturing conditions shown below. Therefore, the storage unit 21 stores data linking the manufacturing conditions in moldings performed in the past with the defect rate of the molded product molded under those manufacturing conditions as teacher data for machine learning by the learning unit 22.

[0031] 4(a), memory unit 21 stores a plurality of manufacturing conditions A, B, C..., which are made up of (1) the softening point of the synthetic resin constituting resin powder P, (2) the average particle size of the resin particles constituting resin powder P, (3) the percentage of moisture contained in resin powder P when performing powder slush molding, and (4) the heating temperature of molding die 90 when performing powder slush molding. Furthermore, memory unit 21 also stores, for each of the manufacturing conditions A, B, C..., the result of (5) the defect occurrence rate of molded product F molded under these manufacturing conditions.

[0032] Here, the defect rate of the molded product F is the ratio of the number of defective products to the total number of molded products F manufactured for each of the containing bags R1, R2, R3, .... However, if the containing bags R1, R2, R3, ... are all stored under the same conditions and contain resin powders P with the same physical properties, the defect rate of the molded product F for all of these resin powders P may be used.

[0033] Here, the synthetic resin constituting the resin powder P is a generally known thermoplastic resin used in powder slush molding. Examples of synthetic resins include olefin-based thermoplastic elastomer resin (TPO), thermoplastic polyurethane elastomer resin (TPU), soft polyvinyl chloride resin (PVC), etc. These synthetic resins have a softening point (softening temperature). The average particle size of the resin particles constituting the resin powder P can be measured by a generally known method, such as a measuring method using a sieve specified by JIS.

[0034] As shown in Fig. 4(a), the learning unit 22 performs machine learning using training data including a plurality of manufacturing conditions A, B, C, etc. under which molding was performed in the past, and the defect occurrence rate of molded product F molded by powder slush molding for each of the plurality of manufacturing conditions A, B, C. Specifically, using this training data, machine learning is performed to learn a calculation method for the defect occurrence rate of molded product F molded under the manufacturing conditions, and a learning model (AI model) is constructed. In this manner, the learning unit 22 performs a learning phase by machine learning.

[0035] This calculation method is an algorithm, and the learning model is not particularly limited as long as it is capable of performing such learning, and may be a generally applicable learning model such as a neural network (a network structure composed of specific neurons).

[0036] The training data consists of explanatory variables and objective variables. The explanatory variables are variables that constitute multiple manufacturing conditions A, B, C, etc. Specifically, the explanatory variables are the softening point of the synthetic resin that constitutes the resin powder P, the average particle size of the resin particles that constitute the resin powder P, the percentage of moisture contained in the resin powder P when performing powder slush molding, and the heating temperature of the mold when performing powder slush molding. The objective variable is the defect occurrence rate of molded product F molded by powder slush molding for each of multiple manufacturing conditions A, B, C, etc.

[0037] The prediction unit 23 has a learning model (a trained learning model) that has been machine-learned by the learning unit 22. As shown in FIG. 4(b), the prediction unit 23 inputs the manufacturing conditions for the planned molding to the learning model via the input device 10, thereby causing the learning model to calculate the defect occurrence rate of the molded product F. The prediction unit 23 predicts the defect occurrence rate calculated by the learning model as the defect occurrence rate of the molded product F to be molded, and transmits this result to the output device 30.

[0038] The manufacturing conditions to be molded are (1) the softening point of the synthetic resin constituting the resin powder P to be used, (2) the average particle size of the resin particles constituting the resin powder P to be used, (3) the percentage of moisture contained in the resin powder P when the planned powder slush molding is performed, and (4) the heating temperature of the molding die 90 when the planned powder slush molding is performed. Here, (3) may be an actual value measured by the moisture meter 40.

[0039] In this way, according to this embodiment, a combination of past manufacturing conditions consisting of the softening point of the synthetic resin, the average particle size of the resin particles, the proportion of moisture contained in the resin powder, and the heating temperature of the molding die 90, and the defect occurrence rate of the molded product F molded in accordance with these past manufacturing conditions is used as training data. Therefore, the learning unit 22 can accurately machine-learn a calculation method (algorithm) for the defect occurrence rate of the molded product F molded under any manufacturing conditions, and a machine-learned learning model can be constructed. As a result, if the prediction unit 23 inputs any manufacturing conditions to be molded into the learned learning model, the defect occurrence rate of the molded product to be molded can be accurately predicted.

[0040] In this embodiment, the above-mentioned training data (1) to (5) are used as training data for machine learning. However, by further using conditions related to the molding die 90, conditions of other equipment, disturbance conditions, etc. as training data as necessary, the accuracy of predicting the defect occurrence rate of the molded product F can be further improved.

[0041] A prediction method using the prediction device 20 will be described below with reference to the flow diagram shown in Fig. 5. First, in step S51, a plurality of past manufacturing conditions and the corresponding defect occurrence rates of molded product F are input to the prediction device 20 from the input device 10. As a result, the prediction device 20 inputs the same as training data into the memory unit 21.

[0042] Next, in step S52, a method for calculating the defect rate of molded product F under any manufacturing condition is machine-learned by the learning unit 22 using the multiple past manufacturing conditions and the corresponding defect rates of molded product F stored in the memory unit 21 as teacher data (learning phase). As a result, a learning model machine-learned by the learning unit 22 is constructed, which constitutes the prediction unit 23.

[0043] Next, in step S53, the manufacturing conditions of the molded product F to be molded are input from the input device 10 to the prediction device 20. As a result, the manufacturing conditions of the molded product F to be molded are input to the trained learning model of the prediction unit 23.

[0044] Finally, in step S54, the prediction unit 23 calculates the defect occurrence rate of the molded product F to be molded, and outputs the result to the output device 30 as the predicted defect occurrence rate of the molded product F. Here, by changing the manufacturing conditions and the resin powder to be used, or adjusting the arrival time and storage period of the resin powder, based on the defect occurrence rate output to the output device 30, it is possible to suppress the occurrence of defective molded products F.

[0045] Second Embodiment The prediction device 20 according to the second embodiment will be described below with reference to Figs. 6 to 8. Fig. 6 is a block diagram of the prediction device 20 according to the second embodiment. The prediction device 20 of the second embodiment differs from that of the first embodiment in that a moisture estimating unit 24 is newly provided. Therefore, detailed description of the configuration of the prediction device 20 of the second embodiment that is similar to that of the first embodiment will be omitted.

[0046] 6, the prediction device 20 according to the second embodiment further includes, as software, a moisture estimation unit 24. The moisture estimation unit 24 estimates the proportion of moisture contained in the resin powder P when powder slush molding is performed from the humidity of the storage location during the storage period of the resin powder P.

[0047] In the first embodiment, the proportion of moisture contained in the resin powder P when powder slush molding is performed is actually measured, and the prediction unit 23 uses this actually measured moisture proportion to calculate the defect occurrence rate of the molded product F. However, in the second embodiment, the prediction unit 23 uses the moisture proportion estimated by the moisture estimation unit 24 to the moisture proportion contained in the resin powder P when powder slush molding is performed to calculate the defect occurrence rate of the molded product F, and predicts this as the defect occurrence rate of the molded product F.

[0048] The percentage of moisture contained in the resin powder P is approximately the same when the resin powder P arrives from the manufacturing plant of the resin powder P, but the percentage of moisture contained in the resin powder P changes depending on the change in humidity of the storage location during the storage period of the resin powder P. Therefore, the moisture estimation unit 24 estimates the percentage of moisture contained in the resin powder P when powder slush molding is performed from the humidity of the storage location during the storage period of the resin powder P.

[0049] Therefore, for example, the moisture estimation unit 24 may use the humidity of the storage location every certain time (e.g., 24 hours) during the storage period. A functional formula (recurrence formula of a numerical sequence) of the moisture ratio contained in the resin powder P before the certain time has elapsed and the humidity of the storage location during that period may be created to calculate the moisture ratio after the certain time has elapsed. Such a functional formula may be created by conducting experiments in advance for each type of resin powder P.

[0050] By repeating this calculation, the percentage of moisture contained in the resin powder P after the storage period has elapsed since arrival (i.e., when powder slush molding is performed) can be measured. The percentage of moisture contained in the resin powder P, which is the first term of the functional formula (sequence formula), is the percentage of moisture in the resin powder P when the resin powder P arrives, but it may also be the percentage of moisture measured in a regular inspection during the storage period, for example. In this case, the percentage of moisture in the resin powder P can be measured from the humidity of the storage location after the measurement in the regular inspection.

[0051] Furthermore, the humidity of the storage location during the storage period used by the moisture estimation unit 24 may be an actually measured value, or may be predicted, for example, from the relationship with the humidity of the outside air. Furthermore, if the timing of powder slush molding is later (for example, several days later) than the timing for predicting the defect rate, the moisture estimation unit 24 may calculate the moisture ratio of the resin powder P using, for example, the humidity of the outside air (humidity for several days later predicted together with a weather forecast, etc.). This makes it possible to predict the defect rate of molded products F to be molded in the future.

[0052] Thus, moisture estimation unit 24 is not particularly limited as long as it estimates the percentage of moisture contained in resin powder P when powder slush molding is performed from the humidity of the storage location during the storage period of resin powder P. Therefore, for example, moisture estimation unit 24 may estimate the percentage of moisture contained in resin powder P when powder slush molding is performed as shown in the following Figs. 7 and 8.

[0053] Fig. 7(a) is a diagram for explaining machine learning by the second learning unit 24a of the moisture estimation unit 24 shown in Fig. 6, and Fig. 7(b) is a diagram for explaining utilization of the second learning model of the moisture calculation unit 24b of the moisture estimation unit 24 shown in Fig. 2. In this embodiment, the first learning unit and the first learning model (first AI model) correspond to the learning unit 22 and the learning model shown in Fig. 4.

[0054] As shown in Fig. 7(a) and (b), the moisture estimation unit 24 includes a second learning unit 24a and a moisture calculation unit 24b. As described above, the inventors have found that the moisture ratio of the resin powder P when powder slush molding is performed is greatly affected by the storage conditions including at least the humidity of the storage location. Therefore, the memory unit 21 stores data linking the storage conditions implemented in the past with the moisture ratio of the resin powder P measured under those storage conditions as teacher data for machine learning by the second learning unit 24a of the moisture estimation unit 24.

[0055] 7(a), memory unit 21 stores a plurality of storage conditions a, b, c... as storage conditions consisting of (I) humidity of the storage location, (II) temperature (air temperature) of the storage location, (III) type of synthetic resin (variable (numerical value) specifying the synthetic resin), and (IV) average particle size of the resin particles that make up resin powder P during a past storage period. Furthermore, memory unit 21 also stores, for each of storage conditions a, b, c..., (V) the percentage of moisture contained in resin powder P measured as a result of storage under these storage conditions.

[0056] As shown in FIG. 7(b), the learning unit 22 performs machine learning using a plurality of storage conditions a, b, c, etc. stored in the past and the proportions of moisture (actual measured values) contained in the resin powder P stored under each of the plurality of storage conditions a, b, c as training data. Specifically, using this training data, a method of calculating the proportion of moisture contained in the resin powder P under the storage conditions a, b, c, etc. is machine-learned to construct a second learning model (second AI model). In this way, the learning unit 22 performs a learning phase by machine learning. Note that this calculation method is an algorithm, and the learning model is not particularly limited as long as it can perform such learning, and may be, for example, a commonly known learning model such as a neural network (a network structure composed of specific neurons).

[0057] The teacher data is composed of explanatory variables and objective variables. The explanatory variables are variables that constitute the storage conditions a, b, c, .... Specifically, the variables are composed of (I) the humidity of the storage location, (II) the temperature (air temperature) of the storage location, (III) the type of synthetic resin (variable (numerical value) that identifies the synthetic resin), and (IV) the average particle size of the resin particles that constitute the resin powder P during the past storage period. (I) and (II) are values ​​measured at predetermined intervals during the storage period after the resin powder P arrives, and the longer the storage period is, the more data are input under one storage condition. When the storage period is short, less data is input, and a specific numerical value is input in the remaining input field. The objective variable is the percentage of moisture contained in the resin powder P due to the storage conditions a, b, c, ... of the resin powder P.

[0058] The moisture calculation unit 24b has a second learning model (a trained second learning model) that has been machine-learned by the second learning unit 24a. As shown in FIG. 7(b), the moisture calculation unit 24b inputs the storage conditions of the resin powder P to be used to the second learning model via the input device 10, thereby causing the second learning model to calculate the proportion of moisture contained in the resin powder P. The result is transmitted to the prediction unit 23.

[0059] In this case, the planned storage conditions refer to (I) the humidity of the storage location, (II) the temperature (air temperature) of the storage location, (III) the type of synthetic resin (variable (numerical value) that identifies the synthetic resin), and (IV) the average particle size of the resin particles that make up the resin powder P, during the storage period of the resin powder P that will be used in the future. Note that the values ​​of (I) and (II) may be predicted values ​​from the present until the day molding is performed, and may be obtained from a weather forecast by the Japan Meteorological Agency or the like.

[0060] According to this embodiment, the moisture percentage is accurately estimated by moisture estimation unit 24 from the humidity of the storage location during the storage period of resin powder P. Therefore, the moisture percentage contained in resin powder P when powder slush molding is performed can be accurately predicted without actually measuring it immediately before powder slush molding, and as a result, the estimated moisture percentage can be used by prediction unit 23 to improve the accuracy of prediction of the defect occurrence rate of molded product F.

[0061] In particular, a combination of the past storage conditions of the resin powder P and the moisture percentage of the resin powder P measured corresponding to the past storage conditions is used as training data. Therefore, the second learning unit 24a of the moisture estimation unit 24 can accurately machine-learn a calculation method (algorithm) for the moisture percentage contained in the resin powder P stored under the past storage conditions, and a machine-learned learning model can be constructed. As a result, if any storage conditions to be used are input from the input device 10 to the moisture calculation unit 24b of the moisture estimation unit 24 to the learned second learning model, the moisture percentage of the resin powder P to be used can be accurately calculated.

[0062] In this embodiment, the teacher data for machine learning is the teacher data with the above-mentioned (I) to (IV) as explanatory variables, but only the humidity of the storage location (I) during the past storage period may be used as the explanatory variable. However, the adhesion to the resin particles is also highly dependent on (II) the temperature (air temperature) of the storage location, (III) the type of synthetic resin (variable (numerical value) that specifies the synthetic resin), and (IV) the average particle size of the resin particles that compose the resin powder P. Therefore, by using these variables as explanatory variables, the proportion of moisture contained in the resin powder P to be used can be calculated with high accuracy.

[0063] A prediction method using the prediction device 20 will be described below with reference to the flow chart shown in Fig. 8. Steps S51 to S54 are the same as those shown in Fig. 5, and detailed description will be omitted. In parallel with step S51, in step S81, a plurality of past storage conditions and the corresponding moisture percentages contained in the resin powder P are input from the input device 10 to the prediction device 20. As a result, the data are input to the memory unit 21 of the prediction device 20 as training data.

[0064] Next, in step S82, the second learning unit 24a of the moisture estimation unit 24 uses the multiple past storage conditions and the corresponding moisture percentages in the resin powder P stored in the memory unit 21 as teacher data to machine-learn the moisture percentage in the resin powder P under any storage condition (learning phase). As a result, a second learning model learned by machine learning in the second learning unit 24a is constructed, and this constitutes the moisture calculation unit 24b of the moisture estimation unit 24.

[0065] Next, in step S83, the storage conditions of the resin powder P to be used are input from the input device 10 to the prediction device 20. As a result, the storage conditions of the resin powder P to be used are input to the trained second learning model of the moisture calculation unit 24b of the moisture estimation unit 24.

[0066] In step S84, the moisture ratio contained in the resin powder P to be used is calculated by the moisture calculation unit 24b, and the result is input to the prediction unit 23 in step S53. As a result, the prediction unit 23 calculates the defect occurrence rate of the molded product F by using the moisture ratio estimated by the moisture estimation unit 24 as the moisture ratio contained in the resin powder P when powder slush molding is performed, and predicts this as the defect occurrence rate of the molded product F to be molded.

[0067] Although the embodiments of the present invention have been described in detail above, the present invention is not limited to the above-described embodiments, and various design modifications can be made without departing from the spirit of the present invention as described in the claims. [Explanation of symbols]

[0068] 20: prediction device, 22: learning unit, 23: prediction unit, 24: moisture estimation unit, F: molded product, P: resin powder

Claims

1. A prediction device for predicting a defect rate of a molded product to be molded by powder slush molding from a resin powder, comprising: a learning unit that uses as training data a plurality of manufacturing conditions under which molding was previously performed, the manufacturing conditions being composed of a softening point of a synthetic resin constituting the resin powder, an average particle size of resin particles constituting the resin powder, a proportion of moisture contained in the resin powder when performing the powder slush molding, and a heating temperature of a molding die when performing the powder slush molding, and a defect occurrence rate of a molded product molded by the powder slush molding for each of the plurality of manufacturing conditions, and machine learning to learn a method of calculating the defect occurrence rate of the molded product molded under the manufacturing conditions; a prediction unit which has a learning model trained by machine learning in the learning unit, and inputs to the learning model, as production conditions to be molded, the softening point of the synthetic resin constituting the resin powder, the average particle size of the resin particles constituting the resin powder, the proportion of moisture contained in the resin powder when performing the powder slush molding, and the heating temperature of a molding die when performing the powder slush molding, to cause the learning model to calculate a defect occurrence rate of a molded product, and predicts the calculated defect occurrence rate as the defect occurrence rate of the molded product to be molded; A prediction device for predicting a defect rate of a molded product, comprising:

2. The prediction device further includes a moisture estimation unit that estimates a ratio of moisture contained in the resin powder when the powder slush molding is performed from a humidity of a storage location during a storage period of the resin powder, 2. The apparatus for predicting the defect rate of a molded product according to claim 1, wherein the prediction unit predicts the defect rate by using the moisture percentage estimated by the moisture estimation unit as the moisture percentage contained in the resin powder when the powder slush molding is performed.

Images