Method for estimating time delay in network distributed system

Abstract

The invention relates to a method for time delay estimation in a network distributed system, comprising: determining a sensor time delay between a sensor and a control unit based on sensor time information, estimating an actuator time delay between the control unit and an actuator, providing a plurality of control functions, the control functions respectively relating to different estimated total time delays, wherein the estimated total delay comprises a sensor delay, an actuator delay and a control delay, providing a delay correction term based on a comparison between the estimated total delay and the measured total delay, calculating a corrected total delay by using (adding) a delay correction term available at the estimated point in time for the total delay, a control function for which the corrected total latency lies within the latency range of the control function is selected from a plurality of control functions and executed, and a control input is calculated from the selected control function, the control input is transmitted to the actuator via the network, and the control input is applied to the technical system by the actuator.

Description

Background Art

[0001] Optimal control design is often hampered by the presence of unforeseen timing effects in the control chain. The impact of uncertain delays is particularly critical in distributed control systems, where the control algorithm and the system to be controlled (i.e., the physical device or process) are separate and therefore require a network for communication, and where the resources of the computing platform may be shared with other applications executing simultaneously.

[0002] Typically, knowledge of the delays that data flows through the control chain will experience is not available before the actual execution of the control device begins. Most models of computational and communication delays in distributed real-time systems available in the literature rely on existing knowledge of the probability distribution of time effects in order to determine upper and lower bounds on the possible delays. Such estimates are typically provided offline, for example, to verify worst-case scalability, while online estimates are typically performed to adapt to variable workloads at runtime, such as in acceptance testing.

[0003] Furthermore, the interval between the estimated best-case and worst-case delays is typically large, and the actual delay value may vary rapidly across different iterations depending on the operating conditions. Current approaches to delay-aware control design either use models to design robust offline control or are limited to responding only to delay variations. Another notable example in the literature is feedback planning, which uses feedback information from monitoring deadline misses and utilization changes to adapt the task schedule, but is generally not targeted for distributed settings with communication delays.

[0004] There is no defined and fail-safe knowledge or prediction of the delays within the distributed control loops that is available in the control function at the time of calculating the control inputs. The resulting delay from the sensor to the actuator can only be reliably determined after the data has been received by the actuator. This knowledge is only available after the control inputs have been calculated. Summary of the Invention

[0005] A first aspect of the invention relates to a method for delay estimation in a networked distributed system having at least one sensor, a control unit, and actuators for controlling a technical system.

[0006] The method of the present disclosure comprises:

[0007] - determining a sensor delay between the sensor and the control unit based on the sensor time information,

[0008] - estimation of actuator delays between the control unit and the actuators,

[0009] - providing a plurality of control functions, each of which is related to a different estimated total delay, wherein the estimated total delay includes sensor delay, actuator delay and control delay,

[0010] - providing a delay correction term at the actuator based on a comparison between the estimated total delay and the measured total delay, the sensor time information and the total delay,

[0011] - calculating a corrected total delay using a delay correction term available at the estimated point in time for said total delay,

[0012] - selecting from a plurality of control functions and executing a control function for which the corrected total delay lies within the delay range of the control function, and calculating a control input from the selected control function,

[0013] - transmitting the control input to the actuator via a network,

[0014] - Applying the control signal to the technical system via the actuator.

[0015] A second aspect of the present disclosure relates to a control system, which includes at least one sensor, a control unit, and an actuator, and is designed to implement the method according to the first aspect (or its embodiments).

[0016] A third aspect of the present disclosure relates to a computer system, which is designed to implement the method according to the first aspect (or its embodiments).

[0017] A fourth aspect of the present disclosure relates to a computer program, which is designed to implement the method according to the first aspect (or its embodiments).

[0018] A fifth aspect of the present disclosure relates to a computer-readable medium or signal storing and / or containing the computer program according to the third aspect (or its embodiments).

[0019] The technologies of the first to fifth aspects can have advantageous technical effects.

[0020] The disclosed technique can provide an optimized decision-making process by providing knowledge of the delay that the data flow through the control chain will experience before the actual execution of the control device is actually initiated. For example, based on the estimated delay, an appropriate control mode optimized for that delay can be selected. This decision-making process can be repeated with each iteration of the control loop in order to adapt the decision-making process to changing operating conditions. However, such knowledge of the delay will realistically be based on estimates, which inherently introduce a certain degree of inaccuracy, which can be balanced by compensation based on measurements.

[0021] The disclosed technique can implement a low-complexity method for estimating control chain delays every k iterations, paired with a feedback mechanism that uses knowledge of the actual delays measured by the actuators in previous iterations to correct the estimate, thereby providing sufficiently accurate information about the delays.

[0022] The techniques of the present disclosure can be used during control design of various distributed systems that include efficient computing resources and where uncertain time effects play a significant role in the sensor-controller-actuator chain.

[0023] The present invention can be used to improve the design of an existing control function when it must be provided on another distributed system where the control function will experience different delays. If the delays encountered in some cases are similar to the original design, the original control design can be retained, and additional operating modes can be added to cover the additional operating modes.

[0024] The technology disclosed herein can be used for V2V communication, for example, for cooperative driving functions such as networked adaptive cruise control, group start (Gruppenstart), cooperative lane merging, intersection management, etc. One example can be a networked driving function for longitudinal motion control. Other examples are controlling / coordinating unmanned vehicles in restricted areas, such as in logistics centers or production facilities, via a local network (local edge, 5G network, ...); transferring (partially) the control algorithms of robotic arms to a local network, for example for manufacturing systems; and lateral and / or longitudinal motion control of vehicles at transportation hubs or other control areas, which is transferred to a local edge or cloud system (e.g., a roadside unit).

[0025] Furthermore, the disclosed technology can be used for robotics, such as swarm applications or floor conveyors. It can also be used for flexible production lines with remote control via a network, such as centralized wireless control. This allows for equalizing communication delays at the actuators, enabling precise, time-critical production. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 A flow chart is shown which illustrates a method for controlling a distributed system according to a first aspect.

[0027] Figure 2 The distributed system of the present technology is schematically illustrated.

[0028] Figure 3 Possible topologies of the present technology are schematically shown.

[0029] Figure 4 Possible topologies of the present technology are schematically shown. DETAILED DESCRIPTION

[0030] Figure 1 A method for delay estimation in a networked distributed system comprising at least one sensor, a control unit, and an actuator for controlling a technical system is disclosed. The distributed system may include a physical device to be controlled / a physical process to be controlled, a sensor, a digital controller, and an actuator. The controller may be executed on a cloud or edge device or on a dedicated processor with significantly greater computing power than a classic embedded device. The sensor, the controller, and the actuator are interconnected via a networked communication infrastructure. The physical technical device to be controlled / the process to be controlled may be included in the distributed system or excluded from it. The sensor may measure at least one characteristic of the technical system, and the actuator may control at least one characteristic of the technical system.

[0031] In a first step 11, a sensor delay between the sensor and the control unit is determined based on sensor time information. The sensor time information may be generated at the time when the data is generated or at the time when the data is sent from the sensor. The time information may include absolute time information, such as a timestamp, or relative time parameters.

[0032] The term sensor delay (or latency) defines the time interval required for a signal to travel from a sensor to a control unit in the network.

[0033] In another step 12, an actuator delay between the control unit and the actuator is estimated (calculated or obtained). Estimating the actuator delay may include calculating the actuator delay or obtaining a pre-calculated actuator delay. The estimated actuator delay may be derived from at least one of historical data, network observations, runtime estimation, calculated delay, and sensor delay.

[0034] The term actuator delay (or latency) defines the time interval required for a signal to travel from a control unit to an actuator in a network.

[0035] In a further step 13, a plurality of control functions are provided, each involving a different estimated total delay, wherein the estimated total delay includes or consists of sensor delay, actuator delay, and control delay. The control delay includes the time interval required by the control unit to calculate the control routine and the estimated total delay. If the sensor delay and actuator delay are large compared to the control delay, the estimated total delay may consist of the sensor delay and the actuator delay.

[0036] The number of control functions can be defined by the range of delays to be covered. A control function or pattern is calculated for a specific delay range or a specific delay band. The number of control functions can be limited to less than ten. Depending on resource awareness and delay variations, this number can be higher.

[0037] In a further step 14, a delay correction term is provided, which is based on a comparison between the estimated total delay and the measured total delay. The measured total delay can be measured at the actuator using sensor time information and information about the arrival time of the signal at the actuator.

[0038] In a further step 15 , a corrected total delay is calculated by using the delay correction term available to the control unit for the estimated total delay, for example including it therein or adding it thereto.

[0039] In a further step 16, a control function for which the corrected total delay is within the delay range of the control function is selected from the plurality of control functions and executed, and a control input is calculated from the selected control function. In other words, the selected control function relates to the corrected total delay or is designed for the corrected total delay.

[0040] In a further step 17, the control input is transmitted to the actuator via the network. The respective estimated total delay of the transmitted control input can be an integral component of the control input and can be recognized by the actuator. Furthermore, the respective estimated total delay can also be provided together with the respective control input, for example as a stamp.

[0041] In a further step 18 , the control input is applied to the technical system via the actuator.

[0042] The method may be performed in a loop. The actuator provides the new delay correction term to the control unit, which uses the new delay correction term to update the delay estimate and repeats the selection of the control function.

[0043] The method 10 may further comprise the following initial method steps which are carried out before the first method step 11 .

[0044] - Receiving detection data and time information of the sensor at the control unit.

[0045] - Calculating the time delay correction term at the actuator and transmitting the time delay correction term to the control unit.

[0046] The method steps described above are then performed. In detail, the steps of estimating actuator delays, calculating the corrected total delay, selecting a control function and transmitting a control input are performed at the control unit.

[0047] Control functions can be classified according to best performance or best resource requirements, and the selection of a control function from a plurality of control functions can begin with the control function with the highest classification and continue with control functions with lower classifications until a control function is identified for which the corrected total delay is within the delay range of the control function. Such use of classification can reduce control delays and thereby improve the robustness and reliability of the distributed system.

[0048] Control units, sensors, and actuators can be synchronized in time, allowing for dynamic compensation of delays between sensors and control units. If a controller or control device is synchronized with one or more actuator / sensor devices, it is possible to dynamically compensate for potentially variable delays between sensors and control units. Furthermore, estimation schemes for computational delays can be integrated. In this case, when generating control inputs, the control unit can provide the transmitted input values ​​with its own time information, such as a timestamp. This can significantly reduce the overall delay that must be accounted for in control design. Furthermore, the performance of closed-loop control systems can be improved.

[0049] The sensor can measure at least one property of the technical system and output detection data to the control unit, and the actuator can receive a control input from the control unit and activate at least one property of the technical system as a function of the control input.

[0050] The following is a diagram showing a control or computing system Figure 2 The features of the present disclosure described in or 3 are also applicable to the method described above.

[0051] Figure 2 A control or computer system 20 is schematically shown in which the techniques of the present disclosure may be used to control a distributed system. The computer system 20 may be implemented in hardware and / or software. Figure 2 , explaining the total delay.

[0052] Distributed systems with edge / cloud computing resources offer numerous advantages, such as more efficient computing resources, more information about multiple input sources, centralized maintenance, and greater flexibility in resource allocation. This design is particularly well-suited for implementing modern control systems, whose increasing complexity requires moving away from traditional embedded solutions. However, one side effect of such systems is higher latency uncertainty in data transmission.

[0053] From the perspective of the controlled system, the total delay experienced by data (due to computation and communication) in the control chain is the main driver of its stability and performance characteristics at runtime, especially when this delay is variable and undetermined.

[0054] The control chain in a networked / distributed environment consists of the following steps:

[0055] 1. A device or technical system 21 is sampled by a sensor 22 and its output is sent to a control unit 23. The control unit 23 is executed on a node (cloud / edge / region / embedded), which may be different from the node / nodes where the sensor 22 and actuator 24 are located.

[0056] 2. The detected data is used as input for the control equations in order to update the internal state (if any) and derive corresponding control commands, which are sent to the actuators 24 .

[0057] 3. The actuator 24 receives the control command and applies the corresponding action to the technical system 21 .

[0058] In such distributed control systems, control chain delays can be balanced through appropriate design of the control devices. Since delays often vary dynamically, a variety of control modes can be provided, each optimized for a specific delay condition. However, their direct applicability is complicated by the inherent uncertainty of the time effects at each of the three steps mentioned above, which are caused not only by computational delays but also by delays introduced through the communication network.

[0059] The entire chain delay can only be determined when the actuator receives data (step 3 above), which of course occurs after the control value is calculated (step 2 above). However, releasing this knowledge before the control function begins (i.e., at the beginning of step 2 above) opens the door to optimized control design. In particular, it becomes possible to preselect the most appropriate control mode to use the chain delay value for the current iteration of the control loop.

[0060] The total delay of the system 20 includes the sensor delay between the sensor 22 and the control unit 23, the actuator delay between the control unit 23 and the actuator 24, and the control delay representing the time required by the control unit 23 to perform the control function, or is composed of the sensor delay between the sensor 22 and the control unit 23, the actuator delay between the control unit 23 and the actuator 24, and the control delay representing the time required by the control unit 23 to perform the control function.

[0061] Figure 3 A control or computer system 20 is schematically shown in which the techniques of the present disclosure may be used to control a distributed system. The computer system 20 is designed to execute commands according to Figure 1 The computer system 20 may be implemented in hardware and / or software. Figure 3The system shown in can be viewed as being designed to perform Figure 1 10. A computer program of the method.

[0062] The delay estimator with actuator feedback presented here is used to solve an optimization problem for a distributed system 20, where the goal is to select the control mode that maximizes / minimizes a given objective function at each iteration (e.g., minimizing mean squared error, minimizing stabilization time, maximizing control performance weighted by runtime complexity, etc.). The delay estimator can also be applied to other scenarios where knowing the runtime delay is of great benefit for making better decisions.

[0063] In its general form, the control means is implemented as a series of control functions ("control modes") m1, ..., m N Each of the control functions is optimized for a specific chain interval or total delay. i The execution time is affected by Limits and explicitly assigns intervals to chain delays In the interval, the control mode can be used with the average performance P i Work. Such performance refers to the goal to be pursued: maximize / minimize. The term "performance" here can be any application / system characteristic that aims to optimize a specific application / specific system.

[0064] Based on performance value P i (and possibly based on other information from the objective function), each control mode can be assigned a (possibly dynamic) preference order o i , that is, as long as the current delay is not within its allowed limit, the preference o i Select the control mode. In general, any control mode m i It is composed of the following components

[0065]

[0066] This system can be safety-critical and potentially destabilized by variable network delays: that is, if the vehicle controller is tuned to a specific delay but experiences a different delay, system performance may degrade, leading to instability. Unfortunately, delay values ​​are subject to a combination of sources of uncertainty due to the computing platform and network communication, making them only partially or nondeterministically known at the time of control command calculation. Better estimates of delays at runtime could be used to select the correct control mode optimized for the system's current state.

[0067] In the following example, a distributed system 20 has a sensor 22, a digital control unit 23, an actuator 24, and optionally a device to be controlled / a process to be controlled or a technical system 21. The control unit 23 is executed on a cloud or edge device or on a dedicated processor with a much greater computing power than a classic embedded device. The sensor 22, the control unit 23, and the actuator 24 are connected to each other via a networked communication infrastructure or network 25. The control unit 23 is implemented as a series of control functions 23a, 23a, ctrl m1, ..., ctrl m2. N Each of the control functions is optimized for a specific delay or a specific delay interval (hereinafter referred to as a control "mode"). Each of these functions is explicitly marked (for example, with an integer).

[0068] The sensors 22, control unit 23, and actuators 24 have a common understanding of time (synchronous clocks), meaning they are time-synchronized. This can be achieved using modern mechanisms (such as the Precision Time Protocol) and can be repeated regularly to avoid the effects of clock drift. Data traversing the control chain is timestamped at the beginning of each communication (sensor to control unit and control unit to actuator). The timestamp generated at the sensor level is transferred to the corresponding control command, which is sent to the actuator 24.

[0069] The local actuator 24 has limited computing power in a distributed system. It can compare the current time with the time-stamped control input to determine the actual sensor-to-actuator delay and, possibly, calculate a correction term based on a simple calculation between the estimated and actual values ​​of the delay chain. The actuator-level application must also be able to send data back to the control node via the network.

[0070] The estimator is implemented on the same node as the control unit 23 or in close connection with the control unit, but with negligible communication delay. At each iteration of the control chain, the estimator is executed strictly before the actual execution of the control function. The estimator can be implemented as a standalone task or as an output module of the control task. The estimator or estimation routine can be implemented in hardware and / or software.

[0071] Figure 2 The structure of the system 20 is shown, in which the component of the chain delay is emphasized. The estimate is provided with a top at the bottom, that is, if τ x is the exact value of any delay, then is its estimate. The proposed mechanism has the following steps, which are performed at each iteration k=0, 1, 2, ... of the periodic control process before executing the control function:

[0072] Measure the delay τ from sensor to control device sc,k .

[0073] Estimating the control device to actuator delay

[0074] For each control mode m i , estimates the response time of the control function taking into account possible interference caused by other tasks executed simultaneously in the node and the additional execution time caused by the estimation process itself

[0075] Combine all the above mentioned components with the correction term ε available in step k provided by the logic at the actuator level k Combined to obtain an estimate of the total control chain delay for each control mode

[0076] Use the estimated delay to select the most appropriate control mode m for the current iteration k i The selected control mode m is then executed i And send the control command to the actuator. If the control command reaches the actuator, the logic at the actuator level will calculate the estimated delay Compared with the actual measured delay τ chain，k Compare, and at the earliest step (ε k+1 ) sends the updated correction term back to the control device for the next step.

[0077] Depending on the chosen planner and communication protocol on the control node (cloud / edge / region / embedded), the expression of the delay estimate can be different. At the start of the estimator execution for the kth iteration, the first element calculated is the sensor-to-controller delay τ sc,k . In the assumption that the clocks of the sensor and control nodes are synchronized, i.e., t s,k is the time when the sensor data has been sampled and t e,k is the starting time point of the k-th estimator execution, from which we can derive

[0078] τ sc,k =t e,k -t s,k (1)

[0079] The next component is estimated and calculated as an upper (and possibly lower) limit. The delay is estimated based on the network state and / or the history of observed delays. This calculation can either be performed by a service external to the application and provided as input to the estimator, or it can be performed by an internal service that has system knowledge about the network state (e.g. through predicted quality of service). Such a service can also estimate the delay based on the delay in previous cycles. Depending on the actual protocol and infrastructure, the calculation may require different schemes and different precisions. In specific settings, the delay in the previous iteration can be used. sc,k history in order to estimate Consider the general expression f for communication estimation c (·), so that the additional modeled uncertainty from the network is considered In the case of Compute upper and lower bounds:

[0080]

[0081] Response time estimation is mode m i is an explicit function of , since each mode can have a different execution time (e.g. due to simpler vs. more complex control designs). In general, the calculation needs to take into account the control mode m i The execution time and the interval Possible interferences due to tasks with higher priority (or due to budget allocations, depending on the planner) may occur in the scheduler.

[0082] It is important to also consider the time required to execute the estimator, i.e., C est (m i ) (either previously measured or added at the end of the calculation by runtime measurements). In its general form, the response time is given as a (recursive) function f r (·) is explained, and the function is related to the execution time, the inherent influence of the planner (Sched) and the response time itself.

[0083] In our calculations, the response time will also be bounded upwards and downwards, which yields or

[0084]

[0085] The most modern algorithms can be used to implement equations (3a-b), such as the classic iterative algorithm or an approximation scheme with polynomial complexity. If this is not guaranteed in the design, the response time can also be used to check the scalability of the selected control mode.

[0086] Therefore, the limit of the total chain delay is and is calculated as follows:

[0087]

[0088] And the estimated median For example it can be obtained as an average:

[0089]

[0090] In equations (4a-b), ε k is the correction term available at the control unit in step k, which has been calculated at the actuator level based on the delay error estimate of the previous step. It has a general expression for the case of information one step late, as follows:

[0091]

[0092] f ε A possible implementation of (·) in the form of a classical (time-discrete) PID design is for example:

[0093]

[0094] where K P , K I , K D is the PID gain, IF(z) is the integral formula, DF(z) is the derivative formula, and T f is the derivative filter period (the last three elements depend on the chosen discretization method). The PID gains must be coordinated in order to obtain stable behavior in the closed-loop regulation loop during delay estimation.

[0095] In order to optimize the testing across the possible control modes, the following algorithm can be provided. It prioritizes the testing of the control mode with the best performance, i.e. from the one with the highest rank value o i The algorithm starts with the control mode of , and passes through the other modes in descending order. The algorithm stops at the first control mode that satisfies the delay constraint.

[0096] 1. Measure the delay τ from the sensor to the control device sc，k (Equation 1)

[0097] 2. Calculate the estimated control device to actuator delay and (Equation 2)

[0098] 3. For each control mode m i , by index o i Sorting

[0099] 4. Calculate the estimated response time and (Equation 3a-b)

[0100] 5.Optional: Check m i Plannability

[0101] 6. Calculate the estimated link delay and (Equation 4a-c)

[0102] 7.if and

[0103] 8. Select m i And end the process

[0104] 9.End if

[0105] 10.End

[0106] exist Figure 3 The control system 20 shown in FIG. 1 and according to the present disclosure can be summarized using the following sequence of actions. These actions can also be considered a method for controlling a distributed system, such as the control system 20.

[0107] The sensor 22 samples the device at step k∈[0,1,2,3,…] and obtains detection data 26, which is expressed as y k Mark. Data y k is provided with a time stamp at the source, ie the data is assigned to a corresponding (absolute) point in time t s,k , at which point in time the detected data has been acquired. The detection data 26 may be generated periodically.

[0108] The sensor data 27 with the detection data 26 and the time information t s，k ,{t s，k ,y k} are sent to the control unit 23 via the network 25. The transmission or sending of the sensor data 26 can be performed periodically. The network 25 can be a wireless network, such as a mobile network such as a 5G network, the Internet, WLAN, etc., or a wired connection, such as CAN, etc. The control unit 23 is located at the edge, cloud, or district. Therefore, there is a certain sensor-to-control device delay or latency τ sc .

[0109] Synchronously with the sensor 22, the actuator 24 sends the time delay correction term 28 to the control unit 23. To calculate the time delay correction term 28, the actuator 24 may have a calculation unit 24a or may have access to a calculation unit outside the actuator 24.

[0110] Before the actual control function is started at the control unit 23, the estimator 29 of the control unit 23 is activated either periodically or in an event-controlled manner, i.e. upon receipt of the sensor data 27 and the delay correction term 28. It then performs the following steps:

[0111] The estimator is based on the current time and the timestamp t of the detected data. s,k Calculate τ sc (Equation 1).

[0112] The estimator estimates upper and lower bounds based on information about the network state and (Equation 2).

[0113] The estimator estimates the upper and lower limits based on the information of the control node and (Equations 3a-b).

[0114] Combine all the above mentioned components with the correction term ε k Add to get the delay and (Equations 4a-c).

[0115] Then choose the best mode (in terms of performance) m i , so that the estimated time delay is within the limits of the mode.

[0116] After these estimation steps, we then use the detected data y k Execute the selected control mode m i .

[0117] At the end of control execution, the control command u is generated k , i.e., the control input 30 to the actuator 24. Then, the control input 30 is sent to the actuator 24 via the network 25 and is compared with the sensor timestamp t s,k and the estimated delay value associated.

[0118] When the actuator 24 receives the control command u k When t , the actuator applies the control command to the device 21. The computing unit 24a at the actuator 24 uses its current time t a,k and sensor timestamps to calculate the actual chain delay, i.e., τ chain，k =ta，k -t s，k Based on this value, the actuator calculates a new updated correction value (Equation 5-6). The loop then restarts from step 1.

[0119] In an alternative implementation, the correction term is calculated by the estimator 29 rather than at the actuator level. In this case, the actuator 24 only has to send the value τ chain，k In addition, the estimator 29 must convert the calculated estimate is stored in a memory in order to correctly calculate the correction term by comparing the calculated estimate with the corresponding actual delay τ chain，k Make a comparison.

[0120] Figure 4 A control system 40 is schematically shown in which the techniques of the present disclosure may be used to control a distributed system. The control system 40 is designed to execute Figure 1 The control system 40 may be implemented in hardware and / or software.

[0121] Figure 4 The context of controlling the (longitudinal) movement of one or more vehicles 41 via a network 42, for example, by edge devices such as roadside units, is schematically shown. The vehicles 41, such as passenger cars, commercial vehicles, electric bicycles, etc., communicate with a control unit 43 via the network 42. The control unit 43 may correspond to a Figure 3 The control unit 23 of the vehicle 41 is implemented accordingly. The sensor 22 and the actuator 24 are implemented in the vehicle 41. The vehicle 41 or a part of the vehicle 41 may correspond to a control unit 23 of the vehicle 41. Figure 3 System 21. Figure 4 In the case shown in , the control goal is to maintain a reference distance 44 from another vehicle 45. The vehicle 45 may be a lead vehicle 41 that follows the vehicle 45 in an autonomous driving mode.

[0122] Figure 4 The illustrated structure is one possible application where the present disclosure can provide significant advantages. With delay-free adaptive control, the closed-loop control system can become destabilized due to network delays or exhibit poor performance. This results in the following constraint: only large distances from surrounding vehicles are possible. Due to the safety-critical nature of this system, a collision is highly likely to occur due to poor performance. If the vehicle controls are set to a specific delay, such as a worst-case scenario, even smaller delays can degrade the resulting performance and lead to instability.

[0123] Therefore, the lack of knowledge about the current delay in the control algorithm has traditionally hindered good performance. In particular, when the control algorithm is located in a zone, edge, or cloud environment, the actual loop delay τ when calculating the control input is unknown or uncertain by design due to the network connection in between. Moreover, the corresponding value of the delay is of a random nature and typically varies over time due to network overload. This dilemma can be overcome by using the proposed pattern-based control scheme with feedback from local actuators. For the relevant delay range, a variety of control patterns Ctrl m1, ... Ctrl m can be designed. N Since information about the current or actual delay is available before the possible control signals for the actuator 24 are calculated in the control unit 43 , the optimal control signal for the current magnitude of the delay can be selected. This decouples the trade-off between robustness of the delay determination and the required performance.

Claims

1. A method for delay estimation in a network distributed system (20), the network distributed system having at least one sensor (22), a control unit (23; 43) and an actuator (24) for controlling a technical system (21), the method comprising: - determining (11) a sensor delay between said sensor (22) and said control unit (23; 43) based on sensor time information, - estimating (12) an actuator delay between said control unit (23; 43) and said actuator (24), - providing (13) a plurality of control functions (23a), each of which relates to a different estimated total delay, wherein the estimated total delay comprises sensor delay, actuator delay and control delay, - providing (14) a delay correction term (28) based on a comparison between the estimated total delay and the measured total delay, - calculating (15) a corrected total delay using a delay correction term (28) available at the estimated point in time for said total delay, - selecting and executing (16) a control function from a plurality of control functions (23a) for which the corrected total delay lies within the delay range of the control function, and calculating a control input (30) from the selected control function, - transmitting (17) the control input (30) to the actuator (24) via a network (25, 42), - applying (18) the control input (30) to the technical system (21) via the actuator (24).

2. The method according to claim 1 , further comprising the following initial method step (10): - receiving detection data (26) and time information of the sensor (22) at the control unit (23; 43), - calculating the time delay correction term (28) at the actuator (24) and transmitting the time delay correction term (28) to the control unit (23; 43), - performing at said control unit (23; 43) the steps of estimating actuator delays, calculating a corrected total delay, selecting a control function and transmitting said control input (30).

3. The method according to claim 2, wherein the control unit (23; 43) transmits the control input (30) to the actuator (24) together with the time information of the sensor (22) and the estimated total delay.

4. The method of any one of claims 1 to 3, wherein the estimated actuator delay is derived from at least one of historical data, network observations, runtime estimates, computational delays, and sensor delays.

5. The method according to any one of claims 1 to 4, wherein the time information comprises absolute time information, such as a timestamp or a relative time parameter.

6. A method according to any one of claims 1 to 5, wherein the control functions are classified according to best performance, and wherein the control functions are selected from a plurality of control functions (23a) starting with the control function with the highest classification and continuing with the control functions with lower classifications until a control function is identified for which the corrected total delay is within the delay range of the control function.

7. The method according to claim 1, wherein the control unit (23; 43), the sensor (22) and the actuator (24) are synchronized in time.

8. The method according to any one of claims 1 to 7, wherein the total delay is composed of the sensor delay and the actuator delay if the sensor delay and the actuator delay are large compared to the control delay.

9. A method according to any one of claims 1 to 8, wherein the sensor (22) measures at least one characteristic of the technical system (21) and outputs detection data (26) to the control unit (23; 43), and wherein the actuator (24) receives a control input (30) from the control unit (23; 43) and manipulates at least one characteristic of the technical system (21) according to the control input (30).

10. A control system (20, 21, 40) designed to implement the method (10) according to any one of claims 1 to 9, wherein the distributed system (20; 40) comprises at least one sensor (22), a control unit (23; 43) and an actuator (24).

11. The control system (20, 21, 40) according to claim 10, wherein the distributed system (20; 40) comprises a technical system (21) which is contained in a closed-loop control circuit having the sensor (22), the control unit (23; 43) and the actuator (24).

12. A computer system (20, 21, 40) designed to carry out the method (10) according to any one of claims 1 to 9.

13. A computer program designed to implement the method (10) according to any one of claims 1 to 9.

14. A computer-readable medium or signal storing and / or containing a computer program according to claim 13.

Images