Communication and interaction method of digital twin system in compressor blisk machining process

Abstract

The invention provides a communication and interaction method of a digital twinning system in a compressor blisk machining process, and relates to the technical field of digital twinning and aeronautical manufacturing, and the method comprises the steps: collecting vibration, temperature and acoustic emission data in the compressor blisk machining process in real time, analyzing machine tool machining data, carrying out the alignment and noise reduction of the collected data, and obtaining a digital twinning result; then time-frequency domain mixed feature extraction is carried out, and time-frequency domain mixed feature data are optimized and uploaded in a graded mode; a deterministic transmission network is constructed, data transmission priorities of 5G communication and LoRa dual channels are dynamically adjusted according to a network load, and intelligent switching between 5G communication and a LoRa protocol is realized based on a switching decision model; analyzing normal processing data of the whole stage of the processing process, and establishing a processing data thread; and predicting the abnormal processing risk in the processing process, and updating the model parameters. According to the method, the communication and interaction efficiency of the digital twin system in the machining process of the compressor blisk in the aero-engine can be improved.

Description

Technical Field

[0001] The present invention relates to the field of digital twin and aviation manufacturing technology, and in particular to a communication and interaction method of a digital twin system for a compressor integral blade disk processing process. Background Art

[0002] The surface accuracy of the blades of the integral blisk must reach the micron level. During the machining process, the digital twin system needs to monitor multi-dimensional data such as vibration, cutting force, and temperature in real time. The sensor sampling frequency often reaches the kHz level, and a single device generates MB of data per second. The MQTT / HTTP protocol of the current traditional digital twin system adopts a "request-response" model, which has high transmission latency and is difficult to carry high-frequency data streams. The real-time feedback delay of the signal may result in abnormal working conditions in the machining process not being suppressed in time, causing vibration marks on the blade surface or tool chipping, resulting in scrapped workpieces. At the same time, unprocessed raw data is directly uploaded to the cloud, occupying bandwidth and increasing the cloud computing load. The wasted bandwidth leads to the obstruction of critical data transmission, and the cloud processing delay affects process optimization decisions and increases storage costs. Summary of the Invention

[0003] In view of this, an embodiment of the present application provides a communication and interaction method for a digital twin system of a compressor integral blade disk processing process, which supports integrated collaborative control of real-time synchronous acquisition of multimodal data, edge intelligent preprocessing, deterministic low-latency transmission, and cloud-based dynamic simulation, and can improve the communication and interaction efficiency of the digital twin system of a compressor integral blade disk processing process in an aircraft engine.

[0004] The present application provides the following technical solution: a communication and interaction method for a digital twin system of a compressor blisk machining process, comprising: Real-time collection of vibration, temperature, and acoustic emission data during compressor blade machining. Analyzing machine tool processing data. Performing multi-source time series data alignment and noise reduction on the vibration, temperature, and acoustic emission data. Extracting time-frequency domain hybrid features from the processed data to obtain time-frequency domain hybrid feature data. Using an improved LZMA2 compression algorithm, the data is optimized and uploaded in stages. A deterministic transmission network is built based on the TSN-OPC UA architecture. The data transmission priority of the 5G communication and LoRa dual channels is dynamically adjusted according to the network load of the transmission network. A switching decision model based on reinforcement learning is used to achieve intelligent switching between 5G communication and LoRa protocols. The normal processing data in the machine tool processing data of the entire compressor integral blade disk processing process is analyzed to establish a processing data thread; based on the processing data thread, a time series prediction model is used to predict the processing abnormality risk in the processing process, and the model parameters are updated through real-time data flow.

[0005] According to one embodiment of the present application, real-time collection of vibration, temperature, and acoustic emission data during the machining of a compressor blade disk and analysis of the machine tool machining data include: A vibration-temperature-acoustic emission multimodal sensor array is constructed to collect vibration, temperature, and acoustic emission data during the machining of the compressor integral blade; and the machining data is analyzed using the TCP / IP protocol stack.

[0006] According to one embodiment of the present application, multi-source time series data alignment and noise reduction processing are performed on the vibration, temperature, and acoustic emission data, including: A dynamic time warping algorithm is used to construct a distance matrix and dynamic planning path search to accurately align asynchronous time series data from multiple sources of vibration, temperature, and acoustic emission; principal component analysis is combined to eliminate common-mode interference between multiple sensors.

[0007] According to one embodiment of the present application, time-frequency domain mixed feature extraction is performed on the processed data to obtain time-frequency domain mixed feature data, including: The processed data is synchronously segmented and a sliding window is used to intercept data signal segments. The time domain features of each segment are calculated and counted in the time domain. The frequency domain features of each frequency band are extracted by fast Fourier transform in the frequency domain. The time-frequency joint features of the time and frequency domains are calculated simultaneously. The time domain features, the frequency domain features and the time-frequency joint features are fused to form a mixed feature vector, and key features are selected as the time-frequency domain mixed feature data through a maximum correlation minimum redundancy feature selection algorithm.

[0008] According to an embodiment of the present application, the time-frequency domain mixed feature data is optimized and uploaded in stages using an improved LZMA2 compression algorithm, including: The time-frequency domain mixed feature data is binary serialized, and the serialized feature data is subjected to lossy-lossless hybrid compression using an improved LZMA2 algorithm. The compression ratio is dynamically adjusted through a feature importance weight matrix, and the compressed data packets are hierarchically uploaded through the MQTT protocol.

[0009] According to an embodiment of the present application, a deterministic transmission network is constructed based on the TSN-OPC UA architecture, including: Based on the OPC-UA architecture, the TSN network is used for time synchronization and deterministic transmission, including: building a TSN network infrastructure that complies with the IEEE802.1AS-2020 standard, configuring industrial switches that support time-aware shaping, and achieving μs-level time synchronization across the entire network through the master clock based on the PTPv2 protocol; deploying the UA PubSub model on the OPC UA server side, encapsulating the machine tool processing data and the optimized time-frequency domain mixed feature data into UADP messages, and marking the TSN traffic category.

[0010] According to an embodiment of the present application, dynamically adjusting the data transmission priority of 5G communication and LoRa dual channel according to the network load of the transmission network includes: Adopting an adaptive bandwidth allocation algorithm to dynamically adjust data transmission priority based on network load, including: Real-time monitoring of RTT delay, packet loss rate, and available bandwidth of 5G communication and LoRa dual channels; based on the monitoring data, the LSTM network is used to predict network load trends within a set time period in the future; Construct a Q-learning model to establish the state space and action space. The state space includes transmission mode switching, compression ratio adjustment, and priority reassignment. When the prediction indicator exceeds the set network congestion threshold, the transmission priority of non-critical data is dynamically downgraded, while the transmission level of abnormal vibration segments is increased.

[0011] According to one embodiment of the present application, normal processing data in the machine tool processing data of all stages of the compressor blisk processing process is analyzed to establish a processing data thread, including: Integrate multimodal data streams including spindle power, tool three-dimensional vibration, workpiece temperature and acoustic emission signals based on a timestamp alignment mechanism; A dynamic time warping algorithm is used to process the multimodal data stream to eliminate the timing drift between sensors, and the machining process is divided into three stages: rough machining, semi-finishing, and finishing. Time domain features, frequency domain features, and time-frequency joint features are extracted for each window to construct a multidimensional hybrid feature vector. The improved t-SNE algorithm is used to perform dimensionality reduction and visualization on the multidimensional mixed feature vector, and DBSCAN clustering is combined to identify stable working condition data clusters in each processing stage; a three-layer data thread structure is established, including the original data layer, feature engineering layer and working condition label layer.

[0012] According to an embodiment of the present application, the time series prediction model adopts an LSTM-Transformer hybrid model.

[0013] According to one embodiment of the present application, it also includes: dynamically adjusting the sampling frequency of the sensor according to the simulation results of the digital twin system, and selecting a reduced-order model for fast simulation and high-precision simulation.

[0014] Compared with the traditional method, the beneficial effects that can be achieved by at least one of the above-mentioned technical solutions adopted in the embodiments of this specification include at least the following: The present invention proposes a method for constructing an edge intelligent perception and preprocessing module considering the LZMA2 lossy-lossless hybrid compression strategy. This method can maintain the integrity of data features while reducing the amount of data transmission. The present invention proposes a method for constructing an adaptive industrial communication module combined with a dual-channel Q-learning bandwidth allocation algorithm, which can dynamically adjust the 5G / LoRa transmission mode according to the network status to ensure the reliability of critical data transmission in the event of sudden congestion. The present invention proposes a method for constructing a cloud-based intelligent analysis and decision-making module based on an LSTM-Transformer hybrid prediction model and a reduced-order / high-precision model dynamic switching strategy, which improves the simulation speed while ensuring sufficient simulation accuracy. The method proposed in the present invention can effectively improve the communication and interaction efficiency of the digital twin system of the integral blade processing process of the compressor of an aero-engine, and solve the technical problems of low data transmission efficiency and large errors in traditional methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0016] Figure 1 It is a flow chart of the communication and interaction method of the digital twin system of the compressor integral blade disk processing process according to an embodiment of the present invention. DETAILED DESCRIPTION

[0017] The embodiments of the present application are described in detail below with reference to the accompanying drawings.

[0018] The following describes the embodiments of the present application through specific examples, and those skilled in the art can easily understand other advantages and effects of the present application from the contents disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. The present application can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present application. It should be noted that, in the absence of conflict, the features in the following embodiments and embodiments can be combined with each other. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without making creative work are within the scope of protection of this application.

[0019] An embodiment of the present invention provides a communication and interaction method for a digital twin system of a compressor blisk machining process, comprising: Real-time collection of vibration, temperature, and acoustic emission data during compressor blade machining. Analyzing machine tool processing data. Performing multi-source time series data alignment and noise reduction on the vibration, temperature, and acoustic emission data. Extracting time-frequency domain hybrid features from the processed data to obtain time-frequency domain hybrid feature data. Using an improved LZMA2 compression algorithm, the data is optimized and uploaded in stages. A deterministic transmission network is built based on the TSN-OPC UA architecture. The data transmission priority of the 5G communication and LoRa dual channels is dynamically adjusted according to the network load of the transmission network. A switching decision model based on reinforcement learning is used to achieve intelligent switching between 5G communication and LoRa protocols. The normal processing data in the machine tool processing data of the entire compressor integral blade disk processing process is analyzed to establish a processing data thread; based on the processing data thread, a time series prediction model is used to predict the processing abnormality risk in the processing process, and the model parameters are updated through real-time data flow.

[0020] An embodiment of the present invention proposes an efficient communication and interaction method for a digital twin system of a compressor integral blade disk processing process, including: an edge intelligent perception and preprocessing module uses a multimodal sensor array to collect vibration, temperature and acoustic emission data in real time during the processing process, uses the TCP / IP protocol stack to efficiently analyze machine tool parameters, and realizes multi-source data alignment and noise reduction through dynamic time warping and principal component analysis, further extracts time-frequency domain mixed features, and uses an improved LZMA2 compression algorithm to optimize data volume and upload in stages, providing lightweight, high-fidelity preprocessed data streams for edge computing; an adaptive industrial communication module builds a deterministic transmission network based on the TSN-OPC UA architecture, and dynamically adjusts the transmission priority and compression strategy of the 5G / LoRa dual channel by combining the LSTM network and the Q-learning model to ensure continuous transmission of critical data; a cloud-based intelligent analysis and decision-making module integrates multimodal processing data streams, establishes phased data threads based on dynamic time warping and the improved t-SNE algorithm, uses the LSTM-Transformer hybrid model to predict processing anomaly risks, and optimizes model parameters through incremental learning to perform a dynamic switching strategy between reduced-order models and high-precision finite element simulations.

[0021] In some embodiments of the present invention, Figure 1 As shown, an efficient communication and interaction method for a digital twin system of a compressor blisk processing process according to this embodiment is as follows: 1. Building Edge Intelligent Perception and Preprocessing Modules Step (1) constructs a vibration-temperature-acoustic emission multimodal sensor array, which is specifically implemented as follows: For data collection during blisk machining, three sensors are selected: vibration sensors, temperature sensors, and acoustic emission sensors. The sensor layout on the machine tool spindle side is as follows: a triaxial accelerometer (X, Y, and Z directions) is installed on each of the front and rear spindle bearing blocks to capture radial runout and axial movement. A thin-film NTC sensor is embedded in the spindle front end, within 5 mm of the bearing raceway, to monitor friction temperature rise. Four acoustic emission sensors are distributed evenly around the spindle-toolholder interface to detect tool wear and chipping. Single-axis vibration sensors are installed at the ends of the X, Y, and Z guideways to monitor guideway clearance anomalies. A temperature sensor is integrated into the ball screw nut seat to compensate for thermal expansion errors. The sensor layout on the blisk side is as follows: six triaxial sensors are distributed evenly around the blade root flange to cover blade bending and torsional modes. An acoustic emission array is placed circumferentially one-third of the way from the blade tip to capture crack growth signals. A micro-thin-film NTC is embedded in the bottom of the mortise and tenon groove and secured by laser welding. A circular MEMS array is installed on the outer edge of the disc to simultaneously collect dynamic strain under centrifugal force.

[0022] Step (2) is to parse the machine tool processing parameters and processing data through the TCP / IP protocol stack. The specific implementation is as follows: To collect machining parameters during blisk machining, dual network cards are enabled on the machine tool's CNC system. The primary network card is configured for EtherNet / IP, with an IP address set to enable CIP communication services. The secondary network card is configured for Modbus TCP, with an IP address set and port 502 open. An M12 X-coded industrial Ethernet connector is used to connect to the protocol conversion gateway via a Cat6A shielded twisted-pair cable. The Modbus TCP / IP protocol stack hybrid parsing system implemented in C++ adopts a multi-threaded architecture design. First, a TCP connection is established, and a request frame that complies with the Modbus protocol specification (including fields such as transaction ID, protocol ID, function code, and CRC checksum) is constructed. The read instruction is sent through the send() function. In the data receiving thread, select() is used to implement non-blocking I / O multiplexing, and the returned response frame is strictly checked (including transaction ID matching, length verification, and CRC checksum). In particular, for the response data of the reading holding register, functions such as modbus_get_float() are called to convert the data type. The system adopts a double buffering mechanism to manage the collected data. The main thread obtains the parsed processing parameters through the thread-safe queue, and finally stores the structured data in the ring buffer for use by the upper-level application. Step (3) aligns multi-source time series data and eliminates common-mode interference between sensors. The specific implementation is as follows: In step (3.1), the dynamic time warping (DTW) algorithm accurately aligns the multi-source asynchronous time series data of vibration (10kHz), temperature (1Hz), and acoustic emission (100kHz) by constructing a distance matrix and performing dynamic programming path search: First, each sensor signal is preprocessed by sliding window normalization (window width δ = 50ms), and the Euclidean distance matrix D∈R^(n×m) between the vibration signal x(t) and the acoustic emission signal y(t) is calculated; then, through recursion formula (1): min(

[0023] Generate a cumulative distance matrix C; finally output a time-stretched synchronized data stream, so that the timing deviation between multiple sensors is ≤100μs. At the same time, derivative dynamic time warping (DDTW) is used to optimize phase difference alignment, providing time-consistent input data for subsequent PCA analysis.

[0024] Step (3.2): During the machining of the integral blade disk, principal component analysis (PCA) eliminates common-mode interference between multiple sensors through orthogonal transformation: first, the n×3-dimensional data matrix composed of vibration, temperature and acoustic emission signals is Z-score standardized (μ=0, σ=1), the covariance matrix Σ=(X^TX) / (n-1) is calculated and eigenvalue decomposition is performed, and the eigenvectors corresponding to the first k largest eigenvalues ​​are extracted to construct the projection matrix W∈R^(3×k); the original data is transformed into the principal component space through T=XW, and the first two principal components with a contribution rate greater than 95% are retained, while the component corresponding to the smallest eigenvalue (<5% variance contribution rate) is identified as common-mode noise and eliminated, and the denoised data is finally output.

[0025] Step (4) extracts the mixed features of the data in the time-frequency domain and constructs the feature space for the overall blade processing state. The specific implementation is as follows: First, the preprocessed multi-source sensor data is synchronously segmented, and a sliding window is used to intercept signal segments. Statistical features (including root mean square, crest factor, pulse index, kurtosis, etc.) are calculated in the time domain. The energy proportion and centroid frequency of each frequency band are extracted by fast Fourier transform in the frequency domain. The acoustic emission signal is additionally subjected to wavelet packet decomposition to obtain the relative energy entropy of 16 sub-bands. Simultaneously, the time-frequency joint features (including the instantaneous frequency variance of the short-time Fourier transform and the cross-term suppression feature of the Wigner-Ville distribution) are calculated. Finally, a hybrid feature vector is fused and the key features are screened out using the maximum relevance minimum redundancy (mRMR) feature selection algorithm.

[0026] Step (5) compresses the data volume while preserving the feature integrity, and uploads the data in stages. The specific implementation is as follows: During the blisk processing process, an improved LZMA compression algorithm achieves efficient data compression and feature preservation at the edge. First, the time-frequency domain mixed feature data is binary serialized and then compressed using an improved LZMA2 algorithm. Lossless mode is used for key features, while lossy quantization is enabled for secondary features. The compression ratio is dynamically adjusted using a feature importance weight matrix. The compressed data packets are uploaded in stages via the MQTT protocol: under normal operating conditions, the compressed feature vectors are transmitted at QoS1. When an anomaly is triggered, the original waveform fragments are uploaded at QoS2.

[0027] 2. Building an Adaptive Industrial Communication Module Step (1) uses the OPC-UA architecture and TSN for time synchronization and deterministic transmission. The specific implementation is as follows: Build a TSN network infrastructure that complies with the IEEE 802.1AS-2020 standard, configure industrial switches that support time-aware shaping, and achieve μs-level time synchronization across the entire network through the master clock based on the PTPv2 protocol. Deploy the UA PubSub model on the OPC UA server side, encapsulate processing parameters and sensor data into UADP messages, and mark the TSN traffic category. Use the 802.1Qbv time gating mechanism to allocate fixed time slots for critical data to ensure jitter <10μs. At the same time, use OPC-UA metadata modeling to achieve semantic description of devices, and cooperate with TSN frame preemption and traffic shaping to ensure that the transmission of non-real-time data does not interfere with real-time streams.

[0028] Step (2) uses a bandwidth adaptive allocation algorithm to dynamically adjust the data transmission priority according to the network load. The specific implementation is as follows: First, network probes are deployed to monitor the RTT delay, packet loss rate, and available bandwidth of the 5G / LoRa dual-channel in real time. An LSTM network is used to predict network load trends over the next period of time. A Q-learning model is constructed to establish the state space, and the action space includes transmission mode switching, compression ratio adjustment, and priority reassignment. When network congestion is detected, the transmission priority of non-critical data is dynamically downgraded, while the transmission level of abnormal vibration segments is increased.

[0029] Step (3) involves intelligent switching between 5G communication and LoRa protocol to ensure data continuity. The specific implementation is as follows: Through the deployed dual-mode communication terminals, key indicators such as RSRP, SINR of the 5G network and RSSI of LoRa are continuously collected. When the 5G signal degrades, the switching decision model based on reinforcement learning is triggered to perform session migration. The key processing parameters are transmitted through LoRa's ABP activation mode to reduce the feature data. At the same time, non-real-time data enters the local SSD cache queue. During the switching process, dual-channel data packet sequence number synchronization and CRC32 check retransmission mechanism are used to ensure that key instructions are not lost. The system automatically switches back after 5G is restored, and resends the missing data during the LoRa period through differential transmission.

[0030] 3. Building a cloud-based intelligent analysis and decision-making module Step (1) analyzes the normal processing data of the compressor blade disk during all stages of the processing and establishes a processing data thread. The specific implementation is as follows: Based on the timestamp alignment mechanism, multimodal data streams such as spindle power, three-dimensional tool vibration, workpiece temperature and acoustic emission signals are integrated; the dynamic time warping (DTW) algorithm is used to eliminate the timing drift between sensors, and the machining process is divided into three stages: rough machining, semi-finishing and finishing through a sliding window; time domain features, frequency domain features and time-frequency features are extracted for each window to construct a multidimensional mixed feature vector; the improved t-SNE algorithm is used for dimensionality reduction and visualization, combined with DBSCAN clustering to identify stable working condition data clusters in each machining stage; finally, a three-layer data thread structure is established: the original data layer, the feature engineering layer and the working condition label layer.

[0031] Step (2) uses the time series prediction model to predict the risk of machining anomalies during the machining process and updates the model parameters through real-time data streams. The specific implementation is as follows: Based on the processing data thread established in step (1), combined with the multi-dimensional sensor data streams such as vibration, acoustic emission, and spindle power of abnormal processing status, after dynamic time warping (DTW) aligning the timestamps, a sliding window is used to extract mixed features such as time domain, frequency domain, and time-frequency domain. These features are input into a hybrid model composed of LSTM and Transformer. The nonlinear correlation between features is captured through the attention mechanism, and risk indicators such as chatter probability and tool wear within the next 3 seconds are output. When the digital twin system detects that the prediction residual continuously exceeds the set 3σ principle threshold, the incremental learning mechanism is automatically triggered. The deterministic transmission based on the TSN network ensures that the parameter update instructions are transmitted first, ensuring that the model can complete the adjustment of key feature weights in a short time when the processing conditions suddenly change.

[0032] Step (3), dynamically adjust the sensor sampling frequency and simulation model to perform fast simulation and high-precision simulation, and its specific implementation is as follows: Step (3.1): When the high-precision finite element simulation shows that the current operating conditions are stable, the vibration sensor sampling rate is automatically reduced from high frequency to low frequency, retaining only basic temperature sampling; when the simulation detects the risk of vibration, the high-speed sampling mode is immediately triggered, and the backup laser Doppler vibrometer is activated for local fine monitoring; this dynamic adjustment is implemented through the 802.1Qbv time-aware shaping protocol of the TSN network to implement command issuance, combined with the adaptive caching strategy of the edge node, so that the system can still maintain the integrity of critical data when the 5G bandwidth fluctuates.

[0033] Step (3.2): When the sensor data is judged to be in a stable working condition by the LSTM-Transformer hybrid model constructed in step (2), the reduced-order model based on the intrinsic orthogonal decomposition is automatically activated, and only the dominant mode is retained, so that a millisecond-level response is achieved at the edge computing node for real-time process parameter adjustment. When an abnormal risk is detected, the high-precision finite element model is immediately switched to, and the cloud GPU cluster is called through the 5G-TSN network to perform coupled simulation of the cutting force field and temperature field. At the same time, the transfer learning technology is used to use the working condition characteristics accumulated by the reduced-order model as the initial conditions for the high-precision simulation, so as to achieve a dynamic balance between simulation accuracy and efficiency.

[0034] The present invention constructs an edge intelligent perception and preprocessing module that considers the LZMA2 lossy-lossless hybrid compression strategy, which can reduce the amount of data transmission while maintaining the integrity of data features. The present invention proposes an adaptive industrial communication module combined with a dual-channel Q-learning bandwidth allocation algorithm, which can dynamically adjust the 5G / LoRa transmission mode according to the network status, and ensure the reliability of critical data transmission in the event of sudden congestion. The present invention proposes a cloud-based intelligent analysis and decision-making module based on an LSTM-Transformer hybrid prediction model and a reduced-order / high-precision model dynamic switching strategy, which improves the simulation speed while ensuring sufficient simulation accuracy. The method proposed in the present invention can effectively improve the communication and interaction efficiency of the digital twin system of the integral blade processing process of the compressor of an aero-engine, and solve the technical problems of low data transmission efficiency and large errors in traditional methods.

[0035] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in the present application should be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. A communication and interaction method for a digital twin system of a compressor blisk machining process, characterized in that: include: Real-time collection of vibration, temperature, and acoustic emission data during compressor blade machining. Analyzing machine tool processing data. Performing multi-source time series data alignment and noise reduction on the vibration, temperature, and acoustic emission data. Extracting time-frequency domain hybrid features from the processed data to obtain time-frequency domain hybrid feature data. Using an improved LZMA2 compression algorithm, the data is optimized and uploaded in stages. A deterministic transmission network is built based on the TSN-OPC UA architecture. The data transmission priority of the 5G communication and LoRa dual channels is dynamically adjusted according to the network load of the transmission network. A switching decision model based on reinforcement learning is used to achieve intelligent switching between 5G communication and LoRa protocols. The normal processing data in the machine tool processing data of the entire compressor integral blade disk processing process is analyzed to establish a processing data thread; based on the processing data thread, a time series prediction model is used to predict the processing abnormality risk in the processing process, and the model parameters are updated through real-time data flow.

2. The communication and interaction method of the digital twin system for the compressor blisk machining process according to claim 1 is characterized in that: Real-time collection of vibration, temperature, and acoustic emission data during compressor blade machining, analyzing machine tool processing data, including: A vibration-temperature-acoustic emission multimodal sensor array is constructed to collect vibration, temperature, and acoustic emission data during the machining of the compressor integral blade; and the machining data is analyzed using the TCP / IP protocol stack.

3. The communication and interaction method of the compressor blisk processing digital twin system according to claim 1 is characterized in that: Performing multi-source time series data alignment and noise reduction processing on the vibration, temperature, and acoustic emission data, including: A dynamic time warping algorithm is used to construct a distance matrix and dynamic planning path search to accurately align asynchronous time series data from multiple sources of vibration, temperature, and acoustic emission; principal component analysis is combined to eliminate common-mode interference between multiple sensors.

4. The communication and interaction method of the digital twin system for the compressor blisk machining process according to claim 1 is characterized in that: The processed data is subjected to time-frequency domain mixed feature extraction to obtain time-frequency domain mixed feature data, including: The processed data is synchronously segmented and a sliding window is used to intercept data signal segments. The time domain features of each segment are calculated and counted in the time domain. The frequency domain features of each frequency band are extracted by fast Fourier transform in the frequency domain. The time-frequency joint features of the time and frequency domains are calculated simultaneously. The time domain features, the frequency domain features and the time-frequency joint features are fused to form a mixed feature vector, and key features are selected as the time-frequency domain mixed feature data through a maximum correlation minimum redundancy feature selection algorithm.

5. The communication and interaction method of the digital twin system for the compressor blisk machining process according to claim 1 is characterized in that: The time-frequency domain mixed feature data is optimized and uploaded in stages using an improved LZMA2 compression algorithm, including: The time-frequency domain mixed feature data is binary serialized, and the serialized feature data is subjected to lossy-lossless hybrid compression using an improved LZMA2 algorithm. The compression ratio is dynamically adjusted through a feature importance weight matrix, and the compressed data packets are hierarchically uploaded through the MQTT protocol.

6. The communication and interaction method of the digital twin system for the compressor blisk machining process according to claim 1 is characterized in that: Building a deterministic transmission network based on the TSN-OPC UA architecture, including: Based on the OPC-UA architecture, the TSN network is used for time synchronization and deterministic transmission, including: building a TSN network infrastructure that complies with the IEEE802.1AS-2020 standard, configuring industrial switches that support time-aware shaping, and achieving μs-level time synchronization across the entire network through the master clock based on the PTPv2 protocol; deploying the UA PubSub model on the OPC UA server side, encapsulating the machine tool processing data and the optimized time-frequency domain mixed feature data into UADP messages, and marking the TSN traffic category.

7. The communication and interaction method of the digital twin system for the compressor blisk machining process according to claim 1, characterized in that: Dynamically adjust the data transmission priority of 5G communication and LoRa dual channel according to the network load of the transmission network, including: Adopting an adaptive bandwidth allocation algorithm, the data transmission priority is dynamically adjusted according to the network load, including: Real-time monitoring of RTT delay, packet loss rate, and available bandwidth of 5G communication and LoRa dual channels; based on the monitoring data, the LSTM network is used to predict network load trends within a set time period in the future; Construct a Q-learning model to establish the state space and action space. The state space includes transmission mode switching, compression ratio adjustment, and priority reassignment. When the prediction indicator exceeds the set network congestion threshold, the transmission priority of non-critical data is dynamically downgraded, while the transmission level of abnormal vibration segments is increased.

8. The communication and interaction method of the digital twin system for the compressor blisk machining process according to claim 2, characterized in that: Analyze the normal processing data in the machine tool processing data of the entire compressor blisk processing process and establish a processing data thread, including: Integrate multimodal data streams including spindle power, tool three-dimensional vibration, workpiece temperature and acoustic emission signals based on a timestamp alignment mechanism; A dynamic time warping algorithm is used to process the multimodal data stream to eliminate the timing drift between sensors, and the machining process is divided into three stages: rough machining, semi-finishing, and finishing. Time domain features, frequency domain features, and time-frequency joint features are extracted for each window to construct a multidimensional hybrid feature vector. The improved t-SNE algorithm is used to perform dimensionality reduction and visualization on the multidimensional mixed feature vector, and DBSCAN clustering is combined to identify stable working condition data clusters in each processing stage; a three-layer data thread structure is established, including the original data layer, feature engineering layer and working condition label layer.

9. The communication and interaction method of the compressor blisk processing digital twin system according to claim 1, characterized in that: The time series prediction model adopts the LSTM-Transformer hybrid model.

10. The communication and interaction method of the digital twin system of the compressor blisk processing process according to claim 2, characterized in that: Also includes: According to the simulation results of the digital twin system, the sampling frequency of the sensor is dynamically adjusted, and the reduced-order model is selected for fast simulation and high-precision simulation.

Images