An automated packaging apparatus remote monitoring system

Abstract

The application provides an automatic packaging equipment remote monitoring system, and relates to the technical field of industrial equipment monitoring.The system comprises a plurality of synchronous sensing units for collecting synchronous raw data of each branch of a pneumatic network, and a data processing device.The data processing device is internally provided with a digital twin model, which can predict a theoretical pulsation reference waveform according to a process beat parameter obtained from an equipment controller, and can separate a pure residual signal by performing cancellation processing on the synchronous raw data and the theoretical waveform.After confirming the occurrence of a micro-leakage event, further inversion calculation is performed based on time delay information of the multi-point residual signal and a pre-stored pneumatic network topology matrix to determine the branch identification where the leakage occurs, and an alarm instruction is sent.The application can significantly improve the reliability and positioning accuracy of micro-leakage detection and improve the production recovery efficiency.

Description

Technical Field

[0001] This application relates to the field of industrial equipment monitoring technology, and more specifically, to a remote monitoring system for automated packaging equipment. Background Technology

[0002] Automated packaging equipment, such as automatic strapping machines or carton sealing machines, typically includes a pneumatic network consisting of multiple branches to drive different actuators in cyclical operations. To ensure packaging quality and production efficiency, the operating status of the pneumatic network needs to be monitored to prevent insufficient pressure due to gas leaks. A common monitoring method is to install pressure or flow sensors on the pneumatic network's pipelines, using real-time signal changes to determine the network's integrity.

[0003] However, during the operation of automated packaging equipment, the periodic movements of its internal actuators generate strong pressure pulsations. These normal operational pulsations are highly similar in signal characteristics to minute leaks caused by aging pipe joints or worn seals. This makes it difficult for existing single-point sensing-based monitoring methods to effectively distinguish between the two types of signals, often misreporting normal operational pulsations as leaks, or missing detections because pulsation interference masks the true minute leaks, resulting in low detection reliability. Furthermore, even when anomalies are detected, single-point sensing solutions lack spatial dimension information to pinpoint the specific branch where the leak occurred, leading to a heavy reliance on manual, segment-by-segment inspections for troubleshooting, resulting in low efficiency in fault handling and production recovery.

[0004] Therefore, in the pneumatic network of automated packaging equipment, how to improve the detection accuracy and positioning efficiency of micro-leakage events to cope with the challenges brought by the complexity of multiple branches and dynamic operating condition interference, so as to achieve more reliable real-time monitoring and fault response, is a technical problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this application provides a remote monitoring system for automated packaging equipment, comprising:

[0006] Multiple sensing units are arranged at the inlet and outlet of each branch of the target aerodynamic network to synchronously collect the inlet pressure, outlet pressure and flow rate of each branch under a unified clock reference, and generate synchronous raw data.

[0007] A data processing device is communicatively connected to the sensing unit and the controller of the automated packaging equipment. The data processing device stores a digital twin model and is configured as follows:

[0008] Receive process cycle parameters from the controller of the automated packaging equipment;

[0009] Based on the process cycle parameters and the digital twin model, the theoretical pulsation reference waveform for the current working condition is calculated.

[0010] The original synchronous data and the theoretical pulsating reference waveform are differentially processed to generate a residual signal;

[0011] The occurrence confidence level of the micro-leakage event is determined based on the residual signal, and micro-leakage event information is generated in response to the occurrence confidence level being greater than or equal to a preset threshold.

[0012] Based on the micro-leakage event information, cross-correlation calculations are performed on the residual signals of different branches to obtain time delay information. Inverse calculations are then performed using the pre-stored pneumatic network topology matrix to determine the branch identifier where the micro-leakage event occurred. Based on the branch identifier, a micro-leakage event alarm command is sent to the automated packaging equipment controller.

[0013] As an optional implementation, it further includes: an automatic time calibration device configured to initiate and manage the time calibration process of the sensing unit and the data processing device, and to determine and store time calibration values ​​corresponding to each of the sensing units;

[0014] The sensing unit is further configured to: record the actual arrival timestamp of the standard physical pulse it receives during the time calibration process, and provide the actual arrival timestamp to the automatic time calibration device;

[0015] The data processing device is further configured to: calculate the theoretical arrival time of the standard physical pulse to each of the sensing units during the time calibration process, and provide the theoretical arrival time to the automatic time calibration device;

[0016] The automatic time calibration device is further configured to: compare the actual arrival timestamp with the theoretical arrival time to determine the time calibration value;

[0017] The sensing unit is further configured to correct the time information using the time calibration value when generating the synchronous raw data.

[0018] As an optional implementation, the automatic time calibration device is further configured to:

[0019] The process cycle time parameters are obtained from the controller of the automated packaging equipment, and a preset idle process time period is identified based on the process cycle time parameters.

[0020] As an optional implementation, the automatic time calibration device is further configured to:

[0021] During the preset idle process period, a calibration control command is sent to the controller of the automated packaging equipment to drive the actuator in the target pneumatic network to generate the standard physical pulse.

[0022] As an optional implementation, the data processing device is further configured to:

[0023] In response to the occurrence of a confidence level lower than a preset threshold, the residual signal is defined as a model error signal;

[0024] The model error signal is used to adaptively correct one or more model parameters in the digital twin model.

[0025] As an optional implementation, the adaptive correction of one or more model parameters in the digital twin model includes:

[0026] The process cycle time parameters are parsed into actuator action events, and the pre-stored pneumatic network topology matrix is ​​used to determine one or more target branches associated with the actuator action events.

[0027] Extract the piecewise error signal that is temporally associated with the target branch from the model error signal;

[0028] Using the segmented error signal, only the dynamic physical parameters related to the target branch in the digital twin model are updated.

[0029] As an optional implementation, extracting the piecewise error signal that is temporally associated with the target branch from the model error signal includes:

[0030] Perform multi-scale wavelet transform on the model error signal to obtain wavelet coefficients distributed in the time-frequency domain;

[0031] Using the pneumatic network topology matrix, one or more adjacent branches adjacent to the target branch are identified, and based on the topological relationship between the adjacent branches and the target branch, the time-frequency characteristics of the crosstalk signal caused by the actuator action event in the adjacent branches are predicted.

[0032] Based on the aforementioned time-frequency features, a wavelet domain mask is constructed;

[0033] The wavelet coefficients are filtered using the wavelet domain mask, followed by inverse wavelet transform to reconstruct and obtain the purified segmented error signal.

[0034] As an optional implementation, the data processing device also stores a packaging process feature library, which stores reference time-frequency features corresponding to various preset packaging action types of the automated packaging equipment.

[0035] The predicted time-frequency characteristics of the crosstalk signal caused by the actuator action event in the adjacent branch include:

[0036] Identify the corresponding packaging action type from the action events of the actuator;

[0037] Extract the reference time-frequency features corresponding to the identified packaging action type from the packaging process feature library;

[0038] The time-frequency characteristics of the crosstalk signal are calculated based on the reference time-frequency characteristics and the topological relationship between the adjacent branch and the target branch.

[0039] As an optional implementation, obtaining the time delay information includes:

[0040] From the plurality of sensing units, the sensing unit that receives the residual signal with the strongest energy or the earliest time is determined as the reference sensing unit.

[0041] For at least one other sensing unit besides the reference sensing unit, the residual signal of the reference sensing unit and the residual signal of the other sensing unit are respectively subjected to Fourier transform to obtain their respective spectral signals.

[0042] The cross-power spectrum is calculated based on the obtained spectral signal, and the cross-power spectrum is phase-transformed and weighted to obtain a whitened spectrum.

[0043] Perform an inverse Fourier transform on the whitened spectrum to obtain the cross-correlation function;

[0044] Identify the peak position of the cross-correlation function to determine the time delay between the reference sensing unit and the other sensing units;

[0045] The one or more time delays thus determined are combined to form the time delay information.

[0046] As an optional implementation, the step of performing inversion calculations based on a pre-stored aerodynamic network topology matrix to determine the branch identifier where a micro-leakage event occurred includes:

[0047] Based on the aerodynamic network topology matrix, the target aerodynamic network is discretized into multiple candidate leak points;

[0048] For each of the candidate leak points, the theoretical time delay when the leak occurs at that candidate leak point is calculated using the aerodynamic network topology matrix;

[0049] The calculated theoretical time delay is compared with the time delay information to calculate the likelihood value of the leak location corresponding to each candidate leak point;

[0050] Identify the candidate leak point with the highest likelihood value of the leak location, and determine the branch where the candidate leak point is located as the branch identifier of the micro-leak event.

[0051] Compared to existing technologies, the remote monitoring system provided in this application, by constructing and utilizing a digital twin model linked to the equipment's process cycle, can accurately predict and separate dynamic pressure pulsation interference generated during normal equipment operation. This clearly highlights minute leakage signals that were previously masked by strong noise backgrounds, fundamentally improving the accuracy and reliability of leak detection. Based on this, the system utilizes spatial and temporal information collected by multiple synchronous sensing units deployed on various network branches, combined with a preset network topology, to achieve rapid and accurate location of the leak source. This ability to combine high-reliability detection with high-efficiency location effectively avoids production interruptions caused by false alarms or missed alarms, and replaces time-consuming manual inspections with automated remote diagnostics, significantly enhancing the stability and economy of automated packaging equipment operation, and achieving more intelligent fault response and predictive maintenance. Attached Figure Description

[0052] Figure 1 A schematic diagram of a remote monitoring system for automated packaging equipment provided in an embodiment of this application;

[0053] Figure 2 This is a flowchart of a remote monitoring method executed by a data processing device in an embodiment of this application;

[0054] Figure 3 A flowchart illustrating an adaptive correction method provided in an embodiment of this application;

[0055] Figure 4 This is a flowchart illustrating a method for extracting a segmented error signal that is temporally associated with the target branch, as provided in an embodiment of this application. Detailed Implementation

[0056] The technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments.

[0057] The "and / or" mentioned in this article is merely a description of the relationship between related objects, indicating that there can be three kinds of relationships. For example, A and / or B can represent three situations: A exists alone, A and B exist simultaneously, and B exists alone.

[0058] See Figure 1 The diagram shown is a schematic of a remote monitoring system for automated packaging equipment provided in an embodiment of this application. The system includes:

[0059] Multiple sensing units 100 are arranged at the inlet and end of each branch of the target aerodynamic network to synchronously collect the inlet pressure value, end pressure value and flow rate value of each branch under a unified clock reference, and generate synchronous raw data.

[0060] The data processing device 200 is communicatively connected to the sensing unit 100 and the automated packaging equipment controller 300, and the data processing device 200 stores a digital twin model.

[0061] In specific implementation, multiple sensing units 100 are respectively arranged at the inlet and outlet of each parallel branch in the target aerodynamic network. In this embodiment, each sensing unit 100 can be an integrated intelligent sensing node, which internally encapsulates a pressure sensing chip, a flow sensing chip, a microcontroller (MCU), and a synchronization clock slave module supporting the IEEE 1588 PTP protocol. Under the coordination of a unified master clock, all sensing units 100 can achieve microsecond-level synchronization under a unified clock reference, and synchronously collect the inlet pressure value, outlet pressure value, and instantaneous flow value of each monitoring point, generating synchronized raw data with high-precision timestamps.

[0062] For example, the sensing unit 100 can be implemented as an integrated intelligent sensing node to meet the requirements of high precision and high synchronization in industrial settings. The node's hardware configuration may include a MEMS pressure sensing chip for measuring the pressure difference across a branch and a flow sensing chip based on the hot-film principle, supplemented by a signal conditioning circuit to amplify and filter the weak analog signal output by the sensors. The conditioned signal is sent to a high-precision analog-to-digital converter, such as an ADC with a sampling rate of at least 10 kS / s and a resolution of 24 bits, to achieve high-fidelity digitization. The node's core processing module is used to execute data processing and communication protocols. The microcontroller (MCU) inside the sensing unit 100 can execute a preset firmware program. This firmware program first exchanges messages with the network master clock through its Ethernet physical layer chip supporting the IEEE 1588 PTPv2 protocol, calibrating and maintaining the local clock error within 1 microsecond. Driven by this synchronous clock, the MCU generates a sampling interrupt at a fixed high frequency, for example, every 200 microseconds. In response to each interrupt, the MCU drives the ADC to perform analog-to-digital conversion on the pressure and flow signals, and immediately reads the precise time of the current synchronous clock as the timestamp of that set of sampled data. Finally, the MCU encapsulates the pressure value, flow value, high-precision timestamp, and the unique ID of this sensing unit 100 into a UDP data packet, and sends it to the data processing device 200 in real time via the Ethernet interface.

[0063] In this embodiment, the data processing device 200 can be an edge computing gateway or industrial computer (IPC) deployed on the production site. It establishes a high-speed communication connection with all the sensing units 100 via industrial Ethernet to aggregate the synchronous raw data in real time. Simultaneously, the data processing device 200 also communicates with the automated packaging equipment controller 300 via industrial communication protocols such as OPC UA or Modbus-TCP. For example, the automated packaging equipment controller 300 can be a PLC or IPC.

[0064] The data processing device 200 has a digital twin model pre-stored in its internal memory. This model is a fluid dynamics simulation model pre-established based on the physical structure of the target pneumatic network, such as pipe length, diameter, valve type, and cylinder parameters. It is used to describe the pressure and flow response of the pneumatic network to the actions of various actuators under ideal conditions without leakage.

[0065] For example, the data processing device 200 may employ an embedded industrial computer (IPC) to meet the real-time computing needs of complex models and algorithms. This industrial computer may run a Linux operating system with real-time patches, and its core monitoring application is designed with a modular architecture, mainly including a data access module, a controller interface module, a digital twin simulation engine, and signal processing and diagnostic modules.

[0066] For example, a digital twin model can be a pre-built fluid dynamics simulation model based on the lumped parameter method.

[0067] In the model building phase, a mathematical framework describing the network topology is first established based on the physical design drawings of the target pneumatic network, including information such as pipe length, diameter, and connection relationships, as well as the specifications of the components deployed on it, such as valve models and cylinder parameters. In this framework, each pipe segment can be equivalently represented as a circuit unit composed of gas resistance (R), gas sensing (I), and gas capacitance (C) elements to characterize the gas flow friction loss, inertial effect, and compressibility, respectively; while the action of each actuator is modeled as a time-varying flow consumption source with a specific gas consumption curve.

[0068] In the model parameterization stage, in order to ensure that the model can accurately reflect the real physical characteristics, an offline system identification method can be used.

[0069] Specifically, on a leak-free reference pneumatic network, an automated packaging device performs all its standard processes, acquiring a complete set of high-fidelity synchronous raw data using the multiple sensing units 100. Subsequently, an optimization algorithm, such as a least-squares fitting algorithm, is initiated to repeatedly compare the output waveform of the simulation model under the same process excitation with the acquired data, automatically adjusting key physical parameters in the model, such as the air resistance and capacity coefficient of each pipe section, and the air consumption curve parameters of each actuator, until the root mean square error between the simulation output and the actual data is minimized. Through this process, a set of accurately identified model parameters that can highly reproduce the real physical process is obtained.

[0070] This complete model, including the network topology, mathematical equations, and precisely identified parameters, is solidified and stored in the data processing device 200. During system operation, when process cycle parameters are received, the simulation engine drives the static model to perform forward calculations, thereby obtaining a highly realistic theoretical pulsation reference waveform under the current operating conditions.

[0071] Reference Figure 2 This is a flowchart of a remote monitoring method executed by a data processing device in an embodiment of this application. The data processing device is configured to execute the following steps S101 to S105, wherein:

[0072] S101: Receive process cycle parameters from the controller of the automated packaging equipment;

[0073] S102: Based on the process cycle parameters and the digital twin model, calculate the theoretical pulsation reference waveform for the current working condition;

[0074] S103: Perform differential processing on the original synchronous data and the theoretical pulsating reference waveform to generate a residual signal;

[0075] S104: Determine the occurrence confidence of the micro-leakage event based on the residual signal, and generate micro-leakage event information in response to the occurrence confidence being greater than or equal to a preset threshold;

[0076] S105: Based on the micro-leakage event information, perform cross-correlation calculation on the residual signals of different branches to obtain time delay information, and perform inversion calculation in combination with the pre-stored pneumatic network topology matrix to determine the branch identifier where the micro-leakage event occurred, and send a micro-leakage event alarm command to the automated packaging equipment controller based on the branch identifier.

[0077] In specific implementation, the controller interface module of the data processing device 200 subscribes to and receives process cycle parameters from the automated packaging equipment controller 300 in real time via the OPC UA protocol (S101). After receiving the parameters, the digital twin simulation engine immediately drives the internally pre-stored static digital twin model based on the principle of fluid dynamics to perform simulation and calculate the theoretical pulsating reference waveform (S102). At the same time, the data access module sends the synchronous raw data collected from each sensing unit 100 to the signal processing and diagnostic module. This module executes step S103, which performs vectorized point-by-point subtraction of these real-time data with the theoretical waveform to generate a residual signal.

[0078] Next, when performing step S104 to determine a micro-leakage event, this module can employ a detection method based on short-time energy. For example, the root mean square (RMS) value of the residual signal is calculated within a sliding time window. When this RMS value persists for a period of time, such as 1 second, exceeding a noise threshold set based on historical data statistics, the confidence level is deemed sufficient, and micro-leakage event information is generated. Once the micro-leakage event information is generated, the system initiates the location procedure in step S105.

[0079] For example, for cross-correlation calculation, the signal processing and diagnostic module can use a standard cross-correlation function algorithm based on Fast Fourier Transform (FFT), which is conventional in the field, to calculate the delay between different residual signals, thereby obtaining time delay information. For inversion calculation, the module can use an iterative search algorithm based on the least squares method.

[0080] For example, the algorithm first virtually divides all pipelines into several 10-centimeter-long line segments based on a topology matrix as candidate leak points. Then, it iterates through each candidate point, calculates the theoretically expected multipath time delay when a leak occurs at that point, and calculates the Euclidean distance between the theoretical delay group and the actual measured delay group as the error. Finally, the branch containing the candidate point that minimizes the error is identified as the branch where the micro-leak event occurred. After determining the branch identifier, the data processing device 200 generates a structured alarm command and sends it to the automated packaging equipment controller 300 through the controller interface module to trigger subsequent response actions.

[0081] For example, the aerodynamic network topology matrix can be a structured data file, such as an XML or JSON file, which records in detail all the physical and logical relationships of the target aerodynamic network in the form of multiple information tables. This file is generated once during system deployment based on the actual equipment piping layout.

[0082] For example, the topology matrix file may include the following core information:

[0083] Node Information Table: This table defines the attributes of all key functional points (nodes) in the network. Each row represents a node and includes at least the following information columns:

[0084] Node ID: A unique identifier, such as "Sensor-A1-Inlet", "Actuator-MainPress", or "Junction-Tee-01"; Node Type: Defines the nature of the point, such as "sensor unit", "actuator", or "tee / four-way connector"; Branch ID: Specifically indicates which branch in the pneumatic network the node belongs to, such as "Branch-A" or "Branch-B"; 3D Spatial Coordinates: Records the X, Y, and Z coordinates of the node in the device coordinate system for more accurate distance calculation.

[0085] Pipeline Information Table: This table defines the attributes of the physical pipelines (edges) connecting the nodes. Each row represents a pipeline segment and includes at least the following information columns:

[0086] Pipeline ID: A unique pipeline identifier; Start Node ID: The node ID connecting one end of the pipeline; End Node ID: The node ID connecting the other end of the pipeline; Pipeline Length: The actual physical length of this pipeline segment, for example, 1.5 meters; Pipeline Inner Diameter: The inner diameter of this pipeline segment, for example, 8 millimeters; Material Sound Velocity: The pressure wave propagation speed preset according to the pipeline material and the working medium (compressed air).

[0087] Event Mapping Table: This table is used to establish the association between upper-level control logic and lower-level physical entities. Each row represents a mapping relationship and includes at least the following information columns:

[0088] Process cycle time parameter / event name: A clearly defined instruction from the automated packaging equipment controller, such as "MainPress_Engage"; Associated actuator ID: The node ID of the "actuator" type that directly corresponds to this instruction, such as "Actuator-MainPress".

[0089] Through this structured information organization, the data processing device 200 can efficiently perform various complex queries and calculations. For example, the device first queries the event mapping table based on the received process cycle parameters to find the corresponding actuator ID; then, it queries the node information table using this ID to determine the branch ID to which it belongs, i.e., the target branch; finally, by querying the pipeline information table and the node information table, it can easily find all other branches directly or indirectly connected to the target branch, i.e., adjacent branches, providing accurate and structured data support for subsequent calculations.

[0090] In this way, by constructing and utilizing a digital twin model linked to the equipment's process cycle, the system can accurately predict and separate dynamic pressure pulsation interference generated during normal equipment operation. This clearly highlights minute leakage signals that were previously masked by strong noise backgrounds, fundamentally improving the accuracy and reliability of leak detection. Based on this, the system utilizes spatial and temporal information collected by multiple synchronous sensing units deployed on various network branches, combined with a pre-defined network topology, to achieve rapid and accurate location of the leak source. This ability to combine high-reliability detection with high-efficiency location effectively avoids production interruptions caused by false alarms or missed alarms, and replaces time-consuming manual inspections with automated remote diagnostics, significantly enhancing the stability and economy of automated packaging equipment operation, and achieving more intelligent fault response and predictive maintenance.

[0091] To further improve the positioning accuracy of the system described in this invention and eliminate the static timestamp offset caused by the difference in physical wiring length of each sensing unit 100, the inherent response delay difference of internal circuits and sensing elements, as an optional implementation, this application also includes: an automatic time calibration device configured to initiate and manage the time calibration process of the sensing unit and the data processing device, and to determine and store the time calibration values ​​corresponding to each of the sensing units.

[0092] The sensing unit is further configured to: record the actual arrival timestamp of the standard physical pulse it receives during the time calibration process, and provide the actual arrival timestamp to the automatic time calibration device;

[0093] The data processing device is further configured to: calculate the theoretical arrival time of the standard physical pulse to each of the sensing units during the time calibration process, and provide the theoretical arrival time to the automatic time calibration device;

[0094] The automatic time calibration device is further configured to: compare the actual arrival timestamp with the theoretical arrival time to determine the time calibration value;

[0095] The sensing unit is further configured to correct the time information using the time calibration value when generating the synchronous raw data.

[0096] In practice, the automatic time calibration device is not a standalone hardware entity, but rather can be integrated and run within the data processing device 200 as a dedicated software module. This automatic time calibration device is configured to initiate and manage a time calibration process involving the sensing unit 100 and the data processing device 200, a process designed to determine and store a unique time calibration value for each sensing unit 100.

[0097] After initiating the calibration process, a standard physical pulse must first be applied to the target aerodynamic network.

[0098] An exemplary approach could be to temporarily connect an external calibration pulse generator, such as a high-speed switching valve driven by a signal generator, to a known reference point in the piping network. This pulse generator produces a steep, distinct pressure pulse at a specified time.

[0099] All sensing units 100 in the network are configured to monitor sudden changes in the pressure signal during this calibration process. When they detect that the rising edge slope of the standard physical pulse exceeds a preset threshold, their internal MCU immediately latches and records the current PTP synchronization time as the actual arrival timestamp of the pulse. Subsequently, each sensing unit 100 uploads its recorded actual arrival timestamp to the data processing device 200, which then forwards it to the automatic time calibration device.

[0100] After receiving the actual arrival timestamps of each sensing unit 100, the data processing device 200 calculates the theoretical arrival time of the pulse to each sensing unit based on the pre-stored aerodynamic network topology matrix information.

[0101] Specifically, the data processing device 200 retrieves the known position coordinates of the calibration pulse generator and the position coordinates of each sensing unit 100 from the topology matrix, and calculates the straight-line physical distance between them. Then, based on the material sound velocity stored in the topology matrix, it calculates the theoretical time required for the pressure pulse to travel this distance.

[0102] The automatic time calibration device compares the actual arrival timestamp of each received sensor unit with the corresponding theoretical arrival time calculated by the data processing device 200. Typically, it performs a subtraction operation, for example, time calibration value = theoretical arrival time - actual arrival timestamp, to determine the time calibration value specific to each sensor unit 100.

[0103] After determining the time calibration values ​​for all sensing units, the automatic time calibration device sends these calibration values ​​to the corresponding sensing unit 100. Upon receiving its own time calibration value, the sensing unit 100 stores it in its internal non-volatile memory.

[0104] After completing the calibration process described above, the firmware configuration of the sensing unit 100 is updated. In its subsequent routine operation, when it generates synchronization raw data, it performs a final time correction step: that is, it adds the original sampled PTP timestamp to the locally stored time calibration value to obtain a corrected final timestamp that better reflects the actual physical event's occurrence. This corrected synchronization raw data is then sent to the data processing device 200 for subsequent leak detection and location, thereby greatly improving the accuracy of the final location result.

[0105] As an optional implementation, the generation of the standard physical pulses can be fully automated, requiring no external device access or manual intervention. This further leverages the deep communication and collaboration capabilities between the system of the present invention and the automated packaging equipment controller 300.

[0106] Specifically, to achieve intelligent selection of calibration timing, the automatic time calibration device, as a functional module within the data processing device 200, is further configured to continuously acquire and parse the process cycle parameters from the automated packaging equipment controller 300. By analyzing these parameters, preset idle process periods in the equipment's production flow can be clearly identified. For example, it can identify the interval between completing a full packaging action and the start of the next action, or the standby state when the equipment is waiting for materials.

[0107] After identifying this safe and interference-free preset idle process period, the automatic time calibration device is further configured to proactively send a predefined calibration control command to the automated packaging equipment controller 300. Upon receiving the command, the automated packaging equipment controller 300, according to preset logic, drives one or more preset, fast-response actuators in the target pneumatic network, such as a large-diameter solenoid valve or a quick-release valve of a main pressure cylinder, to perform a rapid opening and closing action within a very short time, such as 50 milliseconds. This action instantly consumes or releases a certain amount of compressed air, thereby generating an ideal standard physical pulse for calibration purposes throughout the pneumatic network.

[0108] For example, this process can be implemented based on a client / server model of OPC UA (Open Platform UA). The automated packaging equipment controller 300 internally runs an OPC UA server. This server publishes its internal key variables, such as a status register representing the current production step number, e.g., DB1.DINT10, as a node in the OPC UA address space. This node has a unique node ID, e.g., ns=2;s="MachineStatus.CycleStep".

[0109] The data processing unit 200 runs an OPC UA client. During system initialization, this client connects to the controller's OPC UA server and subscribes to the aforementioned status nodes. Using a subscription model instead of a polling model ensures that once the status register value in the controller changes, the server immediately and proactively pushes the new value to the data processing unit 200, thereby achieving efficient and real-time parameter acquisition.

[0110] Upon receiving the raw process cycle parameter, such as an integer value "20", the data processing device 200 needs to parse it. For this purpose, the data processing device 200 internally stores an event definition file, which could be a JSON or XML configuration file. This file defines the mapping between each integer value and a specific actuator action event with a clear physical meaning. For example, the file might include the following mapping: {"20": "MainPress_Engage", "35": "SideClamp_Release", "99": "Cycle_End"}. After receiving the raw parameter "20", the data processing device 200 can query the event definition file to parse it into the event "MainPress_Engage" (main clamping action, mold closing), which can be understood by subsequent logic modules.

[0111] Furthermore, after acquiring and parsing meaningful event sequences, the automatic time calibration device can clearly identify the preset idle process periods in the equipment production process based on these events through a state machine logic.

[0112] For example, this state machine may include at least three states: a production running state, a potentially idle state, and a confirmed idle state. Its workflow is as follows:

[0113] When the system is running normally, the state machine is in the production running state.

[0114] When the automatic time calibration device resolves an event that represents the end of a complete packaging process, such as the Cycle_End event, it starts an internal timer and switches the state machine to a potentially idle state.

[0115] In the potential idle state, the device continues to wait for the next event that represents the start of a new process, such as the Cycle_Start event.

[0116] Scenario 1: If the Cycle_Start event is received before the timer reaches a preset continuous production threshold, it indicates that the device is engaged in high-frequency continuous production. In this case, the device will reset the timer and switch the state machine back to the production running state; calibration will not be performed this time.

[0117] Scenario 2: If the timer reaches the continuous production threshold but no Cycle_Start event is received, the system determines that the device has entered a true, stable pause or waiting phase. At this time, the device switches its state machine to the confirmed idle state.

[0118] The duration of this confirmed idle state is identified as the preset idle process period. Only in this state is the automatic time calibration device authorized to send calibration control commands to the controller to ensure that the calibration process does not conflict with any normal production operations. When the Cycle_Start event is received again, the state machine switches back to the production running state, ending the current idle period.

[0119] In this way, the system can utilize the equipment's own tooling and operating intervals to achieve fully automatic, non-intrusive online time calibration, greatly improving the system's intelligence level and maintenance convenience.

[0120] To further improve the long-term operational accuracy of the system described in this invention and to address the slow drift in characteristics caused by factors such as physical wear, aging, or environmental changes, as an optional implementation, the data processing device is further configured as follows:

[0121] In response to the occurrence of a confidence level lower than a preset threshold, the residual signal is defined as a model error signal;

[0122] The model error signal is used to adaptively correct one or more model parameters in the digital twin model.

[0123] The core idea of ​​this process lies in the intelligent reuse and dual-role definition of the residual signal. In a conventional leak diagnosis process, a significant residual signal is considered evidence of a leak. However, in this embodiment, the data processing device 200 is configured to execute a more refined judgment logic:

[0124] When the confidence level of the micro-leakage event calculated based on the residual signal is lower than a preset threshold, the system determines that it is currently in a healthy or leak-free state. In this case, the non-zero, weak residual signal no longer has the physical meaning of a leakage signal, but is redefined as a model error signal. This error signal precisely reflects the subtle deviation between the current digital twin model and its corresponding real physical entity, that is, the degree of model drift.

[0125] Once the residual signal is defined as the model error signal, the data processing device 200 initiates a background-running adaptive correction algorithm. This algorithm uses the model error signal as input to continuously and subtly adjust and correct one or more model parameters in the digital twin model. Through this online, data-driven feedback correction, the digital twin model can continuously improve itself, automatically compensate for the effects of factors such as equipment aging, and maintain a high degree of accuracy in predicting real physical processes.

[0126] This adaptive correction mechanism enables the monitoring system of this invention to evolve from a static diagnostic system into a dynamic intelligent system that can evolve together with the monitored equipment, thereby ensuring extremely high monitoring reliability and accuracy throughout the entire equipment lifecycle.

[0127] Furthermore, in order to address the problem of uneven aging of physical characteristics caused by uneven workload in different branches of the target aerodynamic network, and to achieve a more accurate and efficient model parameter correction, the adaptive correction process can be an event-driven targeted correction method.

[0128] As an optional implementation method, see [link to implementation details]. Figure 3 The flowchart of an adaptive correction method provided in this application embodiment includes steps S201 to S203, wherein:

[0129] S201: The process cycle parameters are parsed into actuator action events, and one or more target branches associated with the actuator action events are determined using the pre-stored pneumatic network topology matrix.

[0130] S202: Extract the piecewise error signal that is temporally associated with the target branch from the model error signal;

[0131] S203: Using the segmentation error signal, update only the dynamic physical parameters related to the target branch in the digital twin model.

[0132] In practice, the data processing device 200 precisely focuses each model correction on the specific branch where a physical action has just occurred. The specific implementation process is as follows:

[0133] First, upon receiving a process cycle time parameter, the data processing device 200 does not broadly modify the entire digital twin model. Instead, it first performs a target recognition step. The received process cycle time parameter, for example, "MainPress_Engage," is parsed into a specific actuator action event. Then, using the event mapping table and node information table pre-stored in the pneumatic network topology matrix, as described above, a precise mapping query is performed: the event mapping table finds the actuator ID associated with the event, such as "Actuator-MainPress," and then the node information table is used to find its corresponding branch ID, such as "Branch-A." In this way, the system binds a higher-level process instruction to a specific physical target branch, "Branch-A."

[0134] Next, error extraction is performed. In the signal processing and diagnostic module of the data processing device 200, a data segment that is closely related in time to the target branch that was just identified is extracted from the continuous, global model error signal stream.

[0135] Specifically, based on the timestamp of the actuator action event, a time window of a predetermined length can be extracted after the timestamp, and the error signal portions from each sensing unit on the target branch "Branch-A" within this window can be combined into a segmented error signal specifically for this correction.

[0136] Finally, a targeted update is performed. This piecewise error signal is input into an adaptive correction algorithm, which could be, for example, a recursive least squares algorithm module.

[0137] Importantly, the algorithm's output—the calculated parameter corrections—only applies to the dynamic physical parameters in the digital twin model that are directly related to the identified target branch "Branch-A". For example, the algorithm updates the pipe resistance coefficients of each segment on "Branch-A" or updates the gas consumption model parameters of "Actuator-MainPress", while all model parameters related to other unrelated branches such as "Branch-B" and "Branch-C" remain unchanged during this update cycle.

[0138] In this way, through this event-driven, topology-linked targeted correction mechanism, this application can accurately track and compensate for the unique aging trajectory of each independent branch, improving the efficiency and accuracy of model adaptive correction, and enabling the digital twin model to maintain extremely high fidelity during long-term operation.

[0139] To address the technical problem that the dynamic coupling effect between branches of the aerodynamic network may cause crosstalk components from adjacent branches to be mixed in the segmented error signal, thus affecting the correction accuracy, as an optional implementation method, see [link to implementation details]. Figure 4 The flowchart of a method for extracting a segmented error signal that is temporally associated with the target branch, provided in an embodiment of this application, includes steps S301 to S304, wherein:

[0140] S301: Perform multi-scale wavelet transform on the model error signal to obtain wavelet coefficients distributed in the time-frequency domain;

[0141] S302: Using the pneumatic network topology matrix, identify one or more adjacent branches that are adjacent to the target branch, and based on the topological relationship between the adjacent branches and the target branch, predict the time-frequency characteristics of the crosstalk signal caused by the actuator action event in the adjacent branches.

[0142] S303: Construct a wavelet domain mask based on the aforementioned time-frequency features;

[0143] S304: Filter the wavelet coefficients using the wavelet domain mask, then perform inverse wavelet transform to reconstruct and obtain the purified segmented error signal.

[0144] In practical implementation, the signal processing and diagnosis module of the data processing device 200 employs more sophisticated signal processing techniques when performing the error extraction step. First, the module performs a multi-scale wavelet transform on the model error signal or its segment within the target time window.

[0145] For example, the Complex Morlet wavelet, which is suitable for analyzing transient impact signals, can be selected as the basis function to decompose the one-dimensional time-domain error signal into a two-dimensional time-frequency matrix that can simultaneously display time and frequency information, namely the wavelet coefficient matrix.

[0146] Next, the system will execute a crosstalk prediction process based on a physical model and guided by topological information.

[0147] Specifically, the data processing unit 200 uses the aerodynamic network topology matrix to perform a graph theory search algorithm, such as Dijkstra's algorithm, to identify the shortest physical path from the target branch where the actuator's action event occurs to the sensor on each adjacent branch. Subsequently, by accumulating the path lengths and combining them with preset sound velocity information, the theoretical propagation delay of the pressure wave is calculated. Based on a simplified attenuation and dispersion model, the energy attenuation and main frequency band of the crosstalk signal after propagation can be estimated. By synthesizing these calculation results, the system can generate a complete time-frequency characteristic prediction for each adjacent branch, including time, frequency, and amplitude information about its crosstalk signal.

[0148] Based on this prediction, the wavelet domain mask can be constructed. This mask is essentially a two-dimensional weight matrix of the same size as the wavelet coefficient matrix. First, the entire mask matrix is ​​initialized to the pass value "1". Then, all predicted crosstalk signal time-frequency features are iterated over, and the weight values ​​in the time-frequency regions corresponding to these features in the mask matrix are modified to the suppression value "0".

[0149] Finally, by performing element-wise multiplication of the original wavelet coefficient matrix with the constructed wavelet domain mask, all signal components predicted as crosstalk can be accurately suppressed. Performing an inverse wavelet transform on the filtered new wavelet coefficient matrix then reconstructs and obtains the final, purified piecewise error signal. This high-purity signal will serve as input for subsequent targeted update steps to ensure the accuracy of model parameter correction.

[0150] For example, taking action event A as an example, the complete calculation process of this procedure is explained in detail.

[0151] Deployed on the target branch Branch-A, its actions will cause dynamic coupling crosstalk to its nearest neighboring branch Branch-B.

[0152] After extracting time-series data (e.g., 1024 sampling points) related to the action event from the global model error signal, the data processing device 200 calls its internal digital signal processing library to perform a continuous wavelet transform on the one-dimensional sequence. In a specific embodiment, the selected mother wavelet can be a Complex Morlet wavelet with good time-frequency locality, such as cmor 1.5-1.0, where 1.5 is the bandwidth parameter and 1.0 is the center frequency. The scale sequence of the transform can be set as a logarithmically evenly spaced sequence from 2 to 128, which can cover the entire frequency band from the high-frequency impact generated by the actuator action to the low-frequency fluctuations of secondary oscillations in the pipeline network. The result of this transform is a complex matrix of 128 rows (scale axis) × 1024 columns (time axis), i.e., the wavelet coefficients. Taking the modulus of each element of this matrix yields the time-spectrum for analysis and visualization.

[0153] Based on action event A, the data processing unit 200 queries the event mapping table and node information table in the pneumatic network topology matrix to confirm that the action source is located in the target branch, Branch-A. Then, by querying the connection relationships in the pipeline information table, it identifies Branch-B as the adjacent branch of Branch-A.

[0154] Subsequently, the system initiates a pathfinding algorithm, such as Dijkstra's algorithm, to calculate the shortest physical path from the action source node on Branch-A to the sensor node on Branch-B in the network graph defined by the topology matrix. The system then sums the pipe lengths of all segments along this path to obtain the total propagation distance, for example, 2.5 meters. Based on the preset sound velocity for the pipe material in the topology matrix, for example, 340 meters per second, the theoretical propagation delay is calculated to be approximately 2.5 / 340 ≈ 7.4 milliseconds.

[0155] Simultaneously, based on an attenuation and dispersion model, signal energy loss and frequency band changes can be estimated. For example, the model can predict that the energy of the crosstalk signal will attenuate to less than 5% of the original main signal energy, and that since high-frequency components attenuate faster in fluids, their main energy will be concentrated in the lower frequency band of 100-300 Hz. Finally, a specific time-frequency characteristic prediction is given for Branch-B crosstalk synthesis: "A weak crosstalk signal is expected to appear in the frequency range of 100-300 Hz within a time window of 7.4 ms ± 2 ms after the main action."

[0156] For example, the data processing device 200 creates a wavelet domain mask matrix in memory that is the same size as the aforementioned wavelet coefficient matrix (128 × 1024) and initializes all its element values ​​to 1.0 (representing complete pass).

[0157] Next, the system converts the predicted crosstalk time-frequency features from the previous step into index coordinates of the mask matrix. For example, the frequency range [100Hz, 300Hz] might correspond to rows 60 to 80 of the scale axis, while the time window [5.4ms, 9.4ms] might correspond to columns 27 to 47 of the time axis. Then, the system modifies all element values ​​within the rectangular region defined by [rows 60:80, columns 27:47] in the mask matrix from 1.0 to 0.0 (representing complete suppression). If there are crosstalk predictions for multiple adjacent branches, this process is repeated to etch multiple suppression regions onto the mask.

[0158] Furthermore, the signal processing and diagnostic module of the data processing device 200 performs an element-wise matrix multiplication on the original wavelet coefficient matrix obtained in the first step and the wavelet domain mask matrix constructed in the previous step. After this operation, the energy of the wavelet coefficients originally located in the crosstalk region is basically cleared to zero, while the coefficients in the main signal region remain unchanged. Finally, the module calls the Inverse Continuous Wavelet Transform (Inverse CWT) function, using this purified new wavelet coefficient matrix as input, to reconstruct the signal, and finally outputs a one-dimensional time series, which is the required purified piecewise error signal.

[0159] For example, the goal of attenuation and dispersion models is to quickly calculate, based on known physical principles, how the waveform characteristics of a disturbance signal generated in a target branch will evolve after propagating to adjacent branches.

[0160] The attenuation model can be implemented as a frequency-dependent energy attenuation function. The core of this model is that it treats the attenuation coefficient α of the pressure wave as a function of the frequency f, i.e., α(f). According to the acoustic theory of fluids inside pipes, attenuation mainly stems from the viscosity of the pipe wall and heat conduction losses. A suitable simplified relationship is that the attenuation coefficient is approximately proportional to the square root of the frequency.

[0161] When performing prediction, the data processing unit 200 performs the following calculations: First, it knows the initial frequency band of the original disturbance signal, for example, a pulse generated by the action of an actuator. Then, based on the aerodynamic network topology matrix, it obtains the physical distance L that the signal needs to propagate. For each frequency component in the original frequency band, an attenuated amplitude can be calculated. Since the attenuation coefficient α(f) of the high-frequency component is much larger than that of the low-frequency component, after propagating over a long distance L, the high-frequency components in the original signal will be disproportionately and more drastically attenuated.

[0162] The dispersion model can be implemented as a phenomenological description of the consequences of the aforementioned attenuation model. That is, the system does not need to calculate complex, frequency-varying phase velocities, but can directly model its final effect.

[0163] Considering both attenuation and dispersion effects, the data processing device 200 can make the following prediction: a sharp pulse signal whose energy is mainly concentrated in the 500-1000 Hz frequency band at the source (target branch) will, after propagating a distance of 2.5 meters to reach the adjacent branch, have its main energy transferred and concentrated to a lower frequency band, such as 100-300 Hz. Simultaneously, the model can also include a simplified pulse broadening parameter to describe the temporal dispersion effect of the signal; that is, a sharp pulse with a duration of 10 milliseconds may broaden into a smoother waveform with a duration of 25 milliseconds after propagation.

[0164] Ultimately, the model outputs a complete time-frequency characteristic of the crosstalk signal, including the theoretical propagation delay, the predicted energy attenuation ratio, the predicted center frequency band, and the predicted time width. This specific, quantified characteristic description will serve as the accurate basis for subsequently constructing the wavelet domain mask.

[0165] As an optional implementation, the data processing device also stores a packaging process feature library, which stores reference time-frequency features corresponding to various preset packaging action types of the automated packaging equipment.

[0166] The predicted time-frequency characteristics of the crosstalk signal caused by the actuator action event in the adjacent branch include:

[0167] Identify the corresponding packaging action type from the action events of the actuator;

[0168] Extract the reference time-frequency features corresponding to the identified packaging action type from the packaging process feature library;

[0169] The time-frequency characteristics of the crosstalk signal are calculated based on the reference time-frequency characteristics and the topological relationship between the adjacent branch and the target branch.

[0170] Furthermore, in order to further improve the prediction accuracy of the crosstalk signal, this application enables the data processing device 200 not only to understand the physical layout of the pipeline network, but also to deeply understand the physical "personality" of each packaging action.

[0171] To this end, the data processing unit 200 also pre-stores a structured packaging process feature library in its internal memory. This feature library is generated during the offline calibration phase before system deployment, and its purpose is to establish a precise signal fingerprint for each unique packaging action. During the calibration process, each preset packaging action type, such as a high-speed "tightening action" or a slow "corner protection positioning action," is triggered individually and repeatedly on a reference pneumatic network, and a reference-level sensor located close to the action source is used to acquire the purest raw pressure pulsation waveform of the action. After the acquired signal is processed by signal averaging, wavelet transform, etc., its key reference time-frequency features are extracted and parameterized. For example, the reference features of a "tightening action" can be stored as a set of parameters, specifically {Action type: "TENSION", Reference center frequency: 850Hz, Reference bandwidth: 300Hz, Reference energy: 1.0, Reference duration: 12ms}. The entire feature library consists of parameterized fingerprints of multiple such action types.

[0172] When the system enters online monitoring mode and performs crosstalk prediction, its internal process is optimized into a highly intelligent three-step deduction:

[0173] The first step is to identify the action type. After receiving the process cycle parameters and parsing them into actuator action events, the data processing unit 200 further queries the event mapping table in the pneumatic network topology matrix or a dedicated process knowledge base. The purpose of this query is to identify the specific packaging action type corresponding to the current action event. For example, the event "MainPress_Engage" is identified as type "TENSION".

[0174] The second step is to extract the reference features. Once the action type is identified as "TENSION", the data processing device 200 will immediately access the packaging process feature library and extract the corresponding reference time and frequency feature parameter set, which includes information such as center frequency, bandwidth, energy and duration.

[0175] The third step is propagation prediction based on a reference. The data processing unit 200 uses the highly realistic reference time-frequency characteristics extracted in the previous step as the initial disturbance source for the simulation calculation, rather than a general, hypothetical standard pulse. It then initiates the aforementioned physical propagation model and performs calculations in conjunction with information provided by the aerodynamic network topology matrix. For example, based on the reference center frequency of 850Hz and the pipe resistance coefficient along the propagation path, it applies a frequency-dependent attenuation model to calculate that after propagating to adjacent branches, the main energy of the signal will attenuate and concentrate at a new, lower center frequency, such as 180Hz. Simultaneously, it applies a dispersion model; for example, it calculates that a pulse with a reference duration of 12ms will be broadened to 30ms after propagation. Ultimately, the result is no longer a fuzzy estimate, but a specific, quantified prediction of the time-frequency characteristics of the crosstalk signal.

[0176] In this way, by deeply integrating the packaging process knowledge base with the physical propagation model, the wavelet domain mask constructed by the system can achieve unprecedented accuracy. Its target is no longer any possible noise, but rather crosstalk signals with specific fingerprints, occurring at specific times and in specific frequency bands, and which can be accurately predicted. This optimizes the purification of segmented error signals, providing the highest quality data input for subsequent adaptive model correction, and elevating the intelligence and reliability of the entire system to a new level.

[0177] As an optional implementation, obtaining the time delay information includes:

[0178] From the plurality of sensing units, the sensing unit that receives the residual signal with the strongest energy or the earliest time is determined as the reference sensing unit.

[0179] For at least one other sensing unit besides the reference sensing unit, the residual signal of the reference sensing unit and the residual signal of the other sensing unit are respectively subjected to Fourier transform to obtain their respective spectral signals.

[0180] The cross-power spectrum is calculated based on the obtained spectral signal, and the cross-power spectrum is phase-transformed and weighted to obtain a whitened spectrum.

[0181] Perform an inverse Fourier transform on the whitened spectrum to obtain the cross-correlation function;

[0182] Identify the peak position of the cross-correlation function to determine the time delay between the reference sensing unit and the other sensing units;

[0183] The one or more time delays thus determined are combined to form the time delay information.

[0184] As an optional implementation, the step of performing inversion calculations based on a pre-stored aerodynamic network topology matrix to determine the branch identifier where a micro-leakage event occurred includes:

[0185] Based on the aerodynamic network topology matrix, the target aerodynamic network is discretized into multiple candidate leak points;

[0186] For each of the candidate leak points, the theoretical time delay when the leak occurs at that candidate leak point is calculated using the aerodynamic network topology matrix;

[0187] The calculated theoretical time delay is compared with the time delay information to calculate the likelihood value of the leak location corresponding to each candidate leak point;

[0188] Identify the candidate leak point with the highest likelihood value of the leak location, and determine the branch where the candidate leak point is located as the branch identifier of the micro-leak event.

[0189] To further improve the robustness and accuracy of leak location in complex situations where the signal-to-noise ratio of micro-leakage signals is extremely low and measurement data may have slight inconsistencies, the location procedure in step S105 can be implemented using a more advanced combination of algorithms.

[0190] After the positioning program is started, it first performs a high-precision time delay information acquisition process. The first step of this process is to preprocess the multiple residual signals to determine the reference sensing unit.

[0191] Specifically, the data processing device 200 can calculate the total energy or peak amplitude of each residual signal within a short time window, and dynamically select the sensing unit corresponding to the signal with the strongest energy or the earliest arrival time as the reference sensing unit for this positioning calculation.

[0192] Next, for other relevant sensing units in the network besides the reference sensing unit, the system pairs their respective residual signals with the residual signal of the reference sensing unit and performs generalized cross-correlation-phase transform (GCC-PHAT) calculations on each pair. The core of this calculation is to discard amplitude information that is easily contaminated by noise and focus on the most reliable phase relationship between signals.

[0193] In practice, a Fast Fourier Transform (FFT) is first performed on the two time-series signals to obtain their complex spectra. Then, their cross-power spectra are calculated and normalized using a phase-transform weighting function—that is, dividing the cross-power spectrum by its own modulus. Mathematically, this weighting operation whitens the spectrum, physically suppressing noise-dominated frequency components while retaining only the phase spectrum information that accurately reflects the time delay. Subsequently, an Inverse Fast Fourier Transform (IFFT) is performed on this weighted and whitened spectrum to obtain a cross-correlation function with extremely sharp peaks.

[0194] Finally, a peak search algorithm, such as maximum value lookup or parabolic interpolation, is used to accurately identify the peak position of the function and convert it into a precise time delay between the two sensing units. After completing the pairing calculations for all relevant sensing units, the system obtains a set of high-precision time delay information relative to the same reference point.

[0195] After obtaining this set of high-precision time delay information, the system then initiates an inversion calculation program based on maximum likelihood estimation to achieve highly robust localization. This program first generates a series of discrete candidate leak points along all paths of the pipeline network based on the aerodynamic network topology matrix, at preset physical steps, such as 5 centimeters, thereby transforming the continuous pipeline network space into a discrete set of points that can be searched.

[0196] The program then enters a traversal calculation loop. For each candidate leak point, it uses the pipe length and sound velocity information stored in the topology matrix to calculate the theoretical propagation time required for the pressure wave to reach the reference sensing unit and all other relevant sensing units when the leak source is precisely located at that point, thus obtaining a set of theoretical time delays.

[0197] Next, the program compares this set of theoretical time delays with the time delay information actually calculated by the GCC-PHAT algorithm to evaluate the "credibility" of the current candidate point.

[0198] In practice, the likelihood value of the leak location can be obtained by calculating the Gaussian probability density between two sets of time delay vectors; in short, the smaller the difference between the theoretical value and the actual value, the higher the likelihood value of the candidate point.

[0199] After traversing all candidate leak points and calculating their respective likelihood values, the program performs a maximum value search operation to find the candidate leak point with the global maximum leak location likelihood value. Finally, the system queries the node information table of the topology matrix to determine the branch to which the optimal candidate point belongs, and outputs the branch ID as the final branch identifier of the micro-leak event. By combining the robust time delay estimation algorithm of GCC-PHAT with the robust localization algorithm of maximum likelihood, this embodiment can achieve accurate and reliable localization of micro-leak events even in extremely harsh noisy environments.

[0200] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A remote monitoring system for automated packaging equipment, characterized in that, include: Multiple sensing units are arranged at the inlet and outlet of each branch of the target aerodynamic network to synchronously collect the inlet pressure, outlet pressure and flow rate of each branch under a unified clock reference, and generate synchronous raw data. A data processing device is communicatively connected to the sensing unit and the controller of the automated packaging equipment. The data processing device stores a digital twin model and is configured as follows: Receive process cycle parameters from the controller of the automated packaging equipment; Based on the process cycle parameters and the digital twin model, the theoretical pulsation reference waveform for the current working condition is calculated. The original synchronous data and the theoretical pulsating reference waveform are differentially processed to generate a residual signal; The occurrence confidence level of the micro-leakage event is determined based on the residual signal, and micro-leakage event information is generated in response to the occurrence confidence level being greater than or equal to a preset threshold. Based on the micro-leakage event information, cross-correlation calculations are performed on the residual signals of different branches to obtain time delay information. Inversion calculations are then performed using the pre-stored pneumatic network topology matrix to determine the branch identifier where the micro-leakage event occurred. Based on the branch identifier, a micro-leakage event alarm command is sent to the automated packaging equipment controller. The data processing device is further configured to: in response to the occurrence of a confidence level lower than a preset threshold, define the residual signal as a model error signal; Using the model error signal, one or more model parameters in the digital twin model are adaptively corrected; The adaptive correction includes: parsing the process cycle parameters into actuator action events, and using the pre-stored pneumatic network topology matrix to determine one or more target branches associated with the actuator action events; Perform multi-scale wavelet transform on the model error signal to obtain wavelet coefficients distributed in the time-frequency domain; Using the pneumatic network topology matrix, one or more adjacent branches adjacent to the target branch are identified. Based on the topological relationship between the adjacent branches and the target branch, the time-frequency characteristics of the crosstalk signal caused by the actuator action event in the adjacent branches are predicted. Based on the time-frequency characteristics, a wavelet domain mask is constructed. The wavelet coefficients are filtered using the wavelet domain mask, followed by inverse wavelet transform, to reconstruct and obtain the purified piecewise error signal. Using the segmented error signal, only the dynamic physical parameters related to the target branch in the digital twin model are updated.

2. The remote monitoring system for automated packaging equipment according to claim 1, characterized in that, Also includes: An automatic time calibration device is configured to initiate and manage the time calibration process of the sensing unit and the data processing device, and to determine and store time calibration values ​​corresponding to each of the sensing units. The sensing unit is further configured to: record the actual arrival timestamp of the standard physical pulse it receives during the time calibration process, and provide the actual arrival timestamp to the automatic time calibration device; The data processing device is further configured to: calculate the theoretical arrival time of the standard physical pulse to each of the sensing units during the time calibration process, and provide the theoretical arrival time to the automatic time calibration device; The automatic time calibration device is further configured to: compare the actual arrival timestamp with the theoretical arrival time to determine the time calibration value; The sensing unit is further configured to correct the time information using the time calibration value when generating the synchronous raw data.

3. The remote monitoring system for automated packaging equipment according to claim 2, characterized in that, The automatic time calibration device is further configured as follows: The process cycle time parameters are obtained from the controller of the automated packaging equipment, and a preset idle process time period is identified based on the process cycle time parameters.

4. The remote monitoring system for automated packaging equipment according to claim 3, characterized in that, The automatic time calibration device is further configured as follows: During the preset idle process period, a calibration control command is sent to the controller of the automated packaging equipment to drive the actuator in the target pneumatic network to generate the standard physical pulse.

5. The remote monitoring system for automated packaging equipment according to claim 1, characterized in that, The data processing device also stores a packaging process feature library, which stores reference time-frequency features corresponding to various preset packaging action types of the automated packaging equipment. The predicted time-frequency characteristics of the crosstalk signal caused by the actuator action event in the adjacent branch include: Identify the corresponding packaging action type from the action events of the actuator; Extract the reference time-frequency features corresponding to the identified packaging action type from the packaging process feature library; The time-frequency characteristics of the crosstalk signal are calculated based on the reference time-frequency characteristics and the topological relationship between the adjacent branch and the target branch.

6. The remote monitoring system for automated packaging equipment according to claim 1, characterized in that, The obtained time delay information includes: From the plurality of sensing units, the sensing unit that receives the residual signal with the strongest energy or the earliest time is determined as the reference sensing unit. For at least one other sensing unit besides the reference sensing unit, the residual signal of the reference sensing unit and the residual signal of the other sensing unit are respectively subjected to Fourier transform to obtain their respective spectral signals. The cross-power spectrum is calculated based on the obtained spectral signal, and the cross-power spectrum is phase-transformed and weighted to obtain a whitened spectrum. Perform an inverse Fourier transform on the whitened spectrum to obtain the cross-correlation function; Identify the peak position of the cross-correlation function to determine the time delay between the reference sensing unit and the other sensing units; The one or more time delays thus determined are combined to form the time delay information.

7. The remote monitoring system for automated packaging equipment according to claim 6, characterized in that, The inversion calculation, which combines the pre-stored aerodynamic network topology matrix, determines the branch identifiers where micro-leakage events occur, including: Based on the aerodynamic network topology matrix, the target aerodynamic network is discretized into multiple candidate leak points; For each of the candidate leak points, the theoretical time delay when the leak occurs at that candidate leak point is calculated using the aerodynamic network topology matrix; The calculated theoretical time delay is compared with the time delay information to calculate the likelihood value of the leak location corresponding to each candidate leak point; Identify the candidate leak point with the highest likelihood value of the leak location, and determine the branch where the candidate leak point is located as the branch identifier of the micro-leak event.

Images