Sorting system for sorting objects in a material flow according to object classes and method for sorting objects conveyed in a material flow according to object classes

Abstract

The invention relates to a sorting system (10) for sorting objects in a material stream of objects according to object classes and a method for sorting objects in a material stream of objects according to object classes. The sorting system (10) comprises: - a conveying device (20) for conveying the objects in the material stream of objects; - a detection device (40) for detecting objects in the material stream of objects; - a classification device for classifying the objects detected by the detection device (40) into at least two object classes; - a correlation analysis device for determining the value of at least one spatiotemporal correlation criterion between the objects detected by the detection device (40);and - a plurality of sorting actuators (60) for selectively sorting objects from the material stream, wherein the plurality of sorting actuators (60) comprises at least one primary sorting actuator (62) and at least one secondary sorting actuator (64) such that the at least one secondary sorting actuator (64) is arranged downstream of the primary sorting actuator (62) with respect to the material stream such that a first and a second object in the material stream of objects can pass the primary and the secondary sorting actuator (62; 64) successively in such a way that they are selectively sorted by either the primary or the secondary sorting actuator (62;64) can be sorted out of the material stream, wherein the sorting system is designed (10) to control the at least one primary sorting actuator (62) and the at least one secondary sorting actuator (64) depending on the object class of the first object, the object class of the second object and the value of a spatiotemporal correlation criterion determined by the correlation analysis device between the first object and the second object.;

Description

[0001] The present invention relates to a sorting system for sorting objects in a material stream of objects according to object classes and to a method for sorting objects conveyed in a material stream according to object classes.

[0002] Various sorting systems are known from the prior art, which perform sorting functions to separate different conveyed objects. Conventional sorting systems use pneumatic or mechanical actuators to sort out one or more specific objects. Pneumatic actuators are characterized by a shorter reaction time for sorting out a specific object compared to mechanical actuators, but they are energy inefficient, where reaction time refers to the time between actuating an actuator and its effect taking place. Furthermore, pneumatic actuators typically generate an air cone, making it difficult to rule out unintentional influence on other objects.Mechanical actuators are characterized by a longer reaction time for sorting out a specific object compared to pneumatic actuators, although they are significantly more energy-efficient. However, due to the comparatively long reaction time of a mechanical actuator, unintended influence on other objects is difficult to rule out. This unintended influence can result in bycatch, where objects are sorted out that should not have been sorted out according to a predetermined sorting logic.

[0003] WO 2019 / 023545 A2 describes systems and methods for sorting recyclable items and other materials. In one embodiment, a system for sorting objects comprises: at least one image sensor; a controller comprising a processor and memory, wherein the controller receives image data acquired by the image sensor; and at least one pusher coupled to the controller, wherein the at least one pusher is configured to receive an actuation signal from the controller. The processor is configured to detect objects moving on a conveyor and to detect at least one target object moving on a conveyor by processing the image data and determining an expected time at which the at least one target object will be within a diversion path of the pusher.The control system selectively generates the activation signal based on whether a detected object in the image data includes at least one target object.

[0004] WO 2016 / 097014 A1 describes a system for sorting waste articles, wherein the waste article is provided with means for its identification, the system comprising: at least one sorting unit, the sorting unit being connected to at least one sensor system for identifying the waste article or the means of identification, the sensor system being configured to detect one or more means of identification and being configured to provide a control system with a signal when a waste article of a desired type is detected; at least one sorting conveyor for conveying the waste articles to the sorting unit, the sorting unit being equipped with two or more sorting devices for sorting the waste articles, the sorting devices being arranged to respond to the detection of a waste article of a desired type by the sensor system.at least one signal received from the control system.

[0005] It is therefore an object of the present invention to provide an improved sorting system for sorting objects and an improved method for sorting objects, each of which in particular enables precise sorting and furthermore in particular reduces the risk of bycatch.

[0006] The problem described above is solved by the subject matter of the independent claims; preferred embodiments are the subject matter of the dependent claims.

[0007] A first aspect of the invention relates to a sorting system for sorting objects in a material stream according to object classes. The sorting system comprises a conveying device for conveying or transporting the objects in the material stream. The conveying device can, in particular, comprise a belt, chute, groove, and / or roller conveyor, wherein the material stream can also be formed, at least partially, by a section-by-section free fall of the objects.

[0008] The sorting system further comprises a detection device for detecting objects in the material flow. The detection device can, in particular, be an optical detection device and can, in particular, include an optical sensor, such as a camera, e.g., a line scan camera or, preferably, an area scan camera. Furthermore, the optical detection device can, in particular, include laser-based sensors and / or sensors based on infrared or X-ray radiation. Additionally or alternatively, the detection device can, for example, include a proximity sensor, in particular a capacitive proximity sensor, and / or a weight sensor.However, the present sensor system is not limited to the aforementioned types of detection devices, but may instead include further sensors and / or sensor combinations that enable object-specific object classification and / or position detection of an object with sufficient temporal resolution, i.e., suitable for a sorting system.

[0009] In other words, the present sorting system involves sensor-based sorting between objects to be sorted and those not to be sorted, where the sorting criterion is decoupled from the force effecting the sorting. This contrasts with, for example, the following sorting methods, where the sorting criterion is not decoupled from the force effecting the sorting: density sorting, magnetic separation, flotation, electrostatic sorting, etc., as highlighted in Wotruba, Hermann. "State of the art of sensor-based sorting". BHM Berg und Hüttenmännische Monatshefte, Vol. 153, June 2008, pp. 221-224.

[0010] The sorting system further includes a classification device for classifying the objects detected by the detection device into at least two object classes. Classification includes, in particular, assigning one of the at least two object classes to the detected objects. Classification can, in particular, comprise one or more individual steps, such as determining the object class of a detected object. Determining an object class can, in turn, include, in particular, comparing and / or assigning a detected object to one or more object templates, whereby the respective object template is predetermined and assigned to a specific object class.

[0011] The sorting system also includes a correlation analysis device for determining the value of at least one spatiotemporal correlation criterion between the objects detected by the detection device. The spatiotemporal correlation criterion is a criterion that establishes a spatial and temporal relationship between the detected objects. The value of the at least one spatiotemporal correlation criterion can be determined by detecting, in particular by optical detection, one or more positions of an object in the material flow at time intervals. Alternatively or additionally, the value of the at least one spatiotemporal correlation criterion can be determined, in particular, by calculating or estimating the velocity, acceleration, and / or trajectory of an object in the material flow.

[0012] The sorting system further comprises a plurality of sorting actuators, for example a two-dimensional arrangement of a plurality of sorting actuators, for the selective sorting of objects from the material stream, wherein the plurality of sorting actuators comprises at least one primary sorting actuator and at least one secondary sorting actuator such that the at least one secondary sorting actuator is arranged downstream of the primary sorting actuator with respect to the material stream in such a way that a first and a second object in the material stream of objects can pass the primary and the secondary sorting actuator one after the other in such a way that they can be selectively, and if necessary individually, sorted out of the material stream by the primary or the secondary sorting actuator.In other words, a primary sorting actuator and a secondary sorting actuator are arranged along the trajectory of an object, with the secondary sorting actuator being located downstream of the primary sorting actuator. In further terms, a primary sorting actuator and a secondary sorting actuator are arranged such that an object, along its trajectory in the material flow, passes both the primary and the secondary sorting actuator in that order, unless it is rejected by either the primary or the secondary sorting actuator. Passing the sorting actuator is understood as passing within the operating range of the sorting actuator, within which the sorting actuator is configured to reject an object if it is controlled to do so.

[0013] The sorting system is designed to control at least one primary sorting actuator and at least one secondary sorting actuator depending on the object class of the first object, the object class of the second object, and, if applicable, the object class of one or more further objects, and the value of a spatial-temporal correlation criterion determined by the correlation analysis device between the first object and the second object, and, if applicable, one or more further objects.In other words, the sorting system is specifically designed to control at least one primary sorting actuator and at least one secondary sorting actuator depending on the object class assigned by the classification unit to the first and second objects, as well as, if applicable, one or more further objects, and the value of a spatiotemporal correlation criterion determined by the correlation analysis unit between the first and second objects, and, if applicable, one or more further objects. For the sake of simplicity, the following text will generally refer to a first object and a second object. However, it is understood that this does not exclude or even includes the consideration of further objects, particularly with regard to the detection, classification, and determination of a value for at least one spatiotemporal correlation criterion.

[0014] By means of an object's object class, i.e., the object class assigned to an object in particular by the classification device, the sorting system can advantageously classify the object as either to be sorted or not to be sorted. In further exemplary embodiments, and especially when more than two object classes are used for classifying the objects, the sorting system can advantageously classify an object, in addition to or as an alternative to being sorted or not to be sorted, also as being to be sorted in a specific way and / or sorted by a specific sorting actuator.

[0015] By additionally controlling the sorting actuators for sorting out objects, in particular for sorting out objects to be sorted, depending on the value of a spatial-temporal correlation criterion between two objects, it is advantageously possible for a sorting actuator of the majority of sorting actuators to only sort out an object to be sorted out if, due to the value of the spatial-temporal correlation criterion, sorting out an object that should not be sorted out is unlikely or can be avoided.

[0016] This allows for the advantageous provision of a sorting system for improved sorting of objects from a material stream, in particular enabling precise sorting of objects to be sorted, which further advantageously reduces or even eliminates the risk of bycatch.

[0017] This will be illustrated by the following example according to an embodiment of the present invention. For this purpose, a first object and a second object are considered, which are conveyed in the material stream along with other objects. The first object is detected by the detection device, in particular optically, capacitively, and / or by its weight. For example, an optical pattern, a material, a material type, one or more colors, a contour, a texture, a weight, a size, and / or similar characteristics of the first object are detected. Based on the detection of the first object by the detection device, the classification device assigns a first object class to the first object. The first object class is assumed, by way of example, to be an object class that identifies the first object as being to be sorted out by the majority of sorting actuators.The second object, like the first, is also detected by the detection device. Based on this detection, particularly the optical, capacitive, and / or weight-based detection of the second object, the classification device assigns it a second object class. This second object class is assumed to be one that identifies the second object as not to be sorted out by the majority of sorting actuators.

[0018] For this example, it is further assumed that both the first and second objects are conveyed in the material flow in such a way that they both pass through the primary and secondary sorting actuators, unless they are rejected by one of the sorting actuators. Depending on the spatial and temporal relationship between the first and second objects, a specific value is determined by the correlation analysis device with respect to a spatial and temporal correlation criterion between the first and second objects.

[0019] In this example, it is assumed that the first object is conveyed downstream of the second object in the material flow, so that the first object arrives at the operating area of ​​the primary sorting actuator before the second object. Depending on the spatial and temporal distance, in particular the difference in velocity and / or acceleration between the first and second objects, a corresponding value of a spatiotemporal correlation criterion between the first and second objects is determined by the correlation analysis device.

[0020] If, based on the value of one spatial-temporal correlation criterion determined by the correlation analysis device, it is determined that if the first object is sorted out by the primary sorting actuator, the second object would also be sorted out, or there is a high probability that the second object would be sorted out as well, the sorting system can be specifically designed to control the primary sorting actuator in such a way that it does not sort out the first object, even though, based on the object class of the first object, the first object should be sorted out. This advantageously prevents the erroneous sorting out of an object that, based on its object class, should not be sorted out, thus reducing or preventing unwanted bycatch.

[0021] If, in the example, it is further determined based on the value of one spatiotemporal correlation criterion as determined by the correlation analysis device, it is determined that if the first object is sorted out by the secondary sorting actuator, the second object would not be sorted out, or there is only a low probability that the second object would also be sorted out, the sorting system can be designed, in particular, to control the secondary sorting actuator in such a way that it sorts out the first object. This can be the case, in particular, if the first object has a higher velocity than the second object, and both objects have, for example, essentially the same acceleration. Since the secondary sorting actuator is located downstream of the primary sorting actuator, the first object may have a different time interval to the second object at the operating range of the secondary sorting actuator than at the operating range of the primary sorting actuator.The secondary sorting actuator therefore gives the sorting system a second chance to sort out an object that should be sorted out, without causing unwanted bycatch.

[0022] This advantageously enables the reliable sorting of objects that should be sorted based on their object class, while reducing the probability of unintentionally sorting out objects that should not be sorted based on their object class, or even preventing such unwanted sorting altogether. Thus, the present sorting system can advantageously reduce or prevent unwanted bycatch, thereby improving the sorting of objects within a material stream and the removal of objects from a material stream.

[0023] In exemplary embodiments, the plurality of sorting actuators can, in particular, include at least one tertiary sorting actuator, such that the at least one tertiary sorting actuator is arranged downstream of the secondary sorting actuator with respect to the material flow, such that a first and a second object in the material flow can pass successively through the primary, secondary, and tertiary sorting actuators in such a way that they can be selectively, and optionally individually, sorted out of the material flow by the primary, secondary, or tertiary sorting actuator. In other words, the at least one tertiary sorting actuator is for the at least one secondary sorting actuator what the at least one secondary sorting actuator is for the at least one primary sorting actuator. In this way, the plurality of sorting actuators can comprise a plurality of sorting actuators arranged downstream of one another.In other words, the plurality of sorting actuators can comprise a plurality of sorting actuators arranged in a downstream cascade, wherein any two sorting actuators arranged successively along the material flow or in the conveying direction of the material flow each constitute at least one primary sorting actuator and at least one sorting actuator as defined above. This advantageously provides a multitude of opportunities where a spatiotemporal correlation criterion has a sufficient value to control a sorting actuator in such a way that it reliably sorts out an object to be sorted without causing by-catch of objects that should not be sorted.

[0024] In further exemplary embodiments, the majority of sorting actuators can, in particular, comprise object-class-specific sorting actuators, wherein at least one object-class-specific primary sorting actuator and at least one object-class-specific secondary sorting actuator are provided for each object class to be sorted from the material stream, as well as optionally further object-class-specific sorting actuators, in the sense of the previously defined at least one primary and secondary sorting actuators. An object-class-specific sorting actuator is configured to sort out only one or more predetermined object classes within its area of ​​operation, for example, sorting them into a predetermined object-class-specific collection device or object-class-specific collection unit.Sorting out an object that should be sorted out, but which does not belong to one of the object classes assigned to the sorting actor, constitutes bycatch.

[0025] In exemplary embodiments, the correlation analysis device can be configured to determine a value of a spatiotemporal correlation criterion between the objects in question only for those first, second, and, if applicable, subsequent objects, in particular to permanently determine which objects, according to an initial spatiotemporal correlation assessment, for example, based on an initial value of a spatiotemporal correlation criterion, pass one or more identical sorting actuators, especially within a predetermined time. This advantageously reduces computational effort and thereby improves the efficiency of the sorting system.

[0026] The following text uses various terms repeatedly, the understanding of which is facilitated by the definitions below.

[0027] Conveyor system: In this context, a conveying system is understood to mean, in particular, an arrangement along which objects are transported or conveyed. In exemplary embodiments, the conveying system may, in particular, be or comprise a conveyor belt, an incline, a free fall, and / or a ramp to transport or convey a flow of objects. In further exemplary embodiments, the conveying system may, for example, comprise at least a section of a conveying medium, such as water, to transport a flow of objects within and / or with it. The conveying system may, in particular, be arranged at the inlet and / or outlet of a facility that produces, incinerates, or otherwise processes or uses objects.The present conveying device may, in particular, be part of a waste sorting plant, a waste incineration or waste recycling plant, and may also be a conveyor belt or other conveying device during or after harvesting.

[0028] Object: The term "object," as used here, describes an object conveyed or transported in or with a material flow. Depending on its object class, the object represents, for example, an object or good to be sorted out or separated from the material flow. The object could be, for example, a vegetable, fruit, grain, meat, fish, stone, ore, waste, glass, a metal part, a plastic part, bulk material, or any other object to be sorted. The material flow includes, in particular, objects from at least two different object classes, and especially from two, three, four, or even more different object classes.

[0029] Object class: The object class, as used here, allows the objects in the material stream to be divided into different, and therefore distinguishable, object classes. Object classes can be distinguished, for example, by color, pattern, weight, size, contour, type, material, and / or other distinguishing characteristics. For example, a first object class of a first object can be distinguished from a second object class of a second object with the second color, and so on, based on a first color that is different from a second color.

[0030] Bycatch: In this context, bycatch refers to an object that, despite its object class (based on which it was classified as not to be sorted out or as not to be sorted out with respect to a particular sorting actuator), was nevertheless sorted out by that specific sorting actuator. Bycatch, in other words, describes an object that, despite its specific object class (according to which the sorting logic of the sorting system interprets it as not to be sorted out or not to be sorted out by a particular sorting actuator), is nevertheless sorted out. The terms bycatch and unwanted bycatch, as used here, mean the same thing.

[0031] If a direction or angle is given with the addition "essentially" or "approximately", this addition refers in particular to a deviation from the direction or angle in question in the range of 0° to 5°.

[0032] If the value of a physical quantity is given with the addition "essentially" or "approximately", this addition refers in particular to a deviation from the value in question in the range of 0% to 10%.

[0033] In In preferred embodiments of the sorting system, the spatiotemporal correlation criterion between the first object and the second object can include at least a velocity and a position of the first object, and additionally at least one of the following: includes a relative velocity of the second object relative to the first object and a position of the second object, or includes a relative position of the second object relative to the first object and a velocity of the second object, or includes a relative velocity of the second object relative to the first object and a relative position of the second object relative to the first object, or includes a time interval of the second object to the first object with respect to the position of the first object.

[0034] It is understood that the value of a spatiotemporal correlation criterion can be determined between more objects than just the first and second objects. A spatiotemporal correlation criterion between the first object, the second object, and a further object includes, in addition to the spatiotemporal correlation criterion between the first and second objects mentioned above, at least one of the following: a relative velocity of the further object relative to the first or second object and a position of the further object, or a relative position of the further object relative to the first or second object and a velocity of the further object, or a relative velocity of the further object relative to the first or second object and a relative position of the further object relative to the first or second object, or a time distance of the further object to the first or second object with respect to the position of the respective first or second object.

[0035] For a spatiotemporal correlation criterion between the first object, the second object, the next object, and a further object, the same principles apply as defined for the next object. Therefore, considering only a first object and a second object represents an exemplary simplification.

[0036] Based on the value of the spatiotemporal correlation criterion, the temporal encounter of the first and second objects at the primary and secondary sorting actuators can be advantageously determined. In particular, the time interval between the encounter of the first and second objects at the operating area of ​​the primary and secondary sorting actuators can be determined. This temporal encounter can be determined, in particular, by calculating and / or estimating the encounter of the first and second objects at the operating area of ​​the respective sorting actuator. This, in turn, can advantageously increase the accuracy of sorting objects in the material flow, thereby reducing the probability of bycatch.Estimating the arrival of a single object in a manner uncorrelated with other objects is exemplified in document WO 2015 / 128174 A1.

[0037] In preferred embodiments, the value of the spatiotemporal correlation criterion is determined continuously or at several discrete time points upstream of the at least one primary sorting actuator, preferably upstream of each individual sorting actuator. This advantageously allows for the determination of any changes in the value of the spatiotemporal correlation criterion, for example, when two objects conveyed or transported in the material stream collide. This, in turn, advantageously increases the accuracy of sorting objects in the material stream, particularly by reducing the probability of bycatch.

[0038] By selecting several or all of the following characteristics: a relative velocity of the second object relative to the first object and a position of the second object, a relative position of the second object relative to the first object and a velocity of the second object, a relative velocity of the second object relative to the first object and a relative position of the second object relative to the first object, and a time interval of the second object to the first object with respect to the position of the first object, By including objects within a spatiotemporal correlation criterion, particularly over a specific period or across multiple discrete points in time, the accuracy of object sorting can be advantageously increased. The value of a spatiotemporal correlation criterion, as determined by the correlation analysis device, can be specifically based on the objects detected by the detection device in the material flow, particularly based on the various parameters identified by the detection device. Objects in the material flow that were detected at specific times can be determined. In In exemplary embodiments, the value of a spatiotemporal correlation criterion can comprise a single value, multiple values ​​(vector), or a series of values, for example, a time series of values ​​or vectors.

[0039] In preferred embodiments of the sorting system, the detection device can be configured to detect a continuous area along the material flow, wherein the continuous area preferably extends upstream of the at least one primary sorting actuator to downstream of the at least one secondary sorting actuator.

[0040] By configuring the detection device, and in particular the optical detection device, to cover a continuous area along the material flow, it is advantageously possible to determine the value of the spatiotemporal correlation criterion at a multitude of time points, especially in a region extending upstream of the at least one primary sorting actuator to downstream of the at least one secondary sorting actuator. This advantageously improves the quality or significance of the spatiotemporal correlation criterion. Consequently, the accuracy of object sorting can be advantageously improved, and in particular, the probability of bycatch can be advantageously reduced.

[0041] Furthermore, the ability to monitor a continuous area encompassing both primary and secondary sorting actuators allows for the recording of sorting results. This means that both the sorting of objects by the actuators and, if applicable, the failure to sort objects can be recorded. The sorting result influences the control of sorting actuators located downstream of the actuator to which the respective result relates. For example, if an object intended for sorting by a primary sorting actuator is not sorted despite the primary sorting actuator's control setting, any estimation and / or calculation that assumed the object should have been sorted by the primary sorting actuator can be advantageously re-evaluated and determined, particularly using the correlation analysis function.The present sorting system enables the sorting of the object to be sorted by means of at least one secondary sorting actuator arranged downstream of the primary sorting actuator, or by a further sorting actuator arranged downstream. This advantageously improves the sorting of objects and, in particular, achieves a progressively higher sorting quality along the material flow, whereby objects to be sorted are correctly removed and objects that should not be removed remain in the material flow.

[0042] Furthermore, the detection of a continuous area in which the primary and secondary sorting actuators are arranged makes it advantageous to control a sorting actuator based on the value of the spatial-temporal correlation criterion in such a way as to reduce the probability of unwanted bycatch.

[0043] In preferred embodiments of the sorting system, the sorting system can be designed to control one sorting actuator of the plurality of sorting actuators in such a way that it changes the value of the spatial-temporal correlation criterion between the first object and the second object, preferably without sorting out either of the first object or the second object.

[0044] In other words, the sorting system can be designed to control one sorting actuator of the plurality of sorting actuators in such a way that it influences or disturbs the value of the spatial-temporal correlation criterion, in particular without sorting out either the first or the second object.

[0045] For example, in a case where the first and second objects always have the same speed and therefore essentially the same trajectory in the material flow, the first object cannot be sorted out due to insufficient time between them without risking the unwanted sorting of the second object. In this case, the primary sorting actuator of the sorting system can, for example, be moved into a sorting position, but this position is reached so far in advance of the first object that the first object only arrives at the sorting actuator's operating area when the sorting actuator returns to its home position. In this process, the first object may be deflected in a way that prevents it from being sorted, thus affecting or disrupting its speed and / or trajectory.This allows the value of the spatiotemporal correlation criterion to be advantageously influenced or disrupted without actually removing either object. Similarly, the speed and / or trajectory of the second object can be influenced or disrupted, for example, by moving the primary sorting actuator into a sorting position too late to remove the second object.

[0046] By changing, influencing, or disturbing the value of the spatial-temporal correlation criterion, a downstream sorting actuator can easily sort out an object to be sorted, particularly by means of a non-sorting control of a sorting actuator, thereby advantageously reducing the probability of the unwanted sorting out of an object that should not be sorted out.

[0047] In preferred embodiments of the sorting system, the detection device can comprise a sensor for detecting a two-dimensional detection section, wherein the two-dimensional detection section extends along the material flow and transversely to the material flow. in particular to detect objects which are conveyed perpendicular to the material flow offset from each other, wherein the sensor preferably comprises an area camera.

[0048] Capturing a two-dimensional detection section with the detection device, particularly the optical detection device, makes it advantageous to track a detected object, even if the object has a velocity perpendicular to the conveying direction of the material flow. This allows a comparatively accurate value for the spatiotemporal correlation criterion to be determined by the correlation analysis device, which in turn improves sorting and, moreover, reduces the probability of bycatch.

[0049] An area camera as a sensor, which the detection device includes, advantageously makes it possible to continuously detect objects along the material flow, and in particular to detect objects over a range extending upstream of the at least one primary sorting actuator to downstream of the at least one secondary sorting actuator, and thus in particular to detect a sorting success.

[0050] In preferred embodiments of the sorting system, the plurality of sorting actuators can comprise a first array and a second array of sorting actuators, wherein the first array comprises a plurality of primary sorting actuators and the second array comprises a plurality of secondary sorting actuators, wherein for each primary sorting actuator of the first array a secondary sorting actuator of the second array is provided such that the secondary sorting actuator of the second array is arranged downstream of the primary sorting actuator of the first array with respect to the material flow such that a first and a second object in the material flow of objects can pass successively through the primary sorting actuator of the first array and the secondary sorting actuator of the second array in such a way that they can be selectively, and if necessary individually, sorted out of the material flow by the primary sorting actuator of the first array or by the secondary sorting actuator of the second array.

[0051] It is not strictly necessary for the first array of sorting actuators to have the same number of sorting actuators as the second array. Rather, the second array of sorting actuators is arranged downstream of the first array in such a way that an object passing a primary sorting actuator of the first array along the conveying direction of the material flow can be sorted out by a secondary sorting actuator of the second array.

[0052] In exemplary embodiments, the plurality of sorting actuators can comprise further arrays of sorting actuators, with each further array being arranged analogously to the second array with respect to the first array in relation to the upstream preceding array.

[0053] By arranging several arrays of sorting actuators, spaced apart from each other along the conveying direction, the sorting system is advantageously provided with a two-dimensional arrangement of sorting actuators, so that the sorting system can take into account two-dimensional trajectories of the objects in the material flow, and whereby a variety of consecutive possibilities or opportunities for the safe sorting of objects to be sorted out are provided, while the probability of bycatch can be advantageously reduced.

[0054] In exemplary embodiments, one or more sorting actuators of an array of sorting actuators can be arranged transversely to the conveying direction relative to the respective sorting actuators of an upstream array of sorting actuators. In other words, using the example of primary and secondary sorting actuators, one or more secondary sorting actuators of a second array of sorting actuators can be arranged transversely to the conveying direction relative to one or more respective primary sorting actuators of a first array. This advantageously allows for the consideration of two-dimensional trajectories of the objects in the material flow, providing a multitude of consecutive possibilities or opportunities for the reliable sorting of objects to be rejected.

[0055] In preferred embodiments of the sorting system, the sorting system is designed to control the at least one primary sorting actuator and the at least one secondary sorting actuator depending on the object class of the first object, the object class of the second object, the value of the spatiotemporal correlation criterion between the first and second objects determined by the correlation analysis device, and a state or position of the at least one primary sorting actuator and / or the at least one secondary sorting actuator. The state or position of the at least one primary sorting actuator and / or the at least one secondary sorting actuator can, in particular, include a current state as well as a future state or position of the at least one primary sorting actuator and / or the at least one secondary sorting actuator. The future state or position can, for example, be a planned state or position.a planned position, based on a sorting out of the first object and / or the second object planned by the sorting system, each by the primary sorting actuator or the secondary sorting actuator.

[0056] By taking the state or position of each sorting actuator into account during the control process, specific physical requirements for the sorting actuator can be advantageously considered. For example, how quickly or slowly a sorting actuator can be moved into a state or position that allows it to sort out an object, particularly due to actuator-specific inertia. Furthermore, the sorting system's control system can advantageously account for situations where a sorting actuator is unable to sort out one of the first and / or second objects due to an already deflected position or because it is returning from a sorting state. This can advantageously increase the accuracy of object sorting and reduce the probability of bycatch.

[0057] Furthermore, by considering the state or position of a sorting actuator, a sorting actuator cycle specific to that actuator can be taken into account. In particular, the sorting actuator cycle can include, in addition to the sorting position in which the sorting actuator sorts out an object within its operating range, a state of movement towards the sorting position in which the sorting actuator is deflected or activated intermittently or temporarily, but there is uncertainty as to whether an object within its operating range during this state will be reliably sorted out or not.Furthermore, the sorting actuator cycle can, in particular, include a state of movement back from the sorting position, in which the sorting actuator is deflected or activated intermittently or temporarily, but there is uncertainty as to whether an object located within its operating range during this state will be reliably sorted out or not. In addition, the sorting actuator cycle can, in particular, include a reset state, in which the sorting actuator is not deflected or activated, but cannot yet be reactivated to sort out an object due to physical constraints.

[0058] In In preferred embodiments of the sorting system, each sorting actuator of the plurality of sorting actuators can have an actuator-specific switching inertia period, wherein the sorting system is designed to determine a time interval between the first object and the second object at the operating area of ​​the at least one primary sorting actuator, in particular to calculate or estimate it on the basis of the spatial-temporal correlation criterion, or to track it on the basis of the detection device, wherein the sorting system is designed to compare the time interval with the actuator-specific switching inertia of the at least one primary sorting actuator, and wherein the sorting system is designed to control the at least one primary sorting actuator on the basis of the comparison.

[0059] The actuator-specific switching inertia of a sorting actuator can be determined or predetermined, in particular physically, by the sorting actuator itself, as well as physically by the control or actuation of the respective sorting actuator by the sorting system. The actuator-specific switching inertia can, in particular, characterize the physical and control-related inertias of the sorting actuator in question. For example, the actuator-specific switching inertia can include a duration from the sending of a control signal to the sorting actuator in question until the receipt of a control signal from the sorting actuator in question, and / or a duration from the receipt of a control signal until the physical actuation of a flap, a trapdoor, a nozzle valve, or similar device, and / or, in particular, a minimum duration during which the sorting actuator remains in the sorting position, provided it is controlled or operated in a sorting-mode.

[0060] The actuator-specific switching inertia duration can therefore, in particular, include an activation duration of a sorting actuator, from a basic state of the sorting actuator in which the sorting actuator does not sort out passing objects from the material flow, through an activation phase in which the sorting actuator sorts out objects within an area of ​​influence of the sorting actuator, back to the basic state.

[0061] The activation period between an initial ground state of the sorting actuator, before reaching the sorting position, and a ground state after reaching the sorting position, can correspond in exemplary embodiments to a cycle duration of the sorting actuator, in particular to a minimum cycle duration of the sorting actuator, but is not limited to this.

[0062] In further exemplary embodiments, the sorting actuator can be reactivated upon reaching the basic state after having temporarily reached the sorting position, or can only be reactivated after waiting a further period of time, for example due to an actuator-specific inertia, such as a period for a reset.

[0063] A minimum activation duration is, in particular, a duration for which the sorting actuator maintains at least the sorting position. In other words, the minimum activation duration of a sorting actuator can describe the minimum duration within which the sorting actuator, if it is controlled in a sorting-mode (i.e., activated), is configured to sort out an object within its area of ​​operation. Put another way, the minimum activation duration of a sorting actuator describes the minimum duration for which the sorting actuator, if it is or becomes controlled in a sorting-mode, maintains a sorting-mode state. The default state before and after activation according to the minimum activation duration is, therefore, a state in which the sorting actuator is configured not to sort out an object.This can occur, for example, if a trapdoor is not fully open enough for an object to fall in or through, if a pneumatic nozzle does not provide sufficient impulse to an object to reject it (e.g., because the opening of a pneumatic sensor nozzle is not yet sufficiently open), or if a flap is not folded to deflect an object for rejection. In exemplary embodiments, the default state can be, in particular, a state in which the sorting actuator is configured to deflect an object within its operating range, but without rejecting it. Depending on the type of sorting actuator, sorting actuators can also remain in the rejecting position beyond the minimum activation duration, i.e., have an activation duration greater than the minimum activation duration or have an activation duration equal to the minimum activation duration.The former is particularly advantageous for sorting out more than one object in the sorting position, as can be the case, for example, with pneumatic sorting actuators. The latter can be the case, for example, with intermittent mechanical sorting actuators, and there are also mechanical sorting actuators whose activation time can be extended beyond a minimum activation period.

[0064] By comparing the actuator-specific switching inertia of a sorting actuator with the time interval between the first and second objects within the sorting actuator's operating range, it can be advantageously determined whether, taking the actuator-specific switching inertia into account, the second object can be sorted out within the activation period in which the first object is also sorted out, or whether an undesired sorting of the second object occurs within the activation period in which the first object is also sorted out. Based on this comparison, the respective sorting actuator can be controlled in such a way that precise sorting of the objects to be sorted is enabled, and the probability of bycatch is advantageously reduced.

[0065] In preferred embodiments of the sorting system, in the case where the time interval, i.e., in particular the time interval between the first object and the second object at the operating area of ​​the at least one primary sorting actuator, is shorter than or equal to the actuator-specific switching inertia of the at least one primary sorting actuator, The sorting system must be designed not to control at least one primary sorting actuator in such a way that it sorts out the first object if sorting out the second object is not intended due to its object class, and the sorting system must be designed to control at least one primary sorting actuator in such a way that it sorts out the first object if sorting out the first object and the second object is intended due to their object class.

[0066] In other words, if the time interval is shorter than or equal to the actuator-specific switching inertia of the at least one primary sorting actuator, the sorting system can be designed to control the at least one primary sorting actuator in such a way that it does not sort out the first object, i.e., in particular allows it to pass, if the second object cannot be sorted out due to its object class, and to control the at least one primary sorting actuator in such a way that it sorts out the first object if sorting out the first object and the second object is intended due to their object class.

[0067] By controlling the sorting actuator based on a comparison of the time interval between the first and second objects at the sorting actuator's operating area with the actuator's specific switching inertia, whereby the sorting actuator is controlled in such a way that it does not sort out the first object if the second object is not to be sorted out, and if the time interval is shorter than or equal to the actuator's specific switching inertia, an unwanted sorting out of the second object can advantageously be prevented.

[0068] In further exemplary embodiments, the sorting system can be designed to compare the time interval between the first and second objects with the actuator-specific switching inertia in various ways. For example, a comparison can be made where the absolute value of the time interval between the first and second objects is compared with the actuator-specific switching inertia of the sorting actuator in question. Furthermore, a comparison can be made, for example, as to whether the sorting actuator can return to its initial state based on the actuator-specific switching inertia after the first object to be sorted has been removed and before the second object, which is not to be removed, arrives at or encounters the operating area of ​​the sorting actuator in question.

[0069] This allows for a flexible and advantageous way to check whether an object to be sorted out can be safely removed, while advantageously avoiding the incidental collection of objects that cannot be sorted out.

[0070] In exemplary embodiments, the sorting system can be designed to extend the actuator-specific switching inertia by an additional term when comparing the time interval between the first and second objects at the operating area of ​​a respective sorting actuator with the actuator-specific switching inertia of the respective sorting actuator, and to perform the comparison based on this. This is particularly advantageous for sorting actuators that can remain in the sorting position beyond their actuator-specific switching inertia, especially beyond their minimum activation time, in order to sort out more than one object in the sorting position, as can be the case, for example, with pneumatic sorting actuators or mechanically deflecting sorting actuators.

[0071] In preferred embodiments of the sorting system, the sorting system can be designed to determine a time interval between the first object and the second object at an operating area of ​​the at least one primary sorting actuator and at an operating area of ​​the at least one secondary sorting actuator. wherein the sorting system is designed to compare the time interval at the operating area of ​​the at least one primary sorting actuator with the actuator-specific switching inertia of the at least one primary sorting actuator and to compare the time interval at the operating area of ​​the at least one secondary sorting actuator with the actuator-specific switching inertia of the at least one secondary sorting actuator, and wherein the sorting system is designed to control the at least one primary sorting actuator on the basis of the comparisons.

[0072] In exemplary embodiments, the time interval between the first and second objects at the operating area of ​​the at least one primary sorting actuator and at the operating area of ​​the at least one secondary sorting actuator can be determined, in particular by measuring, estimating, or tracking. For example, the time interval between the objects at the operating area of ​​the respective sorting actuators can be measured, estimated, or, in particular, tracked to that point by means of the detection device, in particular optical and / or capacitive detection and / or weight measurement. Estimating the time interval between the objects at the respective operating area of ​​the sorting actuators can be performed, in particular, based on the spatiotemporal correlation criterion, especially by the sorting system.

[0073] In preferred embodiments of the sorting system, in the case where the respective time interval is shorter than, or equal to, the actuator-specific switching inertia of the at least one primary sorting actuator and the at least one secondary sorting actuator, the sorting system can be designed to control the at least one primary sorting actuator in such a way that the at least one primary sorting actuator deflects at least one of the first object and the second object without sorting it out, if sorting out the first object is provided for due to its object class and sorting out the second object is not provided for due to its object class.

[0074] This allows the value of the spatiotemporal correlation criterion with respect to the first and second objects to be advantageously influenced and, in particular, changed by means of at least one primary sorting actuator, especially without sorting out either of the two objects. Furthermore, this allows the time interval between the first and second objects downstream of the at least one primary sorting actuator, i.e., for example, at the operating area of ​​at least one secondary sorting actuator, to be advantageously influenced and, in particular, changed. This allows the sorting of objects that are difficult to separate, which in particular comprise at least one object to be sorted and one object that should not be sorted out, to be advantageously influenced. This allows the probability of reliably sorting out objects to be sorted out and of avoiding the unwanted sorting out of objects that should not be sorted out to be advantageously increased.

[0075] In In preferred embodiments of the sorting system, the sorting system can be designed by means of a sorting detection device, in particular the detection device, to record the sorting out of an object as a sorting result, wherein the sorting system is preferably designed to assign a return value to the sorting result, whereby the sorting system is updated on the basis of the return value.

[0076] InIn further preferred embodiments of the sorting system, the sorting system can be designed, by means of the sorting detection device, in particular the detection device, to detect the sorting of an object with a specific sorting actuator from the plurality of sorting actuators as a sorting result, wherein the sorting system is preferably designed to control a sorting actuator arranged downstream of a specific sorting actuator depending on the object class of the first object, the object class of the second object, the value of a spatial-temporal correlation criterion between the first object and the second object determined by the correlation analysis device, and the sorting result.

[0077] By recording the sorting result, it is therefore advantageous to control sorting actuators, especially those located downstream of the sorting process, depending on the sorting result.

[0078] Since the control of the sorting actuators depends in particular on the value of a spatial-temporal correlation criterion between objects, it is possible to evaluate whether the control of the sorting actuators is carried out in a suitable or successful manner based on the value of one spatial-temporal correlation criterion, i.e., in particular, whether objects to be sorted are reliably sorted out, and objects not to be sorted out are reliably not sorted out.

[0079] Since the control of the sorting actuators is also dependent on the object class of the respective objects, it is particularly possible to assess whether the control of the sorting actuators is carried out correctly based on the object class, i.e., in particular, whether the classification device assigns the correct object class to the objects recognized optically, capacitively and / or by weight.

[0080] The return value represents a possibility for evaluation, in particular a suitable possibility for evaluation for the automation of the sorting system, for example for a self-learning sorting system.

[0081] The return value can be positive, for example, if an object to be sorted out is sorted out, and / or negative, for example, if an object not to be sorted out is sorted out.

[0082] Based on the return value, the sorting system can be updated, for example, by updating the control of the sorting actuators depending on the value of the spatial-temporal correlation criterion, for example, by considering a certain value of the spatial-temporal correlation criterion as sufficient or insufficient to prevent bycatch, and by updating the decision logic of the sorting system to control the sorting actuators depending on the value of the spatial-temporal correlation criterion.

[0083] Additionally or alternatively, the sorting system can be updated by blocking individual spatiotemporal correlation criteria from a large number of spatiotemporal correlation criteria, i.e., excluding them from application, or by designating a specific spatiotemporal correlation criterion as preferred or to be used exclusively. This can be particularly useful when updating the sorting system for a specific sorting scenario, such as waste sorting, in order to advantageously reduce the risk of bycatch as the system's operating time increases.

[0084] Alternatively or additionally, the sorting system can be updated based on the return value in such a way that an object class library or an object class assignment logic is updated, on the basis of which the classification system classifies the objects.

[0085] In exemplary embodiments, the return value can in particular represent a cost function on the basis of which the sorting system is configured as a self-learning sorting system.

[0086] In further exemplary embodiments, the return value can be used in particular to detect signs of aging or wear in the hardware, especially the sorting actuators, in order to update the algorithm or sorting logic of the sorting system on this basis. This allows the sorting system to advantageously have a particularly long operating time and service life.

[0087] In exemplary embodiments, one or more sorting actuators of the plurality of sorting actuators can be configured to sort out an object to be removed transversely to the conveying direction, i.e., in particular laterally to the conveyor. In other words, one or more sorting actuators of the plurality of sorting actuators can be configured to move an object transversely to the conveying direction in order to remove it. This allows a sorting result to be acquired in a conveniently simple manner by a sorting detection device, in particular by the detection device of the sorting system, such as an optical detection device.

[0088] InIn further exemplary embodiments, the sorting system can have a collection device, wherein the collection device can, for example, comprise a collection basin to receive several sorted objects, or a plurality of collection units to receive individual objects at the respective collection unit. The collection device can, in particular, be communicatively coupled to a detection device, especially the sorting detection device, which can correspond to or be included by the detection device of the sorting system, and can, in particular, be configured to communicate a collection signal, dependent on the collected object class of an object, to a detection device, especially the sorting detection device, such as the detection device of the sorting system, or to the sorting system itself.This makes it advantageously easy to record the sorting result of an object being rejected.

[0089] For example, the sorting system can include a primary collection unit for the primary sorting actuator, a secondary collection unit for the secondary sorting actuator, and additional corresponding collection units for any further sorting actuators. In particular, the sorting system can include one collection unit per sorting actuator or per array of sorting actuators. This advantageously allows for the recording of a sorting result specific to a single sorting actuator or array of sorting actuators. This, in turn, advantageously enables precise adjustment of the sorting actuator control based on the sorting result.

[0090] Furthermore, the detection device and / or the collection device can be configured to detect a malfunction and / or failure of a sorting actuator. A malfunction and / or failure of a sorting actuator can be detected, for example, by the detection device by recording that a sorting actuator does not move, does not move, or does not move completely, despite the activation of the sorting actuator in question. A malfunction and / or failure of a sorting actuator can be detected, for example, by the collection device by recording that, despite multiple activations, no objects enter the collection device or unit assigned to the sorting actuator in question. InIn preferred embodiments, the sorting system can additionally be configured to take into account malfunctions and / or failures of individual sorting actuators when calculating the spatiotemporal correlation criterion and to assign the decision to reject an object to another sorting actuator that the object passes through. This advantageously allows the system to utilize the multitude of consecutive possibilities or opportunities for the reliable rejection of objects without causing a system outage for repair and / or maintenance.

[0091] In In preferred embodiments of the sorting system, the plurality of sorting actuators may comprise at least one pneumatic sorting actuator and at least one mechanical sorting actuator; or the plurality of sorting actuators may comprise exclusively mechanical sorting actuators.

[0092] By incorporating both pneumatic and mechanical sorting actuators, the sorting system enables comparatively energy-efficient sorting using the mechanical actuators, while the pneumatic actuators allow for easy control of an object's trajectory and, in particular, its speed. This results in a flexible sorting system that ensures reliable sorting with low energy consumption. Furthermore, the pneumatic actuators offer a particularly short response time for sorting objects. A combination of both pneumatic and mechanical actuators also allows for object-class-specific sorting.For example, small and / or light objects can be sorted by a pneumatic sorting actuator, while large and / or heavy objects can be sorted by a mechanical sorting actuator. Similarly, small and / or light objects can be sorted by a centrifugal flap of a mechanical sorting actuator, while large and / or heavy objects can be sorted by a trapdoor of a mechanical sorting actuator.

[0093] In contrast, a sorting system consisting exclusively of mechanical sorting actuators enables particularly energy-efficient sorting. While mechanical sorting actuators generally have a longer response time compared to pneumatic sorting actuators, making it difficult to manage high material flow densities with an arrangement consisting solely of mechanical sorting actuators, those incorporating solenoids and / or a voice coil motor offer exceptionally short response times. Therefore, deploying multiple mechanical sorting actuators with solenoids and / or voice coil motors can advantageously achieve both energy-efficient sorting and short response times.A sorting system equipped in this way is advantageously suited to reliably and energy-efficiently sorting particularly large mass flows or particularly dense material flows.

[0094] A pneumatic sorting actuator can, in particular, comprise a nozzle and, especially, a valve. In the activated state of the pneumatic sorting actuator, air or an air jet flows into a predetermined area. The air or air jet can push or draw one or more objects in a predetermined direction. Since the air or air jet from the nozzle spreads out conically with increasing distance from the nozzle, and since the pneumatic sorting actuator requires a supply of pressurized air or pressurized gas to push or deflect an object, the pneumatic sorting actuator is comparatively energy-inefficient, particularly compared to a mechanical sorting actuator for sorting an object.

[0095] A mechanical sorting actuator can be, for example, a flap, an actuated conveyor belt, or a trapdoor. While the trapdoor, when activated, sorts out an object by causing it to fall through or into it, and the actuated conveyor belt sorts out an object by moving or accelerating it in a specific direction, the flap can cause an object to be sorted out in various ways.

[0096] On the one hand, the flap, when activated, can apply a pulse to an object to sort it out. To sort out a subsequent object, the flap cannot remain in the activated state but must first return to a deactivated state, specifically to a default state.

[0097] On the other hand, the flap, similar to a pneumatic sorting actuator, can deflect an object for sorting purposes by being deflected into a sorting position. If the flap remains in the sorting position, several objects within its effective range will be sorted out, since a separate deflection of the flap is not required for each object to be sorted.

[0098] Unlike a pneumatic sorting actuator, the flap establishes a physical contact, specifically a fixed contact, between the flap itself and the object by deflecting the flap. The impulse application or deflection by a mechanical sorting actuator can therefore be comparatively more targeted than with a pneumatic sorting actuator, making the operation of a mechanical sorting actuator for sorting an object more energy-efficient.

[0099] Another aspect of the invention relates to a method for sorting objects conveyed in a material stream according to object classes, wherein the method comprises the steps: Providing a plurality of sorting actuators for the selective sorting of objects from the material stream, wherein the plurality of sorting actuators comprises at least one primary sorting actuator and at least one secondary sorting actuator such that the at least one secondary sorting actuator is arranged downstream of the primary sorting actuator with respect to the material stream in such a way that a first object and a second object in the material stream can pass the primary and the secondary sorting actuator successively in such a way that they can be selectively, and optionally individually, sorted out of the material stream by the primary or the secondary sorting actuator; detecting, in particular optically detecting, objects in the material stream; classifying the detected objects into at least two object classes; determining the value of a spatiotemporal correlation criterion between the detected objects;Controlling at least one primary sorting actuator and at least one secondary sorting actuator depending on the object class of the first object, the object class of the second object, and the value of a spatiotemporal correlation criterion between the first object and the second object.

[0100] The method for sorting objects conveyed in a material stream according to object classes, as described in the further aspect, can be carried out in particular by means of the sorting system for sorting objects in a material stream according to object classes as described in the first aspect. The exemplary, preferred, and alternative embodiments of the present sorting system, as well as their respective effects, therefore relate equally to the present sorting method and vice versa.

[0101] InIn preferred embodiments of the sorting method, the step of controlling the at least one primary sorting actuator and the at least one secondary sorting actuator depending on the object class of the first object, the object class of the second object and the value of a spatiotemporal correlation criterion between the first object and the second object may in particular comprise: Controlling the at least one primary sorting actuator and the at least one secondary sorting actuator depending on the object class of the first object, the object class of the second object, the value of the spatiotemporal correlation criterion determined by the correlation analysis device between the first object and the second object, and a state of the at least one primary sorting actuator and / or the at least one secondary sorting actuator.

[0102] By taking the state or position of each sorting actuator into account during the control process, specific physical requirements for the sorting actuator can be advantageously considered. For example, how quickly or slowly a sorting actuator can be moved into a state or position that allows it to sort out an object, particularly due to actuator-specific inertia. Furthermore, the sorting system's control system can advantageously account for situations where a sorting actuator is unable to sort out one of the first and / or second objects due to an already deflected position or because it is returning from a sorting state. This can advantageously increase the accuracy of object sorting and reduce the probability of bycatch.

[0103] Furthermore, by considering the state or position of a sorting actuator, a sorting actuator cycle specific to that actuator can be taken into account. In particular, the sorting actuator cycle can include, in addition to the sorting position in which the sorting actuator sorts out an object within its operating range, a state of movement towards the sorting position in which the sorting actuator is deflected or activated intermittently or temporarily, but there is uncertainty as to whether an object within its operating range during this state will be reliably sorted out or not.Furthermore, the sorting actuator cycle can, in particular, include a state of movement back from the sorting position, in which the sorting actuator is deflected or activated intermittently or temporarily, but there is uncertainty as to whether an object located within its operating range during this state will be reliably sorted out or not. In addition, the sorting actuator cycle can, in particular, include a reset state, in which the sorting actuator is not deflected or activated, but cannot yet be reactivated to sort out an object due to physical constraints.

[0104] In In preferred embodiments of the sorting method, the method may further include the following steps: Determining the time interval between the first object and the second object at an area of ​​operation of the at least one primary sorting actuator; comparing the time interval with an actuator-specific switching inertia period of the at least one primary sorting actuator; and controlling the at least one primary sorting actuator based on the comparison.

[0105] The operating area of ​​a respective sorting actuator represents the region within which the sorting actuator is configured to sort out an object, provided it is controlled in a sorting manner. The step of determining the time interval can, in particular, include a step of estimating the time interval between the first and second objects, within the operating area of ​​at least one primary sorting actuator, based on the value of the spatiotemporal correlation criterion.Alternatively or additionally, the step of determining the time interval can, in particular, include a step of tracking the first object and the second object, or a step of tracking the time interval between the first object and the second object, wherein the tracking preferably extends to the respective operating area, for example, by having the detection device, in particular the optical detection device, also detect the operating area of ​​the respective sorting actuator. Furthermore, the step of determining the time interval can be based on the detected objects, i.e., in particular the first and second objects. This includes detecting the objects at an operating area of ​​a sorting actuator. Based on the determined time interval, the value of the spatiotemporal correlation criterion between the first object and the second object can be determined in exemplary embodiments.

[0106] The predetermined minimum activation duration of a sorting actuator is the duration for which the sorting actuator switches from a default state to a sorting state, in which the sorting actuator is specifically configured to sort out an object within its operating area—that is, to sort out an object located within the sorting actuator's operating area. In other words, the minimum activation duration of the sorting actuator describes, in particular, the duration for which the sorting actuator remains in at least the sorting state.

[0107] The minimum activation time is specific to each sorting actuator. Depending on the sorting actuator, the duration for which the actuator remains in the sorting state can be extended. In other words, the minimum activation time can be extended, for example, by controlling a pneumatic sensor to keep the nozzle open longer than specified by the minimum activation time, or by allowing a trapdoor or flap to remain in the deflected state longer than specified by the minimum activation time. However, the duration for which the sorting actuator is configured to sort out an object is not extendable for every sorting actuator.If, for example, sorting by a mechanical sorting actuator depends in particular on the impulse applied to the object, extending the end position reached during deflection of the mechanical sorting actuator does not necessarily lead to further sorting by means of an impulse, since this depends on the deflecting movement of the exemplary mechanical sorting actuator itself.

[0108] In preferred embodiments of the sorting method, in the case where the time interval is shorter than or equal to the actuator-specific switching inertia of the at least one primary sorting actuator, the method may further comprise the following steps: Controlling at least one primary sorting actor in such a way that it does not sort out the first object if sorting out the second object is not intended due to its object class, and controlling at least one primary sorting actor in such a way that it sorts out the first object if sorting out the first object and the second object is intended due to their object class.

[0109] In preferred embodiments of the sorting method, the method may further comprise the following steps: Determining the time interval between the first object and the second object at an area of ​​influence of the at least one primary sorting actuator and at an area of ​​influence of the at least one secondary sorting actuator; comparing the time interval at the area of ​​influence of the at least one primary sorting actuator with an actuator-specific switching inertia of the at least one primary sorting actuator, and comparing the time interval at the area of ​​influence of the at least one secondary sorting actuator with an actuator-specific switching inertia of the at least one secondary sorting actuator; and controlling the at least one primary sorting actuator based on the comparisons.

[0110] In exemplary embodiments, the at least one primary sorting actuator and the at least one secondary sorting actuator can be controlled based on the comparisons such that the sorting actuator in whose operating area a low probability of bycatch is determined, based on the comparisons, is controlled to perform the sorting action. Alternatively or additionally, the at least one primary sorting actuator and the at least one secondary sorting actuator can be controlled based on the comparisons such that the purity of the unsorted objects, i.e., the proportion of objects not to be sorted among the unsorted objects, is as high as possible.

[0111] In preferred embodiments of the sorting method, in the case where the respective time interval is shorter than, or equal to, the actuator-specific switching inertia of the at least one primary sorting actuator and the at least one secondary sorting actuator, the method may further comprise the following steps: Controlling the at least one primary sorting actor such that the at least one primary sorting actor deflects at least one of the first object and the second object without sorting it out, if sorting out the first object is provided due to its object class and sorting out the second object is not provided due to its object class.

[0112] The following describes embodiments of the invention in more detail with reference to the accompanying figures. It is understood that the present invention is not limited to these embodiments and that individual features of the embodiments can be combined to form further embodiments within the scope of the accompanying claims.

[0113] They show: Figures 1a, 1b, 1c are exemplary sorting systems according to the prior art; Figures 2a, 2b, 2c, 2d are exemplary sketches of sorting systems according to an embodiment of the present invention; Figure 3 is an exemplary arrangement of sorting actuators according to an embodiment of the present invention; Figure 4a is an actuator position / time diagram of a sorting actuator according to an embodiment of the present invention; Figures 4b, 4c are exemplary actuator position / time diagrams of a sorting actuator according to an embodiment of the present invention; Figure 5a is an exemplary actuator position / time diagram; Figures 5b, 5c are exemplary actuator position / time diagrams of sorting actuators according to an embodiment of the present invention; Figure 6a is an exemplary actuator position / time diagram; Figures 6b, 6c are exemplary actuator position / time diagrams of sorting actuators according to an embodiment of the present invention; Figure 7a is an exemplary actuator position / time diagram;Figures 7b, 7c show exemplary actuator position / time diagrams of sorting actuators according to an embodiment of the present invention; Figure 8 shows a decision tree according to an embodiment of the present invention; Figure 9 shows a sorting flowchart according to an embodiment of the present invention; Figure 10 shows a sorting flowchart according to an embodiment of the present invention; and Figure 11 shows a sorting flowchart according to an embodiment of the present invention.

[0114] Figures 1a, 1b and 1c Show exemplary sorting systems 1 according to the state of the art.

[0115] This shows Fig. 1a For example, a conveyor belt as a conveying device 2 for conveying objects in a material flow. The in Fig. 1b The conveying device shown (2) includes, for example, a chute for conveying the objects. The in Fig. 1cThe conveying device shown, number 2, comprises a section along which the objects are conveyed by falling.

[0116] The in the Figures 1a, 1b and 1c The conveying device 2 shown in each image conveys objects 8 and 9 of at least two different object classes: objects 9 with a round object class, which are shown accordingly, and objects 8 with a rectangular object class, which are shown accordingly. The respective conveyed objects 8 and 9 are detected by the optical detection device 4 and sorted out by the sorting actuator 6 according to their object class.

[0117] As through the Figures 1a, 1b and 1c As illustrated, the sorting actuator 6 is configured to sort out the objects 8 to be sorted by means of an air jet, and thereby separate them from the material flow, which otherwise mainly includes the objects 9 not to be sorted out. As shown in the Figures 1a, 1c and 1cTo further clarify, the sorting actuators 6 can, in particular, comprise pneumatic sorting actuators 6, which are arranged downstream of the respective detection devices 4 on the sorting system 1. This provides the sorting actuators 6 with sufficient time to be activated after optical detection by the optical detection device 4, once an object 8 to be sorted out has been detected by the optical detection device 4. Line and / or area scan cameras can, in particular, be used for the optical detection device 4.

[0118] Figures 2a, 2b , 2c and 2d show exemplary sketches of sorting systems 10, each according to an embodiment of the present invention.

[0119] The Figures 2a, 2b and 2d Figure 1 shows an example of a sorting system 10 with a conveyor system 20, wherein the conveyor system 20 includes a slide along which the objects 80, 90 are conveyed. Fig. 2cFigure 1 shows an example of a conveying device 20 along which the objects 80, 90 are conveyed at least partially by falling, i.e., by free fall. In further exemplary embodiments, the conveying device 20 can comprise a belt conveyor, as shown in Figure 20. Fig. 1a As shown. In further exemplary embodiments, the conveyor device can comprise 20 grooves, so that the sorting system 10 can be designed as a groove conveyor system.

[0120] Objects 80 and 90 are in the Figures 2a, 2b , 2c and 2dThe objects 80 and 90 are represented merely as examples of black spheres or balls. They are conveyed by the conveyor 20, at least partially, along the conveying direction FR, which is shown as an example with dashed lines. The objects 80 and 90 can, for example, move essentially in the conveying direction FR, i.e., have a trajectory that is essentially parallel to the conveying direction FR. Furthermore, the objects 80 and 90 can also move perpendicular to the conveying direction FR, so that the trajectory of such an object is inclined, at least partially, relative to a parallel to the conveying direction FR. The multitude of objects 80 and 90 forms a section of a material flow that is conveyed, at least partially, essentially along the conveying direction 20. If the sorting system 10 is designed as a groove conveyor, the objects 80 and 90 are advantageously conveyed essentially along the conveying direction FR.This enables, for example, very precise control of the sorting actuators 60 in order to sort out an object 80 to be sorted at an area of ​​influence 70 of a sorting actuator 60, since it can advantageously be ensured that the objects 80, 90 reach the area of ​​influence 70 of the sorting actuators 60.

[0121] As in the Figures 2a, 2b , 2c and 2dAs shown, a plurality of sorting actuators 60 are arranged on the conveyor 20 to remove objects 80 to be sorted from the material stream. The plurality of sorting actuators 60 includes, in particular, several primary sorting actuators 62, several secondary sorting actuators 64, and, by way of example, at least several tertiary sorting actuators 66. In further exemplary embodiments, the plurality of sorting actuators 60 can include further sorting actuators 60 arranged downstream, i.e., sorting actuators 60 arranged following the conveying direction FR. However, each sorting-controlled sorting actuator 60 inherently carries the risk of also removing an object 90 that should not be removed as byproduct.

[0122] As through the Figures 2a, 2b , 2c and 2dAs shown, the secondary sorting actuators 64 are arranged downstream of the primary sorting actuators 62, and the tertiary sorting actuators 66 are arranged downstream of the secondary sorting actuators 64, so that a plurality of possibilities for sorting out objects 80 are provided in a cascading manner along the conveying direction FR.

[0123] How to proceed through the Figures 2a, 2b , 2c and 2d As shown, the sorting actuators 60 can be arranged in an array, in particular in an array which extends essentially transversely to the conveying direction FR. Figures 2a, 2b , 2c and 2d each shows a first array of primary sorting actors 62, a second array of secondary sorting actors 64, and Fig. 2aFigure 66 additionally shows an example of a third array of tertiary sorting actuators 66. The sorting actuators 60 of the respective array can be arranged at a specific position along the conveying direction FR, or offset from each other along the conveying direction FR, and thus be arranged both transversely and offset from each other along the conveying direction FR.

[0124] In the Figures 2a, 2b and 2c Only mechanical sorting actuators 60 are shown. This shows Fig. 2a For example, a mechanical sorting actuator 60, which includes a pushing pin, in order to exert a pulse on an object 80, 90 by activating it in a position to be sorted and to sort out the object 80, 90. Fig. 2bFigure 60 shows an example of a mechanical sorting actuator 60, which includes a trapdoor or drop flap which, in its sorting position, i.e., when it is open, sorts out an object 80, 90 by dropping or letting it fall out. Fig. 2c Figure 60 shows an example of a mechanical sorting actuator which includes a flap which, in its sorting position, i.e., by being deflected, deflects and sorts out an object 80, 90 so that it enters a sorting area. Fig. 2d Figure 60 shows an example of a pneumatic sorting actuator 60, which includes, in particular, a nozzle for the discharge of pressurized air. By the discharge of air from the pneumatic sorting actuator 60, an object 80, 90 in a sorting position of the pneumatic sorting actuator 60 can be deflected and sorted out by the pneumatic sorting actuator 60 in such a way that it enters a sorting area. Figures 2a, 2b , 2c and 2dThe sorting of an object 80 to be sorted by a sorting actuator 60 is shown as an example.

[0125] As through the Fig. 2a As illustrated, the sorting actuator 60 can be used according to Fig. 2a , i.e., the pushing pin, by extending the duration in which it occupies the sorting position, does not sort out any further objects 80, 90, since it does not generate a new impulse while remaining in the sorting position. In contrast, the trapdoors or drop flaps, in particular, according to Fig. 2b , the flaps according to Fig. 2c , as well as the pneumatic sorting actuator 60 according to Fig. 2d , that if the duration in which the respective sorting actuator assumes the sorting position is extended, i.e., in particular if it is extended beyond the minimum activation period, further objects 80, 90 can be sorted out if they enter the area of ​​influence 70 of the respective sorting actuator 60.

[0126] The sorting actuators 60, as in the Figures 2a, 2b , 2c and 2d The examples shown represent exemplary sorting actuators 60, so that exemplary embodiments may additionally or alternatively include, in particular, both pneumatic sorting actuators 60 and mechanical sorting actuators 60. Figures 1a, 1b and 1c further examples of pneumatic sorting actuators are shown 60.

[0127] As in the Figures 2a, 2b , 2c and 2d As shown, the sorting system 10 can in particular comprise an optical detection device 40, which in turn detects a two-dimensional detection section 45. The detection section 45 extends in particular essentially in one plane along the conveying direction FR and transversely to the conveying direction FR. In other words, the detection section 45 preferably covers an area along the material flow along which the objects 80, 90 are conveyed by the conveying device 20. Preferably, and as shown in the Figures 2a, 2b , 2c and 2dAs shown, the detection section 45 comprises a particularly continuous area upstream of the primary sorting actuators 62 to downstream of the secondary or tertiary sorting actuators 64, 66. If further sorting actuators 60 are provided downstream of the secondary or tertiary sorting actuators 64, 66, the detection section 45 preferably extends downstream to the sorting actuators 60 arranged further downstream. Alternatively, for example, further optical detection devices 40 can be arranged on the conveyor 20 to enlarge the two-dimensional detection section 45, particularly in the conveying direction FR.

[0128] Through the detection section 45, as in the Figures 2a, 2b , 2c and 2dAs shown, objects 80 and 90 can advantageously be tracked up to the respective operating ranges 70 of the sorting actuators 60, thus enabling a particularly precise determination of the value of a spatiotemporal correlation criterion between objects 80 and 90. Furthermore, this advantageously allows for precise control of the sorting actuators 60 by the sorting system 10, which in particular reduces the probability of bycatch and increases the purity of the unsorted objects. In addition, the two-dimensional detection section 45 advantageously enables the recording of a sorting result. The sorting result can, in particular, be the result that an object 80 to be sorted out has been sorted out at a specific sorting actuator 60. The sorting result can also be the result that an object 90 not to be sorted out has been sorted out.

[0129] The optical detection device 40 therefore advantageously enables the quality of the sorting system 10 to be recorded and thus evaluated by means of the two-dimensional detection section 45. The sorting result recorded by the optical detection device 40 can be used to update the sorting system 10, in particular to update the algorithm or the control logic on the basis of which the sorting system 10 controls the sorting actuators 60, in particular by means of sorting.

[0130] Although in the Figures 2a, 2b , 2c and 2dUnless explicitly shown, the sorting system 10 may, in particular, comprise a control and / or regulating unit. The control and / or regulating unit of the sorting system 10 may, in particular, comprise a computer, one or more processors for performing calculations, a memory for storing data, especially updated data, and one or more input and / or output units, such as a keyboard, a mouse, a control terminal, a display, or similar devices.

[0131] The sorting system 10 communicates with the optical detection device 40, the sorting actuators 60, and in particular with a classification device and a correlation analysis device, which are located in the Figures 2a, 2b , 2c and 2d not shown, connected.

[0132] The correlation analysis device is configured, in particular, to determine the value of at least one spatiotemporal correlation criterion between objects 80 and 90, specifically based on the optical detection of objects 80 and 90 by the optical detection device 40. The spatiotemporal correlation criterion relates one object 80 or 90 both spatially and temporally to at least one other object 80 or 90. The spatiotemporal correlation criterion between two objects preferably includes at least one velocity and one position of a first object and additionally at least one of the following: includes a relative velocity of the second object relative to the first object and a position of the second object, or includes a relative position of the second object relative to the first object and a velocity of the second object, or includes a relative velocity of the second object relative to the first object and a relative position of the second object relative to the first object, or includes a time interval of the second object to the first object with respect to the position of the first object.

[0133] In further exemplary embodiments, the spatiotemporal correlation criterion can in particular include an acceleration and / or a relative acceleration of the respective objects.

[0134] The classification device is specifically configured to classify the optically detected objects 80, 90 into at least two object classes. In other words, the classification device is specifically configured to assign each object 80, 90 to one of at least two object classes. As described in the Figures 1a, 1b and 1c As shown by way of example, the classification device can perform a classification in particular based on the geometric shape in order to distinguish an object 8 to be sorted from an object 9 that is not to be sorted.

[0135] The classification device is configured to assign an object class to each of objects 80 and 90, so that a distinction can be made between objects 80 to be sorted and objects 90 not to be sorted, in order to control the sorting actuators 60 accordingly, either to sort or not to be sorted, via the sorting system 10. In further exemplary embodiments, the classification device can be configured to assign more than two different object classes to the objects, so that, based on the different object classes, a distinction can be made, for example, between objects not to be sorted and objects to be sorted by various sorting actuators.

[0136] Figure 3 Figure 60 shows an exemplary arrangement of sorting actuators according to an embodiment of the present invention.

[0137] In Fig. 3Two grooves arranged side by side along the conveying direction FR are shown as examples, along which objects 80, 90 are conveyed. The grooves provide means for guiding the objects 80, 90 along the conveying direction FR, whereby movement of the objects 80, 90 transversely to the conveying direction FR is restricted. Three sorting actuators 60 arranged one behind the other are shown along the conveying direction FR, i.e., one primary sorting actuator 62, one secondary sorting actuator 64, and one tertiary sorting actuator 66 per groove. The sorting actuators 60, as in Fig. 3The sorting actuators 60 each comprise a lifting magnet. Elastic membranes are located above the sorting actuators 60, each representing an operating area 70 of the respective sorting actuator 60. The elastic membranes advantageously seal the conveying device 20 and also enable the transmission of a shock, i.e., an impulse. The membranes can, for example, be fixed on one side by a screw connection and preferably have a freely movable end. When a plunger or shaft, which is connected to a plunger core of a sorting actuator 60, is extended, as shown in Fig. 3As shown, the plunger or shaft strikes the membrane and deflects it, thereby deflecting and, in particular, sorting out the object above it. The membrane is shown in its removed state at the first of the sorting actuators 60 in the conveying direction FR, i.e., at the primary sorting actuator 62. In further exemplary embodiments, the sorting actuator 60 can include a voice coil motor that deflects the objects in the same way as described in connection with Fig. 3 In the manner described above, it distracts and, in particular, filters out.

[0138] Since solenoids and voice coil motors exhibit a comparatively short response time compared to other mechanical actuators, sorting actuators 60 that incorporate solenoids and / or voice coil motors advantageously enable the use of relatively fast mechanical sorting actuators 60 in a sorting system 10, particularly for pulse-like sorting. This allows for the precise sorting of material flows containing relatively small objects, a capability typically only achievable with pneumatic sorting actuators 60. Compared to pneumatic sorting actuators, mechanical sorting actuators 60 that include solenoids and / or voice coil motors also significantly reduce the energy consumption of the sorting system 10.

[0139] As an optical detection device 40, for the in Fig. 3The illustrated example phototransistors are integrated into the respective conveyor device 20, i.e., into the respective groove, whereby the phototransistors preferably serve as a light barrier and detect passing objects. This advantageously enables a particularly simple sensor system.

[0140] Figure 4a Figure 1 shows an actuator position / time diagram of a sorting actuator 60 according to an embodiment of the present invention, by means of which an exemplary operating principle of a sorting actuator 60 is described.

[0141] Fig. 4a Figure 1 shows an example sorting actuator cycle of a simplified sorting actuator model, where the sorting actuator 60 is in its initial position (actuator position "Ready"). The sorting actuator cycle begins from this initial position with the activation of the sorting actuator 60. From activation onwards, the sorting actuator 60 is activated for a period of time t, possibly taking into account a control-related delay.a deflected (actuator position: "Activate"), whereby the sorting actuator 60 does not yet exert any influence on its surroundings. In the case of intermittent mechanical actuators, this is achieved, for example, by the actuator moving up below the surface of the conveying device 20; in the case of pneumatic actuators, this includes, for example, the time required to open the valve and subsequent air inertia.

[0142] Subsequently, the sorting actuator continues to ramp up for a period of time (actuator position: "Ramping up"), during which the sorting actuator 60 already exerts an influence on its surroundings, but successful sorting cannot yet be expected. For intermittent mechanical actuators, this occurs, for example, when the sorting actuator 60 ramps up above the surface of the conveyor 20, which alone is not enough to transmit the full impulse. For pneumatic sorting actuators 60, this includes, for example, the time until the compressed air cone has sufficiently developed. If an object is struck during this time, it is unpredictable which fraction the object will end up in, i.e., whether it can be reliably sorted out.

[0143] Subsequently, a sorting actuator 60 fulfills its intended function for a time th (actuator position: "sorting out"), i.e., during this time one or more objects that are in the area of ​​influence 70 of the sorting actuator 60 are very likely to be sorted out.

[0144] The aim of the sorting system 10 and in particular the control of the sorting actuators 60 by the sorting system 10 is to hit an object 80 classified as to be sorted out as early as possible in this phase. InIn exemplary embodiments, the time or duration th can be part of a minimum activation period of a sorting actuator 60, starting from a basic state of the sorting actuator 60 in which the sorting actuator 60 does not reliably sort out passing objects from the material stream, via the actuator position "Sort out" in which the sorting actuator 60 sorts out objects within an area of ​​operation 70 of the sorting actuator 60, back to the basic state in which the sorting actuator 60 does not reliably sort out passing objects from the material stream.

[0145] The sorting actuator 60 then retracts for a time or duration td (actuator position: "Shutdown"), during which time it continues to exert an effect on its environment. As with tu, it is unpredictable which fraction an object will end up in if it crosses the operating area 70 of the sorting actuator 60 during this time interval. For example, intermittent mechanical sorting actuators 60, this time results from the retraction of the sorting actuator 60; for example, pneumatic sorting actuators 60, it results from the dissipation of the compressed air cone.

[0146] InIn further exemplary embodiments, the times or durations tu and td, in addition to the time or duration th, can each be a part of the minimum activation period of a sorting actuator 60, starting from a basic state of the sorting actuator 60 in which the sorting actuator 60 does not reliably sort out passing objects from the material stream, via the actuator position "Sort out" in which the sorting actuator 60 sorts out objects within an area of ​​operation 70 of the sorting actuator 60, back to the basic state in which the sorting actuator 60 does not reliably sort out passing objects from the material stream.

[0147] The sorting actuator 60 then resets for a period of time tr. During this time, the sorting actuator 60 cannot be used for sorting. After this time has elapsed, the sorting actuator 60 is ready for use again. In the case of intermittent mechanical actuators, this time results, for example, from the sorting actuator 60 resetting to its initial position, according to the actuator position "Ready"; in the case of pneumatic actuators, it results, for example, from inertia in the valves.

[0148] The time or duration t a + t u This represents a prediction gap for sorting actuator 60. If sorting actuator 60 is to sort out an object at time t1, then an algorithm or the sorting logic of the sorting system 10 must be in place by time t1-t at the latest. a -t u Make this decision and control the sorting actuator 60 accordingly.

[0149] The time or duration t u + t h+ t drepresents the intervention time or duration of the sorting actuator 60, i.e., the time during which the sorting actuator 60 exerts an effect on its environment, and in particular on an object within the area of ​​influence 70 of the sorting actuator 60. The time t a + t u + t h + t d+ t r describes the cycle time of the sorting actuator 60, which specifies the frequency with which a sorting actuator 60 can be cyclically activated and subsequently deactivated.

[0150] The lengths of the times or durations t a , t u, t h, t d, t r are particularly dependent on the type of sorting actuator (e.g., pneumatic, intermittent mechanical, or similar), are therefore sorting actuator-specific and are considered predetermined. For some sorting actuators 60, e.g., a pneumatic sorting actuator 60, t h, that is, the time of the sorting actuator 60 in the sorting position, which is also not fixed but can be extended as desired. The time or duration t h This can, in particular, be part of a minimum activation period and be extendable specific to the sorting actuator in order to allow the sorting out of objects 80, 90 encountered or arriving at the operating area 70 of the sorting actuator 60 beyond the minimum activation period. Furthermore, a position of the respective sorting actuator should preferably be specified and provided to the algorithm or sorting logic of the sorting system 10.

[0151] The Figures 4b and 4c based on the Fig. 4a , so that the statements regarding Fig. 4a correspondingly also for the Figures 4b and 4c apply. Figures 4b and 4c However, in comparison to each other and to Fig. 4a find out how the times or durations change. a , t u , t h, t d, tr sorting actuator-specific changes can be made, and in particular the duration in the sorting position t can also be changed. h in preferred embodiments it can be extendable, as for example in flaps, as in the Figures 2b and 2c shown, or as may also be the case with pneumatic sorting actuators, as in the Figures 1a, 1b and 1c shown.

[0152] While in Fig. 4b If a particularly shortened time or duration tr is shown, then for individual sorting actuators 60, in particular, even a negligible reset time t may be required. r to achieve this, so that the sorting actuator 60, for example, is already shut down shortly or directly after the time or duration t d is ready to be reactivated and, in particular, deflected.

[0153] Figures 5a, 5b and 5c show exemplary actuator position / time diagrams of sorting actuators, where Fig. 5aThis exemplifies an actuator position of a single sorting actuator 60, and the Figures 5b and 5c in particular, the actuator positions of several sorting actuators 60 according to an embodiment of the present invention. The illustrations in the Figures 5a, 5b, 5c , 6a, 6b, 6c , 7a, 7b and 7c These can be exemplified as groove sorters, where all objects 80, 90 along the conveying direction FR can pass one or more sorting actuators 60. The ones with the Figures 5a, 5b, 5c , 6a, 6b, 6c , 7a, 7b and 7c However, the relationships described are not limited to a groove sorter, but rather illustrate the case where two objects pass a sorting actuator 60 or several sorting actuators 60 arranged one after the other downstream, unless they are sorted out beforehand.

[0154] While in Fig. 5a For example, a single sorting actuator 60 is shown to sort out the two objects 80, 80 to be sorted out. Figures 5b and 5c Two sorting actuators 60 are provided spaced apart in the conveying direction FR. For the Figures 5a, 5b and 5c The example assumes mechanical sorting actuators 60, which apply a pulse to sort out an object 80, 90. For the Figures 5a, 5b and 5c The object 80 that is closer to the sorting actuators 60, 62, 64 is designated as the first object 80, and the object 80 that is further away from the sorting actuators 60, 62, 64 is designated as the second object 80.

[0155] Fig. 5a Present an exemplary case of two objects 80, both of which are to be sorted out based on their object class, where the time of arrival of each object 80 is plotted on the time axis by points P1A1 and P2A1. The nomenclature for the diagrams of Figures 5a to 7cThe time frame is chosen such that the time of arrival of an object is represented as P1 for the first object and as P2 for the second object, and the arrival time at the operating area 70 of the respective sorting actuator 60, 62, 64 is represented as A1 for the individual sorting actuator 60 or the primary sorting actuator 62, and as A2 for the secondary sorting actuator 64. Accordingly, the first object 80 conveyed in the conveying direction FR towards the sorting actuator 60 arrives at the operating area 70 of the individual sorting actuator 60 at time P1A1. The second object 80 conveyed later in the conveying direction FR towards the sorting actuator 60 arrives at the operating area 70 of the individual sorting actuator 60 at time P2A1. As in Fig. 5a As shown, the time interval between the two objects 80, 80 arriving at the operating area 70 of the sorting actuator 60 is greater than one sorting actuator cycle of the sorting actuator 60.

[0156] As through Fig. 5aAs shown, only one of the first object 80 and one of the second object 80 can be reliably sorted out by the sorting actuator 60. The second object 80 passes through the sorting actuator 60, even though it should also be sorted out due to its object class.

[0157] As through the Figures 5b and 5c As illustrated by the two distances P1A1 - P2A1 and P1A2 - P2A2, the objects 80 and 80 have approximately the same speed and therefore arrive at the respective sorting actuators 62 and 64 with approximately the same time interval.

[0158] The Figures 5b and 5cThe diagram illustrates two methods for successfully sorting the two objects 80 and 80, using two sorting actuators 60 (a primary sorting actuator 62 and a secondary sorting actuator 64) as examples. The solid lines in the diagram represent the position and movement of the primary sorting actuator 62. The dashed lines in the diagram represent the position and movement of the secondary sorting actuator 64. This representation of the position and movement of the primary and secondary sorting actuators 62 and 64 applies to the Figures 6b, 6c , 7b and 7c analog.

[0159] Success criterion in the case according to Figure 5b The primary sorting actuator 62, which was used to sort out the first object 80, no longer has any influence on the second object 80 at the time it passes through. Success criterion in the case according to Figure 5c The primary sorting actuator 62, which is intended to sort out the second object 80, does not, upon its activation, exert any further influence on the first object 80 that passed shortly before. The time reserves available for both cases are denoted as "tres" in the Figures 5b and 5c drawn in the Figures 5a, 5b and 5c was, as was particularly in the Figures 6a to 7c also, assuming that an object should always be hit in the middle of the duration in which the respective sorting actuator 60, 62, 64 is in the sorting position, i.e., in the middle of th.

[0160] Is the behavior of the respective sorting actuator 60, 62, 64, as exemplified in the Figures 5a, 5b and 5c As shown above the surface of the conveyor device 20, axially symmetric to the point of impact, in this case to the midpoint of th, both strategies for sorting out the two objects to be sorted 80, 80, as shown in the Figures 5b and 5cThe same time reserve tres has been shown. The minimum time interval between two sorting operations (tres = 0) cannot be increased by increasing the number of sorting actuators used, but is directly dependent on the intervention time (t). u+ t h+ t d ) of the sorting actuators. However, by adding more sorting actuators, the number of objects 80, 90 that can be sorted out consecutively with minimal intervals can be advantageously increased, since the effect of the reset time tr can be reduced or eliminated.

[0161] In the exemplary case of the Figures 5a, 5b and 5cIn the case of pneumatic sorting actuators 60, using two sorting actuators 60 instead of one sorting actuator 60 would not provide a direct advantage, since the sorting actuator 60 can remain in the sorting position for a longer period of time, i.e., the duration th can be extended to such an extent that a single sorting actuator 60 sorts out the two objects 80, 80 to be sorted out.

[0162] Figures 6a, 6b and 6c show exemplary actuator position / time diagrams of sorting actuators 60, where Fig. 6a for example, represents the state of a single sorting actuator 60, and the Figures 6b and 6c in particular reproduce states of several sorting actuators 60 according to an embodiment of the present invention.

[0163] Figures 6a, 6b and 6c The figures show, by way of example, a first object to be sorted out 80, followed by an object not to be sorted out 90, with a pulsed mechanical sorting actuator 60 ( Fig. 6a) and an arrangement according to the invention of two intermittent mechanical sorting actuators 60, each with a primary sorting actuator 62 and a secondary sorting actuator 64 ( Figures 6b and 6c ).

[0164] As shown by the diagram in Figure 6bAs shown, the first object 80 can be sorted out by the primary sorting actuator 62 at time P1A1, in the center position with respect to duration th, in the sorting position. However, as illustrated by time P2A1, the second object 90 arrives at the operating area 70 of the primary sorting actuator 62 while it is still partially deflected. With a mechanical, intermittent sorting actuator 60, it can be assumed with a high probability that the second object 90 will not be sorted out; nevertheless, an undesirable influence on the second object 90 cannot be ruled out. If, however, mechanical flaps or a pneumatic sorting actuator 60 are used, the sorting out of the second object 90 cannot be excluded.In other words, there is a certain probability that the second non-sortable object 90 will be unwanted, as bycatch, sorted out together with the first sortable object 80.

[0165] As through the Figures 6b and 6c As illustrated by the two distances P1A1 - P2A1 and P1A2 - P2A2, the first object 80 has a higher speed than the second object 90, so that the two objects 80, 90 arrive at the secondary sorting actuator 64 with a larger time difference than at the primary sorting actuator 62.

[0166] As shown by the diagram in Figure 6cAs shown, the first object 80 can be sorted out by the secondary sorting actuator 64 at time P1A2, midway through the sorting phase with respect to duration th. As illustrated by time P2A2, the second object 90 arrives at the operating area 70 of the secondary sorting actuator 64 while it is in the reset phase, and therefore no passing object is sorted out.

[0167] As a comparison of Figures 6b and 6cAs becomes clear, based on the speed difference between the two objects 80 and 90, a sorting actuator 60, 62, 64 can be selected and controlled to remove the object 80 to be removed, thus advantageously reducing the probability that the object 90, which should not be removed, is also removed. The speed difference between the two objects 80 and 90 can be determined, ascertained, or estimated, in particular, as the value of a spatiotemporal correlation criterion between the two objects 80 and 90. As can be seen from the Figures 6b and 6c Therefore, by way of example, a sorting system 10 which takes into account a spatial-temporal correlation criterion between objects 80, 90 and which also has sorting actuators 60, 62, 64 spaced apart along the conveying direction FR, advantageously reduces the risk of bycatch and therefore improves the sorting of objects 80, 90 from a material stream.

[0168] For the in the Figures 6a, 6b and 6c The scenario shown, a sorting system 10 comprising at least one primary sorting actuator 62 and one secondary sorting actuator 64, which is arranged downstream of the primary sorting actuator 62, has an advantage over a system with a single sorting actuator 60, particularly if the time intervals of the objects arriving at the operating area 70 of the respective sorting actuators 60, 62, 64 are greater for one sorting actuator than for the other. Therefore, if it is found that two objects 80, 90 always have the same time interval between them at the respective operating areas 60 of the sorting actuators 60, 62, 64, it can be particularly advantageous if one of the sorting actuators 60, 62, 64 is controlled in such a way that it deflects or influences at least one of the first and second objects, but without sorting it out.

[0169] The example according to the Figures 6a, 6b and 6cThis applies equally to mechanical sorting actuators 60 and pneumatic sorting actuators 60.

[0170] Figures 7a, 7b, and 7c show exemplary actuator position / time diagrams of sorting actuators 60, where Fig. 7a for example, represents the state of a single sorting actuator 60, and the Figures 7b and 7c in particular reproduce states of several sorting actuators 60 according to an embodiment of the present invention.

[0171] The Figures 7a, 7b and 7c The following are examples of the sorting of a non-sortable object 90, followed by a sortable object 80, using a pulsed mechanical sorting actuator ( Fig. 7a ) and an arrangement according to the invention with two intermittent mechanical sorting actuators ( Figures 7b and 7c ), again.

[0172] As shown by the diagram in Figure 7bAs shown, the second object 80 can be sorted out by the primary sorting actuator 62 at time P2A1, in the center position with respect to duration th, in the sorting position. However, as illustrated by time P1A1, the first object 90 arrives at the operating area 70 of the primary sorting actuator 62 while the actuator is already partially deflected, i.e., already deflected relative to the surface of the conveying device 20, such as a chute or groove. Regardless of the type of sorting actuator 60, the risk that the first object 90, which should not be sorted out, will be sorted out as bycatch cannot be ruled out.

[0173] As through the Figures 7b and 7cAs illustrated by the two distances P1A1 - P2A1 and P1A2 - P2A2, the first object 90 has a higher speed than the second object 80, so that the two objects 80, 90 arrive at the secondary sorting actuator 64 with a larger time difference than at the primary sorting actuator 62.

[0174] As shown by the diagram in Figure 7c As shown, the second object 80 can be sorted out at the secondary sorting actuator 64 at time P2A2, in the sorting position, with respect to duration th. As illustrated by time P1A2, the first object 90 arrives at the operating area 70 of the secondary sorting actuator 64 while it is not yet deflected relative to the surface of the conveyor 20, and therefore, with a high degree of probability, will not sort out any passing object.

[0175] As can also be seen from a comparison of the Figures 7b and 7cAs becomes clear, based on the speed difference between the two objects 80 and 90, a sorting actuator 60, 62, 64 can be selected and controlled to remove the object 80 to be removed, thus advantageously reducing the probability that the object 90, which should not be removed, is also removed. The speed difference between the two objects 80 and 90 can be determined, ascertained, or estimated, in particular, as the value of a spatiotemporal correlation criterion between the two objects 80 and 90. As can be seen from the Figures 7b and 7c Therefore, as illustrated by example, a sorting system 10 which takes into account a spatial-temporal correlation criterion between objects 80, 90 and which also has sorting actuators 60, 62, 64 spaced apart along the conveying direction FR, advantageously reduces the risk of bycatch and therefore improves the sorting of objects 80, 90 from a material stream.

[0176] Figure 8 shows an exemplary decision tree according to an embodiment of the present invention, which can in particular serve as a basis for the algorithm or sorting logic of the sorting system 10.

[0177] For the in Fig. 8 The following assumptions shall initially apply to the decision tree shown. 1. "Each object is sorted out by a maximum of one sorting actuator." 2. "The optimal activation time of the sorting actuator is determined solely by the object assigned to the sorting actuator as being sorted out."

[0178] The first assumption makes it possible to assign a specific object to be sorted to a sorting actuator, and allows the decision problem to be reformulated accordingly: The task of the algorithm of the sorting system 10 is now, in particular, to determine for all objects 80, 90 that are located between the beginning of the optically detected area and the end of the optically detected area, i.e., in particular in the detection section 45, whether, when and at which sorting actuator these should be sorted out.

[0179] The second assumption further simplifies the problem insofar as the activation time is now fixed by the selection of the respective sorting actuator 60. It is now sufficient for the algorithm or sorting logic of the sorting system 10 to determine, for all objects 80, 90 conveyed through the detection section 45, whether and at which sorting actuator 60 they are to be sorted out. Such a determination of the activation time could, for example, be such that a sorting actuator 60 is controlled and activated by the sorting system 10 in such a way that an object 80 is sorted out essentially in the middle of the duration t. h , in which the respective sorting actuator 60 is in the sorting position, is sorted out.

[0180] When determining an activation time, i.e., a time at which the sorting system 10 controls a sorting actuator 60 to sort out, and / or when determining the time t hIn the sorting actuator 60, which is in the sorting position, information about the geometric dimensions of the respective object 80, 90, as detected by the detection device 40, and in particular an optical detection device 40, can also be included. For pneumatic sorting actuators 60, the activation time at which the sorting system 10 controls the sorting actuator 60 to sort can, for example, be adapted to the geometric dimensions of the object 80 to be sorted by means of a range-wise estimated object velocity. Accordingly, in this case, the geometric dimensions of the object 80, 90 along the conveying direction FR are a further influencing factor on the activation time at which the sorting system 10 controls a sorting actuator 60 to sort.

[0181] Using the aforementioned assumptions, the problem of sorting system 10, which is addressed by means of the algorithm or sorting logic of sorting system 10, can be solved using a decision tree, as shown in Fig. 8The process is represented and visualized. Here, the individual levels represent the respective objects and the possibilities for each level to have an object sorted out by one of the downstream sorting actuators 60, including the possibility that no sorting occurs. The algorithm or sorting logic of the sorting system 10 can be configured, in particular, to evaluate each path and, after a completed or partially completed evaluation, to select the most successful path in order to control the sorting actuators 60 according to the selected path. Other exemplary algorithms or sorting logics can, for example, evaluate only a subset of the paths in the decision tree. Additionally or alternatively, other exemplary algorithms or sorting logics can, for example, not determine the most successful path, but merely select one successful path from a plurality of successful paths, e.g.the first successful path found for sorting out all objects to be sorted 80, in which sorting out objects not to be sorted 90 is preferably avoided.

[0182] As through Fig. 8 Using the example of "Object 1" and "Object 2," the respective objects can be sorted according to A0: not at all, according to A1: at the primary sorting actuator 62, according to A2: at the secondary sorting actuator 64, or according to A3: at the tertiary sorting actuator 66. For this simple example of only two objects 80 and 90, with three sorting actuators 60 arranged sequentially downstream, there are already 16 paths to be evaluated. The evaluation of the paths can be carried out in particular based on the considerations according to the diagrams, as shown in the Figures 5a to 7cThis is illustrated. The complexity increases with an increasing number of objects 80, 90, so that a successful path is selected, for example, by the sorting system 10 based on a predetermined maximum number of consecutive objects 80, 90 in the material flow.

[0183] Figure 9 shows an exemplary flowchart for sorting according to an embodiment of the present invention.

[0184] The in Fig. 9 The exemplary flowchart shown illustrates a general rule-based sorting procedure, analogous to Fig. 8 The further Figures 10 and 11 can be considered special cases of the in Fig. 9 The relationship shown should be considered. The diagram in Fig. 10The diagram shows another flowchart for sorting according to an embodiment of the present invention and is based on the rule that an object 80, 90, provided it is an object 80 to be sorted, is always sorted out at the first available sorting actuator 60, i.e., at the first sorting actuator 60 ready for use in the conveying direction FR at a predicted arrival time. This can be seen as an exemplary simplification of the decision tree from Fig. 8 This can be viewed as a scenario where only one object is considered at a time, meaning the decision tree has only one level, and the decision between the sorting actors (60) is made immediately according to the aforementioned rule. The one in Fig. 10 The depicted internal cycle is therefore repeated for each object 80, 90. With reference to the decision tree from Fig. 8This is to be understood as meaning that the decision tree is traversed from top to bottom, with an intermediate decision for a single object 80, 90 being made immediately at each level, and paths that do not correspond to this decision being ignored in the further course of the process.

[0185] At the in Fig. 10 In the flowchart shown, objects 90 that are not to be sorted are not considered or are ignored as an example.

[0186] Fig. 10 is essentially based on the considerations of the flowchart from Fig. 9 , so that the corresponding result is obtained Fig. 9 also for Fig. 10 This applies. As already mentioned, the flowchart is based on... Fig. 10 The rule is that an object 80 to be sorted is always sorted at the first available sorting actuator 60, i.e., the first sorting actuator 60 ready for use at a predicted arrival time in the conveying direction FR. The prediction can be analogous to the considerations used to determine how to... Figures 5a to 7chighlighted, take place, in particular depending on the position of a respective sorting actuator 60 at the arrival time of an object 80.

[0187] By using the first available sorting actuator 60, a comparatively short prediction phase is advantageously achieved, meaning that a successful path can be determined quickly, since the first available sorting actuator 60 is the one closest to the last detection of a relevant object 80. This ensures reliable sorting of the objects 80 to be sorted, while the computational effort is advantageously low, so that a sorting system 10 that is both accurate and resource-efficient can be provided. Depending on the position of an object 80, a detection section 45 of the detection device 40, and the determination of a new path, it may be possible, for example, to make a second sorting attempt if the first attempt fails.

[0188] The in Fig. 10The reproduced rule is repeated iteratively from the first object 80, i.e. the next object 80 in the direction of transport FR to the sorting actuators 60, to the last object 80.

[0189] In the Fig. 10 The inner loop shown, from "Object i is not sorted out by a sorting actuator already in motion?" to "i = number of objects?", determines the future behavior of the sorting actuators 60, i.e., their movement and position, as shown in particular in the Figures 4a to 7c Illustrates, predicts.

[0190] If a sorting actuator 60 is selected by the above-mentioned rule as sorting out an object 80 at a specific time, then this sorting actuator 60 is for the prediction of the future behavior during its cycle duration, which in particular the times or durations t a, t u, t h, t d and t rwhich can include, is no longer available and therefore cannot be used for sorting out objects 80 that pass through the sorting actuator 60 within the time interval defined by this.

[0191] It should be noted that blocking a sorting actuator 60 for the entire cycle duration of the sorting actuator 60 may not be necessary, for example, in the case of pneumatic actuators and in the case of objects 80 arriving in quick succession, since the sorting actuator 60 may remain in the sorting position for a longer period of time.

[0192] As through Fig. 11 As illustrated, the failure to consider non-sortable objects can lead to 90, as mentioned at the beginning. Fig. 10 mentioned, this can be advantageously counteracted by using the inner loop, as in Fig. 10shown, in addition, the sorting actuators 60 are blocked even for non-sortable objects 90 until the respective non-sortable objects 90 have passed through the effective areas 70 of the respective sorting actuators 60.

[0193] In light of the figures discussed above, and especially considering the decision tree from Fig. 8 Further exemplary flowcharts and rules for controlling the sorting actuators 60 by the sorting system 10 can be formed.

[0194] In exemplary embodiments, spatiotemporal correlations can arise due to different arrival time differences of objects 80, 90 at different sorting actuators 60 (see in particular the Figures 5a to 7c), as well as future-oriented spatiotemporal correlations, by increasing the number of levels of the decision tree during the search for a suitable path. As already mentioned, ideally the entire decision tree would be evaluated for all objects 80, 90 without an intermediate decision for individual sorting actors 60 and thus without repeatedly building the decision tree for different objects 80, 90. To avoid having to build the complete decision tree, in exemplary embodiments the decision tree can be restricted to a depth of, for example, three, then an intermediate decision can be made, and subsequently, based on this intermediate decision, another decision tree of the exemplary depth of three can be built for the next three objects 80, 90.In this case, at least spatiotemporal correlations between three consecutive objects 80, 90 would be taken into account. While the depth of the decision tree can be increased accordingly, it should be noted that this increases the complexity and computation time. Whereas a model, as exemplified by the... Figures 10 and 11 shown, therefore in particular a fast calculation of a large number of paths for successive objects 80, 90 enables, a deeper model in particular allows an accurate consideration of spatial-temporal correlations of several objects 80, 90, which in particular can reduce the probability of bycatch.

[0195] In further exemplary embodiments, the sorting system 10 can implement the sorting logic for controlling the sorting actuators 60, in particular using a stochastic model predictive controller. The goal of model predictive control (MPC) is to find the most optimal sequence of manipulated values, i.e., values ​​that serve as input for the system to be controlled. For this purpose, the controller determines an optimal sequence of manipulated values ​​for the system under consideration over a prediction horizon comprising N time steps at each call. Typically, the first entry or the first M entries of the resulting optimal sequence of manipulated values ​​are used to control the system. The length of the prediction horizon is greater than the time interval between two successive calls to the MPC.On the next call to the controller, the optimal control frequency is calculated again, now encompassing a prediction horizon shifted compared to the previous call. Thus, each call solves an optimization problem comprising N time steps to determine the control sequence. The system to be controlled is generally considered a dynamic, nonlinear, time-varying system. x k +1 = ak ( x k , u k , w k ) modeled, whereby xk the state of the system, UK the system's input signal (which is specified, for example, by the control and / or regulation unit) and wkThe system noise describing modeling uncertainties (at each discrete time k) is denoted. The state of a system here describes the smallest set of variables that completely determine the future behavior of the system if their values ​​at the current time, as well as all current and future input values ​​and disturbances, are known. The (nonlinear, time-varying) function ak maps the previous state, system input, and system noise to the new state at time k + 1. The variables x k , u k and w k These are primarily vector quantities. The future behavior of the system can be modeled using the equation above. As its name suggests, the MPC uses this model to make a decision for optimal control variables based on the simulated future behavior of the system. If the possible control values ​​at any given time are from a discrete (finite) set of control variables, optimization can be performed, for example, by a decision tree search, in which a decision tree is constructed based on the possible control variables until the prediction horizon is reached. Costs are then calculated for each path, and the path with the lowest costs is used as the optimal sequence of control variables. The costs represent the efficiency with which each path fulfills the given control task. They are typically a function of the states and control variables. In a simple case, they are, for example,The result is derived from the squared deviation of the system's trajectory from a target trajectory at any given time (e.g., zero for all times, if desired). For stochastic MPCs, a distinction is made between: Open-loop feedback: The state is predicted purely based on the system model to calculate the optimal actuator sequence; therefore, no state feedback occurs.

[0196] Closed-loop feedback with directly accessible states: This approach assumes that measurements can be taken on the real system at every time step. It also assumes that the complete system state can be measured directly and without uncertainties. The prediction of future behavior takes into account the impact of the availability of such measurements, particularly on system uncertainty. Problems of this type, especially with discrete-value states and manipulated variables, are also known as Markov decision processes (MDPs).

[0197] Closed-loop feedback with inaccessible states: This approach also considers the possibility of future measurements of the real system and incorporates their effects into the prediction of future behavior. However, it is not necessary to measure the entire state perfectly; rather, measurements can be taken from specific points. z k with this via the generally nonlinear, time-variant measurement equation z k = hk ( x k , v k ) with measurement noise vk These problems are related. Problems of this kind, particularly with discrete states and manipulated variables, are also known as partially observable Markov decision processes (POMDP). Such a stochastic controller includes, in addition to the actual controller, a state estimator that estimates the states that cannot be measured directly from the given measurement data.

[0198] Closed-loop feedback, with its inaccessible states, typically best models processes in real-world systems, but it is also the most difficult to solve and therefore requires approximation approaches. In contrast, the other two classes themselves represent simplifications or approximations of the overall system to be simulated, but depending on the chosen system model, costs, and the type of manipulated variables, they can be solved closed-loop in a sufficiently short computation time. For a discussion of these approaches and possible solutions to the resulting optimization problems, see Bertsekas, Dimitri P., "Dynamic Programming and Optimal Control", 4th edition, Vol. 1, Athena Scientific, 2017, and Weißel, Florian, "Stochastic Model Predictive Control of Nonlinear Systems", KIT Scientific Publishing, 2009.

[0199] The following describes how an exemplary embodiment comprising a stochastic multi-stage controller (MPC) for a groove sorter can be implemented. This assumes that images from an area scan camera, which also captures the area of ​​an array of sorting actuators, are available. Multi-target tracking (MTT) or predictive tracking, as described in WO 2015 / 128174 A1, is also used as part of the controller.

[0200] The condition includes the multi-target tracking state. x k MTT< =[ x k (1)T< x k (2)T< ···] T< with the states of the individual objects x k ( i )< . For example, when using a constant-velocity model (CV model), this consists of the object's position and velocity 80, 90, i.e. x k ( i )= [ xu ( i )< ẋ k ( i )< ] T< where ẋ k ( i )< the time derivative of xk ( i)< is denoted. If the associated system model describing the object movement is linear and time-invariant, as is the case, for example, when using a CV model, then it is a linear, time-invariant system with continuously variable states. Assuming no interactions between objects 80 and 90, i.e., if collisions between objects 80 and 90 are not considered (i.e., collisions are neglected), the descriptive system matrices are then directly derived from the chosen model (e.g., CV model), and the individual objects 80 and 90 move independently of each other. If no interaction occurs between an object 80 or 90 and a sorting actuator 60, the subsystem describing the object 80 or 90 is autonomous, meaning no further inputs are active. u k , such as forces or acceleration, on object 80, 90. Furthermore, the state includes the sorting actuators 60. x k ACTORS< =[ x k ( A1)< xk ( A 2)< ···] T< with the states of the individual actuators x k ( Ai )< . The states change accordingly according to the actuator positions of the sorting actuator 60. Fig. 4 and additionally include, for example, the activation time in order to predict a change in the actuator position of sorting actuator 60. The subsystem is therefore nonlinear. Uncertainties in the actuator model can also be modeled in this step using additional noise terms. The input of the respective actuator model is an activation at the predetermined activation time. The actuator model is independent of the object models, just as individual actuator states or actuator positions are independent of each other. A time-continuous and / or value-continuous representation is particularly suitable for the actuator model, e.g., in the form of the position and velocity of a mechanical actuator.

[0201] However, there is an interaction between the sorting actuators 60 and the objects 80, 90, which is also modeled in the system model as explained below. For simplicity, it is assumed that in the event of a "hit," i.e., an object 80, 90 arrives at sorting actuator 60 while the sorting actuator 60 is in the sorting position, the corresponding object 80, 90 is removed. x k MTT< is removed. To continue tracking objects 80 and 90, especially when an object 80 intended for sorting is not sorted and / or when an object 90 not intended for sorting is sorted, the use of track scores and an implementation of multitarget tracking, as described in Pfaff, Florian "Multitarget Tracking Using Orientation Estimation for Optical Belt Sorting", KIT Scientific Publishing, 2019, is also recommended. Alternatively, a change in the state of the object 80 or 90, for example, due to a collision, can be modeled. If the contact is calculated in continuous time, in order to prevent non-negligible errors due to time discretization, this first requires calculating the arrival time or an arrival time distribution of objects 80 and 90 at the respective operating areas 70 of the sorting actuators 60.The overall model is thus modeled by a nonlinear, time-invariant system with both continuous and discrete-valued state components and takes into account object movements as well as the sorting actuators 60 and the interactions between sorting actuators 60 and objects 80, 90. This is a hybrid system, as it describes both interacting continuous and discrete dynamic components and thus exhibits, among other things, a switching character. This is primarily due to the sorting actuator-object interaction. Depending on the level of modeling of the sorting actuator 60, it can itself also be considered a hybrid system. Individual subsystems of the overall model can be assumed to be linear. The input to the overall system consists of the activation times for the individual sorting actuators 60 by the sorting system 10. These are to be determined by the control unit of the sorting system 10.

[0202] The states of objects 80 and 90 are not fully accessible or observable, since, in the example using a camera, only the positions, but not, for example, the velocities of objects 80 and 90, can be measured. Furthermore, camera measurements typically provide the x- and y-positions of objects 80 and 90, of which, in the case of a groove sorter, only the x-positions along the conveying direction FR need to be directly processed, assuming that the groove along which object 80 or 90 is conveyed is known. Additionally, the position measurements typically exhibit a random but time-invariant error, which is included in the measurement noise. v The aforementioned factors can be captured, for example, by a linear measurement model for the respective object 80, 90. However, it should be noted that the measurements are typically not assigned to the states of the individual objects 80, 90. As is common in multi-target tracking, an association problem must be solved. This can be done using the methods presented, for example, in WO 2015 / 128174 A1. A controller of this embodiment can therefore, in particular, integrate the methods presented in WO 2015 / 128174 A1.

[0203] In order to correctly assign incoming measurements to the object states even during or after the sorting of an object 80, 90, especially in the case of a detection section 45 which includes one or more of the sorting actuators 60, further steps can be taken in particular: To model the removal of an object 80, 90 from the material flow (as well as in a slightly modified form using track scores), actual measurements of such an object 80, 90 could first be identified, but then not assigned to a specific state. For modeling, for example, a collision process, it would be possible to consider how the three-dimensional movement of the object 80, 90 or objects 80, 90 affects the planar measurement taken by the camera.

[0204] In both cases, ambiguous or roughly approximate images can be advantageously avoided, for example, by a lateral sorting process according to the invention.

[0205] For the actuators, it can be assumed, for example, that complete measurements of the actuator state of the sorting actuator 60 are available by directly querying the sorting actuators 60. The measurement model of the overall model is a nonlinear model. It should also be noted that with a discrete-time actuator model, due to the varying clock speeds at which new measurements arrive and the frequency at which the control unit of the sorting system 10 simulates the system behavior according to the system equation (this could be chosen to be much higher, since the sorting actuators 60 preferably operate much faster than the clock speed of the optical detection device 40, especially the camera), new measurements are not available at every time k. This is not the case with a continuous-time model.

[0206] The system has mixed discrete-continuous control variables, represented by the index of the actuators to be activated and the activation times. These are to be determined by the control unit of the sorting system 10.

[0207] The costs of a cost function for potentially updating the sorting system 10 and / or for determining cost-reducing control variables by the MPC include, for example, Penalty terms for incorrectly not sorted objects 80 and / or penalty terms for incorrectly sorted non-sorted objects and / or penalty terms for temporal deviations and / or regarding the accuracy of an object 80, 90, such as a deviation of the arrival time of an object 80, 90 at the area of ​​operation 70 of a sorting actuator 60 from the desired or previously estimated or determined arrival time of the respective object 80, 90, and when using uncertainties for arrival times, e.g. probability mass of the predicted arrival time distribution outside the interval th in which the sorting actuator 60 is in a sorting position and / or penalty terms for small time intervals between an effect of a sorting actuator 60 and preceding or subsequent objects 80, 90 that pass the sorting actuator 60.The penalty terms can be passed on, for example, in the form of a time reserve, or, if a predicted arrival time distribution is given, in the form of a probability or uncertainty, and can be used in particular to update sorting system 10.

[0208] Further penalty terms can be calculated using the opening times of valves of pneumatic sorting actuators 60 as an example, especially because of the energy-intensive compressed air consumption.

[0209] For the stochastic MPC in this exemplary embodiment, analogous to the assumptions made in Fig. 8 It was emphasized that the following assumptions were made to simplify the problem: 1. "Each object is affected by a maximum of one sorting actuator." 2. "The optimal activation time is determined solely by the object assigned to the sorting actuator."

[0210] Assumptions 1 and 2 allow the construction of a decision tree, as the control problem simplifies from a problem with discrete-continuous control variables to a problem with purely discrete control variables. In contrast to classical stochastic MPCs, it is advantageous to determine the tree depth not based on the length of the prediction horizon, but rather on the number of objects under consideration (80, 90).

[0211] Due to the uncertainty analysis using stochastic MPC, it is now possible to include information subject to uncertainty (e.g., in the arrival times and actuator states of the sorting actuators 60) in the decision regarding the control of a sorting actuator 60 by the sorting system 10, and accordingly to select a sorting actuator 60 with a high probability of success in sorting out an object 80. However, if the uncertainty for a successful outcome appears very high for all sorting actuators 60, several sorting actuators 60 can be assigned to an object 80, 90 by relaxing assumption 1. This is particularly advantageous if the sorting actuators 60 involved do not need to be available for further sorting for a certain period of time.In this case, the control and / or regulation unit would be allowed to choose multiple paths of the decision tree simultaneously, should this reduce the costs according to the cost function.

[0212] A potential disadvantage of Assumption 2 is that, for example, slight shifts in the activation times of a sorting actuator 60 can lead to slightly higher time reserves for subsequent objects 80 and 90; however, this type of consideration of interactions between objects 80 and 90 is prevented by Assumption 2. As already mentioned in Fig. 8 As described, assumption 2 also allows for an adjustment of the activation time and / or the time t. h , in which the sorting actuator 60 is in the sorting actuator position, to the spatial extent of the object 80, 90 along the conveying direction FR.

[0213] The following describes the control problem and the stochastic MPC, assuming both assumptions 1 and 2. Based on the explanations above, the control problem can be characterized as follows: nonlinear system model, nonlinear measurement model (with states not fully directly accessible), discrete control variables, generally non-square costs.

[0214] Here, a method from open-loop feedback or closed-loop feedback with states that are not directly accessible is preferably chosen. Open-loop feedback methods exhibit key characteristics, such as... Figures 8 to 11 described, on. The flowcharts and the decision tree according to the Figures 8 to 11 Depending on the specific design, they can be considered a simple form of an open-loop feedback method.

[0215] The control and / or regulating unit of the sorting system 10 can, in particular each time the control and / or regulating unit of the sorting system 10 is called up, perform the following steps: 1. Measuring the current object positions, 2. Querying the current actuator states of the sorting actuators 60, 3. Assigning the measured object positions to the predicted positions (MTT), especially considering the changed behavior due to sorting, 4. Filtering the current object positions (MTT), 5. Simulating the future behavior: Building the decision tree from the control variables for activating the sorting actuators, simulating each path using the system model, in the case of closed-loop feedback with states that are not directly accessible, additionally considering new measurements that arrive every j time steps, calculating the costs for each branch, deciding on a branch, sending the control sequence to the sorting actuators 60 by the sorting system 10, especially by the control unit of the sorting system 10, 6. Using the prediction result for the next j-th time step and the selected control sequence (MTT).

[0216] The step according to point 5 can be adapted in other exemplary embodiments to efficiently calculate the decision tree search (see e.g. Weißel, Florian, "Stochastic Model Predictive Control of Nonlinear Systems", KIT Scientific Publishing, 2009).

[0217] The sorting system 10 configured in this way advantageously enables an optimal, albeit possibly approximate, solution to the problem in terms of the quality criterion / cost.

[0218] The costs chosen here can be significantly more complex than with simple heuristics. The costs can therefore include, in particular: Consideration of both objects to be sorted 80 and objects not to be sorted 90, one or more adjustable trade-offs, no hard decision to consider only objects to be sorted or not to be sorted; and / or reserves or uncertainties regarding the arrival times at an area of ​​operation 70 of a sorting actuator 60 are directly taken into account for both objects to be sorted 80 and objects not to be sorted 90.

[0219] Furthermore, the sorting system configured in particular according to steps 1 to 6 offers the following advantages: that correlated decisions for activating sorting actuators 60 are taken into account based on differences in the arrival times of various objects 80, 90 at the different sorting actuators 60, and / or that correlated decisions for activating sorting actuators 60 based on objects 80, 90 with the same or closely successive arrival times are taken into account completely, i.e., both forwards and backwards in time, and / or that correlated decisions for activating sorting actuators 60 based on the passage of an object 80, 90 at an area of ​​influence 70 of a respective sorting actuator 60 are taken into account, and / or that correlated decisions for activating sorting actuators 60 based on differences in the differences in the arrival times of the objects 80, 90 at the different sorting actuators 60, which arise from different speeds of the objects 80, 90,These factors must be taken into account, and / or that no hard decisions are ever required, because the costs can always be evaluated to determine when it is more advantageous to accept that an object 80 is not entirely successful in terms of a specific criterion, or not at all, in order to maintain the chance of successfully sorting subsequent objects 80, and / or that when choosing a closed-loop feedback method with states that are not directly accessible, the influence of new measurements on the system can also be considered, thus enabling a more realistic system description compared to heuristics or open-loop feedback.

[0220] By relaxing assumption 1 and allowing the assignment of multiple sorting actuators 6 to an object 80, 90, advantageous decisions can be mapped to minimize possible uncertainties, such as the possibility that an object 80 to be sorted is not sorted despite the sorting control of the relevant sorting actuator 60, and furthermore, decisions can also be made depending on possibly subsequent objects 80, 90, for example, that if no object follows, multiple sorting actuators 60 are activated one after the other to increase the probability of successfully sorting out an object 80 to be sorted out.

[0221] In contrast to the exemplary flowcharts, as in the Figures 9 to 11However, as depicted, this represents a comparatively complex algorithm that supposedly results in a greater computational effort and / or longer calculation times. Furthermore, the maximum achievable performance depends on the previously constructed approximate system model of objects 80 and 90, the sorting actuators 60, and the sorting actuator-object interactions, as well as the measurement model.

[0222] The control and / or regulation of the control and / or regulation unit of the sorting system 10 using a stochastic MPC has the following characteristics in common: Models for object movement, sorting actuator, interaction between sorting actuator and object are needed, a general description of the overall system by a hybrid system, definition of costs that reflect the desired result, and a basic process as explained in the last section.

[0223] For all other components of the MPC, significant variations are possible, some of which have already been discussed. The following are possible variations, although this list is not exhaustive: No consideration of uncertainties: A completely deterministic MPC is also possible; Time-variant models (object movement, sorting actuator, sorting actuator-object interaction, measurement model) are useful, for example, if fundamental influencing factors change over time; Different actuator models for the respective sorting actuators 60 (discrete-time or continuous-time, discrete-value or continuous-value), for example, modeling an actuator movement of the sorting actuator 60 as a characteristic curve is a continuous-time and continuous-value representation. In addition, a shift of the system boundary and, for example, the modeling of the mechanics and / or electronics of the sorting actuator 60 and, if applicable, its pneumatics are also possible. The models also differ between the various actuator types, such as mechanical (pulse-like) and pneumatic; Different object-sorting actuator interaction models.Here too, various possibilities exist, ranging from a rudimentary removal of object 80, 90 from the MTT state to detailed modeling, for example, of a collision between several objects 80, 90 or a pneumatic force transmission in continuous time and space; Goal of the optimization problem: In particular, assumptions 1 and 2 need not hold; a joint optimization of activation times and selection of sorting actuators 60 is also possible; Type of control unit: Open-loop feedback or closed-loop feedback with states not directly accessible; Field of view / camera: In exemplary embodiments, and especially in modified form, the use of a line scan camera or an observable area that does not encompass the entire majority of sorting actuators 60 is also possible.

[0224] Simple heuristics for use with the control unit of the sorting system 10 are also possible with a line scan camera and, for example, with a fixed time offset. However, with a fixed time offset, the arrival time difference at all sorting actuators 60 is the same, and no correlations can be exploited in this regard. Furthermore, with simple heuristics, correlated decisions for activating sorting actuators 60 cannot be considered due to differences in arrival times caused by different object speeds, since no speed measurement is possible or the speed is assumed to be the same for all objects 80, 90. Additionally, the depth of the decision tree to be evaluated is very small, as it is determined by the time interval between the line scan camera and an example array of pneumatic sorting actuators 60.Since this interval is deliberately kept small to minimize prediction errors, in most cases only one object (80, 90) can be considered, and in any case, only a very small number of objects (80, 90) can be taken into account. Additionally, the prediction is of comparatively poor quality, as it is based on only a single measurement. Similarly, choosing a fixed time offset is generally inferior to predictive tracking. Because pure prediction is used, and in particular, no camera image can be present in the sorting area, there is no way to react to incorrect sorting.

[0225] If the detection section 45 does not include the sorting actuators 60, a possible failure to sort out an object 80 cannot be detected, which is equivalent to no possibility of reacting to unsorted objects 80.

[0226] In contrast, a detection section 45 of the optical detection device 40, which includes one or more of the sorting actuators 60, for example by providing a camera image section in the sorting area, advantageously enables the identification of a successful sorting operation. In a closed-loop feedback system with states that are not directly accessible, this capability allows for a direct influence on the control and / or regulation of the closed-loop feedback.

[0227] If, for example, a camera image area is still available in the sorting area, this enables the refined estimation or calculation of the state and / or arrival time of an object 80, 90 at the area of ​​influence 70 of a respective sorting actuator 60, especially in the area of ​​the majority of sorting actuators 60.

[0228] Furthermore, the present invention has various advantages when using pneumatic sorting actuators 60, particularly in groove sorters, which can be achieved, for example, with alternating objects to be sorted and not to be sorted 80, 90: Correlated decisions for activating sorting actuators 60 based on the passage of objects 80, 90 can be considered; Correlated decisions for activating sorting actuators 60 based on differences in the arrival times of objects 80, 90 at the different sorting actuators 60, which arise from different speeds of objects 80, 90, can be considered; If assumption 1 is relaxed, it is particularly possible to assign several sorting actuators 60 to one object 80, 90, so that decisions for activating sorting actuators 60 can be made, in particular taking into account minimal uncertainties, or to increase the probability of successfully sorting out an object 80 to be sorted out.However, this also increases the compressed air consumption of pneumatic sorting actuators 60, and is therefore at least energetically inefficient; Depending on the provision of a secondary or tertiary sorting actuator 64, 66 or further sorting actuators, in addition to the primary sorting actuator 62, one or more further opportunities for sorting out an object 80 to be sorted are provided.

[0229] In summary, the control of the sorting actuators 60 should be selected such that the aforementioned correlations are exploited; that is, the use of an area scan camera, detection of the objects 80, 90 in the area of ​​the conveyor 20 in which the sorting actuators 60 are also located, and a stochastic MPC are particularly advantageous. Especially in combination with intermittent mechanical sorting actuators 60, this results in an advantageous combination that makes it possible to sort a comparatively large or dense material flow of objects 80, 90 in an energy-efficient manner.

[0230] Less advantageous, but not entirely out of the question, is the use of a line scan camera for the optical detection device 40, since this would result in all objects 80, 90 being predicted with the same time offsets to the sorting actuators 60. Consequently, neither correlations based on differences in arrival time differences are taken into account, nor can a second chance be realized for unsorted objects 80. However, particularly when using pneumatic sorting actuators, this case of a groove conveyor in combination with a line scan camera offers no advantages from the invention compared to the combination of a line scan camera and a 1D nozzle array. Advantages can be achieved with all other combinations, however.

[0231] In the following, two-dimensional movements of objects 80, 90 over time are permitted, as occur, for example, in a sorting system 10 with a chute or belt sorter as the conveying device 20. For this purpose, the preceding methods and embodiments, designed by way of example for a single spatial direction, are extended to include the additional dimension. Likewise, it is possible to carry out a further extension to model three-dimensional object movements, as may be useful, for example, in drop sorters. With regard to three-dimensional object movements, according to one embodiment of the present invention, sorting in volume with appropriately adapted sorting actuators 60 can be provided.

[0232] The extension of the actuator model of a sorting actuator 60 from Fig. 4This is achieved by a local extension of the sorting actuator 60 as well as by a position of the sorting actuator 60 orthogonal to the conveying direction FR, as is particularly evident in the Figures 2a and 2b illustrated.

[0233] The following is an example of how to extend a logic, particularly in relation to the Figures 10 and 11 executed, based on the rule that an object 80, 90, provided it is an object 80 to be sorted out, is always sorted out at the first sorting actuator 60 available in the direction of object movement, i.e., at the first sorting actuator 60 ready for use at a predicted arrival time, is reproduced: In contrast to the one-dimensional method, not all arrival times at all sorting actuators 60 are determined, since the majority of the sorting actuators 60, as in the Figures 2a and 2bThis is exemplified by the fact that the detection process does not pass by the respective object 80, 90, but only by those objects that are crossed by the predicted trajectory of the object 80, 90. This is repeated for all objects 80, 90 located between the beginning and end of the detection section 45, in particular up to the last array of sorting actuators 60. This procedure, performed for each object 80, 90, includes in particular the following steps: 1. Prediction of the path or trajectory of the object 80, 90 by the plurality of sorting actuators 60, which are arranged in a cascade, in particular downstream of the material flow, using a suitable method; 2. Identification of the sorting actuators 60 that have passed through. Here, tolerances can also be taken into account, for example, which ensure that an object 80, 90 with a correspondingly large selected part of its geometric extent crosses or passes the sorting actuator 60 at its area of ​​effect 70.Accordingly, the geometric dimensions of object 80, 90 are a further influencing factor. 3. Calculation of the arrival times at the sorting actuators 60 passed through using a suitable method.

[0234] Subsequently, analogous to the one-dimensional method, the process iterates over all objects 80, 90, starting with the object furthest downstream of the material flow towards the sorting actuators 60, and assigns the sorting actuators 60 accordingly. In each case, the first sorting actuator 60—in the sense of the sorting actuator 60 with the shortest arrival time—that could sort out an object 80, 90 at the predicted time is selected. All other steps are performed analogously to the one-dimensional variant described above. The method is particularly applicable with a line scan camera and a simple time offset as the arrival time model, assuming no lateral movement.

[0235] In addition to the disadvantages already mentioned for the one-dimensional approach, the two-dimensional approach has the following disadvantages: The actual trajectory of an object 80, 90 is not precisely known; therefore, the predicted trajectory is flawed and uncertain, which can lead to errors in determining the sorting actuators 60 traversed. This is equivalent to disregarding uncertainties regarding the arrival locations of objects 80, 90, particularly concerning the operating ranges 70 of the respective sorting actuators 60; reserves in the arrival locations are not considered; the degree of coverage of a sorting actuator 60, especially an operating range 70 of a sorting actuator 60 and an object 80, 90, is not considered; correlated decisions for activating sorting actuators 60 based on differences in the arrival locations of objects 80, 90 at the various sorting actuators 60, which arise from different speeds of the objects 80, 90, are not considered.

[0236] The extension of the stochastic MPC to the two-dimensional approach is carried out as follows: The multitarget tracking part of the system model needs to be adjusted. The multitarget tracking state x k MTT< =[ x k (1)T< x k (2)T< ···] T< with the states of the individual objects 80,90 x k ( i )< now includes, in addition to the position and / or velocity and possibly other parameters in the conveying direction FR, the corresponding parameters perpendicular to the conveying direction FR. For example, when using a CV model, this now consists of xk ( i )<= [ x k ( i )< ẋ k ( i )< y k ( i )< y k ( i )< ] T< . This presents y A position perpendicular to the conveying direction FR and y a velocity perpendicular to the conveying direction FR. Since objects 80 and 90 move independently of each other in the conveying direction FR and perpendicular to the conveying direction due to the orthogonality of the spatial directions, a separate motion model, e.g., a CV model, can be used for each spatial direction. No further adjustments are necessary to the actuator model of each sorting actuator 60 beyond those previously mentioned for adapting it to the two-dimensional approach. The contact model now also takes into account the spatial extent of the sorting actuators 60, which advantageously means that only those sorting actuators 60 that are passed by object 80 or 90, i.e., by its trajectory, are considered for sorting or non-sorting control. The measurement model must be adapted, in particular, to continue using position measurements in the conveying direction FR and perpendicular to the conveying direction.The control variables for the sorting actuators 60 in the two-dimensional approach remain unchanged. In addition to the one-dimensional variant, the cost function includes, for example, penalty terms for the spatial deviation of the object 80 to be sorted, i.e., for the deviation of the arrival location from a desired hit area within the operating area 70 of a sorting actuator 60 (spatial reserve, e.g., to prevent overlap with other sorting actuators 60 or to prevent an input of angular momentum in the event of partial hits through the operating area 70 of a sorting actuator 60). Accordingly, the spatial extent is a further influencing factor. When using uncertainties for arrival locations, penalty terms can be provided, for example, based on the probability mass of the predicted arrival location distribution outside the operating area 70 of a sorting actuator 60.Penalty terms for small spatial distances between a sorting actuator 60 and preceding or subsequent objects 80, 90 that pass the sorting actuator 60. Either in the form of a spatial reserve or, if a predicted arrival time distribution is given, in the form of a probability. Or, in particular, combined penalty terms for small time intervals and small spatial distances.

[0237] For an exemplary embodiment based on the two-dimensional approach, relaxing assumption 1 appears particularly advantageous, so that an object 80, 90 can be sorted out by several downstream sorting actuators 60. This is especially useful for comparatively large objects 80, 90 and objects 80, 90 whose geometric dimensions, for example, traverse or pass through the operating area of ​​two sorting actuators 60 located side by side transversely to the conveying direction, which can be pneumatic or mechanical, and in particular can be mechanically actuated intermittently.

[0238] In principle, simple heuristics are also possible for the two-dimensional approach, even with a line scan camera, with a fixed time offset and without considering the movement of an object 80, 90 perpendicular to the conveyor direction. This is particularly advantageous because, for example, with sorting actuators 60 arranged perpendicular to the conveyor direction FR, sorting actuators 60 can be selected that are exactly in line with the object position at the line scan camera.

[0239] Nevertheless, with a line scan camera, the local prediction, and thus the selection of a sorting actuator 60, is comparatively worse than with an area scan camera. Furthermore, when using a line scan camera, correlated decisions for activating sorting actuators 60 are not taken into account due to differences in the arrival locations of objects 80 and 90 at the various sorting actuators 60, which arise from the different speeds of the objects 80 and 90. Moreover, the points highlighted for the one-dimensional approach also apply when comparing a line scan camera to an area scan camera.

[0240] Since mechanical sorting actuators 60 and pneumatic sorting actuators 60 behave similarly, especially when operating intermittently perpendicular to the conveying direction FR, there is no significant difference with respect to the component running perpendicular to the conveying direction FR. The same advantages and disadvantages apply as in the one-dimensional analysis. However, it should be noted that in the two-dimensional case, when using pneumatic sorting actuators 60, sorting several objects 80 to be sorted directly one after the other – without switching off the nozzles in between – only occurs in large numbers at very high belt loads. The reasons for this are, firstly, the typically larger proportion of objects 90 not to be sorted compared to objects 80 to be sorted in the material flow, and secondly, the unintentionally increased compressed air consumption when the nozzles are not switched off.In particular, this is also due to the fact that the direct succession of two objects 80, 90 in the same path orthogonal to the conveying direction FR is a rare event in two dimensions, quite unlike in a one-dimensional groove. Therefore, in two dimensions, activation correlations that enable reliable, alternating sorting and passing of objects 80, 90, as well as their distribution among different spatially oriented sorting actuators 60, become especially important. Sorting systems 10 with mechanical sorting actuators 60, as well as those with pneumatic sorting actuators 60, benefit from this.

[0241] For the two-dimensional approach, a combination of an area scan camera and observation of objects 80, 90 in the area of ​​the sorting actuators 60, as well as the use of a stochastic MPC, is particularly advantageous. However, the additional spatial component also yields beneficial effects when using pneumatic sorting actuators 60. Even in the extreme case of using pneumatic sorting actuators 60 with a line scan camera, which in a one-dimensional analysis shows no improvement compared to the combination of a line scan camera and a 1D nozzle array, a slight improvement can now be expected due to a potentially increased spatial resolution.

[0242] Furthermore, the following possibilities and approaches exist for further advantageous embodiments for the reliable sorting of objects 80, 90 in a material stream.

[0243] Besides seemingly simple rule-based heuristics and MPC-based methods, both of which require predefined models of object movement, sorting actuators 60, and possibly a contact model between sorting actuators 60 and objects 80, 90, there is another main class of methods with which the present problem can be solved advantageously. This is a method from reinforcement learning, which, unlike the methods presented so far, does not require any predefined models or not all of the predefined models (object movement, sorting actuator, sorting actuator-object contact).

[0244] Only a definition of costs (similar to MPC, but based solely on observations, not states) is necessary. In these methods, the underlying relationships or models are learned during a training phase. The model attempts to roughly fulfill the task assigned to it, which has been evaluated based on costs, through exploration and exploitation, and subsequently optimizes its performance. The advantage compared to MPC lies in the inherently more precise model definition; however, the disadvantages include the complex procedure, the resource consumption during training, and potentially limited generalizability, resulting in poorer transferability to modified scenarios (e.g., different number, arrangement, or type of sorting actuators, different material, or different transport medium).

[0245] To nevertheless improve the predictive accuracy of the implementations presented here using data-driven methods, learning and / or adaptive methods can be employed for individual components of the heuristic or the MPC. For example, multitarget tracking can be continuously adapted to changing conditions using adaptive methods. This can be achieved either by adaptively adjusting the model parameters of the tracker, as described in document WO 2015 / 128174 A1, or by using a completely data-based model, such as (recurrent) neural networks. Similarly, learning and adaptive methods, such as neural networks and other regressors, can also be used to predict arrival time and location (see, for example, Thumm, Jakob, et al. "Mixture of Experts of Neural Networks and Kalman Filters for Optical Belt Sorting", IEEE Transactions on Industrial Informatics, Vol. 18, No. 6, June 2022, pp. 3724-3733).

[0246] Regarding the spacing of the sorting actuators 60 along the conveying direction FR, it is advantageous to arrange the actuators in series such that an object 80, 90 can generally be optically detected at least once between each array of sorting actuators 60. This particularly facilitates the detection of unsuccessful sorting attempts and, by tracking the objects 80, 90, enables a new, accurate prediction of the arrival time at the subsequent downstream sorting actuators 60.

[0247] The required distance is determined in particular by the expected velocity distribution of objects 80, 90.

[0248] It should also be noted that correlations between decisions regarding the activation of sorting actuators 60 decrease with increasing distance between the sorting actuators 60, or that the consideration of these correlations becomes less reliable due to the increased uncertainties. For example, with comparatively large distances between two sorting actuators 60, it is not possible to determine with a high degree of probability which sorting actuators 60 an object 80, 90 would subsequently pass through. In this extreme case, the multi-actuator system converges to a series of independent sorting systems, each with an array of sorting actuators 60, whose overall performance is generally considered to be lower. For example, such a setup does not guarantee the correct separation of two closely spaced objects 80, 90 to the extent permitted by the described invention.

[0249] In embodiments where, in particular, an area between downstream spaced sorting actuators 60 is not optically detected by the optical detection device 40, it is advantageous to place the sorting actuators 60 as close as possible in time and space along the conveying direction FR, since the uncertainty in the arrival time and location predictions generally increases with increasing distance from the end of an optically detected area.

[0250] Regarding the spacing of the sorting actuators 60 transversely to the conveying direction FR, the highest possible spatial resolution is advantageous, i.e., a narrow, closely packed arrangement of sorting actuators 60. This allows a sorting actuator 60 selected for sorting, with its operating area 70, to be very precisely aligned with the trajectory of the object 80 to be sorted, thus advantageously reducing the risk of bycatch. However, if the uncertainty in the position prediction is significantly higher than the width of the sorting actuators 60, more than one sorting actuator 60 should be activated to sort out a single object 80. This is algorithmically complex (see also the explanations above regarding the relaxation of assumption 1). In practice, the minimum spacing of the sorting actuators 60 transversely to the conveying direction FR depends primarily on the design of the mechanical actuator and / or the fanning of the compressed air cone.It can nevertheless be particularly advantageous to arrange the sorting actuators 60 as closely as possible next to each other transversely to the conveying direction FR. In preferred embodiments, the sorting actuators 60 of an array of sorting actuators 60 are offset from one another, so that several arrays of sorting actuators 60 can form, in particular, a honeycomb pattern of operating areas 70, thereby advantageously enabling reliable sorting of objects 80 to be sorted out, and furthermore, in particular, advantageously reducing the probability of bycatch. Reference symbol list

[0251] 1 Sorting system 2 Conveyor 4 Detection device 6 Sorting actuator 8 Object to be sorted 9 Object not to be sorted 10 Sorting system 20 Conveyor 40 Detection device 45 Two-dimensional detection section 60 Sorting actuator 62 Primary sorting actuator 64 Secondary sorting actuator 66 Tertiary sorting actuator 70 Area of ​​operation 80 Object to be sorted 90 Object not to be sorted FRF Conveyor direction

Claims

1. A sorting plant (10) for sorting objects in a material stream of objects according to object classes, the sorting plant (10) comprising: - a conveyor device (20) for conveying the objects in the material stream of objects; - a detection device (40) for detecting objects in the material stream of objects; - a classification device for classifying the objects detected by the optical detection device (40) into at least two object classes; - a correlation analysis device for determining the value of at least one spatial-temporal correlation criterion between the objects detected by the optical detection device (40); and - a plurality of sorting actuators (60) for selectively sorting out objects from the material stream, wherein the plurality of sorting actuators (60) comprises at least one primary sorting actuator (62) and at least one secondary sorting actuator (64) such that the at least one secondary sorting actuator (64) is arranged downstream of the primary sorting actuator (62) with respect to the material stream such that a first and a second object in the material stream of objects can successively pass the primary and secondary sorting actuators (62; 64) such that they can be sorted out from the material stream selectively by the primary or secondary sorting actuator (62; 64), wherein the sorting plant is designed (10) to selectively control the at least one primary sorting actuator (62) and the at least one secondary sorting actuator (64) depending on the object class of the first object, the object class of the second object, and the value of a spatial-temporal correlation criterion between the first object and the second object determined by the correlation analysis device.

2. The sorting plant (10) according to claim 1, wherein the spatial-temporal correlation criterion between the first object and the second object comprises at least a speed and a position of the first object, and additionally comprises at least one of the following: - a relative speed of the second object relative to the first object and a position of the second object, or - a relative position of the second object relative to the first object and a speed of the second object, or - a relative speed of the second object relative to the first object and a relative position of the second object relative to the first object, or - a temporal distance of the second object from the first object with respect to the position of the first object.

3. The sorting plant (10) according to claim 1 or 2, wherein the sorting plant (10) is designed to control the at least one primary sorting actuator (62) and the at least one secondary sorting actuator (64) depending on the object class of the first object, the object class of the second object, the value of the spatial-temporal correlation criterion between the first object and the second object determined by the correlation analysis device, and a state of the at least one primary sorting actuator and / or the at least one secondary sorting actuator.

4. The sorting plant (10) according to one of the preceding claims, wherein the detection device (40) is configured to detect a continuous region (45) along the material stream, wherein the continuous region (45) preferably extends upstream of the at least one primary sorting actuator (62) to downstream of the at least one secondary sorting actuator (64).

5. The sorting plant (10) according to one of the preceding claims, wherein the sorting plant (10) is designed to control a sorting actuator (60) of the plurality of sorting actuators (60) such that it changes the value of the spatial-temporal correlation criterion between the first object and the second object, preferably without sorting out either of the first object and the second object.

6. The sorting plant (10) according to one of the preceding claims, wherein the detection device (40) comprises a sensor for detecting a two-dimensional detection section (45), wherein the two-dimensional detection section (45) extends along the material stream and transversely to the material stream, in particular to detect objects that are conveyed transversely to the material stream in a way offset from one another, wherein the sensor preferably comprises an area scan camera.

7. The sorting plant (10) according to one of the preceding claims, wherein the plurality of sorting actuators (60) comprises a first array and a second array of sorting actuators (60), wherein the first array comprises a plurality of primary sorting actuators (62) and the second array comprises a plurality of secondary sorting actuators (64), wherein a secondary sorting actuator (64) of the second array is provided for each primary sorting actuator (62) of the first array such that the secondary sorting actuator (64) of the second array is arranged downstream of the primary sorting actuator (62) of the first array with respect to the material stream such that a first and a second object in the material stream of objects can successively pass the primary sorting actuator (62) of the first array and the secondary sorting actuator (64) of the second array such that they can be sorted out from the material stream selectively by the primary sorting actuator (62) of the first array or by the secondary sorting actuator (64) of the second array.

8. The sorting plant (10) according to one of the preceding claims, wherein each sorting actuator (60) of the plurality of sorting actuators (60) has an actuator-specific switching inertia period, wherein the sorting plant (10) is designed to ascertain a temporal distance of the first object and the second object at the region of action (70) of the at least one primary sorting actuator (62), wherein the sorting plant (10) is designed to compare the temporal distance with the actuator-specific switching inertia period of the at least one primary sorting actuator (60), and wherein the sorting plant (10) is designed to control the at least one primary sorting actuator (62) based on the comparison.

9. The sorting plant (10) according to claim 8, wherein, in the case where the temporal distance is shorter than the actuator-specific switching inertia period of the at least one primary sorting actuator (62), or is equal to the actuator-specific switching inertia period of the at least one primary sorting actuator (62), the sorting plant (10) is designed not to control the at least one primary sorting actuator (62) such that it sorts out the first object if sorting out the second object is not intended due to its object class, and the sorting plant (10) is designed to control the at least one primary sorting actuator (62) such that it sorts out the first object if sorting out the first object and the second object is intended due to their object class.

10. The sorting plant (10) according to one of claims 8 or 9, wherein the sorting plant (10) is designed to ascertain a temporal distance of the first object and the second object at a region of action (70) of the at least one primary sorting actuator (62) and at a region of action (70) of the at least one secondary sorting actuator (64), wherein the sorting plant (10) is designed to compare the temporal distance at the region of action (70) of the at least one primary sorting actuator (62) with the actuator-specific switching inertia period of the at least one primary sorting actuator (62) and to compare the temporal distance at the region of action (70) of the at least one secondary sorting actuator (64) with the actuator-specific switching inertia period of the at least one secondary sorting actuator (64), and wherein the sorting plant (10) is designed to control the at least one primary sorting actuator (62) based on the comparisons.

11. The sorting plant (10) according to claim 10, wherein, in the case where the respective temporal distance is shorter than the actuator-specific switching inertia period of the at least one primary sorting actuator (62) and the at least one secondary sorting actuator (64), or is equal to the actuator-specific switching inertia period of the at least one primary sorting actuator (62) and the at least one secondary sorting actuator (64), the sorting plant (10) is designed to control the at least one primary sorting actuator (62) such that the at least one primary sorting actuator (62) deflects at least one of the first object and the second object without sorting it out, if sorting out the first object is intended due to its object class and sorting out the second object is not intended due to its object class.

12. The sorting plant (10) according to one of the preceding claims, wherein the sorting plant (10) is designed, by means of a sorting detection device, to detect the sorting out of an object as a sorting result, and wherein the sorting plant (10) is preferably designed to assign a return value to the sorting result, wherein the sorting plant (10) is updated based on the return value.

13. The sorting plant (10) according to one of the preceding claims, wherein the plurality of sorting actuators (60) comprises at least one pneumatic sorting actuator and at least one mechanical sorting actuator; or wherein the plurality of sorting actuators (60) comprises exclusively mechanical sorting actuators.

14. A method (10) for sorting objects conveyed in a material stream according to object classes, the method comprising the steps of: - providing a plurality of sorting actuators (60) for selectively sorting out objects from the material stream, wherein the plurality of sorting actuators (60) comprises at least one primary sorting actuator (62) and at least one secondary sorting actuator (64) such that the at least one secondary sorting actuator (64) is arranged downstream of the primary sorting actuator (62) with respect to the material stream such that a first object and a second object in the material stream of objects can successively pass the primary and secondary sorting actuators (62; 64) such that they can be sorted out from the material stream selectively by the primary or secondary sorting actuator (62; 64); - detecting objects in the material stream; - classifying the detected objects into at least two object classes; - determining the value of a spatial-temporal correlation criterion between the detected objects; - controlling the at least one primary sorting actuator (62) and the at least one secondary sorting actuator (64) depending on the object class of the first object, the object class of the second object, and the value of a spatial-temporal correlation criterion between the first object and the second object, wherein preferably the step of controlling the at least one primary sorting actuator (62) and the at least one secondary sorting actuator (64) depending on the object class of the first object, the object class of the second object, and the value of a spatial-temporal correlation criterion between the first object and the second object comprises: - controlling the at least one primary sorting actuator (62) and the at least one secondary sorting actuator (64) depending on the object class of the first object, the object class of the second object, the value of the spatial-temporal correlation criterion between the first object and the second object determined by the correlation analysis device, and a state of the at least one primary sorting actuator (62) and / or the at least one secondary sorting actuator (64).

15. The method for sorting according to claim 14, further comprising the steps of: - ascertaining a temporal distance of the first object and the second object at a region of action (70) of the at least one primary sorting actuator (62); - comparing the temporal distance with an actuator-specific switching inertia period of the at least one primary sorting actuator (62); and - controlling the at least one primary sorting actuator (62) based on the comparison, wherein, preferably in the case where the temporal distance is shorter than the actuator-specific switching inertia period of the at least one primary sorting actuator (62), or is equal to the actuator-specific switching inertia period of the at least one primary sorting actuator (62), the method further comprises the steps of: - controlling the at least one primary sorting actuator (62) such that it does not sort out the first object if sorting out the second object is not intended due to its object class, and - controlling the at least one primary sorting actuator (62) such that it sorts out the first object if sorting out the first object and the second object is intended due to their object class.

Images