Method for automatically regulating a container transport device having one or more transport belts for adapting a container density, and container transport device

Abstract

The invention relates to a method for automatically regulating a container transport device having one or more transport belts (17-18) for adapting a container density of containers which are transported on the one or more transport belts. The method comprises, in each case at different times, detection of distance data relating to containers on the one or more transport belts by means of a distance measurement device, transmission of the distance data to a control device by the distance measurement device, processing of the distance data to give control data in the control device, and non-linear regulation of belt speeds of the one or more transport belts by means of the control data. The invention further relates to a container transport device comprising one or more transport belts and to a control device, wherein the container transport device is designed to carry out a method.

Description

[0001] Method for automatically controlling a container transport device with one or more conveyor belts for adjusting a container density and container transport device

[0002] The invention relates to methods for automatically controlling a container transport device with one or more conveyor belts for adjusting a container density of containers according to claim 1 and to a container transport device according to claim 15.

[0003] State of the art

[0004] DE 195 30 626 A1 discloses a method and a device for detecting the occupancy of a conveyor for containers with a storage and / or deflection area, wherein the occupancy is detected in a contactless manner by a distance measurement, which is carried out essentially parallel to the conveying plane, between a reference location upstream of the storage and a rearmost container of the stored container flow.

[0005] Generally, container density control is currently either insufficient or not implemented during container transport on a conveyor belt. This can result in high pressures on the containers, which can lead to glass breakage or scuffing (abrasion, scratches), which is undesirable when transporting sensitive plastic containers (e.g., PET) or other thin-walled or delicate containers, such as aluminum or tinplate cans. Colliding containers can also generate noise and can lead to deviations from the desired distribution of containers on the conveyor belt.

[0006] Task

[0007] The object of the invention is to provide a method and a container transport device that enable gentle and quiet transport of containers on conveyor belts.

[0008] Solution

[0009] The object is achieved by the method according to claim 1 and the container transport device according to claim 15. Preferred embodiments are disclosed in the subclaims.

[0010] A method for automatically controlling a container transport device with one or more conveyor belts for adjusting a container density of containers that are transported on the one or more conveyor belts comprises, at different times, detecting distance data of containers on the one or more conveyor belts by means of a distance measuring device, transmitting the distance data by the distance measuring device to a control device, processing the distance data into control data in the control device and non-linearly controlling belt speeds of the one or more conveyor belts by means of the control data.

[0011] The distance measuring device can comprise various types of sensors designed to detect the position of objects, such as containers. This can be done optically (light barrier) and / or mechanically (jam switch), using a sound sensor, a camera, Hall sensors, and / or proximity sensors.

[0012] The distance data may include a distance from one container to another container, or the distances from one container to multiple other containers, and / or the distances from multiple containers to one container, and / or the distances from multiple containers to multiple other containers, and / or a distance from one container to a railing, and / or the distances from multiple containers to a railing. Alternatively, the distance data may include a number of containers present in a defined area and / or a distribution of the containers within a defined area.

[0013] Each conveyor belt or belts can be assigned an identity.

[0014] A linear or non-linear control can be used for a control in which an output signal is not always proportional to the input signal.

[0015] Here, nonlinear control is used to regulate the speeds of one or more conveyor belts. When multiple conveyor belts are present, each conveyor belt can take the speed and direction of the other conveyor belts into account when regulating its own speed. Furthermore, the speeds can be controlled in such a way that they can respond to a changing number of containers being fed in, allowing the container density on the multiple conveyor belts to be adjusted at any time.

[0016] The distance data acquisition using the distance measuring device can include infrared, sound, sonar, and / or laser, lidar, and / or a mechanical distance measurement, such as one or more jam sensors. It is also possible to determine the distances using an image or a video stream, whereby the distance between two or more containers detected in the image is related to the known container diameter.

[0017] Distance data can be collected continuously or intermittently. For example, the distance between one container and at least one neighboring container can be collected, or the distances between multiple containers can be collected. The distance data can include the information specified in the list above.

[0018] The nonlinear control may include adjusting a speed gradient across the multiple conveyor belts. For example, the speed gradient may increase during a jam in the container transport device. For example, the speed gradient may decrease at distances that exceed a given distance value.

[0019] For conveyor belts moving in a first direction, a first speed gradient can be set, and for conveyor belts moving in a second direction, a second speed gradient can be set. The first and second speed gradients can be the same or different.

[0020] A gradient can describe the change (decline or rise) of a quantity over a specific distance. The quantity can be viewed as the speed of the conveyor belts or the speed of the drive motors, and the distance can be viewed as a first, second, third, etc. conveyor belt. The first conveyor belt can be arranged adjacent to the second, the second adjacent to the third, etc., or the first, second, third, etc. conveyor belts can be connected in series.

[0021] The non-linear control may further comprise adjusting electrically adjustable railings and / or deflector plates and / or aisle plates of the one or more conveyor belts and / or adjusting an inclination of the conveyor belt(s) relative to a horizontal plane. For example, in the case of electrically adjustable railings and / or deflector plates and / or aisle plates divided into sections, the adjustment may comprise adjusting the railings and / or deflector plates and / or aisle plates in sections.

[0022] By adjusting the position, the available space for containers between the electrically adjustable railings and / or deflector plates and / or aisle plates of the conveyor belt(s) can be increased or decreased. Adjusting the position, which requires increasing the available space, can create an additional buffer during ongoing operation of the container transport device.

[0023] The non-linear control may further comprise controlling a plurality of points of one or more elastically designed deflector plates of the one or more conveyor belts in order to adapt a contour of the one or more elastically designed deflector plates.

[0024] This control allows for gentle transport and / or redirection of the containers. The method can further comprise recording current occupancy data regarding the occupancy of one or more conveyor belts with containers using one or more position measuring devices. For example, the recording can comprise recording images with one or more cameras.

[0025] The positioning devices can, for example, record the position data of one or more containers. Based on the number of position data points, the loading of one or more conveyor belts with containers can be determined.

[0026] The position measuring devices may also or alternatively acquire position data, which may include, for example, distance data as specified in the list above.

[0027] The method may further comprise transmitting the current occupancy data to a central computing unit, processing the current occupancy data with an algorithm available on the central computing unit to obtain processed occupancy data, and using the processed occupancy data to control drives of the one or more conveyor belts. Instead of or in addition to the central computing unit, transmitting the current occupancy data to a cloud, a smart device, PLC (Programmable Logic Controller), and / or IPC (InterProcess Communication).Alternatively or additionally, the current occupancy data can be recorded using a smart device, which for example includes a smart camera, and can also be processed into processed occupancy data using an algorithm available on the smart device, and this processed occupancy data can be used to control drives of the one or more conveyor belts.

[0028] For example, processing can be done continuously.

[0029] For example, the processed occupancy data can be compared with a specified occupancy (target occupancy) of the conveyor belt(s) with containers. If the specified occupancy is exceeded, the control system can, for example, require an increase in the drive speed. If the specified occupancy is not met, the control system can, for example, require a reduction or increase in the drive speed.

[0030] The method may further comprise processing the current occupancy data with a further algorithm available on the central computer unit and calculating therefrom a speed for the one or more conveyor belts with the further algorithm.

[0031] Instead of or in addition to the additional algorithm available on the central computer unit, the current occupancy data can be processed using an additional algorithm available on a cloud, on a smart device, on a PLC (Programmable Logic Controller) and / or on an IPC (InterProcess Communication).

[0032] For example, the method may further comprise calculating an adjustment of electrically adjustable railings and / or deflector plates and / or aisle plates of the one or more conveyor belts and / or adjusting an inclination of the conveyor belt(s) relative to a horizontal plane. For example, in the case of electrically adjustable railings and / or deflector plates and / or aisle plates divided into sections, the adjustment may comprise section-by-section adjustment.

[0033] The method may additionally or alternatively comprise predicting and / or learning a strategy for adjusting electrically adjustable railings and / or deflector plates and / or aisle plates of the one or more conveyor belts and / or adjusting an inclination of the conveyor belt(s) relative to a horizontal plane. For example, predicting and / or learning may comprise reaching a predetermined occupancy level of the conveyor belt or reaching predetermined occupancy levels of the conveyor belts.

[0034] If a specified occupancy is exceeded, it can be provided that the space between the electrically adjustable railings and / or deflector plates and / or aisle plates of the conveyor belt(s) can be increased by adjusting them. If the specified occupancy is not met, it can be provided that the space between the electrically adjustable railings and / or deflector plates and / or aisle plates of the conveyor belt(s) can be reduced by adjusting them.

[0035] The additional algorithm can include a neural network or model-based online optimization. With model-based online optimization, the model can learn the behavior of the container transport device in a training program. Using online optimization, the container transport device can be controlled during operation if a deviation from a target behavior occurs (model predictive control).

[0036] For example, the neural network may include an observation layer, which may be followed by a fully connected layer. This may be followed by four ReLU activation functions, followed by fully connected layers. The fifth fully connected layer may be followed by a tanh activation function. The tanh activation function may be followed by a scaling layer. The output of the neural network may be input to a function for calculating a mean square error. An output node of the neural network may include a tanh activation function. The tanh activation function can map suggested actions to the range [-1, 1].

[0037] This can be useful when a scaling layer follows the tanh activation function. For example, it may be easier to map the neural network output to a defined range [min, max] of each belt speed of the conveyor belt(s).

[0038] For example, the speed of a second conveyor belt can have a defined range of [0.8 0.6]m / s and the speed of a third belt can have a defined range of [0.7, 0.5], so that the output of the Tanh activation function maps the value to the range [-1 , 1] and in a subsequent ScalingLayer each belt speed action can be adapted to the respective defined range.

[0039] The neural network can comprise at least one hidden layer, and each of the at least one hidden layer can comprise a ReLU activation function. Of the known activation functions, the ReLU activation function, with its piecewise linear function, is a suitable choice. For example, it can output an input directly if the input is positive and output zero if the input is non-positive.

[0040] An output of the neural network may comprise an array. For example, the array may comprise a respective mapping of the identity of the conveyor belt to its belt speed, or an identity of one of the multiple conveyor belts to its belt speed.

[0041] The neural network may include at least one fully connected layer, which allows the neural network to learn features from all combinations of features of the previous layer.

[0042] Calculations of neurons of the neural network may depend on the outputs of the previous layers, which may have been multiplied by weights and assigned an offset.

[0043] To train a neural network, training data can be generated by an agent through trial and error. At the beginning of training, the agent can try random actions and evaluate the outcome (reward function). It can then adjust these actions until it finds the optimal action. The action can also be added to a noise function to allow the agent to explore uncontested states. The agent can store these states in a fixed-size buffer. For each iteration, the neural network's weights can be updated with a randomly selected mini-batch from the buffer, allowing the data to be distributed independently (this happens because the agent itself generated the training data).

[0044] This data, which can be the input to the neural network, can be uniform areas covering part of a conveyor belt. It can calculate the occupancy, for example, an occupancy rate, of each of these areas and determine whether the containers are evenly distributed. If the containers are not evenly distributed, the agent can adjust the action every n seconds to ensure that the containers are evenly distributed.

[0045] For example, the number of uniform areas can be 5, 11, or 22, or less or more. The data can also be an image or video from the camera showing the occupancy of the areas with containers.

[0046] A container transport device comprises one or more conveyor belts and a control device, wherein the container transport device is designed to carry out a method as described above or below.

[0047] Short character description

[0048] The attached figures represent aspects and / or embodiments of the invention by way of example for better understanding and illustration. It shows:

[0049] Figure 1 schematically shows the layers of a neural network,

[0050] Figure 2 schematically shows an oblique view of several conveyor belts and a subsequent mass transport with a first number of uniform areas and

[0051] Figure 3 schematically shows an oblique view of the plurality of conveyor belts and the subsequent mass transport with a second number of uniform areas.

[0052] Detailed character description

[0053] Figure 1 shows a schematic representation of the layers of a neural network 14 such as can be used for the method and container transport device described above or below. A different or similar neural network may also be used.

[0054] The observation layer 1 is followed by a first fully connected layer 2 so that the neural network 14 can learn the features from all combinations of features of the observation layer 1.

[0055] The first fully connected layer 2 is followed by a first ReLU activation function 3. For example, it can output an input directly if the input is positive and output zero if the input is non-positive. The results of the first ReLU activation function 3 are passed to a second fully connected layer 4, allowing the neural network 14 to learn the features from all combinations of features from the previous layers.

[0056] The second fully connected layer 4 is followed by a second ReLU activation function 5. For example, the second ReLU activation function 5 can also output an input directly if the input is positive and output zero if the input is non-positive. The results of the second ReLU activation function 5 are passed to a third fully connected layer 6, allowing the neural network 14 to learn the features from all combinations of features from the previous layers.

[0057] The third fully connected layer 6 is followed by a third ReLU activation function 7. For example, the third ReLU activation function 7 can also output an input directly if the input is positive and output zero if the input is non-positive. The results of the third ReLU activation function 7 are passed to a fourth fully connected layer 8, allowing the neural network 14 to learn the features from all combinations of features from the previous layers.

[0058] The fourth fully connected layer 8 is followed by a fourth ReLU activation function 9. For example, the fourth ReLU activation function 9 can also output an input directly if the input is positive and output zero if the input is non-positive. The results of the fourth ReLU activation function 9 are passed to a fifth fully connected layer 10, allowing the neural network 14 to learn the features from all combinations of features from the previous layers.

[0059] The fifth fullyConnectedLayer 10 is followed by a Tanh activation function 11. Using the Tanh activation function 11, suggested actions can be mapped to the range [-1 , 1].

[0060] The Tanh activation function 11 is followed by a ScalingLayer 12. The preceding Tanh activation function 11 can make it easier to map the output of the neural network to a defined range [min, max].

[0061] The output of the neural network 14 is input into a function 13 for determining a mean square error.

[0062] Figure 2 schematically shows an oblique view of a system 16 comprising a plurality of conveyor belts 17-22, 23-27, 28 and a subsequent mass transport 31 with a first number of uniform regions 32-42. The first number is 11. A first group of conveyor belts 17-22 and a single conveyor belt 28 can be driven in a first direction and a second group of conveyor belts 23-27 in a second direction, wherein the first and second directions are oriented opposite to one another. The mass transport 31 is shown driven in a third direction, which runs perpendicular to the first and second directions. The mass transport 31 can also be driven in a fourth direction, which is opposite to the third direction. Above the conveyor belts 17-22. 23-27, 28 are provided with railings 29, 30 which are used for guiding and / or deflecting loads on the conveyor belts 17-22.23-27, 28 can serve for the containers transported. The railings 29, 30 can be electrically adjustable railings. Instead of railings or in addition thereto, deflector plates and / or aisle plates can be provided. Figure 3 shows a schematic oblique view of a system 43 whose conveyor belts 17-22, 23-27, 28 and whose subsequent mass transport 31 are provided like the conveyor belts 17-22, 23-27, 28 and mass transport 31 already described with regard to Figure 2. The second number of uniform areas 44-65 is 22 in the illustration. Because the second number is greater than the first number, a better adaptation of the container density, i.e. the occupancy, can be achieved for the second system in which the second number of uniform areas 44-65 is used.

Claims

Claims 1. A method for automatically controlling a container transport device having one or more conveyor belts (17-28) for adjusting a container density of containers transported on the one or more conveyor belts (17-28), the method comprising: at different times respectively: - recording distance data of containers on the one or more conveyor belts (17-28) by means of a distance measuring device, - Transmission of the distance data by the distance measuring device to a control device, - Processing the distance data into control data in the control device, - non-linear control of belt speeds of the one or more conveyor belts (17-28) by means of the control data.

2. The method according to claim 1, wherein the acquisition of the distance data by means of the distance measuring device comprises infrared, sound, sonar and / or laser and / or a mechanical distance measurement, such as one or more jam switches.

3. The method according to claim 1 or 2, wherein the detection of distance data is carried out continuously or in a timed manner, for example a distance of a container to at least one adjacent container is detected or the distances of several containers to one another are detected.

4. The method according to one of claims 1 to 3, wherein the non-linear control comprises: setting a speed gradient across the plurality of conveyor belts (17-28), wherein, for example, the speed gradient increases in the event of a jam in the container transport device, wherein, for example, the speed gradient decreases at distances that exceed a given distance value.

5. The method according to one of claims 1 to 4, wherein the non-linear control further comprises an adjustment of electrically adjustable railings (29, 30) and / or deflector plates and / or aisle plates of the one or more conveyor belts and / or an adjustment of an inclination of the one or more conveyor belts relative to a horizontal plane, wherein, for example, in the case of electrically adjustable railings (29, 30) and / or deflector plates and / or aisle plates divided into sections, the adjustment comprises a section-by-section adjustment.

6. The method according to one of claims 1 to 5, wherein the non-linear control further comprises controlling a plurality of points of one or more elastically designed deflector plates of the one or more conveyor belts to adapt a contour of the one or more elastically designed deflector plates.

7. The method according to any one of claims 1 to 6, further comprising: - detecting current occupancy data of an occupancy of the one or more conveyor belts (17-28) with containers by means of one or more position measuring devices, wherein the detecting comprises, for example, taking images with one or more cameras.

8. The method of claim 7, further comprising: - Transmission of current occupancy data to a central computer unit, - Processing the current occupancy data with an algorithm available on the central computer unit to process occupancy data, - Using the processed occupancy data to control the drives of one or more conveyor belts, for example, where processing takes place continuously.

9. The method of claim 8, further comprising: - Processing the current occupancy data with a further algorithm available on the central computer unit and calculating a speed for the one or more conveyor belts (17-28) using the further algorithm, and for example further calculating an adjustment of electrically adjustable railings and / or deflector plates and / or aisle plates of the one or more conveyor belts, wherein, for example, in the case of electrically adjustable railings and / or deflector plates and / or aisle plates divided into sections, the adjustment comprises section-by-section adjustment.

10. The method according to claim 9, wherein the further algorithm comprises a neural network (14) or a model-based online optimization.

11. The method of claim 10, wherein an output node of the neural network (14) comprises a tanh activation function (11).

12. The method according to claim 10 or 11, wherein the neural network (14) comprises at least one hidden layer and each of the at least one hidden layer comprises a ReLU activation function, wherein, for example, the ReLU activation function is a piecewise linear function that, for example, outputs an input directly when the input is positive and outputs zero when the input is not positive.

13. The method according to any one of claims 10 to 12, wherein an output of the neural network (14) comprises an array, wherein, for example, the array comprises a respective assignment of an identity of the conveyor belt to its belt speed or an identity of one of the plurality of conveyor belts to its belt speed.

14. The method according to any one of claims 10 to 13, wherein the neural network (14) comprises at least one fully connected layer (2, 4, 6, 8, 10), whereby the neural network (14) learns features from all combinations of features of the previous layer.

15. Container transport device comprising one or more conveyor belts (17-28) and a control device, wherein the container transport device is designed to carry out a method according to one of claims 1 to 14.

Images