JP2020070101A - Biting detection system - Google Patents

Abstract

To provide a biting detection system capable of easily setting a threshold value and easily enhancing an accuracy of biting detection.SOLUTION: A biting detection system 1 includes processing parts 50a, 50b, 60a, 60b for processing an object 100 by repeated operations, a sensor 30 for measuring displacement data representing displacement of the processing parts, a prediction part 14 for outputting the displacement data measured by the sensor during a predetermined period into a learned model generated by machine learning to generate predicted data predicted to be measured by the sensor after a predetermined period based on outputting of the learned model, and a determination part 15 for determining biting of the processing parts based on comparison of difference between the displacement data measured by the sensor after the predetermined period and the predicted data with a threshold value.SELECTED DRAWING: Figure 1

Description

  The present disclosure relates to a bite detection system.

  2. Description of the Related Art Conventionally, there is used a packaging machine that wraps an article such as food in a packaging bag, welds and cuts the packaging bag, and packages the article in small portions. In the packaging machine, biting may occur in which the packaging bag is closed while the article is biting.

  The following Patent Document 1 describes a proximity sensor that executes a process of determining whether the position of a detected object is appropriate. The proximity sensor samples the distance to the object to be detected a certain number of times, and every time sampling for one cycle is performed, the sampled value is compared with the value sampled at the corresponding time in the past to detect the object to be detected. The suitability of the position of the object is determined.

Utility model registration No. 3093852

  With the technique described in Patent Document 1, it is possible to accurately determine whether or not any bite has occurred during welding cutting. Here, in order to detect the bite, it is necessary to set a threshold value serving as a reference for comparison between the sampled value and the value sampled at a corresponding time in the past.

  However, the processing unit that processes an object such as a packaging bag may execute a plurality of operations during one cycle of the repeated operation. In that case, it becomes necessary to set a threshold value for detecting a bite for each of a plurality of operations, which makes it difficult to set the threshold value, and the accuracy of the bite detection may not be sufficiently improved.

  Therefore, the present invention provides a trapping detection system in which the threshold value can be easily set and the precision of trapping detection can be easily enhanced.

  A bite detection system according to an aspect of the present disclosure includes a processing unit that processes an object by a repetitive operation, a sensor that measures displacement data that represents displacement of the processing unit, and displacement data that is measured by the sensor during a predetermined period. , A prediction unit that inputs to a learned model generated by machine learning, and based on the output of the learned model, generates prediction data that is predicted to be measured by the sensor after a predetermined period, and a prediction unit after the predetermined period. A determination unit that determines the biting of the processing unit based on a comparison between a difference between the displacement data measured by the sensor and the prediction data and a threshold value.

  According to this aspect, the prediction unit generates prediction data predicted to be measured by the sensor, and compares the difference between the displacement data actually measured by the sensor and the prediction data with a threshold value to determine the bite. Therefore, even when the processing unit performs a plurality of operations during one cycle of the repeated operation, one threshold value may be set regardless of the number of operations, and the threshold value can be easily set. The accuracy of jam detection can be easily increased.

  In the above aspect, a learning unit that performs machine learning of a learning model using learning data including displacement data measured by a sensor in a predetermined period to generate a learned model may be further included.

  According to this aspect, the learned model can generate prediction data that is predicted to be measured by the sensor after a predetermined time period, and even if the displacement data has a complicated waveform, the bite detection is performed. The accuracy of can be improved.

  In the above aspect, the learning unit further includes a position sensor that measures position data representing a repetitive operation of the processing unit, and the learning unit includes displacement data measured by the sensor during a predetermined period and position data measured by the position sensor during a predetermined period. Machine learning of the learning model may be executed using the learning data to generate the learned model.

  According to this aspect, by including the position data representing the repetitive operation of the processing unit in the learning data, the operation can be identified even when the processing unit performs a plurality of operations during one cycle of the repetitive operation. A trained model can then be generated that produces predictive data that is predicted to be measured by the sensor.

  In the above aspect, the prediction unit inputs the displacement data measured by the sensor in the predetermined period and the position data measured by the position sensor in the predetermined period into the learned model, and based on the output of the learned model, the predetermined period. Predictive data that is predicted to be measured by the sensor after may be generated.

  According to this aspect, even when the processing unit executes a plurality of operations during one cycle of the repeated operation, the predicted data predicted to be measured by the sensor by identifying the operation based on the position data. Can be generated and the bite can be detected with higher accuracy.

  In the above aspect, the processing section may include a plurality of welding cutting sections that weld and cut the object a plurality of times in one cycle of the repeated operation.

  According to this aspect, even when the processing unit welds and cuts the object a plurality of times during one cycle of the repeated operation, one threshold value can be set regardless of the number of times of welding and cutting per cycle. However, the threshold value can be easily set, and the accuracy of bite detection can be easily increased.

In the above aspect, the plurality of welding cutting parts weld and cut an object by a rotating operation,
The position sensor may include a rotary encoder that measures the angular positions of the plurality of weld cuts.

  According to this aspect, by measuring the angular positions of the plurality of welding cutting portions by the rotary encoder, even when the welding cutting portion welds and cuts the object a plurality of times during one cycle of the rotation operation. , It is possible to identify the motion based on the angular position, generate prediction data predicted to be measured by the sensor, and detect the bite with higher accuracy.

  In the above aspect, the determination unit may determine the welding cutting portion in which the biting has occurred among the plurality of welding cutting portions based on the position data.

  According to this aspect, even when the plurality of welding cutting portions perform a plurality of operations during one cycle of the repetitive operation, the welding in which the biting has occurred among the plurality of welding cutting portions is performed based on the position data. The cutting portion can be specified, and the bite can be detected with higher accuracy.

  In the above aspect, the determination unit may determine the consumption of the processing unit based on the difference between the displacement data measured by the sensor after the predetermined period and the prediction data.

  According to this aspect, it is possible to determine whether or not maintenance of the machined portion is necessary by utilizing the fact that the displacement data is constantly displaced due to wear of the machined portion.

  According to the present invention, there is provided a trapping detection system in which the threshold value can be easily set and the precision of trapping detection can be easily enhanced.

It is a figure which shows the outline of the bite detection system which concerns on embodiment of this invention. It is a figure which shows the functional block of the master unit which concerns on this embodiment. It is a figure which shows the example of the displacement data and position data measured by the bite detection system which concerns on this embodiment. 6 is a flowchart of a machine learning process executed by the bite detection system according to the present embodiment. It is a figure which shows the example of the displacement data and position data measured by the bite detection system which concerns on this embodiment, and prediction data. 7 is a flowchart of a bite detection process executed by the bite detection system according to the present embodiment.

  Hereinafter, an embodiment according to one aspect of the present invention (hereinafter referred to as “the present embodiment”) will be described with reference to the drawings. In addition, in each of the drawings, components denoted by the same reference numerals have the same or similar configurations.

  FIG. 1 is a diagram showing an outline of a trapping detection system 1 according to an embodiment of the present invention. The bite detection system 1 detects an bite in a packaging machine that wraps an article with the film material 100, welds and cuts while feeding the film material 100, and wraps the article in small portions.

  The bite detection system 1 includes a processing unit that processes an object by repeating the operation. In the present embodiment, the processing unit includes a plurality of welding cutting units 50a, 50b, 60a, 60b that weld and cut the film material 100, which is an object, a plurality of times in one cycle of the repeated operation. The bite detection system 1 rotates the welding cutting parts 50a, 50b, 60a, 60b while feeding the film material 100, performs welding and cutting with the welding cutting part 50a and the welding cutting part 60a as one set, and the welding cutting part Welding and cutting are performed by using 50b and the welding cutting portion 60b as one set.

  The bite detection system 1 includes a proximity sensor 30 that measures displacement data that represents the displacement of the processed portion. In the present embodiment, the proximity sensor 30 measures the displacement data of the metal plate 90 that moves in conjunction with the displacement of the weld cutting parts 50a and 50b. Here, the welding and cutting parts 50a and 50b are slightly displaced in the direction orthogonal to the feeding direction of the film material 100 when welding and cutting the film material 100. This amount of displacement differs depending on whether the film material 100 is normally welded and cut and when foreign matter is caught. The sensor for measuring the displacement data indicating the displacement of the processed portion may be a sensor other than the proximity sensor, and may be a laser distance measuring sensor, a confocal sensor, an ultrasonic sensor, or the like.

  The bite detection system 1 includes a position sensor that measures position data that represents a repetitive operation of the processing unit. In the present embodiment, the bite detection system 1 includes a rotary encoder 31 that measures the angular positions of the weld cutting parts 50a and 50b. The rotary encoder 31 measures the rotation angles of the welding and cutting parts 50a and 50b. When the processing unit performs reciprocating motion such as piston motion instead of rotating motion, the position sensor may measure position data indicating the position of reciprocating motion.

  The displacement data measured by the proximity sensor 30 and the position data measured by the rotary encoder 31 are acquired by the master unit 10 and it is determined whether or not any bite has occurred in the welding cutting parts 50a, 50b, 60a, 60b. It The determination result may be transmitted to a PC (Personal Computer) 20 or a PLC (Programmable Logic Controller) 40 that is a higher-level device. Further, the user may set the master unit 10 via the PC 20 or the PLC 40.

  FIG. 2 is a diagram showing functional blocks of the master unit 10 according to this embodiment. The master unit 10 includes an acquisition unit 11, a storage unit 12, a learning unit 13, a prediction unit 14, a determination unit 15, and a communication unit 16.

  The acquisition unit 11 acquires displacement data indicating the displacement of the processing unit from the proximity sensor 30, and acquires position data indicating the repetitive operation of the processing unit from the rotary encoder 31. The displacement data and the position data may be time-series data, and the acquisition unit 11 may continuously acquire the displacement data and the position data. However, the acquisition unit 11 may intermittently acquire the displacement data and the position data.

  The storage unit 12 stores the learning data 12a, the learned model 12b, and the threshold 12c. The learning data 12a may include displacement data measured by the proximity sensor 30 during a predetermined period, and may include position data measured by the position sensor (the rotary encoder 31 in this embodiment) during the predetermined period. It is predicted that the learned model 12b is generated by machine learning using the learning data 12a and receives the displacement data measured by the proximity sensor 30 in a predetermined period as an input and is measured by the proximity sensor 30 after the predetermined period. It may be a model that produces predictive data. In addition, the learned model 12b is predicted to be measured by the proximity sensor 30 after a predetermined period, with the displacement data measured by the proximity sensor 30 in the predetermined period and the position data measured by the position sensor in the predetermined period as inputs. It may be a model for generating prediction data. The learned model 12b may be composed of, for example, a neural network or a binary tree. In this example, one type of learning data 12a and one type of learned model 12b are stored, but a plurality of types of learning data 12a and learned model 12b may be stored. Of course, a plurality of types of learned models 12b may be selectively used depending on the type of biting. The threshold 12c is used for the determination by the determination unit 15.

  The learning unit 13 executes the machine learning of the learning model using the learning data 12a to generate the learned model 12b. When the operating conditions of the processing unit are changed, such as when the rotational speeds of the plurality of welding cutting units 50a, 50b, 60a, 60b are changed, the learning unit 13 re-learns the learned model 12b. Good. The learned model 12b generated by the learning unit 13 can generate prediction data predicted to be measured by the proximity sensor 30 after a predetermined period, and even when the displacement data has a complicated waveform. It is possible to improve the accuracy of bite detection. Further, by including the position data representing the repetitive operation of the processing unit in the learning data, even when the processing unit performs a plurality of operations during one cycle of the repetitive operation, the operation is identified and the proximity sensor is detected. A trained model 12b may be generated that produces prediction data that is predicted to be measured by 30.

  The prediction unit 14 inputs the displacement data measured by the proximity sensor 30 in a predetermined period into the learned model 12b generated by machine learning, and based on the output of the learned model 12b, the proximity sensor after a predetermined period. Generate prediction data that is predicted to be measured by 30. The prediction unit 14 inputs the displacement data measured by the proximity sensor 30 in a predetermined period and the position data measured by the position sensor (rotary encoder 31) in a predetermined period into the learned model 12b, and outputs the learned model 12b. Based on, the prediction data that is predicted to be measured by the proximity sensor 30 after the predetermined period may be generated. By including the position data in the input of the learned model 12b, even if the processing unit executes a plurality of operations during one cycle of the repeated operation, the operation is identified based on the position data and the proximity sensor 30 is used. It is possible to detect the bite with higher accuracy by generating prediction data that is predicted to be measured by.

  The determination unit 15 determines the biting of the processing unit based on the comparison between the difference between the displacement data measured by the proximity sensor 30 after a predetermined period of time and the prediction data and the threshold value 12c. Here, the difference between the displacement data and the prediction data may be the difference between the two data at the same time. The threshold 12c may be arbitrarily set by the user, but based on the standard deviation of the difference between the normal displacement data and the predicted data, for example, twice the standard deviation is automatically set as the threshold 12c. Good. The threshold 12c may be set based on the maximum value of the difference between the normal displacement data and the predicted data, or may be set based on the average value of the difference between the normal displacement data and the predicted data. Good. The determination unit 15 may use data obtained by converting at least one of the displacement data and the prediction data. For example, the determination unit 15 may use the data obtained by smoothing the displacement data and the prediction data, the data obtained by Fourier transform, or the normalized data.

  According to the bite detection system 1 according to the present embodiment, the prediction unit 14 generates prediction data predicted to be measured by the proximity sensor 30, and the displacement data and the prediction data actually measured by the proximity sensor 30. By comparing the difference between the number of operations with the threshold value 12c to determine the bite, even when the processing unit executes a plurality of operations during one cycle of the repeated operation, one threshold value 12c is irrespective of the number of operations. Is set, the threshold 12c can be easily set, and the accuracy of bite detection can be easily increased.

  The determination unit 15 may determine the welding cutting portion in which the bite has occurred among the plurality of welding cutting portions 50a, 50b, 60a, 60b based on the position data measured by the position sensor such as the rotary encoder 31. Thereby, even when a plurality of operations are performed by the plurality of welding cutting parts 50a, 50b, 60a, 60b during one cycle of the repeated operation, the plurality of welding cutting parts 50a, 50b, It is possible to identify the welding cut portion in which the biting has occurred among 60a and 60b, and it is possible to detect the biting with higher accuracy.

  The determination unit 15 may determine the consumption of the processing unit based on the difference between the displacement data measured by the proximity sensor 30 after the predetermined period and the prediction data. For example, the determination unit 15 calculates the standard deviation of the difference between the normal displacement data and the predicted data, and when the difference between the displacement data and the predicted data is constantly deviated by more than the standard deviation to one side, It may be determined that the processed portion may be worn, or the processed portion and the film material 100 may be misaligned. If the determination unit 15 determines that the processing unit is exhausted, the fact may be notified to the PC 20 or the PLC 40. As described above, it is possible to determine whether maintenance of the machined part is necessary or not by utilizing the fact that the displacement data is constantly displaced due to the wear of the machined part.

  The communication unit 16 is an interface that communicates with the PC 20 and the PLC 40. The communication unit 16 may communicate with an external device other than the PC 20 and the PLC 40, and may be connected to a communication network such as the Internet.

  FIG. 3 is a diagram showing an example of displacement data v and position data w measured by the bite detection system 1 according to the present embodiment. In the figure, the horizontal axis represents time and the vertical axis represents measured values, and a graph showing the time change of the displacement data v and the position data w is shown. The position data w shown in the figure is data indicating the position in the direction orthogonal to the feeding direction of the film material 100 for any of the plurality of welding cutting parts 50a, 50b, 60a, 60b. It is data indirectly indicating the angular positions of 50a, 50b, 60a, 60b. In the figure, the displacement data v is shown by a solid line, and the position data w is shown by a broken line.

  The bite detection system 1 includes displacement data {v (t-n), v (t-n + 1), ..., V (t)} measured by the proximity sensor 30 during a predetermined period and a position sensor (rotary encoder during a predetermined period). The position data {w (t−n), w (t−n + 1), ..., W (t)} measured by 31) may be stored as the learning data 12a. Here, the predetermined period can be set arbitrarily, but may be, for example, an integral multiple of the cycle of the repeated operation of the processing unit. Further, based on the position data w, the displacement data v measured before and after the welding and cutting of the film material 100 by the plurality of welding cutting parts 50a, 50b, 60a, 60b may be included in the learning data 12a. Good.

  FIG. 4 is a flowchart of a machine learning process executed by the jamming detection system 1 according to this embodiment. First, the bite detection system 1 acquires the displacement data measured by the proximity sensor 30 in a predetermined period (S10), and acquires the position data measured by the rotary encoder 31 in a predetermined period (S11).

  Then, the bite detection system 1 executes machine learning of the learning model using the learning data including the acquired data (S12). The bite detection system 1 determines whether the accuracy of the learning model is equal to or higher than the target value (S13), and when the accuracy of the learning model is not equal to or higher than the target value (S13: NO), obtains displacement data (S10), position data. (S11) and execution of machine learning (S12) are repeated.

  On the other hand, when the accuracy of the learning model is equal to or higher than the target value (S13: YES), the bite detection system 1 stores the learned model (S14). With the above, the machine learning process ends.

  FIG. 5: is a figure which shows the example of the displacement data v and the position data w measured by the bite detection system 1 which concerns on this embodiment, and the prediction data p. In the figure, the horizontal axis represents time and the vertical axis represents measured values, and a graph showing the time change of the displacement data v and the position data w and the prediction data p is shown. In the figure, the displacement data v is shown by a solid line, the position data w is shown by a broken line, and the prediction data p is shown by a dotted line.

  The bite detection system 1 includes displacement data {v (t-n), v (t-n + 1), ..., V (t)} measured by the proximity sensor 30 during a predetermined period and a position sensor (rotary encoder during a predetermined period). 31) The position data {w (t−n), w (t−n + 1), ..., W (t)} measured by 31) is input to the learned model 12b, and based on the output of the learned model 12b, Prediction data p predicted to be measured by the proximity sensor 30 after a predetermined period is generated. Here, the predetermined period can be set arbitrarily, but may be, for example, an integral multiple of the cycle of the repeated operation of the processing unit.

  When the difference between the displacement data v actually measured by the proximity sensor 30 and the predicted data p after the predetermined period is larger than the threshold value 12c, the determination unit 15 determines that biting has occurred in the processing unit. In the example of FIG. 5, the difference between the displacement data v actually measured by the proximity sensor 30 and the predicted data p after a predetermined period is relatively large, which suggests that biting has occurred.

  The processing section may include a plurality of welding cutting sections 50a, 50b, 60a, 60b for welding and cutting the object a plurality of times in one cycle of the repeated operation. In this case, the displacement data v indicates the repeated operation of the processing section. It has a plurality of peaks in one cycle. In such a case, if a threshold value is set for each of a plurality of welding and cutting operations, it is difficult to set the threshold value, and as a result, the accuracy of bite detection may decrease. According to the entrapment detection system 1 according to the present embodiment, even when the object 100 is welded and cut a plurality of times during one cycle of the repeated operation, the number of times of welding and cutting per cycle is performed. Regardless of this, it is sufficient to set one threshold value, the threshold value can be easily set, and the accuracy of bite detection can be easily increased.

  Further, the plurality of welding cutting parts 50a, 50b, 60a, 60b weld and cut the target object 100 by the rotation operation, and the position sensor measures the angular positions of the plurality of welding cutting parts 50a, 50b, 60a, 60b. A rotary encoder 31 may be included. By measuring the angular positions of the plurality of welding cutting parts 50a, 50b, 60a, 60b by the rotary encoder 31, the welding cutting parts 50a, 50b, 60a, 60b are used to measure a plurality of objects 100 during one cycle of the rotation operation. Even in the case of performing the double welding and the cutting, it is possible to detect the operation based on the angular position, generate the prediction data predicted to be measured by the proximity sensor 30, and detect the bite with higher accuracy. .. Further, based on the angular positions of the plurality of welding cutting portions 50a, 50b, 60a, 60b, the welding cutting portion in which the bite has occurred can be determined among the plurality of welding cutting portions 50a, 50b, 60a, 60b.

  FIG. 6 is a flowchart of the trapping detection process executed by the trapping detection system 1 according to the present embodiment. First, the bite detection system 1 acquires the displacement data measured by the proximity sensor 30 for a predetermined period (S20), and acquires the position data measured by the rotary encoder 31 for a predetermined period (S21).

  Then, the bite detection system 1 inputs the acquired data to the learned model 12b and generates prediction data (S22). The entrapment detection system 1 determines whether the deviation between the predicted data and the actually measured displacement data is equal to or more than a threshold value (S23). If the deviation is not equal to or more than the threshold value (S23: NO), the displacement data is acquired (S20), the position. Data acquisition (S21) and prediction data generation (S22) are repeated.

  On the other hand, when the deviation between the predicted data and the actually measured displacement data is equal to or greater than the threshold value (S23: YES), the bite detection system 1 detects the bite and the occurrence of the bite in the processing unit is detected by the PC 20 or the PLC 40. (S24). As described above, the jamming detection process is completed.

  The embodiments described above are for facilitating the understanding of the present invention, and are not for limiting the interpretation of the present invention. Each element included in the embodiment and its arrangement, material, condition, shape, size and the like are not limited to the exemplified ones but can be changed as appropriate. Further, the configurations shown in different embodiments can be partially replaced or combined.

[Appendix 1]
Processing parts (50a, 50b, 60a, 60b) for processing the object (100) by repetitive operation,
A sensor (30) for measuring displacement data representing displacement of the processing parts (50a, 50b, 60a, 60b),
The displacement data measured by the sensor (30) in a predetermined period is input to a learned model (12b) generated by machine learning, and based on the output of the learned model (12b), the predetermined period. A prediction unit (14) for generating prediction data predicted to be measured by the sensor (30) after
Based on a comparison between a difference between the displacement data measured by the sensor (30) after the predetermined period of time and the predicted data and a threshold value, the biting of the processing portion (50a, 50b, 60a, 60b) is performed. A determination unit (15) for determining,
A bite detection system (1) provided with.

  DESCRIPTION OF SYMBOLS 1 ... Bite detection system, 10 ... Master unit, 11 ... Acquisition part, 12 ... Storage part, 12a ... Learning data, 12b ... Learned model, 12c ... Threshold value, 13 ... Learning part, 14 ... Prediction part, 15 ... Judgment part, 16 ... Communication part, 20 ... PC, 30 ... Proximity sensor, 31 ... Rotary encoder, 40 ... PLC, 50a, 50b, 60a, 60b ... Welding cutting part, 90 ... Metal plate, 100 ... Film material

Claims (8)

A processing unit that processes the object by repeated operation,
A sensor for measuring displacement data representing the displacement of the processing part,
The displacement data measured by the sensor in a predetermined period is input to a learned model generated by machine learning, and based on the output of the learned model, measured by the sensor after the predetermined period. A prediction unit that generates predicted prediction data,
A difference between the displacement data measured by the sensor after the predetermined period of time and the prediction data, based on the comparison of the threshold value, the determination unit for determining the biting of the processing unit,
An entrapment detection system including. Further comprising a learning unit that performs machine learning of a learning model using learning data including the displacement data measured by the sensor in the predetermined period, and generates the learned model.
The bite detection system according to claim 1. Further comprising a position sensor that measures position data representing the repetitive operation of the processing unit,
The learning unit performs machine learning of the learning model using the learning data including the displacement data measured by the sensor during the predetermined period and the position data measured by the position sensor during the predetermined period. And generate the trained model,
The bite detection system according to claim 2. The predicting unit inputs the displacement data measured by the sensor in the predetermined period and the position data measured by the position sensor in the predetermined period into the learned model, and outputs the learned model as an output. Based on, generating the prediction data predicted to be measured by the sensor after the predetermined period of time,
The bite detection system according to claim 3. The processing unit includes a plurality of welding cutting units that weld and cut the object a plurality of times in one cycle of the repeating operation,
The bite detection system according to claim 3 or 4. The plurality of welding cutting parts weld and cut the object by a rotating operation,
The position sensor includes a rotary encoder that measures an angular position of the plurality of welding cut portions,
The bite detection system according to claim 5. The determination unit determines, based on the position data, the welding cutting portion in which the biting has occurred among the plurality of welding cutting portions,
The bite detection system according to claim 5 or 6. The determination unit determines wear of the processing unit based on a difference between the displacement data measured by the sensor after the predetermined period and the prediction data,
The bite detection system according to any one of claims 1 to 7.

Images