Classification destination determination device, classification destination determination method and program

Abstract

The classification destination determining device extracts a first record indicating a reversal of a rotation direction of the rotating body, extracts a second record in which the opening and closing state is registered as open from a record slightly earlier than the first record, extracts a third record from the first record, in which the second record is registered slightly earlier than the third record, and determines a classification destination of the disassembled parts based on the rotation angle registered in the third record.

Description

Technical Field

[0001] The present invention relates to a classification destination determination device and a technology related thereto. Background Art

[0002] Conventionally, there are known construction machines (dismantling machines) in which a crusher (shearing machine) is installed at the distal end of a working attachment. For example, in a dismantling machine (1) described in Patent Document 1, a crusher (9) is installed at the distal end of a boom (8) constituting a part of an attachment (4) (see Patent Document 1). Figure 1 ).

[0003] However, a car dismantling machine equipped with a shearing machine is used to recover reusable parts (dismantled parts) such as wiring harnesses and small parts contained in scrapped vehicles. In this operation, the car dismantling machine is operated to recover parts from scrapped vehicles and the recovered parts are sorted into one of a plurality of sorting destinations.

[0004] More specifically, in the work area, the following actions are repeatedly performed: a shearing machine grabs a component (wire harness, etc.) from a scrapped vehicle, rotates while grabbing the component, releases the component at one of a plurality of sorting destinations, and returns to the work area.

[0005] As a method of analyzing the above-mentioned work, it is conceivable to film the work and register the predetermined information by manual input while watching the filmed video. However, since this requires a lot of time and effort, it is not necessarily efficient.

[0006] In view of this, the inventors of the present application have studied a method that can analyze the operation using only the operation data (time sequence data) obtained by the construction machinery (such as automobile dismantling machine) in the above operation. One of the problems is that it is difficult to correctly determine the classification destination where the object (such as dismantled parts) has been classified from multiple classification destinations using only the operation data.

[0007] In order to solve the above problem, the inventors of the present application first studied a method for determining the classification destination of the disassembled parts (objects) based on the coordinates of the release point of the shearing machine, but failed to obtain sufficient accuracy. After analyzing the reasons for the failure to obtain accuracy, it was found that when the disassembled parts grasped by the shearing machine are classified into each classification destination, the disassembled parts are thrown to each classification destination by the inertia of the rotation. As a result, it was found that the disassembled parts grasped by the shearing machine are released in front of each classification destination.

[0008] That is, the release point of the shears is likely to be located before the actual sorting destination, and if an attempt is made to identify the sorting destination based on the coordinates of the release point of the shears, it is likely that the identification cannot be performed correctly.

[0009] Prior art literature

[0010] Patent Literature

[0011] Patent Document 1: Japanese Patent Publication No. 2017-141552 Summary of the invention

[0012] Accordingly, an object of the present invention is to provide a technology capable of accurately determining a classification destination of an object based on work data of a construction machine that performs a sorting operation of sorting an object into one of a plurality of classification destinations in response to an operator's operation.

[0013] In order to achieve the above-mentioned purpose, one method disclosed in the present invention is a classification destination determination device for determining the classification destination of an object based on the operation data of an engineering machinery, wherein the engineering machinery includes a rotating body and a holding body that holds the object, and, accompanied by the operation of an operator, can perform a classification operation of classifying the object into one of a plurality of classification destinations, and the classification destination determination device includes: an acquisition unit, which acquires the operation data having a plurality of records registered in chronological order, each record including an associated rotation angle of the rotating body and a holding state of the holding body; an extraction unit, which extracts a first record indicating that the rotation direction of the rotating body has been reversed from the plurality of records based on the rotation angle registered in the operation data, extracts a second record in which the holding state is registered as released from a record registered slightly earlier than the first record, and extracts a third record from the first record, wherein the second record is registered slightly earlier than the third record; and a determination unit, which determines the classification destination of the object based on the rotation angle registered in the third record.

[0014] According to this configuration, the classification destination of the object can be accurately determined based on the work data of the construction machine that performs the classification work of classifying the object into one of a plurality of classification destinations in accordance with the operation of the operator. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 It is a side view showing a dismantling machine according to an embodiment of the present invention.

[0016] Figure 2 This is a schematic plan view for explaining the sorting operation of dismantled parts by a dismantling machine.

[0017] Figure 3This is a diagram showing the operation data collected during the operation period of the classification operation.

[0018] Figure 4 This is a functional block diagram showing a classification destination determination device according to an embodiment of the present invention.

[0019] Figure 5 This is a flowchart showing the classification destination determination processing performed by the classification destination determination program.

[0020] Figure 6 This is a graph showing the explanatory variables and target variables used in random forest machine learning.

[0021] Figure 7 This is a diagram showing how a random forest decision tree is generated based on a teaching data set.

[0022] Figure 8 It is a graph showing the verification results based on the prediction model (classification accuracy based on multiple decision trees).

[0023] Fig. 9 It is a diagram showing the verification results of a comparative example (a prediction model that does not take the rotation angle into consideration).

[0024] Fig.10 It is a diagram showing the verification results of other comparative examples (prediction model with added teaching data). DETAILED DESCRIPTION

[0025] <1. Implementation Method>

[0026] based on Figures 1 to 10 The classification destination determination device according to the embodiment of the present invention will be described. Figure 1 ). In addition, an example of a classification destination determination device is a classification destination determination device 1 (see Figure 4 ).

[0027] like Figure 1 As shown, the dismantling machine 10 includes a lower traveling body 11, an upper rotating body 12 and an attachment 13. A shearing machine 14 is installed at the distal end of the attachment 13, and the shearing machine 14 clamps and removes dismantled parts (wiring harnesses or small parts) from the dismantling object (car, etc.).

[0028] Here, the upper rotating body 12 is an example of a "rotating body". In addition, the shearing machine 14 is an example of a "holding body". Furthermore, the disassembled parts are an example of an "object".

[0029] In accordance with the operator's operation, the dismantling machine 10 grabs and takes out dismantled components such as wire harnesses and small parts from a dismantling target such as an automobile, and sorts the dismantled components into predetermined sorting destinations.

[0030] In detail, Figure 2 As shown, the dismantling machine 10 clamps and holds the dismantled components (wiring harnesses or small components) in the dismantling operation area AR1 using the shears 14, and moves the dismantled components in a predetermined direction ( Figure 2 counterclockwise in the middle).

[0031] Next, the dismantling machine 10 releases the wire harness near the wire harness classification destination AR2 while holding the wire harness by the shears 14. On the other hand, the dismantling machine 10 releases the small component near the small component classification destination AR3 while holding the small component by the shears 14. As a result, the wire harness is classified into the wire harness classification destination AR2, and the small component is classified into the small component classification destination AR3.

[0032] Then, the dismantling machine 10 moves in the direction opposite to the prescribed direction ( Figure 2 The dismantling machine 10 rotates in the clockwise direction and returns to the dismantling operation area AR1. With the operation of the operator, the dismantling machine 10 repeatedly performs the above-mentioned actions.

[0033] During the operation of the above operation, the dismantling machine 10 obtains various operation information at fixed sampling intervals and accumulates them into Figure 3 Specifically, the dismantling machine 10 acquires the position coordinates of the shearing machine 14, the rotation angle of the upper rotating body 12, and the opening and closing information of the shearing machine 14 as the operation information, and accumulates them in the operation data DT1.

[0034] like Figure 3 As shown, the operation data DT1 registers records RC1, RC2, RC3, . . . in which sampling time, position coordinates of the shearing machine 14, the rotation angle of the upper rotating body 12, and opening and closing information of the shearing machine 14 are associated in chronological order.

[0035] An example of the position coordinates of the shearing machine 14 is the value of the X coordinate, the value of the Y coordinate, and the value of the Z coordinate of the shearing machine 14 .

[0036] An example of the rotation angle of the upper rotating body 12 is a value of the rotation angle of the upper rotating body 12 relative to a reference position (for example, a position at which disassembled components are taken out in the disassembly work area AR1 ).

[0037] Furthermore, an example of the opening and closing information of the shearing machine 14 is an opening and closing state based on the output value of a stroke sensor provided in a cylinder for opening and closing the shearing machine 14. The opening and closing state includes "open" indicating that the shearing machine 14 is in a released state, and "closed" indicating that the shearing machine 14 is in a closed state.

[0038] Next, refer to Figure 4 The classification destination determination device 1 is described in detail. The classification destination determination device 1 is a computer for determining the classification destination of the disassembled parts based on the above-mentioned work data DT1. Figure 4 As shown in FIG. 1 , the classification destination determination device 1 includes a control unit 3 and a storage unit 5 .

[0039] The control unit 3 includes a central processing unit (CPU), and executes various calculation processes based on the programs and data stored in the storage unit 5 .

[0040] like Figure 4 As shown, the control unit 3 includes a job data acquisition unit 31 (an example of an acquisition unit), a record extraction unit 32 (an example of an extraction unit), an input data calculation unit 33 (an example of a calculation unit), a prediction model generation unit 34 (an example of a generation unit) and a classification destination determination unit 35 (an example of a determination unit).

[0041] The work data acquisition unit 31 acquires the work data DT1 registered in the storage unit 5 .

[0042] The record extraction unit 32 extracts records that meet predetermined conditions from the work data DT1.

[0043] The input data calculation unit 33 generates input data that is an explanatory variable used by a random forest, which is a machine learning algorithm.

[0044] The prediction model generation unit 34 generates a prediction model for determining a classification destination of a disassembled component using a random forest as a machine learning algorithm.

[0045] The classification destination determination unit 35 determines the classification destination of the disassembled parts during the work period using the prediction model.

[0046] The storage unit 5 has a memory or a magnetic disk device, and stores various programs and data. In addition, the storage unit 5 also functions as a working memory of the control unit 3. In addition, the storage unit 5 can be composed of an information storage medium such as a flash memory and an optical disk, or can be composed of a reading device that reads information from the information storage medium.

[0047] For example, in the above hardware configuration, the storage unit 5 is shown as being built in the classification destination determination device 1 , but the present invention is not limited thereto, and a storage device that can communicate with the classification destination determination device 1 may be externally connected.

[0048] like Figure 4 As shown, the storage unit 5 stores the classification destination determination program PG, work data DT1 , teaching data DT2 , verification data DT3 , and verification result data DT4 .

[0049] The classification destination determination program PG is a program for executing the classification destination determination process ( Figure 5 The process of determining the classification destination of the disassembled parts during the operation is a program of the process of determining the classification destination of the disassembled parts during the operation.

[0050] As described above, the operation data DT1 is data obtained by the dismantling machine 10 during operation by acquiring operation information at fixed sampling intervals and accumulating the data (see Figure 3 ).

[0051] The teaching data DT2 is the input data and output data used for machine learning. Specifically, the teaching data DT2 is data containing the explanatory variables v = (d1, d2, d3, d4, d5, d6, d7) used in the random forest and the target variable (target variable) t. The teaching data DT2 is data that associates the target variable with the task data DT1 used for machine learning in the task data DT1.

[0052] Explanation variables v=(d1, d2, d3, d4, d5, d6, d7) are variables calculated by the input data calculation unit 33. Target variable t is a variable indicating an actual classification destination (a classification destination after the worker visually confirms the video captured during the work).

[0053] Verification data DT3 is data used to verify the prediction model generated by machine learning. Verification data DT3 is data that is not used as teaching data DT2 in random forest learning. That is, verification data DT3 is work data DT1 other than work data DT1 used as teaching data DT2 in work data DT1. Verification data DT3 includes explanatory variables v = (d1, d2, d3, d4, d5, d6, d7).

[0054] The verification result data DT4 is a result obtained by verifying the prediction model generated by machine learning using the verification data DT3.

[0055] Next, refer to Figure 5The processing executed by the classification destination determination program PG will be described in detail. In addition, "step S" will be simply described as "S" hereinafter.

[0056] First, in S1 , the work data acquisition unit 31 acquires the work data DT1 from the storage unit 5 .

[0057] Next, in S2 , the record extraction unit 32 performs exponential smoothing on the rotation angles registered in chronological order in the work data DT1 .

[0058] Next, in S3 , the record extracting unit 32 extracts all records registered at the time when the rotation direction of the dismantling machine 10 is reversed from the predetermined direction as reverse position records (an example of first records) based on the exponentially smoothed rotation angle.

[0059] Specifically, the record extraction unit 32 extracts all records registered at the time when the sign of the slope of the exponentially smoothed rotation angle function reverses (from positive to negative or from negative to positive) as the reverse position records.

[0060] The exponentially smoothed rotation angle has the following characteristics: when the upper rotating body 12 rotates in a predetermined direction, it continues to increase (or decrease), and when the upper rotating body 12 rotates in a direction opposite to the predetermined direction, it continues to decrease (or increase). That is, if the rotation direction of the upper rotating body 12 is reversed, the sign of the slope of the exponentially smoothed rotation angle function is also reversed. In S3, this characteristic is used to determine the reversal position record.

[0061] In addition, the original data of the rotation angle registered in the operation data DT1 is not monotonically increasing or monotonically decreasing data, and contains noise. Therefore, if the original data of the rotation angle is directly used in the extraction of the reversal position record (S3), the reversal position record may be erroneously extracted. Therefore, in this embodiment, in S2, the original data of the rotation angle is exponentially smoothed, and in S3, the reversal position record is determined using the exponentially smoothed rotation angle.

[0062] In S4, the record extraction unit 32 extracts a record in which the opening and closing information of the shearing machine 14 is registered as "open" from the record registered slightly earlier than the reversal position record (the record registered just before the reversal position record) as a release position record (an example of a second record). More specifically, the record extraction unit 32 extracts a record in which the release information of the shearing machine 14 is first registered as "open" from among the predetermined number of records registered before the reversal position record as a release position record.

[0063] In this embodiment, the record registered slightly before the reversal position record does not refer to only the record before the reversal position record, but refers to a predetermined number of records registered before the reversal position record. The predetermined number is appropriately set according to the sampling interval of the operation data DT1.

[0064] In S5, the record extraction unit 32 extracts, from the reversal position records, the reversal position record in which the release position record is registered in the record that is slightly earlier, as the post-release reversal position record (an example of the third record). That is, the record extraction unit 32 extracts the record registered at the time when the rotation direction of the shearing machine 14 is reversed after release as the post-release reversal position record. In other words, the record registered at the time when the rotation direction of the shearing machine 14 is reversed in the closed state is not extracted as the post-release reversal position record.

[0065] In S6 , the input data calculation unit 33 calculates the input data (explanatory variable v=(d1, d2, d3, d4, d5, d6, d7)) used for the random forest as the machine learning algorithm.

[0066] Specifically, the input data calculation unit 33 calculates input data (explanatory variable v) using the work information registered in the release position record extracted in S4 and the work information registered in the post-release reversal position record extracted in S5 .

[0067] In more detail, Figure 6 As shown, the input data calculation unit 33 calculates the angle deviation d1 of the rotation angle at the reversing position, the position coordinates d2, d3, and d4 of the shearing machine 14 at the release position, and the elapsed time d5 from the release position to the reversing position, the moving distance d6, and the moving rotation angle d7 as the explanatory variables v. The reversing position is the position of the shearing machine 14 at the moment when the upper revolving body 12 is reversed. The release position is the position of the shearing machine 14 at the moment when the shearing machine 14 is released. The moving distance d6 is the distance from the release position to the reversing position. The moving rotation angle d7 is the rotation angle of the upper revolving body 12 from the release position to the reversing position. The angle deviation d1 is the angle deviation of the rotation angle calculated from the average rotation angle during the operation time.

[0068] The input data calculation unit 33 calculates the above-mentioned interpretation variable v=(d1, d2, d3, d4, d5, d6, d7) for all the extracted release position records and post-release reversal position records.

[0069] The teaching data DT2 is data in which the explanatory variables v=(d1, d2, d3, d4, d5, d6, d7) and the target variable t are registered in association. As described above, the target variable t is a variable indicating the classification destination (correct classification destination) after the operator visually confirms the video captured during the operation.

[0070] In addition, for the sake of convenience, the teaching data DT2 is also expressed as a teaching data set S = {(v, t)} (see Figure 6 ).

[0071] In S7, the prediction model generation unit 34 generates a prediction model based on random forest. Random forest is a well-known machine learning algorithm, which is an ensemble learning algorithm that combines the results of multiple decision trees (weak classifiers) to perform classification, regression, and clustering.

[0072] In the present embodiment, a classification problem of determining a classification destination of a dismantled component from a plurality of classification destinations is dealt with, and therefore a classification tree is used as a decision tree of a random forest.

[0073] like Figure 7 As shown, the prediction model generation unit 34 randomly samples training data S1, S2, ..., SN for learning each decision tree from the teaching data set S by bootstrapping.

[0074] The prediction model generation unit 34 first randomly extracts M variables for branching from the training data S1 and creates a decision tree T1. The prediction model generation unit 34 repeatedly performs this process on the training data S2, ..., SN, and finally generates N decision trees T1, T2, ..., TN.

[0075] In addition, each decision tree is split to a predetermined depth. In addition, the value of the Gini coefficient is used as the splitting criterion for the root node and the internal node. Furthermore, the target variable t is set as a leaf node.

[0076] In S8, the classification destination determination unit 35 inputs the verification data DT3 to the N decision trees T1, T2, ..., TN, and determines the classification destination of the disassembled parts by majority decision of the classification destinations classified by the decision trees T1, T2, ..., TN, thereby determining the classification destination of the disassembled parts. As a result, the classification destination of each disassembled part during the operation is classified into any one of the disassembly operation area AR1, the wire harness classification destination AR2, and the small parts classification destination AR3.

[0077] Figure 8 The verification result data DT4 (DT41) is obtained by generating N decision trees T1, T2, ..., TN (prediction model) using the teaching data DT2, and verifying the classification accuracy using the verification data DT3.

[0078] Specifically, the first row of the verification result data DT41 is the result of generating a prediction model (N decision trees T1, T2, ..., TN) using data No. 15 as teaching data DT2, and verifying the prediction model using data No. 16 to No. 20 as verification data DT3.

[0079] The second row of the verification result data DT41 is the result of generating a prediction model using data No. 16 as teaching data DT2 and verifying the prediction model using data No. 15, 17 to 20 as verification data DT3. The third to sixth rows of the verification result data DT41 are also the results of verifying the prediction model using the same criteria.

[0080] The verification result data DT41 obtained an average of 91.3% (refer to Figure 8 The existing method (clustering without random forest) only achieved an average classification accuracy of 72%, so it can be seen that the classification accuracy has been greatly improved.

[0081] in addition, Fig. 9 The verification result data DT4 (DT42) is the result of verifying the classification accuracy based on the following prediction model (comparative example), which reduces the number of explanatory variables v compared with the above-mentioned embodiment. In the comparative example, the angle deviation d1 of the rotation angle at the reversal position, the elapsed time d5 from the release position to the reversal position, the moving distance d6, and the moving rotation angle d7 are excluded from the explanatory variables v, and the prediction model is generated using the explanatory variables V = (d2, d3, d4). That is, in the comparative example, the prediction model is generated using only the information related to the release position of the shearing machine 14.

[0082] like Fig. 9 As shown, the verification result data DT42 obtained an average of 89.3% (refer to Fig. 9Thus, similarly to the above-mentioned embodiment, the classification accuracy is improved compared to the existing method (clustering without using random forest).

[0083] However, the classification accuracy of the comparative example is poor compared to the above embodiment. Therefore, it can be seen that the angle deviation d1 of the rotation angle at the reversal position, the elapsed time d5 from the release position to the reversal position, the moving distance d6, and the moving rotation angle d7 contribute to improving the classification accuracy.

[0084] As described above, in actual operation, when the disassembled parts caught by the shears 14 are sorted into the respective sorting destinations, the disassembled parts caught by the shears 14 are released in front of the respective sorting destinations and thrown to the respective sorting destinations. Therefore, there is a high possibility that the release point of the shears 14 is located in front of the actual sorting destination. Therefore, it is inferred that if the sorting destination is to be determined using only the information related to the release position of the shears 14, there is a high possibility that the sorting destination cannot be determined correctly.

[0085] On the other hand, after the shearing machine 14 releases and throws out the dismantled parts, the upper rotating body 12 still rotates slightly due to inertia. Therefore, it is estimated that the reversal position of the rotation reversal of the upper rotating body 12 is likely to coincide with the classification destination of the dismantled parts. Therefore, the above embodiment uses not only the information related to the release position of the shearing machine 14, but also the information related to the rotation angle of the upper rotating body 12 as the input data (explanatory variable) of the random forest.

[0086] If Figure 8 The verification result data DT41 and Fig. 9 By comparing the verification result data DT42 with the above, it is clear that it is effective to consider the information related to the rotation angle of the upper rotating body 12 in the process of determining the classification destination of the disassembled parts. Therefore, it can be said that the above assumption is generally correct.

[0087] Fig.10 The verification result data DT4 (DT43) is the result of verifying the classification accuracy of the prediction model generated by increasing the amount of teaching data DT2.

[0088] Specifically, the first row of the verification result data DT43 is a result of generating a prediction model using data No. 15 and data No. 16 as teaching data DT2 and verifying the prediction model using data No. 17 to No. 20 as verification data DT3.

[0089] In addition, the second row of the verification result data DT43 is the result of generating a prediction model using data No. 15 and data No. 17 as teaching data DT2, and verifying the prediction model using data No. 16, No. 18 to No. 20 as verification data DT3. The third row and subsequent rows of the verification result data DT43 are also the results of verifying the prediction model using the same benchmark.

[0090] The verification result data DT43 obtained an average of 96.0% (ref. Fig.10 Therefore, it can be seen that the prediction model with an increased amount of teaching data DT2 can obtain a higher classification accuracy than the average classification accuracy (91.3%) of the verification result data DT41.

[0091] According to the above embodiment, in the disassembled parts classification work, the classification destination of each disassembled part (wiring harness or small part) during the work can be accurately determined using the N decision trees T1, T2, ..., TN obtained through machine learning using random forest.

[0092] In particular, according to the above embodiment, not only the information on the release position of the shear 14 but also the information on the rotation angle of the upper rotating body 12 is used as input data (explanatory variables) of the random forest. Therefore, high classification accuracy can be obtained in the classification work of the disassembled parts.

[0093] <2. Modifications>

[0094] The classification destination determination device disclosed in the present invention is not limited to the above-described embodiment, and various modifications and improvements can be made within the scope described in the claims.

[0095] For example, in the above-mentioned embodiment, although the case where the prediction model is generated using the random forest is exemplified, the present invention is not limited to this and the prediction model may be generated using a machine learning algorithm other than the random forest.

[0096] In addition, in the above embodiment, although the classification destination of the dismantled parts is determined by using the prediction model based on random forest, the present invention is not limited to this. For example, the classification destination of the dismantled parts may be determined by determining whether the rotation angle registered in the above released reverse position record is within a preset range.

[0097] Specifically, the rotation angle range RG1 corresponding to the wire harness classification destination AR2 and the rotation angle range RG2 corresponding to the small parts classification destination AR3 may be set in advance. Then, when the rotation angle registered in the released after-reversal position record is within the rotation angle range RG1, the classification destination of the disassembled parts is determined to be the wire harness classification destination AR2, and when the rotation angle registered in the released after-reversal position record is within the rotation angle range RG2, the classification destination of the disassembled parts is determined to be the small parts classification destination AR3.

[0098] <3. Modifications>

[0099] The work data acquisition unit 31 acquires work data DT1 whose classification destination is not determined from the dismantling machine 10, and the input data calculation unit 33 generates input data based on the acquired work data DT1. The classification destination determination unit 35 inputs the generated input data into the learned prediction model, thereby determining the classification destination.

[0100] <4. Modifications>

[0101] The sorting destination determination device 1 may be communicatively connected to the dismantling machine 10 via a network such as the Internet, and may acquire the operation data from the dismantling machine 10 .

[0102] (Summary of Implementation Methods)

[0103] A classification destination determination device in one embodiment of the present disclosure is a classification destination determination device that determines the classification destination of an object based on operation data of an engineering machine, wherein the engineering machine includes a rotating body and a holding body that holds the object, and, accompanied by an operator's operation, is capable of performing a classification operation of classifying the object into one of a plurality of classification destinations, the classification destination determination device including: an acquisition unit that acquires the operation data having a plurality of records registered in chronological order, each record including an associated rotation angle of the rotating body and a holding state of the holding body; an extraction unit that extracts, from the plurality of records, a first record indicating that the rotation direction of the rotating body has been reversed, based on the rotation angle registered in the operation data, extracts, from a record registered slightly earlier than the first record (a record registered just before the first record), a second record in which the holding state is registered as released, and extracts, from the first record, a third record, the second record being registered slightly earlier than the third record (just before the third record); and a determination unit that determines the classification destination of the object based on the rotation angle registered in the third record.

[0104] According to this structure, the third record is extracted from the first record indicating that the rotation direction of the rotating body has been reversed, and there is a second record in which the hold state is registered as released in the record registered slightly earlier than the third record, and the classification destination of the object is determined based on the rotation angle registered in the third record. Therefore, the classification destination of the object can be correctly determined based on the operation data of the construction machine, and the construction machine performs a classification operation of classifying the object into any one of a plurality of classification destinations in accordance with the operation of the operator.

[0105] In the above-mentioned sorting destination determination device, preferably, the record further includes position coordinates of the holder, and the determination unit determines the sorting destination of the object based on the rotation angle registered in the third record and the position coordinates registered in the second record.

[0106] According to this configuration, the classification destination of the object is determined based on the rotation angle registered in the third record and the position coordinates registered in the second record. Therefore, the classification destination of the object can be accurately determined.

[0107] In the above-mentioned classification destination determination device, it is more ideal to further include: a calculation unit, which calculates input data based on the third record and the second record, and the input data includes information related to the rotation angle when the rotation direction is reversed and coordinate information related to the release position where the object is released; and a generation unit, which generates a prediction model for determining the classification destination of the object by performing machine learning on the input data and output data representing the classification destination of the object as teaching data, wherein the determination unit uses the prediction model to determine the classification destination of the object.

[0108] According to this structure, the classification destination of the object is determined using a prediction model obtained by machine learning of teaching data, wherein the teaching data includes input data calculated based on the third record and the second record, and output data indicating the classification destination of the object, so that the classification destination can be correctly determined.

[0109] In the above-mentioned classification destination determination device, it is more ideal that the prediction model is a plurality of decision trees generated by a random forest, the input data is an explanatory variable of the random forest, the output data is a target variable of the random forest, and the determination unit uses the plurality of decision trees obtained by machine learning through the teaching data to determine the classification destination of the object.

[0110] According to this structure, the prediction model includes a plurality of decision trees generated by the random forest, and therefore the classification destination of the object can be accurately determined.

[0111] In the above-mentioned classification destination determination device, preferably, the extraction unit performs exponential smoothing on the rotation angle included in the work data, and extracts the first record based on the exponentially smoothed rotation angle.

[0112] According to this configuration, the first record is extracted based on the exponentially smoothed rotation angle, and therefore the first record can be accurately extracted.

[0113] In the above-mentioned classification destination determination device, it is more ideal that the classification operation includes an operation of repeatedly performing the following actions: holding the object in the working area, rotating in a prescribed direction while holding the object, and releasing the object, thereby classifying the object to any one of the multiple classification destinations, and rotating in a direction opposite to the prescribed direction and returning to the working area.

[0114] According to this configuration, the sorting destination can be accurately determined in the sorting work in which holding of the object, turning toward the sorting destination, releasing of the object, and turning toward the work area are repeated.

[0115] In the above-mentioned classification destination determination device, preferably, the operation data is data collected during operation of the construction machine, and the record further includes a sampling time.

[0116] According to this configuration, since the sampling time during the operation is recorded, the time when the rotation direction is reversed and the time when the object is released can be accurately determined.

[0117] In the above-mentioned classification destination determination device, preferably, the rotation angle is a rotation angle relative to a reference position of the construction machine.

[0118] According to this configuration, the rotation angle can be accurately determined.

[0119] In the above-mentioned classification destination determination device, preferably, the information on the rotation angle when the rotation direction is reversed includes the rotation angle registered in the third record.

[0120] According to this configuration, it is possible to accurately identify information on the rotation angle when the rotation direction is reversed.

[0121] In the above-mentioned classification destination determination device, preferably, the coordinate information related to the release position includes the position coordinates of the holding body registered in the second record.

[0122] According to this configuration, coordinate information related to the release position can be accurately specified.

[0123] In the above-mentioned classification destination determination device, preferably, the input data further includes: an elapsed time from the release of the holding body to the reversal of the rotation direction of the rotating body, and a movement distance of the rotating body from the release of the holding body to the reversal of the rotation direction of the rotating body.

[0124] According to this configuration, since the input data includes the elapsed time and the moving distance from the release of the holding body to the reversal of the rotating body, the classification destination can be accurately determined.

[0125] In the above-mentioned classification destination determination device, preferably, the second record is a record in which the release is first registered as the hold state among a predetermined number of records registered before the first record.

[0126] According to this configuration, the release timing of the holding body can be accurately determined slightly before the reversal of the rotating body.

[0127] A classification destination determination method in another embodiment of the present invention is a classification destination determination method in a classification destination determination device for determining a classification destination of an object based on operation data of an engineering machine, wherein the engineering machine includes a rotating body and a retaining body for retaining the object, and is capable of performing a classification operation of classifying the object into any one of a plurality of classification destinations accompanied by an operator's operation. The classification destination determination method obtains the operation data having a plurality of records registered in chronological order, each record being associated with a rotation angle of the rotating body and a retaining state of the retaining body, extracts a first record indicating that a rotation direction of the rotating body has been reversed from the record based on the rotation angle registered in the operation data, extracts a second record in which the retaining state is registered as released from a record registered slightly earlier than the first record, and extracts a third record from the first record, wherein the second record is registered slightly earlier than the third record, and determines the classification destination of the object based on the rotation angle registered in the third record.

[0128] A program in another embodiment of the present disclosure is a program that enables a computer to function as a classification destination determination device that determines a classification destination of an object based on operation data of an engineering machine, wherein the engineering machine includes a rotating body and a retaining body that retains the object, and is capable of performing a classification operation of classifying the object into any one of a plurality of classification destinations accompanied by an operator's operation, and enables the computer to perform the following processing: obtaining the operation data having a plurality of records registered in chronological order, each record being associated with a rotation angle of the rotating body and a retaining state of the retaining body, extracting from the records a first record indicating that a rotation direction of the engineering machine has been reversed based on the rotation angle registered in the operation data, extracting from the records a second record in which the retaining state is registered as released from a record registered slightly earlier than the first record, and extracting from the first record a third record in which the second record is registered slightly earlier than the third record, and determining the classification destination of the object based on the rotation angle registered in the third record.

[0129] Possibility of industrial application

[0130] As described above, the classification destination determination device of the present invention is suitable for correctly determining the classification destination of an object in a classification operation in which the object is classified into any one of a plurality of classification destinations in accordance with an operation by an operator.

Claims

1. A classification destination determination device, characterized in that: Determine the classification destination of objects based on the operation data of construction machinery, The construction machine includes a rotating body and a holding body for holding the object, and can perform a sorting operation of sorting the object into one of a plurality of sorting destinations in accordance with an operation by an operator. The classification destination determination device comprises: an acquisition unit that acquires the operation data in which a plurality of records are registered in chronological order, each record including an associated rotation angle of the rotating body and a holding state of the holding body; an extracting unit that extracts, based on the rotation angle registered in the operation data, a first record indicating that the rotation direction of the rotating body has been reversed from the plurality of records, extracts a second record in which the hold state is registered as released from a record registered slightly earlier than the first record, and extracts a third record from the first record in which the second record is registered slightly earlier than the third record; and The specifying unit specifies a classification destination of the object based on the rotation angle registered in the third record.

2. The classification destination determination device according to claim 1, characterized in that: The record also includes the position coordinates of the holding body, The specifying unit specifies a classification destination of the object based on the rotation angle registered in the third record and the position coordinates registered in the second record.

3. The classification destination determination device according to claim 2, characterized in that Also includes: a calculation unit that calculates input data based on the third record and the second record, the input data including information related to the rotation angle when the rotation direction is reversed and coordinate information related to a release position where the object is released; as well as A generating unit generates a prediction model for determining a classification destination of the object by performing machine learning on teaching data including the input data and output data indicating a classification destination of the object, wherein: The determination unit determines a classification destination of the object using the prediction model.

4. The classification destination determination device according to claim 3, characterized in that: The prediction model is a plurality of decision trees generated by random forests, The input data are explanatory variables of the random forest, The output data is the target variable of the random forest, The determination unit determines a classification destination of the object using the plurality of decision trees obtained by machine learning using the teaching data.

5. The classification destination determination device according to any one of claims 1 to 4, characterized in that: The extraction unit performs exponential smoothing on the rotation angle included in the operation data, and extracts the first record based on the exponentially smoothed rotation angle.

6. The classification destination determination device according to any one of claims 1 to 4, characterized in that: The classification operation includes repeatedly performing the following actions: holding the object in the working area, rotating in a specified direction while holding the object, and releasing the object, thereby classifying the object into one of the multiple classification destinations, and rotating in a direction opposite to the specified direction and returning to the working area.

7. The classification destination determination device according to any one of claims 1 to 4, characterized in that: The operation data is data collected during the operation of the construction machine. The record also contains the sampling time.

8. The classification destination determination device according to any one of claims 1 to 4, characterized in that: The rotation angle is a rotation angle relative to a reference position of the construction machine.

9. The classification destination determination device according to claim 3, characterized in that: The information on the rotation angle when the rotation direction is reversed includes the rotation angle registered in the third record.

10. The classification destination determination device according to claim 3, characterized in that: The coordinate information related to the release position includes the position coordinates of the holding body registered in the second record.

11. The classification destination determination device according to claim 3, characterized in that: The input data further includes: an elapsed time from the release of the holding body to the reversal of the rotation direction of the rotating body; and a movement distance of the rotating body from the release of the holding body to the reversal of the rotation direction of the rotating body.

12. The classification destination determination device according to any one of claims 1 to 4, characterized in that: The second record is a record in which the hold status is first registered as the release among a predetermined number of records registered before the first record.

13. A method for determining a classification destination, characterized in that: A classification destination determination method in a classification destination determination device for determining a classification destination of an object based on operation data of a construction machine, The construction machine includes a rotating body and a holding body for holding the object, and can perform a sorting operation of sorting the object into one of a plurality of sorting destinations in accordance with an operation by an operator. The classification destination determination method, The operation data is obtained in which a plurality of records are registered in chronological order, each record being associated with a rotation angle of the rotating body and a holding state of the holding body, extracting, from the records, a first record indicating that the rotation direction of the rotating body has been reversed, based on the rotation angle registered in the operation data, extracting, from the records registered slightly earlier than the first record, a second record in which the hold state is registered as released, and extracting, from the first record, a third record in which the second record is registered slightly earlier than the third record, A classification destination of the object is determined based on the rotation angle registered in the third record.

14. A computer program product, characterized in that The computer is made to function as a classification destination determination device for determining a classification destination of an object based on operation data of a construction machine, The construction machine includes a rotating body and a holding body for holding the object, and can perform a sorting operation of sorting the object into one of a plurality of sorting destinations in accordance with an operation by an operator. The computer is caused to execute the following processing: The operation data is obtained in which a plurality of records are registered in chronological order, each record being associated with a rotation angle of the rotating body and a holding state of the holding body, Based on the rotation angle registered in the operation data, a first record indicating that the rotation direction of the construction machine has been reversed is extracted from the records, a second record in which the hold state is registered as released is extracted from the records registered slightly earlier than the first record, and a third record in which the second record is registered slightly earlier than the third record is extracted from the first record, A classification destination of the object is determined based on the rotation angle registered in the third record.

Images