A system comprising an intelligent DT communication framework for digital twin (DT) systems and networks with resource constraints

Abstract

The invention relates to a system comprising an intelligent DT communication framework designed for digital twin systems and networks with resource constraints.

Description

[0001] A SYSTEM COMPRISING AN INTELLIGENT DT COMMUNICATION FRAMEWORK FOR DIGITAL TWIN (DT) SYSTEMS AND NETWORKS WITH RESOURCE CONSTRAINTS

[0002] Technical Field of the Invention

[0003] The invention relates to a system comprising an intelligent DT communication (IDTC) framework designed for digital twin (DT) systems and networks with resource constraints. The framework in the invention strikes a balance between accuracy and timing in DT communication, allowing for more synchronization with the real world. In addition, said IDTC framework is not only in the field of Internet of Things (loT) with timeseries, but is also integrated into various DT applications in different industries such as aerospace, automotive and energy.

[0004] State of the Art

[0005] Digital twin (DT) applications are often based on an assumption of creating an accurate, real-time digital copy of a physical system. The digital twin should realistically model the behaviors and status of the physical system. This is important both for accurately reflecting historical data and for predicting future situations. The DT is expected to be in constant communication with the physical system and will be updated in real-time with data from the sensors. In this way, instant data may be used in decision-making processes. However, there is a negative correlation between accuracy and timing relationships of DTs, and there are significant shortcomings in the literature in terms of balance. These shortcomings make it difficult to use DTs effectively, especially in networks with resource constraints. Current studies have generally focused on high-frequency data transfer or timestamp alignment, but they do not fully achieve real-time and accurate synchronization of DTs.

[0006] The first shortcoming is efforts to increase the data transfer frequency of the existing algorithms. High-frequency data transfer causes overuse of the network bandwidth, causing network congestion. This results in the fact that the transmitted data does not arrive on time and the synchronization quality decreases. In addition, increasing data frequency leads to more energy consumption, making resource consumptionunsustainable. In addition, the continuous transfer of data requires more processing power to process the data. This is especially problematic for devices with limited processing capacity. Continuous data transfer increases the energy consumption of both the devices sending data and the network infrastructure. This may be especially problematic in systems which aim for low power consumption, such as loT devices. The data incoming at a high frequency rapidly causes large amounts of data accumulation. This increases data storage costs and compromises the scalability of the storage infrastructure. Data transmitted at high frequencies also makes it easier for malicious actors to monitor or manipulate the data stream. Secondly, timestamp-based mechanisms generally address the DT synchronization problem only as a measurement and an alignment of the timestamp pairs. This approach neglects the need to reflect the real-world changes in DT in real time. This shortcoming is especially problematic in situations where critical decisions need to be made quickly and accurately. Thirdly, some algorithms aim to reduce the load on the network by making low-frequency data updates. However, this method negatively affects the accuracy of DT, causing significant data loss. This prevents DT from accurately capturing changes in the physical system and reduces its reliability. The real-time missing data may cause DT to not accurately reflect the current state. This leads to the inability to make critical decisions in a timely manner. The updates at a low frequency cause DT to miss the rapid changes which occur in the physical system. When DT relies on the existing missing data while simulating the future behaviors, it produces inaccurate predictions. All these shortcomings cause a contradiction between accuracy and timing in DT applications. Especially in cases where resource constraints are present, data transfer at a high frequency increases energy and bandwidth consumption while low-frequency data updates exacerbate timing issues. This contradiction seriously prevents the DTs to operate synchronously with the real world. Technically, the consequences of all these problems may be observed, for example, during the production processes. The delay in data from sensors on a production line makes it difficult for a DT to make instant decisions and prevents potential errors from being detected in advance. This reduces the operational efficiency and increases the costs.

[0007] In the state of the art, there are significant limitations to the application of digital twin (DT) systems. Firstly, approaches aimed at increasing the frequency of data transfer lead to network congestion and performance degradation due to the high bandwidth usage. In addition, the existing synchronization mechanisms based on timestamping ignore thereal-time and complete reflection of dynamic changes in the physical system to the DT, which increases the risk of delays and errors during the critical decision processes. Alternative methods based on low-frequency data updates, on the other hand, damage the accuracy of the DT, causing data loss, and reducing reliability by disrupting the synchronization between the physical system and the DT, despite their efforts to reduce the network load. All these shortcomings prevent the existing systems from operating efficiently, accurately and reliably, especially in networks with resource constraints. Therefore, it is necessary to make an improvement in the relevant technical area.

[0008] Summary and Objectives of the Invention

[0009] The invention discloses a system comprising an intelligent DT communication (IDTC) framework designed for digital twin systems and networks with resource constraints. The framework in the invention strikes a balance between accuracy and timing in DT communication, allowing for more synchronization with the real world. In addition, said IDTC framework is not only in the field of loT with time-series, but is also integrated into various DT applications in different industries such as aerospace, automotive and energy.

[0010] The object of the invention is to improve the synchronization between the physical system and the digital twin by achieving a balance between accuracy and timing in data transmission in the DT systems.

[0011] An object of the invention is to minimize the effects of delays in data transmission and to solve timing problems in advance by predicting the real-time data flow, by using a machine learning-based Predictive Synchronization (PS) algorithm.

[0012] An object of the invention is to proactively address the potential problems by predicting the future state of the DT through aman series modeling methods and deep learning models such as LSTM.

[0013] Another object of the invention is to improve the long-term performance of the DT by continuously updating the prediction models, measuring accuracy errors, and retraining the model by adapting to the varying environmental conditions.An object of the invention is to ensure an efficient use of the network resources and to eliminate the contradiction between accuracy and timing, by dynamically setting the frequency of data transfer through the DT Synchronization Management Protocol (DTSYNC).

[0014] An object of the invention is to improve the network efficiency and enhance the performance, especially in networks where the resource constraints exist, by eliminating the unnecessary loads during data transmission. This makes the synchronization process 3.8 times more efficient and provides a significant advantage in networks with resource constraints.

[0015] Another object of the invention is to optimize resource consumption and synchronization performance in accordance with the system requirements by configuring the data flow on a device or the DT basis.

[0016] Description of the Drawings

[0017] Fig. 1 is the structure of the system of the invention.

[0018] Fig. 2 is a work-flow diagram of the system of the invention.

[0019] Fig. 3 is a work-flow diagram of a PS.

[0020] Description of the References in the Drawings

[0021] 1. Real plane

[0022] 2. DTSYNC

[0023] 3. Data flow

[0024] 4. Device-centric data flow

[0025] 5. DT -centric data flow

[0026] 6. Control flow

[0027] 7. PS

[0028] 8. DT data store

[0029] 9. Data completing model

[0030] 10. Prediction model

[0031] 11. Synchronization, frequency management and optimization

[0032] 12. IDTC (Intelligent Digital Twin Communication)13. Digital twin (DT)

[0033] 1001. Initializing the sensors and the DT.

[0034] 1002. Assigning a sensor sampling frequency ( ) and a basic sampling frequency (fo).

[0035] 1003. Checking commands from DT to sensors.

[0036] 1004. Responding from sensors to DT.

[0037] 1005. Collecting the data g(t) from the sensors.

[0038] 1006. Adding the data which need to be synchronized to a queue.

[0039] 1007. Controlling the queue occupancy.

[0040] 1008. Retrieving the measured data from the queue.

[0041] 1009. Retrieving the prediction value previously obtained by the model from the store.

[0042] 1010. Calculating the error between the predicted value and the measured value.

[0043] 1011. Checking an error threshold value.

[0044] 1012. Updating the prediction model by applying backpropagation.

[0045] 1013. Updating the data with the current model.

[0046] 1014. Obtaining the instantaneous time.

[0047] 1015. Controlling the exact division of the obtained instantaneous time to 1 / ft, 1016. Retrieving the real data from the queue.

[0048] 1017. Completing the missing data, if any.

[0049] 1018. Calculating the new prediction.

[0050] 1019. Saving the new prediction.

[0051] 2001. Defining the queue and the error threshold.

[0052] 2002. Adding the data which need to be synchronized to a queue.

[0053] 2003. Controlling the queue occupancy.

[0054] 2004. Retrieving the real data from the queue.

[0055] 2005. Retrieving the prediction value previously obtained by the model from the store.

[0056] 2006. Calculating the error threshold.

[0057] 2007. Checking the error threshold value.

[0058] 2008. Updating the model by applying backpropagation.

[0059] 2009. Updating the data with the current model.

[0060] 2010. Saving the current time.

[0061] 2011. Controlling the time.

[0062] 2012. Retrieving the real data from the queue.

[0063] 2013. Completing the missing data, if any.2014. Calculating the new prediction.

[0064] 2015. Time-based planning the outcome.

[0065] Detailed Description of the Invention

[0066] The invention relates to a system comprising an intelligent DT communication (IDTC) framework designed for digital twin (DT) systems and networks with resource constraints. The framework in the invention strikes a balance between accuracy and timing in DT communication, allowing for more synchronization with the real world. In addition, said IDTC framework is not only in the field of loT with time-series, but is also integrated into various DT applications in different industries such as aerospace, automotive and energy.

[0067] A system for digital twin (DT) systems and networks with resource constraints according to the invention comprises:

[0068] • A real plane (1) which represents data collected from devices in a real universe, • A DTSYNC (2) which provides data transfer mechanisms and an optimization ft, • Data flow (3), which ensures the synchronization of the measurements obtained from the device by DT,

[0069] • Device-centric data flow (4),

[0070] • DT-centric data flow (5) transmitting measurements from devices (£>) to DT (5) during the last time slot ( ),

[0071] ft

[0072] • A control flow (6) initiated by the DT that sends the action along with the input parameters,

[0073] • A PS (7), which compensates for the delays of DT, which cause the time shift ( ) problem,

[0074] • A DT data store (8), which allows the data to be kept in a historical or time series format,

[0075] • A data completing model (9), which converts the data into time series with an interval 7,

[0076] ft

[0077] • A prediction model which calculates future measurements of data using LSTM (10),• A synchronization frequency management and optimization (11), which initiates the transfer of the measured and predicted data and enables them to be synchronized at regular intervals (^),

[0078] ft

[0079] • An IDTC (12) which is designed to optimize the balance of accuracy and timing in DT systems,

[0080] • A digital twin (DT) (13) which represents the data collected from devices located in a real plane (1) in a virtual plane.

[0081] The operation method of the system according to the invention comprises the process steps of:

[0082] i. Initializing the sensors and the DT (1001 ),

[0083] ii. Assigning a sensor sampling frequency ( ) and a basic sampling frequency (o) (1OO2),

[0084] iii. Checking commands from DT to sensors (1003),

[0085] iv. Responding from sensors to DT (1004),

[0086] v. Collecting the data g(t) from sensors (1005),

[0087] vi. Adding the data which needs to be synchronized to a queue (1006), vii. Controlling the queue occupancy (1007),

[0088] viii. Retrieving the measured data from the queue (1008),

[0089] ix. Retrieving the prediction value previously obtained by the model from the store (1009),

[0090] x. Calculating the error between the predicted value and the measured value (1010),

[0091] xi. Checking an error threshold value (1011),

[0092] xii. Updating the prediction model by applying backpropagation (1012), xiii. Updating the data with a current model (1013),

[0093] xiv. Obtaining instantaneous time information (1014),

[0094] xv. Controlling the exact division of the obtained instantaneous time information to 1 / ft(1015),

[0095] xvi. Retrieving the real data from the queue (1016),

[0096] xvii. Completing the missing data, if any (1017),

[0097] xviii. Calculating the new prediction (1018),

[0098] xix. Saving the new prediction (1019).Problem Formulation:

[0099] During the modeling process, if g(t) and h(t) are equal, then they are in a synchronization process. The functions g(t) and h(t) are a step function changing at certain points in time. Each sample g(t) is assigned a value as the devices D are sampled, and these values are transmitted to D, which causes a value change in h(t). Consequently, these functions have constant values between the sampling / update points, t in both functions represents the actual runtime. Therefore, in order to evaluate the performance of this synchronization, the runtime behavior of both signals should be considered and the performance should be formulated accordingly. To solve this problem, the accuracy of the DT between the times [ti;tj] is calculated as expressed in equation 1 .

[0100] E = J^jJ^ gCt)2- h(t - <£)2dt, tj > ti (1)

[0101]

[0102] In Equation 1 , <|) is the time shift, shown over the DT, of g(t). This represents the timing of the DT and is formulated as in equation 2.

[0103] = argmin J1’ g(t) - h(t - 4>)dt (2)

[0104] 4>

[0105] As g(t) and h(t) are a step function, integrals may be solved by being resolved into sums over these intervals. In this case, E may be written as in equation 3.

[0106] $ g(t)2- h(t - 4>)2dt = Ek=t.(g(tk)2- h(tk- 4>)2)(tk+1 / f- tk) (3)

[0107]

[0108] Similarly, the integral in <p may be resolved as in equation 4.

[0109]

[0110] g(t) - h(t - c|>)dt = Zk=t.(g(tk) - h(tk- <t>))(tk+1 / f - tk) (4)

[0111] Consequently, when the problem is to be formulated, the synchronization problem of the DT may be expressed as in equation 5 while representing the accuracy of E and the time ofargmin E, cp

[0112] ft

[0113] s. t ft < f < f (5)

[0114] f = fo,

[0115] tj > tj

[0116] PS Optimization:

[0117] Optimization of the PS mechanism is a0= 1 in equation 6 to overcome the challenges, whereas a setting ftis performed in equation 7 based on an approach of an additive increase / a multiplier decrease.

[0118] ft= af (6)

[0119] ( max(10h-^2- E < Eth A 0 < 0th

[0120] ccn+l

[0121] max f klO-11,- an^22-) E < EthA an= 1 A 0 > 0

[0122] +Vth

[0123] E < EthA an* 1 A 0 > 0th

[0124] k min f kl,— an) / E > Eth

[0125]

[0126] The cases in equation 7 are respectively as follows:

[0127] • Case 1 : The DT runs with an error lower than the threshold value (E < Eth), i.e., the operation is done correctly. At the same time, the time shift of the DT is minimal ( < <p_th ). This may happen in the following two cases:

[0128] o PS is not used, i.e. an= 1. In this case, the network may include the data stream, a is decreased in order to protect the network resources, and PS is used instead.

[0129] o PS is used, i.e. an< 1. The predictive mechanisms compensate 8Cand 8W, resulting in low

[0130]

[0131] a is decreased in order to protect the network resources and make more use of the PS.

[0132] • Case 2: The DT may operate correctly, PS is not used, and the time shift of the DT is higher than the threshold. In this case, the network cannot meet this data flow. In this case, a and is decreased; PS is used instead.• Case 3: DT may operate with maximum accuracy using PS (an< 1), but delays cannot be compensated ( > 4>th). The window size may cause a bottleneck on the DT side, which means is affected by 8P. a is incremented to protect the DT resources and increase the timeliness.

[0133] • Case 4: DT cannot work correctly. Regardless of 4>, more training is needed for the PS model. The a is incremented to use actual values instead of predictions.

[0134] It is assumed that there are two planes, a real plane (1) and a virtual plane of that real plane, the DT (13). n Device is indicated as D = {D^D , ...,Dnin the real plane. Monitoring these devices is a representation of the sampling of a feature of the environment by being indicated by g(t). Each device is represented by a < X, T, f > tuple indicating the measured data ( = g(T)) of X, the measurement times of T and the measurement frequency of f. In addition, f0is the lowest frequency of the measurements to produce g(t). DT devices located in the virtual plane are indicated as D = {D^D^ ...,Dnand < X,T,f >. T here is a time when a certain data point from D becomes a part of D. X may also be produced by the DT. In this context, the frequency of data transmitted from D toD is defined as ft. The object of the DT is to produce the signal / r(t) which is a copy of the original g(t). In order for all these operations to be carried out, the invention offers an IDTC framework (12) designed to optimize the balance of accuracy and timing in the DT systems. The proposed IDTC framework (12) includes a DT Synchronization Management Protocol (DTSYNC) (2) which provides machine learning-based Predictive Synchronization (PS) (7) for synchronization and the data transfer mechanisms, as well as the optimization of ft.

[0135] PS is recommended to compensate for delays that cause the time shift ( ) problem of the DT. The PS process is done by proactively adding the expected measurements to the DT at regular intervals. Consequently, the accuracy affects E. The proposed PS algorithm is performed by making predictions using data completing process (9) and prediction models (10) of measurements, as shown in diagram 2. Here, the data collected from the devices is converted to the time series with an interval using the data completing model (9). Then, the prediction model (10) uses this default time series data to calculate future measurements of D. The actual data and predictions are stored in a historical manner, while the DT representation is kept in the DT data store (8) in aform of a time series. LSTM is used in the prediction model. To retrain the models in the PS mechanism, the data from a Dare used.

[0136] DTSYNC synchronization (2) defines the mechanisms for creating data (3), controlling streams (6), and executing synchronization by setting ft. The DTSYNC protocol minimizes E and <p based on a certain setting ftand selects an ideal ftaccording to the environment dynamics and PS performance.

[0137] The object of the data flow (3) is to ensure the measurements obtained from the device to be synchronized by the DT. Devices works according to f and data is transferred to ft. In the device-centric data stream (4), the measurements (D) are transmitted from the devices D to the DT during the last time slot (- ), and the PS uses the incoming data.

[0138] ft

[0139] For each device, the PS error is evaluated by the DT. In the DT-centric data flow (5), the distribution is carried out by transmitting the predictions from the previous window to D for confirmation. If the values correspond to the data in D, no further action is taken; if not, a setting is sent. The control flow (6) is a remote procedure call initiated by the DT and sends the action with the corresponding input parameters. When the device receives a call, it initiates a response and generates the corresponding message and output.

[0140] The data flow (3) is used to be synchronized at regular intervals by initiating the ft

[0141] transfer of the measured / predicted data (11). Also, the data transferred are cretaed by a frequency of f and contains the data within the window thereof.

Claims

CLAIMS1. A system for digital twin (DT) systems and networks with resource constraints, characterized in that it comprises:• A real plane (1) which represents data collected from devices in a real universe,• A DTSYNC (2) which provides data transfer mechanisms and an optimization ft ,• Data flow (3), which ensures the synchronization of the measurements obtained from the device by DT,• Device-centric data flow (4),• DT-centric data flow (5) transmitting measurements from devices (£>) to DT (D) during the last time slot (^),ft• A control flow (6) initiated by the DT that sends the action along with the input parameters,• A PS (7), which compensates for the delays of DT, which cause the time shift ( ) problem,• A DT data store (8), which allows the data to be kept in a historical or time series format,• A data completing model (9), which converts the data into time series with an interval -,ft• A prediction model which calculates future measurements of data using LSTM (10),• A synchronization frequency management and optimization (11), which initiates the transfer of the measured and predicted data and enables them to be synchronized at regular intervals (^),ft• An IDTC (12) which is designed to optimize the balance of accuracy and timing in DT systems,• A digital twin (DT) (13) which represents the data collected from devices located in a real plane (1) in a virtual plane.

2. An operation method of a system according to Claim 1 , characterized in that it comprises the process steps of:i. Initializing the sensors and the DT (1001 ),ii. Assigning a sensor sampling frequency ( ) and a basic sampling frequency (Z ) (1002),iii. Checking commands from DT to sensors (1003),iv. Responding from sensors to DT (1004),v. Collecting the data g(t) from sensors (1005),vi. Adding the data which needs to be synchronized to a queue (1006), vii. Controlling the queue occupancy (1007),viii. Retrieving the measured data from the queue (1008),ix. Retrieving the prediction value previously obtained by the model from the store (1009),x. Calculating the error between the predicted value and the measured value (1010),xi. Checking an error threshold value (1011 ),xii. Updating the prediction model by applying backpropagation (1012), xiii. Updating the data with a current model (1013),xiv. Obtaining instantaneous time information (1014),xv. Controlling the exact division of the obtained instantaneous time information to 1 / ft(1015),xvi. Retrieving the real data from the queue (1016),xvii. Completing the missing data, if any (1017),xviii. Calculating the new prediction (1018),xix. Saving the new prediction (1019).

Images