METHOD FOR ADAPTING A PROCESS MODEL FOR OPERATING ONE OF MULTIPLE WORKING MACHINES, IN PARTICULAR HARVESTING MACHINE FOR ROOT CROPS

Abstract

A method is provided by which a particularly good agricultural work result is achieved upon use both at the same location and also at other locations for one or multiple agricultural working machines.

Claims

1. A method for adapting a process model for operating one of multiple working machines which comprise one first agricultural working machine or multiple combined first agricultural working machine, and at least one further working machine, wherein the adaptation takes place on an EDP device, the method comprising: generating, to influence the agricultural work result, at least one first process model output comprising at least one control command for at least one controllable or regulatable functional unit of at least one of the first working machines, in operation, by the first process model running on the first working machine based on at least one first working data set, by which at least one following working data set results, receiving, by the EDP device via at least one interface from the at least one first working machine, which is in particular arranged remotely, during or after the operation thereof, at least a part of a first machine data set comprising: at least a part of the first working data set and at least a part of at least the or one of the following working data sets, and changing, on the EDP device, a basic process model stored in the EDP device, the first process model or the further process model in consideration of: items of information of the first machine data set, a further machine data set of one or multiple of the further working machines, of the first process model and/or a further process model of one or multiple of the further working machines, subsequently providing at least a part of the changed process model is subsequently as the output process model for an operation by the EDP device.

2. The method as claimed in claim 1, wherein the first working data set is formed by first operating parameter data, first machine sensor data, first functional unit data, and/or first working data derived from these data.

3. The method as claimed in claim 1 wherein the following working data set is at least partially formed by following machine sensor data, following functional unit data, following operating parameter data, and/or following working data derived from these data.

4. The method as claimed in claim 1, wherein the output process model is used as the new basic process model.

5. The method as claimed in claim 1, wherein the machine data sets of multiple working machines are each allocated into individual groups for machine-spanning comparability.

6. The method as claimed in claim 1, wherein the machine data sets for adapting the process model are filtered in a filter module.

7. The method as claimed in claim 1, wherein the process model to be adapted is changed in at least one assembly-specific module and this is combined or compiled with further process model modules to form a changed process model.

8. The method as claimed in claim 1, wherein items of adaptation information are recorded via an observer interface of the EDP device for adapting the process model.

9. The method as claimed in claim 1, wherein the process model to be adapted is changed by means of at least one method of artificial intelligence.

10. The method as claimed in claim 2, wherein items of feedback information of an operator or of an observer of the first working machine or one of the further working machines are used for the adaptation.

11. The method as claimed in claim 1, wherein the output process model is validated before the transmission to the first working machine or one of the further working machines in a validation module and/or on a further real or virtual working machine.

12. The method as claimed in claim 1, wherein the output process model is made available at least partially to an or the operator of one of the working machines on a mobile device for app-based generation of action instructions.

13. The method as claimed in claim 1, wherein the EDP device receives items of information to supplement the first or one of the further machine data sets via an auxiliary interface.

14. The method as claimed in claim 1, wherein the machine data sets depicted in a database of the EDP device are supplemented with machine data sets of other working machines depicted in a further database of a further EDP device.

15. An EDP device having at least one computer program product, wherein the computer program product carries out the method as claimed in claim 1.

16. A harvesting machine characterized by a process model; that is changed according to the method as claimed in claim 1.

17. The method as claimed in claim 1, wherein the first agricultural working machine is a self-propelled harvesting machine, or the multiple combined first agricultural working machines are a combination of a tractor with a harvesting machine towed thereby.

18. The method as claimed in claim 1, wherein the EDP device is arranged remotely from the at least one first working machine.

19. The method as claimed in claim 1, wherein the step of subsequently providing at least a part of the changed process model as the output process model for an operation by the EDP device is for transmission to the first working machine and/or one of the further working machines, and/or transmitted in the direction thereof.

20. The method as claimed in claim 2, wherein the first working data set furthermore contains associated feedback, assessment, and/or operating input data.

21. The method as claimed in claim 3, wherein the following working data set furthermore contains associated feedback, assessment, and/or operating input data.

22. The method as claimed in claim 6, wherein the machine data sets for adapting the process model are filtered by assembly in a filter module.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Reference is now made more particularly to the drawings, which illustrate the best presently known mode of carrying out the invention and wherein similar reference characters indicate the same parts throughout the views.

[0054] FIG. 1 shows the control sequence locally on a working machine.

[0055] FIG. 2 shows a schematic diagram of a subject matter according to the invention.

[0056] FIG. 3 shows an explanation of a process model.

[0057] FIG. 4 shows an exemplary embodiment of an adaptation module.

[0058] FIGS. 5 and 6 show analysis data for adapting a regulating mechanism of a process model.

DETAILED DESCRIPTION OF THE INVENTION

[0059] Individual technical features of the exemplary embodiments described hereinafter can also be combined in combination with above-described exemplary embodiments and the features of the independent claims and possible further claims to form subjects according to the invention. If reasonable, functionally identically acting elements are provided with identical reference numerals.

[0060] A process model to be adapted is embedded as follows in a respective working machine, characterized by a box 1. Correspondingly, there are then individual control sequences for each working machine 1, which is equipped with a corresponding system structure, which are identified for the n working machines by 1a, 1b, 1c . . . 1n (see FIG. 2). These control sequences can be provided on working machines of the same type and the same configuration; however, it is provided that different machine types and configurations can also be compared to partially identical assemblies and thus an experience exchange is advantageous on the part of the EDP device.

[0061] In block 2, operating parameter data are provided due to operating parameters, which are defined via a human-machine interface or also detected by sensors. These can be harvesting strategies, harvested crop sizes, types, or various soil consistencies for the field to be processed and operating goals. A queried process feedback of an operator which is triggered via specific events can also be stored in the block 2. Process feedback data or feedback data or items of feedback information are used synonymously for the purpose of this application. For example, harvesting strategies oriented to quality or throughput, optimized for loss or optimized for admixtures come into consideration as harvesting strategies, on the specification of which specific settings of the functional units and the control units thereof are performed. The process feedback data can be performed, for example, in a time-controlled manner and feedback can take place on higher-order goals, for example, the satisfaction of the operator with the harvesting result or also feedback on part-specific goals (for example, how satisfied they are with the separation in the separating device 1 (TG1), in the separating device (TG2) of the screening performance on specific screening belts, etc.). These assessments are preferably conceivable in at least two levels (good/poor). The inputs are preferably input by the operator directly on the machine. The input of process feedback can, on the one hand, take place triggered by the machine controller, for example, after defined harvesting time frames or larger setting changes by the process model in block 7 or they can take place non-triggered, i.e., the driver gives feedback when it suits him. It is relevant that this feedback input is linked in each case to specific process states, i.e., to the sum of items of information describing the respective state of the machine. This can be, for example, via an item of time information or via a direct linkage of the items of feedback information to the sensor data and operating parameters provided at the point in time of the feedback. If a complete set of operating parameters is not provided at the beginning of an operating procedure of the working machine, this can be queried or specified internally or externally.

[0062] While block 2 involves items of specified or in particular also external information, in block 4, sensor data of the sensors provided on the machine are collected. These can be all types of sensors which are provided on a working machine. For example, they are mechanical, optical, or acoustic sensors. These can also be provided at all conceivable positions at which items of process information can be obtained. For example, these are fill level sensors in the area of the screening belts or separating devices or pressure sensors for determining the drive pressure of hydraulic drives. In addition, these can be sensors which directly ascertain the process feedback. The advantage of the sensorial acquisition over the acquisition of the process feedback via inputs of the operator in block 2 is that, on the one hand, a more objective assessment base is provided over the subjective, operator-specific feedback, in particular more high-frequency information, and, on the other hand, process feedback can be continuously collected by means of the sensors. The sensor-based feedback collection additionally has the advantage that the operator is not unnecessarily distracted or fatigued. The sensor data can be updated either as items of raw information and raw data directly in the process model in block 7. They can alternatively or additionally also be processed and interpreted in a signal processing module or interpreter module according to block 5, also called a sensor data interpreter, and converted, for example, into other state variables describing the process. This applies in particular to items of 2D or 3D information which generally have to be interpreted in separate evaluation devices.

[0063] For example, an item of feedback information can first be generated about the presence of a clog or about an imminent clog at a separating device in block 5. A further example of the sensor data provided in block 4 are items of location information and/or speed information, which are obtained via internal position sensors. Data from blocks 2, 4, and 5 are therefore considered to be working data. Items of information derived from these data, in particular items of process state information describing a process state, can also be considered to be working data. These can result due to a categorization or a statistical evaluation of a large number of simultaneous or successive sensor data.

[0064] The items of information compiled in a working data set are temporarily stored in a data set module 3. In the data set module 3, in which preferably all working data sets are stored before they are transmitted to the EDP device, process model outputs and/or measures of an operator which have the result that process model outputs are overridden are each stored associated with the working data sets or separately. It is advantageously configurable which data can be extracted from the sensor data and operating parameters, in order to store them in the data set module 3. Those process data, using which statements can be made about the process state at a defined point in time, in particular the period of time immediately before and after process feedback, are relevant. In particular, these data sets can be allocated and/or compressed with respect to their memory size so that they can be transmitted by means of a telemetry unit 6 to an EDP device, characterized by a box 20 in FIG. 2. Accordingly, the data sets are transmitted according to arrow 14 upon definable events, for example, reaching a minimum size, connection to a communication medium, for example, upon provision of a mobile wireless connection or the presence of a USB stick, via an interface from the system according to arrow 14 to the EDP device 20. In addition, items of information can also be introduced externally into the data set by the telemetry unit 6, for example, items of weather and environmental information according to the arrows 12 and 13. For example, these can be items of environmental information such as soil type and soil consistency, which are transmitted, for example, from external maps.

[0065] It is also advantageous if, in the case of a telemetry unit 6, in particular in the form of a wireless interface, which functions via mobile wireless, WLAN, or Bluetooth, preferably in the close range of the working machine, a process observer 13a (FIG. 2) connects from outside the working machine and, for example, views camera images or monitors other sensor data and gives feedback thereto. Such feedback can also again preferably be stored according to data arrow 12 in the data set module 3 and linked to the respective working data sets which describe the associated process scenario.

[0066] In particular, time stamps or also defined events on the basis of, for example, an event counter are used for the linkage of the individual data set elements.

[0067] A process model is transmitted to the process module 7 via the communication interface or telemetry unit 6 according to the data input arrow 13. This process module 7 can be located centrally on a control device and distributed onto multiple control devices. The process model thus consists of one or multiple software components, which are incorporated in the machine controller or are executed for the machine controller. In particular, the input signals on the part of block 2 (operating parameters) and the sensor data (blocks 4 and 5) are converted into control commands on the basis of the process module 7. These measures for controlling the working machine can be transmitted by the process module 7, on the one hand, to external control units in the block 9, for example, a tractor of a towed harvesting machine, and to internal control and functional units (block 8). The data of the functional units are possibly also added as functional unit data in a respective working data set in the data set module 3, which is not shown in the present case in the graphic.

[0068] In block 8, the individual functional units or assemblies of the working machine and its control units are realized. In particular, these are actuator activations that influence the process result of the working machine, in the example of a potato harvesting machine thus all cleaning and separating devices from the receiving of the crop flow to the bunker or a transfer device. Such functional units are often connected via a CAN bus system or hardwired directly by the control device. A tractor unit is connected, for example, via the TIM protocol (module 9).

[0069] According to the invention, a method for adapting a process model on the level of the EDP device now runs in block 20 as described hereinafter. Firstly, a collection of all machine data sets is performed in a data collection module 21. These are machines of various types in this case (cf. machine type 1, machine type 2, machine type 3, machine type 4, machine type X). The control procedures 1a to 1n run on these machines corresponding to FIG. 1. Different process modules 7 are accordingly implemented on the respective process modules 7 of these working machines, which can be processed at least in parts jointly on the EDP device due to their fundamentally similar structure and due to identical assemblies, however. For example, the respective working machines are various embodiments of potato harvesting machines, beet harvesting machines, each in towed or self-propelled variant, and/or having receptacles for different numbers of potato ridges or beet rows, respectively.

[0070] As an example, a feedback by an operator 15 is shown on the far left only for the machine identified by an operating sequence 1a. The other working machines having the respective process model variants 7a, 7b, 7c, 7d, and 7e could also in principle be provided with items of information by external operators if corresponding interfaces are provided.

[0071] Also by way of example, only the supply of the left working machine in the figure with further items of environmental information, for example items of weather information, from an information source 16 is shown.

[0072] After receiving the data in the data collection module 21, which in particular represents an interface for the connection of the preferably local working machines remote from the EDP device, it can be reconciled using the data set supplementation module 22, to which an auxiliary interface is assigned, whether still further external items of information have to be added to the machine data sets or the respective working data sets. These can be items of weather information, items of classification information, or other external items of information.

[0073] This supplementation is not necessarily to be carried out. In general, the dashed boxes and arrows in the diagram of FIG. 2 represent optional variants, each according to further exemplary embodiments of the invention. In a data set memory 23, in particular designed as a database, the machine data sets of all working machines are then collected. In an optional extraction module 24, the items of data set information of individual machines can be extracted in individual groups, allocated according to assemblies, for example, for the purpose of machine-spanning comparability. That is to say, after the processing steps in the extraction module 24, the data are allocated according to selectable or definable properties, for example assembly or separating focus. Subsequently, the data are prefiltered in the filter module 25 in a next processing step. These filter properties are in particular directly related to the goal of the specific process model development, for example, the process model development for a specific machine type and/or depending on usage goal, harvesting strategy, or usage location or usage year. It is subsequently checked in an assessment module 26 whether the items of data set information are complete and plausible. For this purpose, corresponding properties and requirements for the data sets are stored in the assessment module. These properties are parameterizable and adaptable. The goal is to only let through those data sets for further adaptation of the process model in the adaptation module 28 which are consistent and reasonable.

[0074] Instead of the access to the data of the optional modules 24, 25, and 26, alternatively or additionally, the data of the memory module 23 can be directly accessed in an adaptation module 28. In the adaptation module 28 a process model stored in an output base 27 as the basic process model is adapted by methods of artificial intelligence, in particular by reinforcement learning. The new and adapted process model thus generated can be validated in the following step against defined test cases (validation module 29). Subsequently, an adapted process model or at least an adapted process model assembly is provided. This is then formed in step 31, possibly together with further process model modules, into the machine-typical process model, the output process model. Via a feedback loop, the assembly-specific process model generated in the module 30 can in turn be written back into the output base in the module 27. Alternatively or additionally, other process models can also be adapted in the adaptation module 28, for example, the process model of the first or the following working machine. Furthermore, items of adaptation information, for example, developer specifications, can be updated and taken into consideration via an observer interface in the form of an interface module 32.

[0075] The output process model can optionally be generated in block 33 for the app-based generation of action instructions and thus as a mobile device variant, which can be transmitted to a mobile device 34. By means of this mobile device 34, items of feedback information of the operator 15 can in turn also be made in the direction of the adaptation module 28 or also toward the data set memory 23. The output process model is then completely or partially made available specifically by machine type in the variants 7a to 7n to the respective working machines for an adaptation or change of the process model respectively provided on the working machine and for an in particular following operation.

[0076] It is also conceivable that the EDP device in one embodiment variant is located on a local working machine 1 and the method according to the invention is thus executed there. Accordingly, however, one or multiple communication interfaces are then to be provided, via which the relevant data sets can then be transmitted via the further or other working machines.

[0077] A process model adapted in the scope of the invention to the EDP device is explained in still more detail with its individual components in FIG. 3 on the basis of a working machine. The essential components of a process model in this case are the two process knowledge components 41, in which a library having the process states is provided, and the regulating mechanism module 42 having the regulating mechanism with respect to the process states, the process goals, and the measures M, which are transmitted in the form of control commands in block 43 to the external and internal functional units of blocks 8 and 9. The essential task in this exemplary embodiment of the process model is first to ascertain a known process state, subsequently to select a corresponding measure on the basis of the process state and the process goal from the regulating mechanism, which is then converted into the control commands in block 43.

[0078] Upon the initial operation of the process model, for example, after a working machine start, a process state is not yet defined. Initially at least one process state is ascertained in the block 41 solely from the operating parameters originating from the operating parameter module or block 2. An initial process state is insofar provided in block 44 for the beginning of the operation of a working machine. The operating state or process state defined in block 44 is set as the active process state in block 45. Proceeding from this process state 45 and by means of the process goal in block 46, the most suitable measure is ascertained in the regulating mechanism module 42 for the process state provided in block 45, in order to achieve the transition to the process goal.

[0079] In normal operation, the process state is ascertained continuously from the sensor values and the operating parameters (blocks 2, 3, and 4) on the basis of the defined process states. This process state is initially stored in block 45 as a temporary process state. Such a temporary process state can typically be ascertained at quite high frequency, since sensor values change frequently. In order that the regulating system does not become unstable due to excessively frequent state changes, however, it is reasonable to make this temporary process state more stable by filtering of the state change. This takes place in module 47. Time averaging of the data preferably takes place over time periods between 6 seconds and 600 seconds. The control commands resulting due to the regulating mechanism then implemented in block 43 can be overridden manually according to a specification in block 48 by an operator 15.

[0080] An exemplary embodiment of the adaptation of a regulating mechanism or the process model in the adaptation module 28 on the level of the EDP device is shown in FIG. 4. Initially, the adaptation module 28 receives one or multiple machine data sets or parts thereof. These data can accordingly be provided unfiltered from the database from the module 23 or also filtered according to assemblies or plausible/consistent data according to block 26. All included scenarios, for which feedback is provided, are now extracted from these data sets. A scenario is defined in this case by the presence of at least four items of information. A scenario is a combination of a) a process state according to the first working data set, with b) a process state according to the following data set, with c) an associated process model output or measure which has resulted in the transition from the first process state to a following process state, and with d) associated feedback on the part of either the sensors and/or an operating unit.

[0081] After the extraction of the scenarios in the extraction module 51, a total of n scenarios are then provided, characterized by the above-described four items of information I1.sub.x, I2.sub.x, I3.sub.x, I4.sub.x, with x=1 to n. Each of these extracted scenarios is then passed on to a reinforcement learning agent in module 52. This agent uses the process model to be changed, in which already weighted state transitions due to corresponding measures are contained, and learns further on the basis of these scenarios. While it goes through the scenarios, it adapts the weights or probabilities using which the transitions from one state into another are defined on the basis of measures. After the agent has played through all scenarios for all data sets which are made available to the adapter in the adaptation module 28, the finally ascertained weights are used in the state transitions to thus adapt the measure sequence in the regulating mechanism, for example, by reordering if a measure has changed relevantly in relation to the output model. The ascertainment of weighted state transitions takes place in block 53 and the adaptation of the regulating mechanism then takes place in block 54. By means of a developer interface, it is possible to act from block 32 both on the process of the change of the regulating mechanism and on the extraction of the scenarios. The finally adapted regulating mechanism is then stored in block 55 and can be transferred therefrom for validation to the validation module 29. While the regulating mechanism described in FIG. 4 is generally a regulating mechanism based on a graph or tree structure, it is obvious that such a regulating mechanism can also be provided as a precisely mathematically defined regulating mechanism. It can also be a neural network in which weights are adapted based on items of feedback information. In a more precisely defined, rule-based process model, one or multiple regulators are then precisely regulated in dependence on the process variables on the basis of the sensor data and the operating parameter data. The regulating parameters are then changed as described above, for example, in the adaptation module 28 on the basis of the items of feedback information.

[0082] The illustrations of FIGS. 5 and 6 show on the X axis the possible target value setting options, for example, of a hedgehog web speed (FIG. 5) and a height of a stripping shaft in the same separating device (FIG. 6). The relative usage duration of the target value during the harvesting over a defined period of time is plotted on the Y axis, in the present example over a year. The comparison of the two illustrations makes it clear that the setting ranges can be reasonably limited in working operation, in the example of the stripping shaft (FIG. 6) to a value range of 0% to 20% and in the example of FIG. 5 to a value range of 10% to 70%. This is a result of an observation in the adaptation module 28, in which during the adaptation, the setting ranges used often are weighted more strongly than the unused setting ranges, whereupon a restriction of the setting range results.