EP2417501A1

EP2417501A1 - Load-dependent routing in material flow systems

Info

Publication number: EP2417501A1
Application number: EP10722018A
Authority: EP
Inventors: Georg Baier; Konstantin Keutner
Original assignee: Siemens AG
Current assignee: Siemens AG
Priority date: 2009-04-06
Filing date: 2010-03-10
Publication date: 2012-02-15
Also published as: CN102378947A; WO2010115673A1; US20120029689A1; DE102009033600A1

Abstract

The invention relates to a method for determining the route of transport units, especially in material flow systems (e.g. luggage conveyor systems in airports). According to said method, a prognosis is established as to how many transport units arrive at each module (e.g. points, conveyor track) within a window of time, an evaluation function is defined based on the prognosis for each module, an edge weight is allocated to each module according to its load predicted in the window of time, and a route is determined for each transported unit (e.g. piece of luggage) in a temporally successive manner. The method enables an automation of the fine adjustment (tuning) of an installation according to the current and expected load situation.

Description

description

Load-dependent routing in material flow systems

The invention relates to methods for route finding of transport units, in particular in material flow systems. Furthermore, the invention relates to a device and a material flow system for carrying out the method.

Material flow systems should as far as possible achieve the optimum throughput for the transported goods to be transported. For this, material flow decisions, such as e.g. the positions of points, or whether new goods are loaded, are taken so that it does not come to unbalanced or traffic jams. For this purpose, the current occupancy state of the plant and, if present, information about the planned goods to be loaded can be used for a prognosis on which areas of the plant congestion etc. are to be expected. Then this can be counteracted with suitable control strategies.

On the one hand, large material flow systems, such as in airports, have a very complex structure and, on the other hand, are subject to ever-changing transport requirements. In an airport, the passenger and thus also the amount of luggage is distributed very differently over the day and also over the days of the week. Nevertheless, the material flow system should each achieve a high throughput. A crucial way to influence the performance of the plant is to select the appropriate route to fulfill a transport order. The route is determined by the directional choice for a transport unit on a turnout. This selection is made, for example, by routing tables at the switch. Classic material flow systems have controls that determine the direction depending on the destination of the transport unit. This decision is usually determined by the routing tables at the turnouts.

Adjustment prior to commissioning of a material handling facility (e.g., baggage conveyor at the airport), i. Which routing tables the material flow computer should distribute to the controllers under which system states requires a considerable amount of work because different loading scenarios have to be determined and tested for the system.

The object of the present invention is to provide methods for route finding of transport units in material flow systems, which require only little effort for the startup of the material flow system.

The object is achieved by a method for determining the route of transport units, in particular in material flow systems, comprising the following steps: a) Modeling the material flow system into modules, which respectively represent physical elements of the material flow system, wherein a module has a number of transport units, the module within to reach a definable time window is assigned; b) making a forecast of how many transport units arrive within the time window of each module; and c) creating an evaluation function based on the prediction for each module, each module being assigned an edge weight depending on its predicted load in the time window; and wherein d) a route is determined in chronological succession for each transport unit, the route each having one possible route. Lends a short way, based on the edge weight of the modules, is. The method enables automation of the fine tuning of a plant according to the current and expected load situation. Because the plant is no longer due to expected loading scenarios (ie suspected

Load situations) must be adjusted, errors in the selection of loading scenarios for adjustment can be avoided. The load scenarios are no longer selected by trial and error of presumed load situations. The method is adaptive (to changing plant states) and needs no knowledge of expected load data. The planned routes of the transport units are determined by the current or expected system status and can change over time. This self-configuration (self-tuning) of the plant reduces commissioning efforts and costs. In a decentralized system, the self-configuration allows a so-called "plug and convey" of the system.

A first advantageous embodiment of the invention is that each transport unit is assigned a path in the temporal sequence. As a result, the path of a transport unit in the material flow system is still dynamically changeable.

A further advantageous embodiment of the invention lies in the fact that each transport unit is assigned a path, based on the values, of routing tables allocated by the modules in the time sequence, a routing table being dependent on the time. Thus, it can be time-differentiated for a particular destination to determine which routes are used to this destination. A material flow calculator is not necessary for this. This leads to a relief of the material flow computer.

A further advantageous embodiment of the invention is that the method is repeated in irregularly clocked intervals. As a result, there is no synchronization effort in decentralized systems. Furthermore, an oscillation in the system is avoided.

A further advantageous embodiment of the invention is that the forecast is created based on a cyclic information process and an exponential decay process. This eliminates the need for an explicit reservation and release process, thereby increasing throughput. Plants with explicit reservations and shares tend to have low throughput.

A further advantageous embodiment of the invention is that for the prognosis creation, the current route of a transport unit is fixed in a definable time cycle and for all modules along the route in the expected arrival time window of the transport unit, the forecast is increased by 1, and wherein in the specified time cycle for all modules the forecast is multiplied by 0.5.

A further advantageous embodiment of the invention is that for the prognosis creation, the current route of a transport unit is fixed in a definable time cycle and for all modules along the route in the expected arrival time window of the transport unit, the forecast is increased by 1, and wherein in the specified time cycle for all modules the prognosis is multiplied in time successively by values Si to s _k , for 0.5 <S ₁ <1 (i = 1 ... k), where the Product Si * ... * s _{k is} equal to 0.5. This achieves an easy-to-implement smoothing of the decay process.

A further advantageous embodiment of the invention is that the evaluation function for creating the edge weight is determined by an expected standard throughput time of a transport unit on the module and a penalty component per module, which is determined from the predicted number of transport units in the expected entrance window of the transport unit on the module , composed. The evaluation function is composed of two parts, the expected standard throughput time of a transport unit by the module and a load-dependent waiting or penalty time. The exact design of the evaluation function depends on the specific module, but must be coordinated across all modules of a system. The evaluation function is based on two parameters (expected standard lead time and penalty component). This facilitates the material flow system to run in a state below the maximum load, i. All modules have a prognosis smaller than their maximum load at all future times. Congestion and unbalanced loads are thereby avoided.

A further advantageous embodiment of the invention is that the shortest path for a transport unit is determined by the A * algorithm, the Dijkstra algorithm, the Bellman-Ford algorithm, the Floyd-Warshall algorithm or the Johnson Algorithm. These standard algorithms can be easily implemented on control computers for the system.

A further advantageous embodiment of the invention is that a short or the shortest path for a trans- based on distributed algorithms for determining or approximating shortest paths. This facilitates integration into decentralized material flow systems.

A further advantageous embodiment of the invention is that a module forms a self-contained unit with regard to actuators, sensors and control and includes a self-simulator for determining a load forecast for the module, the module can exchange data with its predecessor and successor modules, and wherein the The load forecast for the module is calculated on the basis of the entry times of the transport units at the module as supplied by the predecessor modules. This facilitates integration into decentralized material flow systems and facilitates system configuration in engineering on the basis of technological objects that represent the modules of the system in terms of software.

A further advantageous embodiment of the invention is that the module forwards to the successor modules the exit times of the transport units predicted by the own simulator on the module. This facilitates integration into decentralized material flow systems and enables the system to be self-configured.

The object is further achieved by a method for route finding of transport units, in particular in material flow systems, comprising the following steps: a) Modeling of the material flow system into modules, which respectively represent physical elements of the material flow system, wherein a module is assigned a time-dependent routing table, wherein the routing table for each Destination point of a transport unit contains the next module on the way to the destination or the information that the destination point is not reachable; and b) updating the routing tables. The updating of the routing tables can take place in a specific time frame (cycled), but also asynchronously. It can thus be determined time-differentiated for a particular destination, which paths are used to this destination. A material flow calculator is not necessary for this. The routing tables are always adapted to the new load situation. The method further enables automation of the fine tuning of a plant according to the current and expected load situation.

A further advantageous embodiment of the invention is that the updating of the routing tables is carried out by exact or approximative algorithms for determining the shortest path. For example, The following algorithms can be used: A * algorithm, Dijkstra algorithm, Bellman-Ford algorithm, Floyd-Warshall algorithm or the Johnson algorithm. Standard programs exist for these algorithms; they can easily be implemented on computer systems or controllers.

A further advantageous embodiment of the invention is that the updating of the routing tables is done by self-simulation. This is an advantage when creating decentralized systems.

A further advantageous embodiment of the invention is that the time-dependent routing table is characterized in that the information about the next module on the way to the destination depends on the time at which the transport Transport unit to be forwarded to this module.

It can thus be determined time-differentiated for a particular destination, which paths are used to this destination.

The object is further achieved by a device and a material flow system suitable for carrying out the method. The processes can be implemented in plants with standard components (switches, conveyor belts, etc.) and based on standard hardware. For communication, cable connections (e.g., LAN, Ethernet), but also wireless connections (e.g., WLAN) may be used as computing devices, PLCs, or e.g. Industrial PCs.

An embodiment of the invention is illustrated in the drawing and will be explained below.

Showing:

1 shows an exemplary architecture for the use of decentralized control components using eigensimulators,

2 shows a plant example with modules of a material flow system,

3 shows an example of a first occupation state of the system of FIG. 2,

4 shows an example of a second occupation state of the installation of FIG. 2,

5 shows an example of a third occupancy state of

Appendix of Figure 2, and

6 shows exemplary forecast time windows of the modules mi and m ₂ . On the one hand, large material flow systems, such as in airports, have a very complex structure and, on the other hand, are subject to ever-changing transport requirements. In an airport, the number of passengers and thus the amount of luggage is distributed very differently over the day and also over the days of the week. Nevertheless, the material flow system should each achieve a high throughput. A crucial way to influence the performance of the plant is to select the appropriate route to fulfill a transport order. The route is determined by the directional choice for a transport unit on a turnout. This selection is usually made by routing tables on the switch. That is, in the table, depending on the destination of the transport unit, the direction to be selected by the switch is deposited.

In order to avoid the occurrence of unbalanced loads in such systems with a complex topology and strongly fluctuating load profile, these routing tables are updated according to the system situation. In general, this task is taken over by a higher-level, central material flow computer.

Classic material flow systems have controls which, depending on the destination of the transport unit

Set direction. This decision is determined by a routing table. At a turnout, the goal may be achieved via both way alternatives. In such cases, the directional decision can also be given as a ratio between the right and left route. A ratio of 2: 1 means that with the same destination, two transport units are always steered to the left and then one to the right. The material flow computer is responsible for the Task to update these tables so that the system is always run in an effective state, depending on the current situation. For this purpose, the material flow computer receives, as the central instance, the up-to-date information about the plant status from the lower-level controllers. Adjustment prior to commissioning of the material flow system, which should distribute routing tables of the material flow computer under which system states to the controls, requires a considerable amount of work, since the different loading scenarios must be determined and tested (system tuning).

By contrast, the method according to the invention is based on the production of a chronologically differentiated forecast of the number of expected transport units for a conveying element (module), e.g. Conveyor, switch or merge. For each transport unit to be transported, a route is determined based on a shortest path (possibly with constraints). The evaluation function for determining the path length depends on the predicted load of the modules in the path. The time is discretized in time windows of the same size. The generated forecast always refers to the number of transport units in a time window.

Creation of the forecast

The basis for creating the forecast is the assignment of a route to each transport unit (in the true sense, a path does not have to be assigned to a transport unit, two transport units with the same destination in close proximity can exchange their remaining routes on a common module). Due to the binding of a path to the transport unit, the waited entry time of the transport unit to the modules along the way to be determined. The transport unit can then be taken into account in the associated time window in the forecast for the module. The prognosis, however, does not represent an attempt to determine precisely which transport units passing a module. In particular, it is not a reservation / release mechanism. The forecast is modeled with the help of a cyclic information and an exponential decay process. That is, in a fixed rhythm, the information is sent for each transport unit to the modules of the path associated with the transport unit, that a transport unit will arrive at the calculated time. The forecast values of the modules, which are the number of expected transport units within a timeframe, are scaled by 1/2 at the same rate. To avoid the sudden halving, the decay process can be implemented more continuously, eg multiply n times by l / v2 within one cycle. If the route assigned to a transport unit is changed, then no release procedure for the modules of the old route takes place. If a module is no longer included in the new path, the proportion with which this transport unit was previously counted on the module is reduced by half by scaling within one cycle. That is to say, after five cycles, a portion of 1/2 ⁵ = 0.03125 of this transport unit is still included in the forecast for this module.

If data are available for future transport requests, these can also be taken into account. The expected transport unit will be like all the others

Transport units assigned a path. Only the start time for the route is different, instead of the current time the expected time of loading is used. evaluation function

Based on the forecast for each module, a weighting function is defined which assigns each module an edge weight depending on its predicted load in the time window in question. The weighting function is monotone increasing, increasing with increasing approach to the module's maximum load. However, since the forecast is not an exact prediction of the future load, the function must be well-defined and finite even for values greater than the maximum load. The aim of the route selection is to drive the material flow plant in a state below the maximum load, i. all modules have a forecast lower than their maximum load at all future times. The evaluation function is composed of two parts, the expected standard throughput time of a transport unit by the module and a load-dependent waiting or penalty time. The exact design of the evaluation function depends on the specific module, but must be coordinated across all modules of a system.

routing

The routing, ie the selection of the concrete path along which a transport unit is directed from the start to the destination, is reduced to determining a shortest path with regard to the load-dependent evaluation function. When determining the shortest path, additional constraints must be taken into account, depending on the system. For example, when determining the path, individual modules can be blocked (omitted) on the basis of properties of the transport unit. The particular route is assigned to the transport unit and both used for direction selection in the material flow system as well as for the generation of the forecast.

Due to the dynamics in material flow systems, the prognosis and thus the evaluation function for a module changes continuously. In order to take account of the changed state of the material flow system, the shortest path (possibly with secondary conditions) for a transport unit is regularly redetermined from the current position of the transport unit to the destination. If the path changes, it will become the new way of the world

Transport unit assigned. Indirectly, through the regular information of all modules contained in the path, the prognosis values of the newly contained or no longer contained modules change. The shortest routes should therefore not all be determined or updated at the same time for the transport units in order to avoid the coincidence of the routes.

1 shows an exemplary architecture for a decentralized control component for use in decentralized material flow systems (decentralized material flow systems do not have a central material flow computer). In decentralized plants, the use of self-simulators of the decentralized control components enables the forecast to be generated efficiently and accurately.

FIG. 1 illustrates the architecture of a system module (conveyor belt, transport path, switch, etc.) with control component SK1 having a self-simulation component ES1 suitable for use in decentralized systems.

If the self-simulator ES1 gets access to the path assigned to a transport unit, or at least the successor module ES2 in the way, it has all the necessary data. to create the forecast. For clarification, we assume that the self-simulator ES1 can read the paths from the system state AZ1 and has access to the loading schedule EP. For example, the path can be stored on an RFID tag on the transport unit, via the sensors SE and the

Control component SKl then enters the path to the system state AZl, to which the intrinsic simulator ES1 has access. The self-simulator ES1 determines the effects on its own forecast for all transport units that concern it and transfers to the neighboring inherent simulators ES2 the exit times and paths of the transport units leaving it virtually in the self-simulation. On his part, he also receives from Nachbareigensimulatoren ES2 virtually to him arriving transport units and their associated routes. Once the maximum forecast horizon has been reached, each self-simulator ES1, ES2 has a histogram of the transport units that will virtually pass it in the near future. This histogram is the prognosis of the self-simulator ESl for the future and will be entered in the future system state AZ2.

The control optimizer SO1 has to determine the paths for the transport units on the basis of the future plant state AZ2 predicted by the self-simulator ES1. That is, the control optimizer SO1 must be able to determine a shortest path from the controlled module M to a given destination with respect to the evaluation function and assign it to a transport unit. Since the path assigned to a transport unit is to be recalculated regularly, it also assigns a validity period to a newly determined path. The validity period can be varied within certain limits in order to avoid possible oscillation. The control optimizer SO1 must determine not only a new way for a transport unit when the validity period of the path has expired, but also if the current route has become inadmissible due to other influences, eg by failure of individual modules. Based on the data supplied by the self-simulator ES1 about the new Ania- zenenzanz AZ2 the control optimizer SOl changes the control parameters SP for the control component SKl of the plant module M. The control optimizer SOl is advantageously with the neighbor control optimizer SO2 of the neighboring modules in conjunction, so that the control parameters of Neighbor modules matched to the module M are changeable. This allows a quick reaction to changing plant conditions.

Decentralized path determination

In principle, the decentralized method for determining the paths presented below finds the shortest paths to the evaluation function, namely when the prognosis remains stable. Since these are due to the newly found

Changing paths again, the paths found are only an approximation to the actual shortest paths under the given load. In practice, the paths found prove to be sufficiently good.

The simplest direct way is to use a distributed labeting algorithm with directional tables (Labein) extended by the temporal dimension. An extended direction table for a module M contains a table for each possible target x in the material flow system: Time window 0 1 dist ™ =

Distance 10 15 Successor module ITIQ m ₇

The table dist ™ contains for all forecast time windows, starting with the current time window, the distance to the target as well as the successor module, over which this distance was determined.

If these tables are available on every module, the determination of a shortest path is very easy. If the transport unit is in the current time window = 0 on the module Tn ₁ and has the destination x, the successor module is the module in the column from table dist ™ ¹ , m ₂ designate the corresponding module. Furthermore, module Ui _{1 determines} the entry time and the entry time window t ₂ into the module m ₂ . These values are transferred together with the previous path [In ₁ ] to the module m ₂ . The module m ₂ determines from the

Column t ₂ of Table dist ™ ² the successor module m ₃ , as well as the entry time and the entry time window ts in the module mj-. These values are transferred together with the previous route, [mι, m ₂ ], to module m ₃ . Module m ₃ and all subsequent modules proceed analogously until a module with end point x is reached. If necessary, the generated path, [ifii, m ₂ , TÜ ₃ , ...], is communicated back to module irii.

The creation of the extended direction tables should first be described on a simplified version. It is assumed that all throughput times of a transport unit by a module are a multiple of the time window length. A module regularly updates its flow tation table on the basis of the direction tables of his direct successor modules. Specifically, the following describes how to update the distance table dist ™ to the destination x for a module m. Let Tn ₁ , ..., m _{k be} the direct successors of module m. The cycle time through module m in the time window t is n _t times the time window length and d _t is the length (value of the evaluation function) of module m in the time window t. Then, the distance from module m to the target x, when started within the time window t, is determined

dist ™ (t) = d _t + mm dist ¹ (t + n _t )

where dist ™ (t) denotes the distance entry in column t of the table dist. The value is ∞ if the target x is over the

Module can not be reached. The successor module in column t of table dist ™ is module m _D for index

(2) j = arg min dist ¹ (t + n _t )

At the beginning of the section it was motivated, why it is not necessary, that the determined ways are always exact shortest ways. Therefore, the updates of the direction tables for all modules can be performed independently of each other in a regular rhythm. If the evaluation functions of the modules did not change, the directional tables contained the exact shortest paths after some time. What are the consequences if the throughput times through the modules are not multiples of the time window length? In the extended direction table method, this assumption was used implicitly over the cycle times. When updating according to equation (1), the time window t + n _{t is used} for the successor modules. This is due to the observation that a transport unit which reaches the module m at any time within the time window t will reach the successor module in the time window t + n _t . This property no longer applies if the cycle times are not multiples of the time window length.

To select the entry time window for the successor modules in equation (1), one may choose a specific start time within the time window t, eg the midpoint. The time window t + n _t is then that which corresponds to the entry of a transport unit into the successor modules when the transport unit has reached the module m exactly at the middle of the time window t. This approach is problematic if the cycle times are less than half the time window length. In a sequence mι _r ..., m _k of modules with cycle times less than half the time window length, the method leads to unusable results. Let Δ be the time window length, x the end point of module m _k and the destination the determination of the distance of module irii to x at start in time window t = 0. When updating the direction tables, each module In _{1 will} be n 0 as n _t receive. This means that the entry dist ™ ¹ is the sum of all evaluations of the modules in the time window 0. If a transport unit in module Tn ₁ is currently 0, the throughput times for three modules can not be greater than

3Δ / 2, but a value greater than 2Δ / 2 = Δ is possible. That is, module m _Λ is then reached in time window 1. Along the sequence m _\ , ..., iT3 _k accumulates this error and thus leverages the effect of the evaluation function.

The accumulation of errors can be reduced and the behavior just described can be prevented if the entry time window into the successor module is determined to be somewhat more complex. Let mi and Tn _{2 be} two consecutive modules and τ the cycle time through the module m _x . FIG. 6 illustrates the temporal position of the forecast time windows of the modules Ta ₁ and m ₂ when the transit time τ is smaller than the time window length Δ.

A transport unit loaded at the beginning of time window 0 in module ΛI _I reaches τ time units later module m ₂ . Therefore, the time slots of module m ₂ are shifted by τ time units compared to the time slots of / ni. The transition time from time window 0 to time window 1 at module m ₂ divides the time window 0 of module m _\ into two parts p and q. Transport units loaded in section p or q in module m _\ reach module m ₂ in the time window 0 and 1, respectively. Assuming uniformly distributed loading times in module Jn ₁ , a proportion of p / Δ or g / Δ of the transport units achieves this Module m ₂ in the time window 0 or 1. Applied to the equation (1) is replaced in the minimum of the term dist ™ ¹ (t + m _t ) by ^ dist ^m '(t + m _t ) + -dist ^m - (t + m _t + ϊ), AA

where P ₁ and q _{± are} the parts into which the time window t of module m is decomposed by the time slots t + m _t and t + m _t + l of module In ₁ . Overall, the following new equation results

(3)

disr (t) = d _t + dist? (t + m _t The method described above corresponds to a substantially direct transmission of the Dij kstra algorithm for determining shortest paths to a distributed method. If this method is used in a dynamic environment, as in the case considered here, then a phenomenon known as "looping" occurs.To avoid "looping", different extensions of the distributed labeling algorithm were used, depending on the focus of the boundary conditions developed, for example, the OSPF routing protocol for the Internet.

Allocation and coordination of paths

The assignment of a route to a transport unit can be done in various ways. The following are examples.

version 1

The simplest variant is the explicit storage of the route at the transport unit. For example, on an RFID tag or in a, assigned to the transport unit, software agents. In decentralized systems, this procedure is unfavorable. More natural is decentralized storage in decentralized systems. Has each one Transport unit via a unique ID within the material flow system, this can be exploited. When calculating the route for a transport unit, eg by means of extended direction tables, each point remembers to which successor module the corresponding ID should be forwarded. This procedure can be very well combined with the cyclic information / decay process. In addition to the ID and the direction, the points also store the last confirmation time of the route. With each information of a transport unit about the route usage for the purpose of forecasting, the confirmation time for the corresponding ID is also updated. If the confirmation time of an ID is longer than the cycle length of the information / decay process, the entry is deleted. Each diverter thus has a routing table indexed by ID, which is used both for the direction selection of the transport unit and for the generation of the forecast.

Variant 2

If every transport unit does not have a unique ID within the system, classic routing tables can be used to add a time component. In principle, forecasting does not require an exact assignment of a route to a transport unit. Only the fact that all transport units with the same destination are distributed after a switch within a time window at a fixed ratio to the successor modules (within a time window, transport units with the same destination can be exchanged in their remaining distance) is utilized. Therefore, it is sufficient if a switch within a time window distributes the transport units with the same destination according to a key to the successor modules. That is, the switch When determining a route, eg by means of extended direction tables, only one counter, including its validity, must be remembered for a transport unit. After the validity expires, the counter is cleared again. In order for the transport unit to continue to be included in the forecast, it must at this time prompt a redefinition of "its" path.

A routing table extended by a temporal component in this case is a stand-alone routing table for each future time slot. The routing table of the current time window is used in the classical sense by the controller for the transport units currently to be routed. The values of the routing table of a time window result from the counters mentioned above. For example, there are two possible successor modules Ia ₁ , m ₂ from module ia to destination x and there are U ₁ and n ₂ counters for this purpose. The routing table of the corresponding time window is to be initialized in such a way that a portion of U ₁ / (n ₁ + n ₂ ) transport units is routed via module Ia ₁ and a fraction of n ₂ / (Ti ₁ ^ n ₂ ) over module m ₂ .

The routing tables for the future time slots are thus suggestions for the routing table to be used by the controller when the corresponding time slot has begun. However, the routing tables may still change until the Future Time window starts. In contrast to the variant with IDs, the way remains longer in the case without IDs. In the first case, the validity of a path corresponds to the cycle length of the information / decay process, which is smaller than the recalculation cycle for the paths. This difference has effects only in special situations, such as redirected transport units or the failure of modules. The method described avoids the described disadvantages of methods which have to be adjusted on the basis of different loading scenarios. At the same time, the process anticipates future plant states and counteracts the formation of unbalanced loads through its path selection. In the sense of an adjustment, corresponding to the concrete plant, only three values are to be understood: the choice of the time window size, the maximum number of time windows to be forecasted in the future and the length of the cycles in which the forecast values are scaled or the information about the currently assigned paths be distributed.

In addition, the process can be used in both centralized and decentralized systems. Especially in decentralized systems, the absence of plant-specific parameters is of importance. As a result, all the advantages of decentralized systems, such as reduced engineering costs, fast commissioning, etc., are maintained, while at the same time effective routing as in centrally controlled systems.

The method makes it possible in a simple way to take account of future transport units to be loaded. This also applies to decentralized systems.

FIG. 2 shows a system example with modules ml-ml 4 of a material flow system (eg baggage conveyor system in an airport). The modules ml - ml 4 can be, for example, conveyor belts, switches or mergers. In the upper section (a) of FIG. 2, a system example with modules ml-ml4 is shown, in the lower section (b) of FIG. 2 respective cycle times for the modules ml-ml4 are indicated and possible routes (routes) R1, R2 from a starting point Si, S2 to a target point ti, t2. The method will be clarified using the example of the simple system shown in FIG. In principle, the picture could be the section from a larger layout, and nothing would change in the process. Two types of transport requests are to be fulfilled: orders from Si to ti (dashed line, Rl) and orders from S2 to t ₂ (solid line, R2).

First, some technical assumptions on the system: The system consists of 14 modules ml - ml 4. All modules ml - ml 4 should run at the same speed of 1 m / s and all pieces of luggage (goods in transit, transport units) have along with the minimum distance to the next Luggage a length of 1 m. The transit time of a transport unit through a module is indicated in the lower section (b) of FIG. 2, eg 4 sec for module ml or mlO and 2 sec for module m3 or m7. Calculating the shortest S ₁ -1I ₁ paths for i = l, 2, with respect to the cycle time, we obtain the paths Rl and R2 marked in the lower section (b) of FIG. 2.

The three plant-related parameters are chosen as follows: the time windows have a size of 10 sec, the maximum number of time windows is sufficient with 4 to cover the longest path in time, and the cycle length for updating the forecast values is 5 sec 2 sec, beginning with time 0, transport units are loaded on Si and S2. These parameters, as well as the throughput times for the modules are selected by way of example and are used to illustrate the method.

FIG. 3 shows the state of the system at the time t = 2. The plant has four transport orders. Two for the transport of the transport units xl and x2, as well as two further transport orders for the transport of the transport units yl and y2. The positions of the transport units x1, x2, y1, y2 are shown schematically in the appendix. The situation at module m3 is to be considered more precisely. In FIG. 3, module m3 is used to specify the table of its forecast values for the first three time windows (for the sake of easier comprehension, the beginning of the time slot is indicated in the first row of the table instead of current numbers). The 1 in the first time window comes from the loading of the transport unit xl, the 3 in the second time window comes from the loadings of the three transport units x2, yl and y2.

FIG. 4 shows the state of the system (esp. Situation on module m3) at time t = 4. Two additional transport units were loaded, x3 and y3. Both transport units contribute to increasing the forecast at the module m3 in the second time window. In the second time window (t = 10) a 5 (for predicted 5 transport units at time t = 10 at module m3) thus appears.

FIG. 5 shows the state of the system at the time t = 5. At time t = 5, the first update takes place. For the sake of simplicity, we assume that all modules first scale their forecast values to 1/2, and then all the transport units repeatedly report their current route selection. After scaling, the table of module m3 contains the values 0.5 and 2.5 for the first two time slots. The expected arrival time of x1 at module m3 continues to fall within the first time window, the remaining modules are all expected in the second time window. That is, the corresponding values increase by 1 or 5, as shown in the table to module m3 in Figure 5.

The module m3 has an arithmetic maximum capacity of 10 transport units (for example luggage) within a time window of 10 sec. Let us assume 80% of this as the maximum desired load, that is 8 transport units per time slot. The predicted utilization of module m3 at time t = 5 in the time frame 10 - 20 is over 90%, x = 7.5 / 8 = 0.934. When applying the evaluation function

(4: xf ( ^χ ) = (x <D

1 - x

this results in penalty costs of £ (7.5 / 8) = 15 for the module m3 in this time window. That is to say, when recalculating the shortest paths from S ₁ to t _± for i = l, 2, the paths R3, R4 shown in FIG. 5 result. The next, at time t = 6, to be embraced transport unit y4 of S2 (starting point) to t ₂ (destination point) then gets the new route R3 assigned while avoiding module m3.

The time t = 5 is also the first time at which the associated paths can be updated. However, the affected transport units x1 and y1 are in a position from which there is only one way to the destination. Therefore, the assigned paths will not change. This example, with its multitude of events at time t = 5, shows that it is advantageous not to make the cycles for re-tuning the associated path non-deterministic. Since the cycle for redetermining the assigned path has no influence on the correctness of the forecast calculation, it can be chosen arbitrarily. It offers itself here- to choose the time to the next redetermination randomized.

The evaluation function forms a core element of the method in avoiding congestion situations in material flow systems. It controls which routes are used for transport requests under the current forecast. The evaluation function consists of two parts,

1. the standard throughput times for the modules, as well

2. a penalty component per module, which is determined from the predicted number e _m (t) of transport goods in the expected entrance time window t of the goods at module m.

P is a route and In ₁ , ..., m _{k are} the modules to be traversed along this route. Then the evaluation function w (P, t) determines for the route P.

w (P, t) = W ₁ (P) + W ₂ (P, t)

In this case, t denotes the time of the start of the transport unit along the path P. The first term W ₁ (P) describes the standard transit time for the path P and w ₂ (P, t) the associated penalty component. The evaluation function w (P, t) is dependent on the time window t because of the penalty component at which the transport unit will enter the first module In _{1 of} the path P.

The first component W ₁ (P) results from the sum

W ₁ (P) = ^ ₁ (In ₁ ) + ... + wι (m _k )

If W ₁ (m) denotes the standard transit time of a transport unit by the module m.

The second part w ₂ (P, t) of the evaluation function, the penalty component, is more complex. w ₂ (P, t) is a function that in Ab- depends on the forecast for the number of goods to be transported in a specific time window, which defines penalty costs for this time slot and module. Therefore, this function is, in contrast to W ₁ (P), in addition to the time t 5 parametπ- Siert at which the transport unit is located at the starting point of the Rou ^¬ P th. Be t = t _lf ..., t _k IrII the expected arrival time window ^¬ the transport unit to the modules, ..., m _k (are along the route P. In the case of a jam-free operation, the time window t _jr j = 1,. .., k, thus the times

10 t + Σ _{1 = i} Wi (IH ₁ ) associated time window). Furthermore denote w ₂ (m, t) is the penalty costs for module m, when it is used in Zeitfens ^¬ ter t. Then the second component results

W ₂ (P, t) = w ₂ (m _lr U ₁ ) + ... + w ₂ (m _k , t _k )

15

Ideally, the forecast should be e _m (t), for the module m in the

Time window t, always be smaller than the maximum possible number u _m of Transportgutern which can accommodate the module m within a time window. The evaluation function 20 w ₂ (m, t) offers a function that approaches infinity when approaching the maximum value. Let x = x _m (t) = e _m (t) / u _m, the relative use of module m in the considered time window t ^¬, then

25 f (x) = ---

1 - x

A suitable function; w ₂ (m, t) = x _m (t) / (1-x _m (t)). Since the prognosis is not determined in the sense of an exact calculation of the expected transport goods, the relative utilization x can also be greater than 1. The function f (x) must be adjusted accordingly, for example, f (x) for x> x ₀ (X _Q = a threshold value <1) must be replaced by the linear Appro ^¬ ximation of f (x) at x = Xg Overall, for the evaluation function W ₂ Im, t), expressed in the values of the prediction e _m (t), in this case

for any threshold x ₀ <1.

In principle, the route for a single transport unit to a destination is determined as the shortest route to the destination, possibly taking into account additional constraints. For example, For example, in many material handling systems, certain conveyors may not carry all types of cargo because of insufficient clearance. Any type of shortest path calculation, e.g. Dij kstra-based labeling algortihmen. Less suitable are algorithms which use precalculated data for runtime improvement, since these have to be constantly recalculated due to the time dynamics of the evaluation function.

As described above, the evaluation function takes into account the expected utilization of the modules at the expected entry time of the transport unit. The expected utilization of the modules changes with the loading of new transport units or the transport of the units in the system. To take this into account, the routes of the transport units are redetermined at regular intervals. To prevent oscillation in the route selection, the routes should not be changed in a fixed rhythm. It is better to throw a coin in a fixed rhythm for a transport unit and only redefine the route at the head. In this case, the probability for head or number does not have to be half each at H, but can be any pair (p, 1-p) of probabilities for 0 <p <l. The choice of a route assigns this route only to the transport unit. In particular, there is no direct link between the choice of the route for a transport unit and the prognosis of the modules contained in the route. The forecast values are indirectly changed by the process described in section 2.2. Therefore, changing a route for a transport unit also does not result in any further updates, such as are necessary when using a reservation / release procedure.

Method for route finding of transport units, in particular in material flow systems (eg baggage conveyor systems in airports), wherein a forecast is generated, how many transport units within the time window arrive at each module (eg switch, conveyor line), a rating function based on the forecast for each module Each module is assigned an edge weight as a function of its predicted load in the time window, and a route is determined in chronological succession for each transport unit (eg luggage). The method enables automation of the fine tuning of a plant according to the current and expected load situation.

reference numeral

SKl, SK2 control component

SE sensor AK actuator

AZl, AZ2 Plant condition

EP load plan

SP control parameters

SOl, SO2 control optimizer ESl, ES2 self-simulator

M, ml - ml 4 module xl - x3 transport unit yl - y3 transport unit

Rl - R4 route

Claims

claims

1. A method for route finding of transport units (xl-x3, yl-y3), in particular in material flow systems, comprising the following steps: a) modeling of the material flow system in modules (M, ml-ml4), each physical elements of the material ^¬ flow system a module (M, ml-ml 4) being assigned a number of transport units (x1-x3, y1-y3) intended to reach the module (M, ml-ml4) within a definable time window; b) creating a forecast of how many Transporteinhei ^¬ th (xl-x3, y3-yl) (within the time window in each module M, ml - to arrive ml 4); c) creating a weighting function based on the forecast for each module (M, ml-ml4), each module (M, ml-ml4) being assigned an edge weight depending on its predicted load in the time window; and wherein (d) for each transport unit xl x3, yl-y3) in succession, a route is determined in time, wherein the route in each case the shortest possible path, based (on the ^edge-¬ weight of the modules M, ml - is ml4).

2. The method of claim 1, characterized in that each transport unit (xl-x3, yl-y3) in the temporal From ^¬ each follow a path is assigned.

3. The method of claim 1 or 2, characterized in that each transport unit (xl-x3, yl-y3) in the temporal From ^¬ each follow a path based on the values of the Mo- assigned routing tables, wherein a routing table is dependent on the time.

4. Method according to one of the preceding claims, characterized in that the method is repeated at irregularly clocked intervals.

5. The method of claim 1, wherein the prediction is generated based on a cyclic information process and an exponential decay process.

6. The method according to any one of the preceding claims, characterized in that for the prognosis creation, the current route of a transport unit (xl-x3, yl-y3) is fixed in a definable time cycle and for all modules (M, ml - ml 4) along the route in the expected time-of-arrival window of the transport unit (x1-x3, y1-y3) the forecast is increased by 1, and in the specified time interval for all modules (M, ml-ml4) the forecast is multiplied by 0.5.

7. The method according to any one of the preceding claims, characterized in that for the prognosis creation, the current route of a transport unit (xl-x3, yl-y3) is fixed in a definable time cycle and for all modules (M, ml - ml 4) along the route in the expected time-of-arrival window of the transport unit (x1-x3, y1-y3), the forecast is increased by 1, and in the specified time interval for all modules, the prognosis is multiplied in time by values Si to s _k , for 0.5 <S ₁ <1 (i = 1 ... k), where the product is Si * ... * s _k = 0.5.

8. The method according to any one of the preceding claims, characterized in that the evaluation function for creating the edge weight from an expected standard transit time of a transport unit (xl-x3, yl-y3) on the module (M, ml - ml 4) and a penalty component per module , which is composed of the predicted number of transport units (x1-x3, y1-y3) in the expected time window of the transport unit (x1-x3, y1-y3) at the module (M, ml-ml4).

9. Method according to one of the preceding claims, characterized in that the shortest path for a transport unit (x1-x3, y1-y3) is determined by the A * algorithm, the Dijkstra algorithm, the Bellman-Ford algorithm, the Floyd - Warshall algorithm or the Johnson algorithm.

10. The method according to claim 1, wherein a short or the shortest path is determined for a transport unit (x1-x3, y1-y3) based on distributed algorithms for determining or approximating shortest paths.

11. Method according to one of the preceding claims, characterized in that a module forms a self-contained unit with regard to actuators, sensors and control and a self-simulator

(ES1, ES2) for determining a load forecast for the module (M, ml-ml4), whereby the module (M, ml-ml4) can exchange data with its predecessor and successor modules, and wherein the load forecast for the module (M, ml - ml 4) is calculated on the basis of the entry times of the transport units (xl-x3, yl-y3) at the module (M, ml-ml 4) provided by the predecessor modules.

12. Method according to one of the preceding claims, characterized in that the module (M, ml-ml4) transmits to the successor modules the exit times of the transport units (x1-x3, y1-y3) predicted by the self-simulator (ES1, ES2) from the module.

13. A method for route finding of transport units (xl-x3, yl-y3), in particular in material flow systems, comprising the following steps: a) Modeling the material flow system into modules (M, ml-ml4), which respectively represent physical elements of the material flow system, wherein a Module (M, ml - ml4) is associated with a time-dependent routing table, wherein the routing table for each destination point of a transport unit contains the next module (M, ml - ml4) en route to the destination or the information that the destination point is not reachable ; and b) updating the routing tables.

14. The method as claimed in claim 13, wherein the updating of the routing tables takes place by means of exact or approximative algorithms for determining the shortest path.

15. The method according to claim 13 or 14, characterized in that the routing tables are updated by self-simulation (ES1, ES2).

16. The method according to any one of claims 13 to 15, wherein the time-dependent routing table is characterized in that the information about the next module on the way to the destination depends on the time at which the transport unit (xl-x3, yl-y3) to this Module should be forwarded.

17. Apparatus for carrying out a method according to one of claims 1 to 16.

18. Material flow system, suitable for carrying out a method according to one of claims 1 to 16.