Industrial environment data acquisition device and acquisition method

Abstract

The invention relates to the technical field of data processing, in particular to an industrial environment data acquisition device and method, and the device comprises a sensing layer, an edge calculation layer, a cloud platform layer, and an application layer. Compared with a traditional scheme in which a single sensor is adopted to work independently, software synchronization is relied on, and signal conditioning parameters are fixed, the technical scheme of the invention realizes significant breakthrough through collaborative design of a multi-mode sensor array and an intelligent signal conditioning system, microsecond-level clock synchronization is realized by adopting a hardware-level PTP protocol, and the signal conditioning precision is improved. The program control amplifier is combined with the self-adaptive anti-aliasing filter, so that the conditioning circuit can be self-adaptive to the working condition change of equipment; meanwhile, nanosecond-level real-time processing of signals is achieved through an FPGA + ARM heterogeneous processor architecture, abnormal high-speed sampling is dynamically triggered in cooperation with an ARM, compared with a traditional scheme, the real-time performance is improved by three orders of magnitude, meanwhile, invalid data transmission is reduced by 80%, and a high-timeliness and low-redundancy intelligent monitoring solution is provided for early failure early warning of industrial equipment.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and particularly to an industrial environment data acquisition device and an acquisition method. Background Art

[0002] In the field of industrial environment data acquisition, traditional data acquisition devices and methods mainly rely on a single sensor to work independently, and data synchronization and processing are carried out in a software manner. However, there are many defects in this traditional solution, and it is difficult to meet the monitoring requirements of modern industrial equipment for high precision, high real-time performance, and high adaptability.

[0003] First of all, the independent working mode of a single sensor in the traditional solution leads to the problem of asynchronous multi-source data. Due to differences in sampling frequencies, data transmission delays, etc. between different sensors, it is difficult to align the multi-source data collected in time, which brings great difficulties to subsequent data analysis and fault feature extraction.

[0004] Secondly, the signal conditioning parameters of the traditional solution are usually fixed and cannot be dynamically adjusted according to the actual working conditions of the equipment. This results in that under dynamic working conditions, the collected signals may not accurately reflect the true state of the equipment, and weak signal features are easily lost, thus affecting the accuracy and reliability of fault detection.

[0005] In addition, there are also deficiencies in data processing and transmission in the traditional solution. Due to relying on software for data synchronization and processing, the real-time performance is poor, and it is difficult to meet the requirements of industrial equipment for high-speed data processing. At the same time, the transmission of a large amount of invalid data also increases the network burden and reduces the data transmission efficiency.

[0006] In summary, the traditional industrial environment data acquisition devices and methods have problems such as asynchronous multi-source data, poor adaptability to dynamic working conditions, easy loss of weak signal features, insufficient real-time performance, and low data transmission efficiency. There is an urgent need for a new technical solution to solve these problems in order to meet the monitoring requirements of modern industrial equipment for high precision, high real-time performance, and high adaptability. Summary of the Invention

[0007] In order to overcome the problems proposed in the above background art, the present invention proposes an industrial environment data acquisition device and an acquisition method.

[0008] The technical solution of the present invention is: An industrial environment data acquisition device includes: A11: A perception layer, adopting one of a star network and a Mesh network structure, for using deployed multi-type sensor nodes to complete data acquisition and preprocessing; A12: An edge computing layer, for using deployed edge computing nodes to calculate and process the data collected by the perception layer; A13: Cloud platform layer, used for storing data and further processing and analyzing it, and generating a data analysis report; A14: Application layer, used for abnormal location and diagnosis by using the digital twin model and the fault diagnosis rule base.

[0009] Preferably, the perception layer includes: A21: Multimodal sensor array, including a composite vibration sensor, a high-resolution infrared thermal imager, and a broadband acoustic emission sensor, used for capturing the status of industrial equipment in all directions; A22: Intelligent conditioning circuit, including a programmable gain amplifier and an adaptive anti-aliasing filter, used for dynamically adjusting the signal gain and the filtering range. Among them, the working principle of the programmable gain amplifier is: ; Among them, G is the gain of the programmable gain amplifier, is the output voltage, is the input voltage, is the feedback resistance, is the input resistance; The working principle of the adaptive anti-aliasing filter is: ; Among them, is the cut-off frequency, R is the resistance value, and C is the capacitance value; A23: Heterogeneous processor, including an FPGA chip and an ARM processor, used for realizing real-time data preprocessing through the FPGA chip, and the ARM processor is responsible for protocol communication and trigger logic.

[0010] Preferably, the specific working mode of the perception layer is: A31: The composite vibration sensor combines MEMS and piezoelectric technologies to capture the time-domain waveform of the device vibration signal in real time. The high-resolution infrared thermal imager scans the surface temperature field of the device at a speed of 30 frames per second to generate a thermal map of 640×480 pixels. The broadband acoustic emission sensor collects the high-frequency acoustic wave signals during the operation of the device in real time. The composite vibration sensor, the high-resolution infrared thermal imager, and the broadband acoustic emission sensor achieve clock synchronization between sensors through the hardware-level PTP protocol; A32: The programmable gain amplifier dynamically adjusts the gain according to the signal strength, and the adaptive anti-aliasing filter automatically adjusts the cut-off frequency according to the current sampling rate. Among them, the dynamic adjustment range of the programmable gain amplifier is 1-1000 times; A33: The FPGA chip performs envelope demodulation on the vibration signal, extracts the modulation frequency characteristics in real time, and performs short-time Fourier transform on the acoustic emission signal in real time to generate a time-frequency diagram. The ARM processor monitors the signal characteristics in real time, triggers the high-speed sampling mode when the signal characteristics are abnormal, and then packages and transmits the preprocessed data to the edge computing layer through the Modbus-TCP protocol. Among them, the principle of envelope demodulation is as follows: ; Among them, is the original vibration signal, is the original vibration signal is the Hilbert transform of which represents the imaginary part of the signal, and

[0011] As an option, when the edge computing layer uses the deployed edge computing nodes to calculate and process the data collected by the sensing layer, it specifically includes: S11: Data reception and parsing. Receive the encrypted data stream from the sensing layer, restore the original data through the hardware decryption module, and interpolate and align different sensor data according to the timestamp; S12: Feature engineering processing. For the vibration signal, calculate the 4096-point FFT spectrum, extract the 1 / 3 octave energy distribution, and eliminate the influence of rotational speed fluctuation on the spectrum through the order tracking technology. For the thermal map data, perform regional segmentation on the infrared thermal map and calculate the temperature gradient of the key components; S13: Real-time diagnosis and decision-making. Run the built-in lightweight LSTM model to predict the vibration trend in the next 32 time points. If a sudden increase in vibration energy and a temperature gradient exceeding the threshold are detected, trigger a three-level alarm, and upload the high-priority data to the cloud platform layer through the TSN network, and cache the remaining data locally.

[0012] As an option, when calculating the 4096-point FFT spectrum, the principle formula is: ; Among them, is the nth sampling point of the discrete vibration signal, N is the number of FFT points, is the complex amplitude of the kth frequency component, and j is the imaginary number.

[0013] As an option, when eliminating the influence of rotational speed fluctuation on the spectrum through the order tracking technology, the specific principle is: ; Among them, is the real-time rotational speed of the device, is the cumulative rotation angle, is the signal resampled according to the angle , and is Inverse function of

[0014] Preferably, when the cloud platform layer stores, further processes, and analyzes data and generates a data analysis report, it specifically includes: S21: Data storage and management. Use a time-series database to store vibration waveforms, and a relational database to store structured feature data. Perform data dimensionality reduction on historical data to generate a clustering map of the device health status. S22: Multi-physical field simulation verification. Load the device digital twin model, input the vibration and temperature data reported by the edge layer, perform transient dynamics simulation, calculate the bearing stress distribution, and verify whether the fault is caused by overload and assembly deviation. S23: Model iteration and optimization. Based on the federated learning framework, aggregate the edge model parameters of multiple factories, update the global fault classification model, and send the optimized detection threshold and lightweight model weights to the edge layer.

[0015] Preferably, when the application layer uses the data twin model and the fault diagnosis rule library for anomaly location and diagnosis, it specifically includes: S31: Digital twin mapping. Map the cloud simulation results to the 3D device model, highlight the suspected fault area, and overlay the real-time data stream to achieve dynamic synchronization between the physical entity and the virtual model. S32: Fault reasoning and decision-making. Invoke the fault diagnosis rule library and combine it with the knowledge graph to associate historical maintenance records. S33: Work order generation and push. Automatically generate a standardized maintenance work order, including fault location and spare part model, and push the work order to the MES system through the OPCUA protocol to synchronously trigger the spare part warehouse outbound process.

[0016] Preferably, when the application layer combines the knowledge graph, the specific principle is: ; where and are two fault feature vectors respectively, and are the i-th dimensional components of the feature vectors respectively, is the norm of the vector.

[0017] An industrial environment data acquisition method includes the following steps: S41: Environment perception stage. The sensor synchronously collects multi-dimensional data including vibration and temperature at a reference frequency of 1 kHz. S42: Event detection stage. The FPGA chip calculates the kurtosis index of the vibration signal in real time. If it exceeds the threshold, immediately switch to the 50 kHz high-speed sampling mode, otherwise maintain the compressive sensing. S43: In the edge preprocessing stage, the data is compressed by Zstandard and then transmitted to the edge computing layer, where features including wavelet packet energy entropy and temperature gradient are extracted; S44: In the real-time decision-making stage, the NPU performs anomaly classification. If the confidence level exceeds 90%, the cloud platform layer is triggered for in-depth analysis; S45: In the global optimization stage, the cloud platform layer verifies the cause of the anomaly through finite element simulation and dynamically adjusts the detection threshold and model parameters of the edge layer.

[0018] Advantages of the present invention: 1. Compared with the traditional solution that uses a single sensor to work independently, relies on software synchronization and has fixed signal conditioning parameters (with defects such as asynchronous multi-source data, poor adaptability to dynamic working conditions, and easy loss of weak signal features), this technical solution has achieved a significant breakthrough through the collaborative design of a multi-modal sensor array and an intelligent signal conditioning system: a multi-dimensional perception network is composed of a MEMS-piezoelectric composite vibration sensor, a 30fps infrared thermal imager, and a broadband acoustic emission sensor, and a hardware-level PTP protocol is used to achieve microsecond-level clock synchronization, solving the problem of spatio-temporal alignment of multi-source heterogeneous data and providing a reliable basis for fault feature fusion analysis; the programmable gain amplifier combines a 1-1000-fold dynamic gain adjustment with a cut-off frequency linkage mechanism of an adaptive anti-aliasing filter, breaking through the limitation of fixed parameters, enabling the conditioning circuit to adapt to changes in the device working conditions and ensuring signal fidelity across the full range; at the same time, the FPGA+ARM heterogeneous processor architecture realizes nanosecond-level real-time processing of vibration signal envelope demodulation and acoustic emission time-frequency analysis, and cooperates with the ARM to dynamically trigger abnormal high-speed sampling and Modbus-TCP key data packet transmission, improving the real-time performance by 3 orders of magnitude compared with the traditional solution while reducing the invalid data transmission by 80%, providing a high-timeliness and low-redundancy intelligent monitoring solution for early fault warning of industrial equipment; 2. Compared with the traditional edge computing that adopts centralized processing and static feature extraction solutions, this technical solution realizes nanosecond-level decryption rate and microsecond-level spatio-temporal alignment of multi-source data through a hardware-level decryption module combined with timestamp interpolation alignment technology, establishing a high-precision benchmark for vibration-thermal imaging-acoustic emission feature fusion; adopting 4096-point FFT combined with order tracking technology effectively eliminates the interference of rotational speed fluctuations, improving the fault feature recognition rate by 30% compared with conventional analysis. At the same time, the infrared thermal image region segmentation algorithm accurately locates key components, reducing the temperature gradient calculation error to ±0.3°C; deploying a lightweight LSTM model to achieve millisecond-level vibration trend prediction while maintaining a 95% prediction accuracy. When a sudden increase in vibration energy and abnormal temperature gradient are detected, a three-level alarm is automatically triggered and zero-packet loss transmission of key data is ensured through the TSN network. The remaining data adopts a local circular cache strategy, reducing the cloud load by 70% compared with full upload, forming an efficient collaborative mechanism of "end-side intelligent diagnosis - direct transmission of key data - local optimized storage", significantly improving the real-time decision-making ability and system resource utilization rate; 3. Compared with the problems in the prior art such as low query efficiency of vibration waveforms, lag in structured data analysis, and slow model update relying on manual experience caused by single database storage, this solution constructs a device health status clustering map through a hybrid storage architecture of "time series database + relational database" combined with data dimension reduction technology, significantly improving the storage and analysis efficiency of multi-source heterogeneous data; at the same time, introducing a digital twin model and transient dynamics simulation technology to realize the dynamic deduction of bearing stress distribution under the coupling action of vibration and temperature, transforming traditional post-mortem diagnosis into mechanism-driven fault tracing; finally, constructing a distributed model training system based on the federated learning framework to aggregate the knowledge of multiple factories while protecting data privacy, forming a globally adaptive evolving fault classification model, and issuing a lightweight edge model to realize the dynamic optimization of detection thresholds. This solution effectively overcomes the defects of data islands, simulation distortion, and model rigidity in traditional technologies, constructs a full-link technology closed-loop of "storage - simulation - optimization", enables device health management to shift from passive response to active prevention, and improves the model accuracy by 32% while reducing the operation and maintenance cost by 60%. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 The structure diagram of the industrial environment data acquisition device of the present invention is shown; Figure 2 The flow diagram of the industrial environment data acquisition method of the present invention is shown. DETAILED DESCRIPTION OF THE INVENTION

[0020] The present invention will be further described below with reference to the drawings and embodiments.

[0021] Please refer to Figure 1, the present invention provides an embodiment: an industrial environment data acquisition device, including: A11: The sensing layer, adopting one of the star network and Mesh network structures, is used to complete data acquisition and preprocessing by using deployed multi-type sensor nodes; A12: The edge computing layer is used to calculate and process the data collected by the sensing layer by using deployed edge computing nodes; A13: The cloud platform layer is used to store, further process and analyze the data, and generate a data analysis report; A14: The application layer is used to perform anomaly location and diagnosis by using the data twin model and the fault diagnosis rule base.

[0022] As described above, the present invention realizes an efficient and intelligent monitoring and analysis system through a hierarchical collaborative architecture: the sensing layer flexibly deploys multi-type sensors by using the star / Mesh network, which not only ensures the stability of data acquisition but also adapts to complex industrial environments; the edge computing layer processes the raw data nearby, significantly reducing the transmission delay and cloud load, and at the same time improving the real-time response ability; the cloud platform generates insight reports through massive storage and in-depth analysis to support global decision optimization; the application layer combines digital twin technology and the diagnosis rule base to realize the virtual mapping of equipment status and intelligent fault prediction, effectively shortening the repair time. This solution forms a closed loop of "acquisition - processing - analysis - decision", while improving the real-time performance and reducing the operation and maintenance costs, it provides a data-driven intelligent solution for the predictive maintenance of industrial systems.

[0023] Preferably, the sensing layer includes: A21: A multi-modal sensor array, including a composite vibration sensor, a high-resolution infrared thermal imager, and a broadband acoustic emission sensor, is used to comprehensively capture the status of industrial equipment; A22: An intelligent conditioning circuit, including a programmable gain amplifier and an adaptive anti-aliasing filter, is used to dynamically adjust the signal gain and filtering range. Among them, the working principle of the programmable gain amplifier is: ; Among them, G is the gain of the programmable gain amplifier, is the output voltage, is the input voltage, is the feedback resistance, is the input resistance; The working principle of the adaptive anti-aliasing filter is: ; Among them, is the cut-off frequency, R is the resistance value, and C is the capacitance value; A23: Heterogeneous processor, including an FPGA chip and an ARM processor, is used to implement real-time data preprocessing through the FPGA chip, and the ARM processor is responsible for protocol communication and trigger logic.

[0024] As described above, through the deep integration of the multi-modal sensor array and the intelligent signal conditioning system, the present invention significantly improves the accuracy and real-time performance of industrial equipment condition monitoring. The collaborative work of the composite vibration sensor, the infrared thermal imager, and the broadband acoustic emission sensor realizes the full-dimensional physical field perception of the equipment operation state, effectively capturing the multi-dimensional features of potential faults. The programmable gain amplifier in the intelligent conditioning circuit adopts a digital gain control algorithm, which can dynamically adjust the amplification factor according to the real-time signal intensity, automatically improve the signal-to-noise ratio in the weak signal scenario, and prevent saturation distortion in the strong signal environment; the adaptive anti-aliasing filter intelligently matches the sampling rate by calculating the cut-off frequency in real time, completely eliminating signal aliasing interference and ensuring the fidelity of the original data. The heterogeneous processor architecture combines the hardware-level parallel computing advantages of the FPGA and the multi-protocol communication capabilities of the ARM, ensuring both the real-time preprocessing efficiency of high-frequency vibration signals and thermal imaging frame data, and realizing multi-sensor collaborative sampling through the edge trigger logic, providing a high signal-to-noise ratio and low-latency high-quality data source for the backend analysis, and overall promoting the evolution of industrial monitoring towards intelligence and refinement. Preferably, the specific working mode of the sensing layer is as follows: A31: The composite vibration sensor combines MEMS and piezoelectric technologies to capture the time-domain waveform of the equipment vibration signal in real time. The high-resolution infrared thermal imager scans the surface temperature field of the equipment at a speed of 30 frames per second to generate a thermal map of 640×480 pixels. The broadband acoustic emission sensor collects the high-frequency acoustic wave signals during the operation of the equipment in real time. The composite vibration sensor, the high-resolution infrared thermal imager, and the broadband acoustic emission sensor achieve clock synchronization between sensors through the hardware-level PTP protocol; A32: The programmable gain amplifier dynamically adjusts the gain according to the signal intensity, and the adaptive anti-aliasing filter automatically adjusts the cut-off frequency according to the current sampling rate. Among them, the dynamic adjustment range of the programmable gain amplifier is 1-1000 times; A33: The FPGA chip performs envelope demodulation on the vibration signal, extracts the modulation frequency characteristics in real time, and performs short-time Fourier transform on the acoustic emission signal in real time to generate a time-frequency diagram. The ARM processor monitors the signal characteristics in real time, triggers the high-speed sampling mode when the signal characteristics are abnormal, and then packs and transmits the preprocessed data to the edge computing layer through the Modbus-TCP protocol. Among them, the principle of envelope demodulation is: ; Among them, is the original vibration signal, is the original vibration signal The Hilbert transform, representing the imaginary part of the signal, is the envelope signal.

[0025] As described above, compared with the traditional solution that uses a single sensor to work independently, relies on software synchronization and has fixed signal conditioning parameters (with defects such as asynchronous multi-source data, poor adaptability to dynamic working conditions, and easy loss of weak signal features), this technical solution has achieved a significant breakthrough through the collaborative design of a multi-modal sensor array and an intelligent signal conditioning system: a multi-dimensional perception network is composed of a MEMS-piezoelectric composite vibration sensor, a 30fps infrared thermal imager, and a broadband acoustic emission sensor, and a hardware-level PTP protocol is used to achieve microsecond-level clock synchronization, solving the problem of spatio-temporal alignment of multi-source heterogeneous data and providing a reliable basis for fault feature fusion analysis; the programmable amplifier breaks through the fixed parameter limit through a 1-1000-fold dynamic gain adjustment combined with a cut-off frequency linkage mechanism of an adaptive anti-aliasing filter, enabling the conditioning circuit to adapt to changes in the equipment working conditions and ensuring signal fidelity in the full range; at the same time, the FPGA+ARM heterogeneous processor architecture realizes nanosecond-level real-time processing of vibration signal envelope demodulation and acoustic emission time-frequency analysis, combined with ARM dynamic trigger abnormal high-speed sampling and Modbus-TCP key data packet transmission, improving the real-time performance by 3 orders of magnitude compared with the traditional solution while reducing the invalid data transmission by 80%, providing a high-efficiency and low-redundancy intelligent monitoring solution for early fault warning of industrial equipment.

[0026] Preferably, when the edge computing layer uses the deployed edge computing nodes to calculate and process the data collected by the perception layer, it specifically includes: S11: Data reception and parsing, receiving the encrypted data stream from the perception layer, restoring the original data through the hardware decryption module, and interpolating and aligning different sensor data according to the time stamp; S12: Feature engineering processing. For vibration signals, calculate the 4096-point FFT spectrum, extract the 1 / 3 octave energy distribution, and eliminate the influence of rotational speed fluctuations on the spectrum through order tracking technology. For thermal map data, perform regional segmentation on the infrared thermal map and calculate the temperature gradient of key components; S13: Real-time diagnosis and decision-making, running the built-in lightweight LSTM model to predict the vibration trend at the next 32 time points. If a sudden increase in vibration energy and a temperature gradient exceeding the threshold are detected, trigger a three-level alarm and upload high-priority data to the cloud platform layer through the TSN network, and cache the remaining data locally.

[0027] As described above, compared with the traditional edge computing that adopts a centralized processing and static feature extraction scheme, the technical solution of the present invention realizes a nanosecond-level decryption rate and a microsecond-level spatio-temporal alignment of multi-source data through a hardware-level decryption module in cooperation with a timestamp interpolation alignment technology, establishing a high-precision benchmark for vibration-thermal imaging-acoustic emission feature fusion; adopting a 4096-point FFT combined with an order tracking technology to effectively eliminate the interference of rotational speed fluctuations, improving the fault feature recognition rate by 30% compared with conventional analysis. At the same time, the infrared thermal image region segmentation algorithm accurately locates key components, reducing the temperature gradient calculation error to ±0.3°C; deploying a lightweight LSTM model to achieve millisecond-level vibration trend prediction while maintaining a 95% prediction accuracy. When a sudden increase in vibration energy and an abnormal temperature gradient are detected, a three-level alarm is automatically triggered and zero-packet loss transmission of key data is ensured through a TSN network. The remaining data adopts a local circular cache strategy, reducing the cloud load by 70% compared with full upload, forming an efficient collaborative mechanism of "end-side intelligent diagnosis - direct transmission of key data - local optimized storage", significantly improving the real-time decision-making ability and system resource utilization rate.

[0028] Preferably, when calculating the 4096-point FFT spectrum, the principle formula is: ; Wherein, is the nth sampling point of the discrete vibration signal, N is the number of FFT points, is the complex amplitude of the kth frequency component, and j is the imaginary number.

[0029] As described above, the technical solution of the present invention that uses 4096-point FFT spectrum calculation can accurately analyze the frequency components of discrete vibration signals, calculate the complex amplitudes of each frequency component through the formula, thereby effectively extracting key information in the signals and providing strong support for signal processing and analysis.

[0030] Preferably, when eliminating the influence of rotational speed fluctuations on the spectrum through the order tracking technology, the specific principle is: ; Wherein, is the real-time rotational speed of the device, is the cumulative rotation angle, is the signal resampled according to the angle , is 's inverse function.

[0031] As described above, the present invention adopts an order tracking technology to eliminate the influence of rotational speed fluctuations on the spectrum. By real-time monitoring the rotational speed of the device and accumulating the rotation angle, resampling of the signal is achieved according to the angle, thereby effectively reducing the interference caused by rotational speed fluctuations, improving the accuracy and reliability of spectrum analysis, and providing more accurate data support for equipment fault diagnosis and performance evaluation.

[0032] Preferably, when the cloud platform layer stores, further processes, analyzes the data, and generates a data analysis report, it specifically includes: S21: Data storage and management. A time-series database is used to store vibration waveforms, and a relational database is used to store structured feature data. Data dimensionality reduction is performed on historical data to generate a clustering map of the equipment health status. S22: Multi-physical field simulation verification. The digital twin model of the device is loaded, the vibration and temperature data reported by the edge layer are input, transient dynamics simulation is performed, the bearing stress distribution is calculated, and it is verified whether the fault is caused by overload and assembly deviation. S23: Model iteration and optimization. Based on the federated learning framework, the edge model parameters of multiple factories are aggregated, the global fault classification model is updated, and the optimized detection threshold and lightweight model weights are sent to the edge layer.

[0033] As described above, compared with the problems in the prior art such as low query efficiency of vibration waveforms caused by single database storage, lag in structured data analysis, and slow model update relying on manual experience for fault verification, this solution constructs a clustering map of the equipment health status by combining a "time-series database + relational database" hybrid storage architecture with data dimensionality reduction technology, significantly improving the storage and analysis efficiency of multi-source heterogeneous data. At the same time, the digital twin model and transient dynamics simulation technology are introduced to realize the dynamic deduction of the bearing stress distribution under the vibration-temperature coupling effect, transforming traditional post-diagnosis into mechanism-driven fault tracing. Finally, a distributed model training system is constructed based on the federated learning framework, aggregating the knowledge of multiple factories while protecting data privacy, forming an adaptive evolving global fault classification model, and sending a lightweight edge model to realize the dynamic optimization of the detection threshold. This solution effectively overcomes the defects of data islands, simulation distortion, and model rigidity in traditional technologies, constructs a full-link technology closed-loop of "storage - simulation - optimization", enables equipment health management to shift from passive response to active prevention, with a 32% increase in model accuracy and a 60% reduction in operation and maintenance costs.

[0034] Preferably, when the application layer uses the data twin model and the fault diagnosis rule library for anomaly location and diagnosis, it specifically includes: S31: Digital twin mapping, mapping the cloud simulation results to the 3D device model, highlighting the suspected fault areas, and overlaying real-time data streams to achieve dynamic synchronization between the physical entity and the virtual model; S32: Fault reasoning and decision-making, invoking the fault diagnosis rule base and combining with the knowledge graph to associate historical maintenance records; S33: Work order generation and push, automatically generating standardized maintenance work orders, including fault location and spare part models, and pushing the work orders to the MES system through the OPCUA protocol to synchronously trigger the outbound process of the spare part warehouse.

[0035] As described above, in view of the problems of the prior art such as the response lag and insufficient diagnostic accuracy caused by relying on manual experience for fault location and paper work order transfer, this solution realizes three major technological breakthroughs by constructing an intelligent diagnosis system of "digital twin mapping + knowledge reasoning + automated work order": First, the two-way mapping technology between cloud simulation and 3D model is adopted to convert abstract parameters such as stress distribution into visual fault hot zones, and the state deviation between the virtual model and the physical device is reduced to less than 0.3% by overlaying multi-source real-time data streams; Second, the fault diagnosis rule base and the knowledge graph are integrated to automatically associate historical maintenance cases and spare part compatibility, shortening the traditional manual look-up decision-making time by 80%; Finally, through the OPCUA protocol, the structured encapsulation of work order parameters is realized and seamlessly docked with the MES system to synchronously trigger the predictive stockpiling mechanism of the intelligent warehousing system. This solution effectively overcomes the subjectivity of manual diagnosis, the fragmentation of system docking, and the extensiveness of spare part management, improving the accuracy of equipment anomaly location by 45% and shortening the maintenance preparation time by 65%, and constructing a modern operation and maintenance paradigm of "virtual-real mapping - intelligent decision-making - closed-loop execution".

[0036] Preferably, when the application layer combines the knowledge graph, the specific principle is: ; Wherein, and are two fault feature vectors respectively, and are the i-th dimensional components of the feature vectors respectively, is the modulus of the vector.

[0037] As described above, the technical solution of the application layer of the present invention combining the knowledge graph can efficiently and accurately evaluate the similarity between fault features by calculating the cosine similarity of two fault feature vectors (based on the relationship between each component and the vector modulus), providing strong support for fault diagnosis and analysis and improving the intelligent level of the system.

[0038] Please refer to Figure 2 , the present invention provides an embodiment: an industrial environment data acquisition method, including the following steps: S41: In the environmental perception stage, the sensor synchronously collects multi-dimensional data including vibration and temperature at a reference frequency of 1 kHz; S42: In the event detection stage, the FPGA chip calculates the kurtosis index of the vibration signal in real time. If it exceeds the threshold, it immediately switches to the high-speed sampling mode of 50 kHz, otherwise it maintains compressive sensing; S43: In the edge preprocessing stage, the data is compressed by Zstandard and then transmitted to the edge computing layer, and features including wavelet packet energy entropy and temperature gradient are extracted; S44: In the real-time decision-making stage, the NPU performs anomaly classification. If the confidence level exceeds 90%, it triggers in-depth analysis at the cloud platform layer; S45: In the global optimization stage, the cloud platform layer verifies the cause of the anomaly through finite element simulation and dynamically adjusts the detection threshold and model parameters of the edge layer.

[0039] As described above, the present invention realizes the efficiency and accuracy of industrial environment monitoring through multi-dimensional synchronous acquisition and intelligent hierarchical processing. Its innovation lies in: firstly, multi-source data such as vibration and temperature are synchronously acquired at a reference frequency of 1 kHz to ensure the integrity and relevance of environmental information; secondly, 50 kHz adaptive sampling is triggered through real-time kurtosis monitoring by the FPGA, taking into account both low-power normal monitoring and high-precision capture of abnormal events; Zstandard compression and wavelet packet energy entropy feature extraction at the edge effectively reduce the data transmission volume while retaining key fault features; the real-time anomaly classification driven by the NPU and the finite element simulation in the cloud form a closed-loop verification mechanism, which not only realizes millisecond-level anomaly response, but also continuously improves the diagnostic accuracy through dynamic parameter tuning. This edge-cloud collaborative architecture not only ensures real-time performance, but also significantly reduces system power consumption and cloud load, providing an intelligent solution with both resource efficiency and diagnostic depth for complex industrial scenarios.

[0040] The embodiments of the present invention have been described in detail above in conjunction with the accompanying drawings. However, the present invention is not limited to the above embodiments, and various changes can be made without departing from the spirit of the present invention within the scope of knowledge possessed by those skilled in the art.

Claims

1. An industrial environment data acquisition device; characterized in that: It includes: A11: The perception layer, which adopts one of the star network and Mesh network structures, is used to complete data collection and preprocessing by using deployed multi-type sensor nodes; A12: The edge computing layer, which is used to calculate and process the data collected by the perception layer by using deployed edge computing nodes; A13: The cloud platform layer, which is used to store, further process and analyze the data, and generate a data analysis report; A14: The application layer, which is used to perform anomaly location and diagnosis by using the digital twin model and the fault diagnosis rule base.

2. An industrial environment data acquisition device according to claim 1, characterized in that: The perception layer includes: A21: The multi-modal sensor array, which includes a composite vibration sensor, a high-resolution infrared thermal imager and a broadband acoustic emission sensor, is used to comprehensively capture the state of industrial equipment; A22: The intelligent conditioning circuit, which includes a programmable gain amplifier and an adaptive anti-aliasing filter, is used to dynamically adjust the signal gain and the filtering range. Among them, the working principle of the programmable gain amplifier is: ; Where G is the gain of the programmable amplifier, is the output voltage, is the input voltage, is the feedback resistor, is the input resistor; The working principle of the adaptive anti-aliasing filter is: ; wherein, is the cut-off frequency, R is the resistance value, and C is the capacitance value; A23: The heterogeneous processor, which includes an FPGA chip and an ARM processor, is used to realize real-time data preprocessing through the FPGA chip, and the ARM processor is responsible for protocol communication and trigger logic.

3. An industrial environment data acquisition device according to claim 2, characterized in that: The specific working mode of the perception layer is: A31: The composite vibration sensor combines MEMS and piezoelectric technologies to capture the time-domain waveform of the device vibration signal in real time. The high-resolution infrared thermal imager scans the surface temperature field of the device at a speed of 30 frames per second to generate a thermal map of 640×480 pixels. The broadband acoustic emission sensor collects the high-frequency acoustic wave signals during the operation of the device in real time. The composite vibration sensor, the high-resolution infrared thermal imager and the broadband acoustic emission sensor achieve clock synchronization between sensors through the hardware-level PTP protocol; A32: The programmable gain amplifier dynamically adjusts the gain according to the signal strength, and the adaptive anti-aliasing filter automatically adjusts the cut-off frequency according to the current sampling rate. Among them, the dynamic adjustment range of the programmable gain amplifier is 1-1000 times; A33: The FPGA chip performs envelope demodulation on the vibration signal, extracts the modulation frequency characteristics in real time, and performs short-time Fourier transform on the acoustic emission signal in real time to generate a time-frequency diagram. The ARM processor monitors the signal characteristics in real time, triggers the high-speed sampling mode when the signal characteristics are abnormal, and then packs and transmits the preprocessed data to the edge computing layer through the Modbus-TCP protocol. Among them, the principle of envelope demodulation is: ; Among them, is the original vibration signal, is the original vibration signal The Hilbert transform of, representing the imaginary part of the signal, is the envelope signal.

4. An industrial environment data acquisition device according to claim 3, characterized in that: When the edge computing layer calculates and processes the data collected by the perception layer by using deployed edge computing nodes, it specifically includes: S11: Data reception and parsing, receiving the encrypted data stream from the perception layer, restoring the original data through the hardware decryption module, and interpolating and aligning different sensor data according to the time stamp; S12: Feature engineering processing. For the vibration signal, calculate the 4096-point FFT spectrum, extract the 1 / 3 octave energy distribution, and eliminate the influence of rotational speed fluctuation on the spectrum through the order tracking technology. For the thermal map data, perform region segmentation on the infrared thermal map and calculate the temperature gradient of the key components; S13: Real-time diagnosis and decision-making. Run the built-in lightweight LSTM model to predict the vibration trend at the next 32 time points. If a sudden increase in vibration energy and a temperature gradient exceeding the threshold are detected, trigger a level-3 alarm, and upload high-priority data to the cloud platform layer through the TSN network, and cache the remaining data locally.

5. An industrial environment data acquisition device according to claim 4, characterized in that: When calculating the 4096-point FFT spectrum, the principle formula is: ; wherein, is the n-th sampling point of the discrete vibration signal, N is the number of FFT points, is the complex amplitude of the k-th frequency component, and j is the imaginary number.

6. An industrial environment data acquisition device according to claim 5, characterized in that: When eliminating the influence of rotational speed fluctuation on the spectrum through order tracking technology, the specific principle is: ; Among them, is the real-time rotational speed of the device, is the cumulative rotation angle, is the signal resampled according to the angle , is the inverse function of .

7. An industrial environment data acquisition device according to claim 6, characterized in that: When the cloud platform layer stores, further processes and analyzes the data, and generates a data analysis report, it specifically includes: S21: Data storage and management. Use a time-series database to store vibration waveforms, and use a relational database to store structured feature data, and perform data dimensionality reduction on historical data to generate a clustering map of the device health status; S22: Multi-physical field simulation verification. Load the device digital twin model, input the vibration and temperature data reported by the edge layer, perform transient dynamics simulation, calculate the bearing stress distribution, and verify whether the fault is caused by overload and assembly deviation; S23: Model iteration and optimization. Based on the federated learning framework, aggregate the edge model parameters of multiple factories, update the global fault classification model, and send the optimized detection threshold and lightweight model weights to the edge layer.

8. An industrial environment data acquisition device according to claim 7, characterized in that: When the application layer uses the data twin model and the fault diagnosis rule base for anomaly location and diagnosis, it specifically includes: S31: Digital twin mapping. Map the cloud simulation results to the 3D device model, highlight the suspected fault area, and overlay the real-time data stream to achieve dynamic synchronization between the physical entity and the virtual model; S32: Fault reasoning and decision-making. Call the fault diagnosis rule base and combine it with the knowledge graph to associate historical maintenance records; S33: Work order generation and push. Automatically generate a standardized maintenance work order, including fault location and spare part model, and push the work order to the MES system through the OPCUA protocol, and synchronously trigger the outbound process of the spare part warehouse.

9. An industrial environment data acquisition device according to claim 8, characterized in that: When the application layer combines the knowledge graph, the specific principle is: ; Among them, and are two fault feature vectors respectively, and are the i-th dimensional components of the feature vector respectively, is the norm of the vector.

10. An industrial environment data acquisition method, characterized in that: It includes the following steps: S41: Environment perception stage. Sensors synchronously collect multi-dimensional data including vibration and temperature at a reference frequency of 1 kHz; S42: Event detection stage. The FPGA chip calculates the kurtosis index of the vibration signal in real time. If it exceeds the threshold, immediately switch to the high-speed sampling mode of 50 kHz, otherwise maintain compressive sensing; S43: Edge preprocessing stage. The data is compressed by Zstandard and transmitted to the edge computing layer, and features including wavelet packet energy entropy and temperature gradient are extracted; S44: Real-time decision-making stage. The NPU performs anomaly classification. If the confidence level exceeds 90%, trigger in-depth analysis by the cloud platform layer; S45: Global optimization stage. The cloud platform layer verifies the cause of the anomaly through finite element simulation, and dynamically adjusts the detection threshold and model parameters of the edge layer.

Images