TWI715133B - Photoelectric sensor - Google Patents

Abstract

Provided is a photoelectric sensor having a simple configuration and determining a state of an object with short time delay. The photoelectric sensor includes: a light projecting part emitting light toward a detection range where the object arrives; a light receiving part acquiring signal values in time series based on reception of the light; a FIFO memory ordering and storing a predetermined number of signal values ​​in an order of acquisition, and periodically updates the predetermined number of signal values ​​with newly acquired signal values; a prediction part, based on a signal value acquired in a first time range, predicting a signal value to be acquired in a second time range after the first time range according to a prediction model including a predetermined parameter; and a determination part, determining the state of the object based on a degree of coincidence between the signal value predicted by the prediction part and a signal value stored in the FIFO memory corresponding to the second time range at a frequency of once every time the FIFO memory is updated once or a plurality of times.

Description

Photoelectric sensor

The present invention relates to a photoelectric sensor having a function of determining the state of an object.

Conventionally, as a sensor for detecting the presence or absence of an object, a photoelectric sensor that irradiates light on the object to detect light transmitted through the object, or detect light shielding by the object, or detect light reflected by the object . In addition, when detecting the state of the object instead of the presence or absence of an object, a vision sensor that captures the object with a camera and performs image analysis is sometimes used.

Regarding a photoelectric sensor, for example, Patent Document 1 below describes a photoelectric sensor configured to store a detection value corresponding to the background level as a zero reset reference value, thereby enabling Arbitrary detection values are displayed using relative values based on the background level.

In addition, the following Patent Document 2 describes an inspection method that scans the surface of a steel plate with laser light, calculates a plurality of characteristic quantities representing the reflected light waveform, and adds them to a neural network that learns the characteristic quantities in advance. And for output with/without flaws.

[Prior Art Literature]

[Patent Literature]

[Patent Document 1] Japanese Patent Laid-Open No. 2001-124594

[Patent Document 2] Japanese Patent Laid-Open No. 2-298840

Sometimes it is necessary to detect the state of an object that is more difficult than the detection of the presence or absence of an object that can be performed with a popular photoelectric sensor, but it has not yet reached the need to be larger and larger than a photoelectric sensor. The degree of multiple capabilities of expensive visual sensors. For example, when recognizing objects with very different shapes or appearances, it is not enough to simply detect the presence or absence of the object, but it has not yet reached the level that requires multiple capabilities of the visual sensor.

Here, it is considered that the state of the object can be determined in principle by combining the conventional photoelectric sensor and the method of analyzing the received light amount waveform as shown in Patent Document 1. However, since it is impossible to obtain an appropriate trigger for acquiring the waveform of only the portion required for determination (the portion corresponding to the size of the defect in Patent Document 2), it takes longer to complete than the portion required for determination. Waveform analysis is performed after acquiring the waveform of the range. For example, it lacks real-time performance when applied to objects successively conveyed on a conveying line.

Therefore, the present invention provides a photoelectric sensor that can determine the state of an object with a simple structure and a small time delay.

One aspect of the photoelectric sensor of the present disclosure includes: a light projection part, Light is emitted from the detection range where the object comes; the light-receiving part obtains the signal value of the time series based on the light-receiving light; the First In First Out (FIFO) memory stores a specified number of items according to the order of acquisition. Signal value, and periodically update a specified number of signal values with the newly acquired signal value; the predicting part uses a prediction model containing specified parameters to predict based on the signal value acquired in the first time range The signal value acquired in the second time range that is later than the first time range; and the determination unit, at a frequency that is performed every time the FIFO memory is updated one or more times, based on the signal value predicted by the prediction unit, The degree of agreement with the signal value stored in the FIFO memory corresponding to the second time range is used to determine the state of the object.

According to the aspect, the frequency is performed once every time the FIFO memory is updated one or more times, based on the signal value predicted by the predicting unit and corresponding to the second time range that is later than the first time range The degree of coincidence of the signal values stored in the FIFO memory is used to determine the state of the object. This allows a simple structure to determine the state of the objects successively conveyed on the conveying line with less time delay.

In the above aspect, when the degree of coincidence is higher than a predetermined value, the determination unit may determine that the state of the object is a specific state.

In the above aspect, the predictive model may be a model generated by machine learning.

According to the aspect, by using the learned model, based on the signal value obtained in the first time range, it is predicted to obtain the signal value in the second time range after the first time range. The signal value obtained can improve the prediction accuracy of the signal value, thereby making it possible to more flexibly determine the state of objects that are successively conveyed on the conveying line.

In the above aspect, it may further include an action control unit that generates a predictive model based on the signal value.

According to this aspect, since the photoelectric sensor can generate the prediction model by itself, it is not necessary to obtain the prediction model from the outside, and the prediction model generated corresponding to the actual object can be used.

In the above aspect, the action control unit can generate a forecast based on the signal value belonging to the variable period when the time-series signal value changes relatively large after the stable period after the time-series signal value changes relatively small. model.

According to the aspect, the signal value generated by the object can be selectively used from the signal value to generate a prediction model.

In the above aspect, the motion control unit can output the prediction model to the outside.

According to the above aspect, since the generated prediction model can be used in other photoelectric sensors, there is no need to repeatedly generate a prediction model for each of a plurality of photoelectric sensors used under the same object and installation conditions .

In the above aspect, the action control unit can output the time-series signal value to the outside.

According to the aspect, the signal value can be output to the outside, so that the external device generates a prediction model. In this way, the photoelectric sensor itself does not need to have computing resources related to the process of generating the predictive model.

In the above aspect, it may further include an action control unit that obtains the prediction model from the outside.

According to the above aspect, by using the prediction model generated by other devices, such as other photoelectric sensors, the generation of the prediction model can be omitted.

According to the present invention, there is provided a photoelectric sensor capable of judging the state of an object with a simple structure and a small time delay.

1: Detection system

10: Photoelectric sensor

10a: detection range

11: Projection Department

11a: Projection element

11b: Drive circuit

12: Light receiving part

12a: Light receiving element

12b: amplifier

12c: sample/hold circuit

12d: A/D converter

13: Processing Department

13a: Motion control section

13b: FIFO memory

13c: Predicted value storage unit

13d: Judgment Department

13e: Forecast Department

14: Operation Department

15: Output section

20: Controller

30: Computer

40: Robot

50: Conveying device

100: Object

P: predicted value

q1~q8: segment

q0: Early stage

q9: last paragraph

r0~r3: signal value, value, predicted value

s0~s3: signal value

S1, S2: The shape of the object

S10~S15: steps

t0~t10: time

W1, W2, W3: Waveform

Fig. 1 is a diagram showing an overview of a detection system including a photoelectric sensor according to an embodiment of the present invention.

Fig. 2 is a diagram showing the configuration of the photoelectric sensor of the present embodiment.

FIG. 3 is a diagram showing an example of the configuration of the processing unit of the photoelectric sensor of the present embodiment.

Fig. 4 is a flowchart of processing in the learning mode and the determination mode of the photoelectric sensor of the present embodiment.

Fig. 5a is a diagram showing an example of signal values measured in the nth cycle (cycle) of the photoelectric sensor of the present embodiment.

Fig. 5b is a diagram showing an example of signal values measured in the n+1 cycle of the photoelectric sensor of this embodiment.

Fig. 6 is a diagram showing another example of signal values measured in the n-th cycle of the photoelectric sensor of the present embodiment.

Fig. 7 is a diagram showing another example of the configuration of the processing unit of the photoelectric sensor of the present embodiment.

Hereinafter, based on the drawings, an embodiment of one aspect of the present invention (hereinafter referred to as "this embodiment") will be described. In addition, in each figure, elements denoted with the same reference numerals have the same or the same configuration.

[Configuration example]

1 to 3, an example of the configuration of the photoelectric sensor 10 of this embodiment will be described. FIG. 1 is a diagram showing the outline of a detection system 1 including a photoelectric sensor 10 of the present embodiment. The detection system 1 includes a photoelectric sensor 10, a controller 20, a computer 30, a robot 40, and a transport device 50.

The photoelectric sensor 10 is a device that detects that the object 100 comes to the detection range 10a of the photoelectric sensor 10 based on the acquired signal value, and determines the state of the object 100. The photoelectric sensor 10 can be a reflective photoelectric sensor, a transmission type photoelectric sensor, or a retro-reflective photoelectric sensor. In addition, the photoelectric sensor 10 may also be a displacement sensor, projecting a laser beam toward the object 100, and obtaining a signal value corresponding to the distance to the object 100 based on the principle of triangulation distance measurement. In addition, the photoelectric sensor 10 may also be a distance measuring sensor that obtains a signal value corresponding to the distance to the object 100 based on the reciprocating time of the light reflected by the object 100. In this specification, "signal value" includes not only the value of the received light amount, but also the signal value corresponding to the distance to the object 100.

The object 100 is an object to be detected by the photoelectric sensor 10, and may be, for example, a finished product of a produced product, or an unfinished product such as a part. The object 100 illustrated in FIG. 1 is an object having a convex shape on a base. In addition, as objects of different types, it is assumed that objects having the same base but not having a convex shape are mixed and transported. When the photoelectric sensor 10 is, for example, a reflective photoelectric sensor, if the object 100 comes to the detection range 10a of the photoelectric sensor 10, the amount of reflected light detected increases. Furthermore, when the object 100 has a shape with protrusions on the base, if there are protrusions of the object 100 in the detection range 10a, the amount of reflected light further increases.

The controller 20 controls the robot 40 and the conveying device 50. The controller 20 may be constituted by, for example, a programmable logic controller (PLC). The controller 20 detects the arrival of the object 100 by the output from the photoelectric sensor 10, and further controls the robot 40 in accordance with the determined state of the object 100.

The computer 30 sets the photoelectric sensor 10, the controller 20, and the robot 40. In addition, the computer 30 obtains the execution result of the control performed by the controller 20 from the controller 20. Furthermore, the computer 30 may include a learning device to generate a determination model for determining the state of the object 100 by the photoelectric sensor 10 through machine learning. Here, the decision model may be composed of, for example, a neural network or a decision tree.

The robot 40 operates or processes the object 100 in accordance with the control of the controller 20. The robot 40 can pick up the object 100 and move it to another place, or can cut or assemble the object 100, for example. Also, the robot 40 can be based on Whether or not the object 100 has protrusions changes the processing content or the moving destination.

The conveying device 50 is a device that conveys the object 100 under the control of the controller 20. The conveying device 50 may be, for example, a belt conveyor, and the object 100 may be conveyed at a speed set by the controller 20.

FIG. 2 is a diagram showing the structure of the photoelectric sensor 10 of the present embodiment. The photoelectric sensor 10 includes a light projecting unit 11, a light receiving unit 12, a processing unit 13, an operation unit 14, and an output unit 15.

<Projection Department>

The light projecting unit 11 emits light to the detection range 10a where the object 100 comes. The light projecting unit 11 may include a light projecting element 11a and a driving circuit 11b. The light projecting element 11a may be composed of an LED (Light Emitting Diode) or a laser diode, and the driving circuit 11b controls the current for making the light projecting element 11a emit light. The drive circuit 11b can cause the light projecting element 11a to emit light intermittently, for example, in a pulse of 0.1 ms. The light emitted from the light projecting element 11a can be irradiated toward the detection range 10a through a lens or fiber (not shown).

<Light receiving part>

The light receiving unit 12 acquires a time-series signal value based on light reception. The light receiving unit 12 may include a light receiving element 12a, an amplifier 12b, a sample/hold circuit 12c, and an analog/digital (A/D) converter 12d. The light receiving element 12a can be composed of a photodiode, which converts the amount of received light into an electrical output signal. The light receiving unit 12 can make the light reflected or transmitted in the detection range 10 a enter the light receiving element 12 a through a lens or an optical fiber not shown. Amplifier 12b will receive the light element The output signal of 12a is amplified. The sample/hold circuit 12c holds the output signal of the light-receiving element 12a amplified by the amplifier 12b in synchronization with the timing of the pulse light emission from the light-emitting unit 11. This reduces the influence of interference light. The A/D converter 12d converts the analog signal value held by the sample/hold circuit 12c into a value of the received light amount as a digital value.

<Processing Department>

The processing unit 13 includes an action control unit 13a, a first in first out (FIFO) memory 13b, a prediction value storage unit 13c, a determination unit 13d, and a prediction unit 13e. The processing unit 13 may be configured as, for example, a computer including a microprocessor, a memory, a program stored in the memory, and the like.

In addition to the processing of the operation prediction model described later, the operation control unit 13a also controls the overall operation of the photoelectric sensor 10 as a whole.

The FIFO memory 13b stores a predetermined number of signal values according to the acquired order, and periodically updates the predetermined number of signal values with the newly acquired signal values. Here, the number of signal values stored in the FIFO memory 13b, that is, the predetermined number, is arbitrary, and may be about 100, for example. In addition to being implemented by dedicated hardware, the FIFO memory 13b can also be implemented on the memory of the processing unit 13 in accordance with the program of the processing unit 13. At this time, the shift of the signal value toward the latter stage of the FIFO memory 13b is not a physical shift of the stored data, but can be performed by updating the access location on the memory.

The prediction value storage unit 13c stores the prediction value predicted by the prediction unit 13e at least temporarily. Here, the predicted value can be used for judgment immediately after being predicted by the prediction unit 13e. The determination of the determination section 13d may be used for the determination of the determination section 13d after the FIFO memory 13b is updated one or more times after being predicted by the prediction section 13e.

The prediction unit 13e predicts the signal value obtained in the second time range after the first time range based on the signal value obtained in the first time range by using the prediction model including the predetermined parameters. Here, the first time range and the second time range do not refer to the range of a fixed time (for example, 0:0:00 to 0:0:01). The first time range is a time range with a prescribed duration, regardless of time. The second time range may be a time range that has a predetermined time relationship with the first time range (a time range immediately after the end of the first time range, etc.) and lasts for a predetermined time. The number of signal values in the second time range predicted by the prediction unit 13e may be single or multiple.

The prediction unit 13e may predict the signal value obtained in the second time range after the first time range based on the signal value obtained in the first time range by using a prediction model generated by machine learning. The prediction model can be composed of learned neural networks or decision trees. When the predictive model is a learned neural network, the prescribed parameters included in the predictive model may include weighted parameters or bias values between nodes. With the learned model, based on the signal value obtained in the first time range, and predict the signal value obtained in the second time range after the first time range, the prediction accuracy of the signal value can be improved, which can be more It can flexibly determine the state of the object 100 that is successively conveyed on the conveying line.

The judging unit 13d performs one or more updates of the FIFO memory 13b at a frequency based on the prediction unit 13e based on the first time range The degree of agreement between the predicted signal value corresponding to the signal value stored in the FIFO memory 13b and the signal value stored in the FIFO memory 13b corresponding to the second time range is used to determine the state of the object. For example, when the object 100 is an object with a protrusion on the base, the predicting unit 13e predicts the signal of the protrusion obtained in the second time range based on the signal value of the base portion obtained in the first time range value. Then, when the degree of coincidence between the signal value actually acquired in the second time range and the predicted value is sufficiently high, the determination unit 13d can determine that the object 100 is in a state where the base is convex. In this way, the frequency is performed every time the FIFO memory 13b is updated one or more times, based on the value predicted by the predicting unit 13e based on the signal value stored in the FIFO memory 13b corresponding to the first time range, The degree of coincidence of the signal value stored in the FIFO memory 13b corresponding to the second time range that is later than the first time range is used to determine the state of the object 100, thereby making it possible to use a simple structure to determine with less time delay The state of the object 100 being conveyed one after another on the conveying line. Thereby, it is possible to utilize a simple structure similar to a popular photoelectric sensor, that is, without image processing or another trigger mechanism, and to determine the state of the object 100 with little time delay.

More specifically, when the degree of coincidence is higher than a predetermined value, the determination unit 13d can determine that the state of the object is a specific state.

In addition, the learning data used to generate the learned model may include the signal value obtained in the first time range for one object and the signal value obtained in the second time range. However, when multiple specific objects are mixed and transported, Including the signal values obtained in the first time range and the signal values obtained in the second time range for a variety of objects. When multiple objects are mixed and transported, the learned model can be based on the first The waveform composed of the signal value obtained in the time range and the predicted value of the signal value obtained in the second time range predicted, and the degree of agreement with the signal values obtained in the second time range corresponding to multiple objects, and judged to be transported The state of the object.

For example, when the first object and the second object are mixed, the learned model predicts the signal value obtained in the second time range based on the signal value obtained in the first time range. When the predicted signal value and the sum When the degree of coincidence of the signal values corresponding to the first-type object is higher than the predetermined value, the determination unit 13d can determine that the state of the conveyed object is a specific state of the first-type object. In addition, the learned model predicts the signal value obtained in the second time range based on the signal value obtained in the first time range, when the predicted signal value and the signal value corresponding to the second object are more than the specified value In this case, the determination unit 13d can determine that the state of the object being conveyed is a specific state of the second type of object.

<Operation part>

The operating unit 14 is used to operate the photoelectric sensor 10, and may include operating switches, displays, and the like. The operator of the photoelectric sensor 10 can use the operation unit 14 to input instructions such as setting the operation mode of the photoelectric sensor 10 or confirm the operation state. Furthermore, the photoelectric sensor 10 of this embodiment may include a learning mode for generating a prediction model and a determination mode as the operation mode. The determination mode is used for determining the object using the generated prediction model. 100 status.

<Output section>

The output unit 15 outputs various data including the result of the judgment made by the judgment unit 13d. In the simplest terms, the output unit 15 can make the judgment result made by the judgment unit 13d. The binary output of the fruit. Furthermore, the photoelectric sensor 10 may include a communication unit instead of the output unit 15 to input and output a large amount of data.

FIG. 3 is a diagram showing an example of the configuration of the processing unit 13 of the photoelectric sensor 10 of this embodiment. The processing unit 13 shifts the signal value of each stage stored in the FIFO memory 13b to the later stage in the first cycle, and stores the digital value of the received light output from the A/D converter 12d in the initial stage q0. Furthermore, in this figure, in order to explain the principle, the number of segments of the FIFO memory 13b is set to 10 segments of q0~q9, but the number of segments of the FIFO memory 13b can be more, for example, it can be about 100 segments.

The first period for updating the FIFO memory 13b may be the same as or different from the period of the pulse light emitted by the light projecting unit 11. In addition, the first cycle for updating the FIFO memory 13b may be the same as the cycle (set as the second cycle) of the pulsed emission of the light projector 11 and the conversion by the A/D converter 12d, or it may be different. . For example, the second period may be fixed to a value inherent to the photoelectric sensor 10 (for example, 0.1 ms). The first cycle can be set from the computer 30 shown in FIG. 1 via the controller 20. The first cycle needs to be determined in such a way that the range of the signal value waveform to be processed at the same time falls into the FIFO memory 13b. The first period is mostly longer than the second period, for example, it may be 1 ms.

The prediction unit 13e predicts the second time range based on the signal value s0, signal value s1, signal value s2, and signal value s3 of q6, q7, q8, and q9 stored in the FIFO memory 13b corresponding to the first time range Obtained signal value r0, signal value r1, signal value r2, and signal value r3. The judging unit 13d performs one or more updates of the FIFO memory 13b at a frequency based on the prediction by the predicting unit 13e The value r0, the value r1, the value r2, and the value r3, and the coincidence degree of the signal values of q2, q3, q4, and q5 stored in the FIFO memory 13b corresponding to the second time range, and determine the object 100 State, and output the determination result to the action control unit 13a in the first cycle.

The predicted value storage unit 13c stores the value r0, the value r1, the value r2, and the value r3 predicted by the prediction unit 13e. The determining unit 13d calculates the sum of the absolute values of the difference between the predicted value r0, the predicted value r1, the predicted value r2, and the predicted value r3, and the signal values of q2, q3, q4, and q5 stored in the FIFO memory 13b, The degree of coincidence is calculated such that the smaller the value, the greater the degree of coincidence. When the degree of coincidence is greater than a predetermined value, it can be determined that the state of the object 100 is a specific state.

The action control unit 13a generates a prediction model based on the signal value before executing the judgment performed by the judgment unit 13d. For example, the motion control unit 13a may perform machine learning of a learning model based on the acquired signal value, generate a learned model, and install the generated learned model in the prediction unit 13e. In this way, since the photoelectric sensor 10 can generate a prediction model by itself, there is no need to obtain a prediction model from outside, and a prediction model generated corresponding to an actual object can be used.

The action control unit 13a can output the prediction model to the outside. As a result, since the generated prediction model can be used in other photoelectric sensors, there is no need to repeatedly generate a prediction model for each of a plurality of photoelectric sensors used under the same object and installation conditions. Therefore, it is possible to efficiently prepare a photoelectric sensor for determining the state of the object.

The action control unit 13a can have relatively small changes in the signal value following the time series After the stable period of the time series, when the signal value of the time series shows a relatively large change period, a prediction model is generated based on the signal value belonging to the change period. Here, the stable period with relatively small fluctuations in the signal value of the time series can be substantially a period during which the signal value of the time series does not fluctuate when the influence of noise is removed. In addition, the fluctuation period during which the signal value of the time series has a relatively large fluctuation can be substantially a period during which the signal value of the time series changes when the influence of noise is removed. In this way, the signal value generated by the object can be selectively used from the signal value to generate a prediction model. Furthermore, specific examples of a stable period in which the change in the signal value of the time series is relatively small, and a specific example of the change period in which the change in the signal value of the time series is relatively large will be explained using FIGS. 5a and 5b.

FIG. 4 is a flowchart of processing in the learning mode and the determination mode of the photoelectric sensor 10 according to this embodiment. First, the photoelectric sensor 10 determines whether it is a learning mode for generating a predictive model (S10). Furthermore, the operation unit 14 can be used to switch between the learning mode and the judgment mode.

When the photoelectric sensor 10 is in the learning mode (S10: YES), the photoelectric sensor 10 obtains the signal value of the time series, and generates a prediction model (S11). The prediction model can be generated by the action control unit 13a using any machine learning method.

On the other hand, when the photoelectric sensor 10 is not in the learning mode (S10: NO), that is, when the photoelectric sensor 10 is in the determination mode, the photoelectric sensor 10 updates the FIFO memory 13b with the new signal value (S12), applying a prediction model to the signal value obtained in the first time range, thereby predicting the signal value in the second time range (S13). Then, based on the measured value of the signal value acquired in the second time range, and the predicted The degree of coincidence of the signal values determines the state of the object 100 (S14).

After that, the photoelectric sensor 10 determines whether to end the determination mode (S15). The end of the determination mode can be generated when the photoelectric sensor 10 is completed or when the self-determination mode is switched to the learning mode. When the determination mode is not ended (S15: No), the photoelectric sensor 10 acquires a new signal value again (S12), and performs signal value prediction (S13) and determination of the state of the object 100 (S14). On the other hand, when the judgment mode is ended (S15: Yes), the processing of the learning mode and the judgment mode is ended.

Fig. 5a is a diagram showing an example of signal values measured in the n-th cycle of the photoelectric sensor 10 of the present embodiment. 5b is a diagram showing an example of the signal value measured in the n+1 cycle of the photoelectric sensor 10 of this embodiment. In FIGS. 5a and 5b, the vertical axis represents the value of the received light amount, and the horizontal axis represents time and the level of the FIFO memory 13b corresponding to the time. As shown in the two figures, the value of the latest received light (value at time t9) is stored in the first section q0 of the FIFO memory 13b, and the value of the first received light (value at time t0) is stored in the last section of the FIFO memory 13b q9. In this example, the FIFO memory 13b stores 10 signal values according to the acquired order.

The shape S1 of the object indicated by the dotted line schematically represents the shape of the object in accordance with the timing of obtaining the value of each received light amount. According to the shape S1 of the object, it can be understood that the object has a convex shape on the base.

The waveform W1 shown by the solid line in FIG. 5a is a waveform composed of the signal value obtained in the nth cycle and stored in the FIFO memory 13b. In addition, the waveform W2 shown by the solid line in FIG. 5b is a waveform composed of the signal value acquired in the n+1 cycle and stored in the FIFO memory 13b. As shown in the two figures, it is stored in the FIFO in the nth cycle The signal value of the memory 13b is shifted to a later stage in the n+1 cycle and stored in the FIFO memory 13b.

Since there is a certain width in the detection range 10a, the reflected light from the upper surface and the lower surface of the step difference is received near the timing corresponding to the step difference of the object 100 to form the signals of the waveform W1 and the waveform W2 The value becomes an intermediate value. The value of the intermediate amount of received light tends to vary greatly depending on the small acquisition timing. Therefore, even for the object 100 of the same shape, the value of the received light amount changes every time. When generating the judgment model, it is preferable to transport the object 100 a certain number of times, and obtain the value of the received light amount repeatedly to obtain an averaging effect.

The action control unit 13a may generate a prediction model based on the signal value belonging to the fluctuation period when a fluctuation period in which the time-series signal value fluctuates relatively large appears after a stable period with relatively small fluctuations in the time-series signal value. In the case of the example of FIG. 5a, the stable period with relatively small changes in the signal value of the time series is from time t0 to t1, and the fluctuation period with relatively large changes in the signal value of the time series is from time t2 to t8. Furthermore, in the case of the example of FIG. 5b, the stable period in which the signal value of the time series has relatively small fluctuation is after time t8, and the fluctuation period in which the signal value of the time series has relatively large fluctuation is from time t2 to t7. The action control unit 13a compares the values stored in adjacent stages from the last stage to the initial stage of the FIFO memory 13b, and when there is an adjacent stage whose difference is greater than the threshold value, it can be determined as being from the adjacent stage The signal value belonging to the variable period is stored in the initial stage near the initial stage side, and it can also be determined that the signal value belonging to the stable period is stored in the last stage of the adjacent stage near the last stage. Specifically, in the case of the example in FIG. 5a, the action control unit 13a can compare the values stored in the last stage q9 and the eighth stage q8, due to its The difference is 0 and is below the threshold. Therefore, comparing the values stored in the eighth segment q8 and the seventh q7, it can be determined that the difference is 2 and above the threshold. Here, the threshold value may be 1, for example. Then, the action control unit 13a can determine that the seventh stage q7 near the first stage q0 side of the eighth stage q8 and the seventh stage q7 where the difference between the stored values is greater than the threshold value is stored in the first stage q0 belonging to the change period The signal value can also be determined as storing the signal value belonging to the stable period from the eighth segment q8 on the last segment side of the eighth segment q8 and the seventh segment q7 to the last segment q9.

The prediction unit 13e predicts the signal value obtained in the second time range after the first time range based on the signal value obtained in the first time range by using the prediction model including the predetermined parameters. In the case of the example shown in FIG. 5a, the first time range is t0~t3, and the second time range is t4~t7. The prediction unit 13e predicts the signal values obtained in the second time range t4~t7 based on the signal values of q9~q6 obtained in the first time range t0~t3 and stored in the FIFO memory 13b. When the object 100 carried during the operation in the judgment mode is a convex object, in the specific shift cycle (nth cycle) of the FIFO memory 13b, (t0 to t3) The degree of agreement between the waveform formed by the value of the received light intensity acquired and the waveform formed by the value of the received light intensity at the time of generating the prediction model becomes higher. If so, it is possible to predict the value of the received light amount in the second time range (t4~t7) with reduced errors. Then, the determination unit 13d performs one or more updates of the FIFO memory at a frequency based on the value predicted by the prediction unit 13e and stores it in the FIFO memory corresponding to the second time range (t4~t7). The degree of coincidence of the signal value of the body 13b (the signal value stored in q5~q2) is used to determine the state of the object. The error can be reduced by the prediction unit 13e When predicting the value of the received light amount in the second time range (t4 to t7), the degree of coincidence becomes high, and the determination unit 13d determines that the object 100 is an object having protrusions.

On the other hand, in the case of the n+1th cycle shown in FIG. 5b, the first time range is t1 to t4, and the second time range is t5 to t8. At this time, the signal value input to the prediction unit 13e is also the value of the received light amount stored in the q9~q6 segment of the FIFO memory 13b. At this time, since the waveform of the value of the received light input input to the prediction unit 13e and the waveform of the value of the received light when the prediction model was generated are time-shifted, accurate prediction cannot be performed, and the value predicted by the prediction unit 13e is The coincidence degree of the signal value stored in the FIFO memory 13b (the signal value stored in q5~q2) corresponding to the second time range becomes smaller. Therefore, in the n+1th cycle, the determining unit 13d does not determine the state of the object 100 as the state of the object having protrusions.

FIG. 6 is a diagram showing another example of the signal value measured in the n-th cycle of the photoelectric sensor 10 of the present embodiment. In this figure, the vertical axis represents the value of the received light amount, and the horizontal axis represents time and the level of the FIFO memory 13b corresponding to the time.

The waveform W3 shown by the solid line in FIG. 6 is a waveform composed of the signal value acquired in the nth cycle and stored in the FIFO memory 13b. In addition, the shape S2 of the object indicated by the dotted line schematically represents the shape of the object in accordance with the timing of obtaining the value of each received light amount. According to the shape S2 of the object, it can be understood that the object has a shape without protrusions but only a base.

Comparing the waveform W3 with the waveform W1 shown in FIG. 5a, it can be seen that there is a difference in the value of the received light amount during the period from time t4 to t6. When the object being transported during operation in the judgment mode is a non-protruding object, if the object is recorded in the FIFO The value of the amount of light received in the first time range (t0~t3) under a certain shift cycle of the memory 13b becomes as shown in Fig. 6, because the value is substantially the same as the value of the amount of light received when the prediction model is generated. , So in the second time range (t4~t7), the predicted value P can be obtained as shown by a dotted chain line. Comparing the predicted value P with the actual measured values of q5 to q2 stored in the FIFO memory 13b shows that the difference is large and the degree of coincidence is small, so the determination unit 13d can determine that the object is not an object with protrusions.

FIG. 7 is a diagram showing another example of the configuration of the processing unit 13 of the photoelectric sensor 10 of the present embodiment. The example of the configuration of the processing unit 13 shown in this figure is different from the example of the configuration of the processing unit 13 shown in FIG. 3 in the following points: the level of the FIFO memory 13b corresponding to the first time range, The level of the FIFO memory 13b corresponding to the second time range overlaps, and the action control unit 13a outputs the time-series light-receiving value to the outside, and generates it by the external computer based on the time-series light-receiving value The input of the predictive model is common to other components.

In this example, every time the FIFO memory 13b is updated one or more times, the prediction unit 13e updates the prediction value stored in the prediction value storage unit 13c. The predicted value stored in the predicted value storage unit 13c is obtained based on the value of the received light amount stored in the q9~q6 segment of the FIFO memory 13b before the 4 shift cycles as the time difference between the first time range and the second time range. Predicted value. Since the received light quantity values of q0 to q3 stored in the FIFO memory 13b during the four shift cycles are replaced with the values of the second time range, the determination can be made accurately. With this configuration, the number of required stages of the FIFO memory 13b is reduced. On the contrary, in the prediction unit 13e, a FIFO memory needs to be provided for the time from solving the prediction value to updating the prediction value by the prediction value The prediction value is stored in the storage unit 13c in advance. However, because the FIFO memory 13b for the value of the received light amount may be used for prediction or determination every multiple stages, the number of stages required in this case is reduced The effect is great. In this way, using the part close to the beginning of the FIFO memory 13b for prediction, the time delay becomes smaller.

The action control unit 13a may also be able to output the time-series signal value to the outside. The signal value output to the outside may also be the signal value stored in the FIFO memory 13b. The signal value can be output to the outside, and the external equipment generates a predictive model. In this way, the photoelectric sensor itself does not need to have computing resources related to the process of generating the predictive model.

The action control unit 13a can obtain a prediction model from the outside. The action control unit 13a may obtain a prediction model generated by an external computer, or may obtain a prediction model generated by another photoelectric sensor. The generation of the prediction model can be omitted by using the prediction model generated by other devices, such as other photoelectric sensors.

Furthermore, the action control unit 13a may not need to output the time-series signal value to the outside, but may generate a model in another photoelectric sensor, or an external computer based on the time-series signal obtained by another photoelectric sensor. Value and use it for the generated model input.

The embodiment described above is an embodiment for easy understanding of the present invention, and is not an embodiment explained to limit the present invention. The various elements included in the embodiment and their arrangement, materials, conditions, shapes, dimensions, etc. are not limited to the exemplified contents, but can be changed as appropriate. In addition, it is possible to partially replace or combine the configurations shown in different embodiments.

[Supplement]

A photoelectric sensor (10) is provided with: a light projecting unit (11) that emits light to a detection range (10a) where an object (100) comes; a light receiving unit (12) that obtains light-receiving light based on the light Time series signal value; FIFO memory (13b) stores a predetermined number of the signal values according to the acquired order, and periodically updates the predetermined number of the signal values with the newly acquired signal value ; The prediction unit (13e), based on the signal value obtained in the first time range, by a prediction model that includes prescribed parameters, and predicts all the values obtained in a second time range that is later than the first time range The signal value; and the determination unit (13d), at a frequency that is performed every time the FIFO memory (13b) is updated one or more times, based on the value predicted by the prediction unit (13e), and The second time range corresponds to the degree of coincidence of the signal values stored in the FIFO memory (13b) to determine the state of the object (100).

1: Detection system

10: Photoelectric sensor

10a: detection range

20: Controller

30: Computer

40: Robot

50: Conveying device

100: Object

Claims (8)

A photoelectric sensor includes: a light projecting part that emits light to a detection range for an object to come; a light receiving part that acquires a signal value based on the time series of the light received by the light; a first-in first-out memory, based on the acquired The predetermined number of the signal values are stored in the order of, and the predetermined number of the signal values are periodically updated by the newly acquired signal values; the predicting part is based on the prediction model containing the predetermined parameters based on the The signal value acquired in the first time range and stored in the first-in-first-out memory is predicted, and the signal value acquired in a second time range after the first time range is predicted; and a determination unit to The frequency that is performed every time the first-in-first-out memory is updated one or more times is based on the signal value predicted by the prediction unit and stored in the first-in-first-out corresponding to the second time range The degree of coincidence of the signal values of the memory is used to determine the state of the object. The photoelectric sensor according to the first item of the patent application, wherein when the degree of coincidence is higher than a predetermined value, the determination unit determines that the state of the object is a specific state. The photoelectric sensor as described in item 1 or item 2 of the scope of patent application, wherein the prediction model is a model generated by machine learning. The photoelectric sensor as described in claim 1 further includes an action control unit that generates the prediction model based on the signal value. The photoelectric sensor according to claim 4, wherein the action control unit displays a relatively large change in the signal value in the time series after a stable period after the relatively small change in the signal value in the time series During the period of change, the prediction model is generated based on the signal value belonging to the period of change. The photoelectric sensor according to claim 4 or 5, wherein the operation control unit can output the prediction model to the outside. In the photoelectric sensor according to the 4th or 5th item of the scope of patent application, the action control unit can output the signal value in the time series to the outside. The photoelectric sensor described in item 1 of the scope of patent application further includes an action control unit that obtains the prediction model from the outside.

Images