Rapid plastometer precision measurement method based on edge calculation

Abstract

The invention discloses a rapid plastometer precision measurement method based on edge calculation. The method comprises the following steps: collecting measurement data, transmitting the measurement data to an edge calculation node, and preprocessing the measurement data; performing dynamic feature extraction and parameter identification to form a real-time measurement feature value; performing error compensation and temperature drift correction by using the improved DESN model to generate a corrected feature vector; utilizing a data consistency test method to obtain an effective measurement data set; performing multi-source measurement data fusion to obtain a fusion processing result data set; calculating a measurement deviation index and outputting a correction result; generating loading response information according to the execution state of the loading control unit and the returned measurement data; and the updated improved DESN model is obtained, real-time processing, error suppression and high-precision feature recognition of the measurement of the plastometer are realized, and the data reliability and the measurement stability are improved.

Description

Technical Field

[0001] This invention relates to the field of edge intelligent measurement, and more particularly to a rapid plasticity meter precision measurement method based on edge computing. Background Technology

[0002] Plasticity testers, as important instruments for measuring the plastic deformation of materials, are widely used in the mechanical property testing of metallic materials, structural health monitoring, fatigue testing of engineering components, and strain behavior analysis during material processing. Current plasticity tester measurement methods primarily rely on transmitting measurements such as strain, voltage, and temperature through acquisition devices to a host computer or centralized server. The host computer then performs calculations such as noise processing, feature extraction, error compensation, and data fusion to ultimately calculate the material's plastic deformation parameters. However, as material testing processes evolve towards high-speed loading, complex working conditions, and high dynamic response characteristics, traditional centralized measurement and processing methods are gradually revealing several limitations.

[0003] In existing technologies, plasticity tester measurement systems typically upload all raw measurement data to a server for centralized processing, resulting in long system processing paths and large data transmission volumes. When the test process requires high sampling rates or simultaneous measurement across multiple sensor channels, the raw data throughput increases dramatically, leading to significant computational delays on the host computer. Traditional plasticity tester data processing methods often employ fixed signal processing procedures, such as simple filtering, static calibration, and linear compensation. These methods are ill-suited to the nonlinear response characteristics of materials during the plastic deformation stage. In practical applications, the plastic deformation process involves complex factors such as abrupt strain rate changes, thermally induced temperature drift, electrical noise disturbances, and load variations. If instantaneous strain rate, material plasticity coefficient, and nonlinear response characteristics cannot be extracted in real time, the true mechanical behavior of the material cannot be accurately reflected. Therefore, the shortcomings of existing technologies in parameter identification and feature extraction directly lead to problems such as accumulated measurement deviations and measurement instability. Summary of the Invention

[0004] One objective of this invention is to propose a rapid and precise measurement method for plasticity gauges based on edge computing. This invention fully utilizes edge computing, dynamic feature extraction technology, multi-source measurement data fusion technology, and an improved DESN model. By completing measurement data preprocessing, real-time feature value construction, error compensation and temperature drift correction, data consistency verification, and multi-channel data fusion at the edge computing node, it achieves rapid processing and high-precision measurement of plasticity gauge measurement data.

[0005] According to an embodiment of the present invention, a rapid plasticity meter precision measurement method based on edge computing includes the following steps:

[0006] A precision measurement system for a plasticity meter is built to collect measurement data and transmit it to an edge computing node, where the measurement data is preprocessed.

[0007] Dynamic feature extraction and parameter identification are performed on the preprocessed measurement data to form real-time measurement feature values;

[0008] Real-time feature values ​​are input into the improved DESN model for error compensation and temperature drift correction, generating a corrected feature vector.

[0009] The modified feature vector is compared with the historical measurement samples, and the data consistency test method is used to obtain the effective measurement dataset.

[0010] Within an edge computing node, multi-source measurement data fusion is performed on the effective measurement dataset to obtain the fused result dataset.

[0011] The fusion processing result dataset is synchronized to the host computer display and storage module through the high-speed communication interface of the edge node. Visual results are generated on the host computer, and the measurement deviation index is calculated and the correction result is output.

[0012] On the host computer, the correction result is converted into a loading adjustment command according to the set sampling period and control logic and fed back to the loading control unit. Loading response information is generated based on the execution status of the loading control unit and the returned measurement data.

[0013] Based on the correction results and loading response information, the improved DESN model in the edge computing node is updated to obtain the updated improved DESN model.

[0014] Optionally, the plasticity gauge precision measurement system includes a plasticity gauge sensing unit, an edge computing node, a signal acquisition module, a data processing module, an error correction module, and a host computer display and control module. The preprocessing includes noise filtering, outlier detection, and signal smoothing.

[0015] Optionally, the formation of the real-time measured feature values ​​specifically includes:

[0016] Based on the preprocessed measurement data, the instantaneous strain rate is calculated and constructed into an instantaneous strain rate sequence in chronological order. The calculation process involves extracting the strain measurement value corresponding to each sampling time from the preprocessed measurement data, selecting two adjacent sampling times in sequence, subtracting the strain measurement value of the previous sampling time from the strain measurement value of the current sampling time, and dividing by the time interval between the current sampling time and the previous sampling time to obtain the instantaneous strain rate.

[0017] Within a preset plastic deformation range, a plastic measurement window containing a predetermined number of continuous sampling times is selected from the preprocessed measurement data. For each sampling time within the plastic measurement window, a plastic response coefficient sequence is calculated. The plastic response coefficient sequence is obtained by reading the load data and plastic-related data corresponding to each sampling time within the plastic measurement window, and calculating them according to a preset relationship between the load data and plastic-related data. The arithmetic mean of the plastic response coefficient sequence is then taken as the material plasticity coefficient.

[0018] At the selected sampling time, the absolute value of the difference between the actual strain measurement value and the strain measurement value of the corresponding linear reference is calculated, and the arithmetic mean of the absolute values ​​of the difference corresponding to all sampling times is taken as the nonlinear response characteristic.

[0019] The instantaneous strain rate sequence, material plasticity coefficient, and nonlinear response characteristics are combined to form real-time characteristic values.

[0020] Optionally, the generation of the modified feature vector specifically includes:

[0021] Real-time feature values ​​are input into the improved DESN model for error compensation and temperature drift correction. The improved DESN model includes an input feature encoding module, a deep reservoir dynamic state generation module, and an adaptive output weight calculation module. The input feature encoding module normalizes the real-time feature values ​​and uses multi-scale wavelet decomposition to form an input encoding vector. The deep reservoir dynamic state generation module introduces a state evolution mechanism approximating the Koopman operator to generate the reservoir state vector at the current moment. The adaptive output weight calculation module calculates the error compensation amount and temperature drift correction amount based on the reservoir state vector at the current moment.

[0022] In the input feature encoding module, the real-time feature values ​​are normalized and multi-scale wavelet decomposition is used to obtain feature components at different scales to form the input encoding vector.

[0023] In the deep reserve pool dynamic state generation module, based on the input encoding vector, a state evolution mechanism approximating the Koopman operator is introduced to generate the reserve pool state vector at the current moment. The generation process is to map the input encoding vector to a high-dimensional feature space to obtain a high-dimensional feature vector, and use the approximate Koopman operator to perform linear evolution processing on the high-dimensional feature vector, and map the evolved high-dimensional feature vector to the reserve pool state vector at the current moment.

[0024] In the adaptive output weight calculation module, based on the current state vector of the reservoir, error compensation amount and temperature drift correction amount are generated. The generation process is to multiply each component in the reservoir state vector with the preset output weight parameters one by one and sum all the products. According to the preset component mapping relationship, it is divided into error compensation amount and temperature drift correction amount.

[0025] The error compensation amount and the temperature drift correction amount are concatenated to form a correction feature vector.

[0026] Optionally, obtaining the effective measurement dataset specifically includes:

[0027] The modified feature vectors are organized according to the sampling time to establish a sequence of modified feature vectors, and the corresponding historical measurement sample feature vectors are matched for each modified feature vector.

[0028] For each modified feature vector in the modified feature vector sequence, a similarity calculation is performed with the corresponding historical measurement sample feature vector. The similarity calculation process involves successively calculating the difference between the values ​​of the modified feature vector and the historical measurement sample feature vector at each corresponding component, squaring each component difference, summing the differences, and taking the square root.

[0029] Based on the data consistency verification method, the similarity value is compared with a preset similarity threshold. When the similarity value is less than the similarity threshold, the measurement point data at the corresponding sampling time is determined to be measurement point data that meets the consistency requirements. When the similarity value is greater than the similarity threshold, the measurement point data at the corresponding sampling time is determined to be measurement point data that does not meet the consistency requirements and is removed, thus forming a valid measurement dataset.

[0030] Optionally, obtaining the fusion processing result dataset specifically includes:

[0031] Within the edge computing node, the effective measurement dataset is organized according to the measurement channel and sampling time to obtain multi-source measurement data, which includes strain measurement data, voltage measurement data, temperature measurement data, and load measurement data.

[0032] Using the sampling time as an index, the measurements of each measurement channel in the multi-source measurement data at the same sampling time are combined into a multi-dimensional measurement feature vector, and a channel correlation matrix is ​​constructed. The construction process involves calculating the numerical differences between the multi-dimensional measurement feature vectors at each sampling time and filling them into the corresponding positions of the matrix according to the correspondence between channels. By traversing the combination relationships of all measurement channels, a channel correlation matrix is ​​formed.

[0033] Based on the channel correlation matrix, a channel correlation-driven feature aggregation operation is performed on the multi-source measurement data to obtain a fused measurement feature vector, which is then used as the fused measurement data corresponding to the sampling time.

[0034] The fused measurement data are combined and stored in the order of sampling time to form a fused processing result dataset.

[0035] Optionally, the output of the correction result specifically includes:

[0036] The fusion processing result dataset is synchronized to the host computer display and storage module through the high-speed communication interface of the edge node. Based on the fusion processing result dataset, the measurement data corresponding to each sampling time is called, and the visualization results are generated according to the time order and the category of the measured physical quantity.

[0037] Based on the measurement data in the fusion processing result dataset, the measurement deviation index is calculated. The calculation process is to take the difference between the measurement data in the fusion processing result dataset and the corresponding reference measurement data, square the difference successively, accumulate them, and divide by the number of sampling times to perform a square root operation to obtain the measurement deviation index.

[0038] Based on the measurement deviation index and the preset correction rules, a correction result is generated. The correction result is obtained by adjusting the direction and magnitude of the fused measurement data in terms of numerical values ​​according to the measurement deviation index. Then, the fused measurement data corresponding to each sampling time is numerically corrected one by one according to the preset correction rules. The corrected fused measurement data of each sampling time is arranged and combined according to the sampling time sequence to obtain the result.

[0039] Optionally, the generation of the loading response information specifically includes:

[0040] On the host computer, based on the set sampling period and control logic, at each sampling moment, a loading adjustment command is generated from the corresponding correction result according to the preset correction rule, forming a loading adjustment command sequence.

[0041] The loading adjustment command sequence is sent to the loading control unit through the control interface, and the loading control unit is controlled to perform loading adjustment operation according to the command sequence at each sampling time.

[0042] The host computer receives the execution status information and returned measurement data from the loading control unit, associates and binds the execution status information and the returned measurement data to form a combined data entry containing the loading execution status and measurement results, thus forming loading response information.

[0043] Optionally, obtaining the updated and improved DESN model specifically includes:

[0044] The correction results and loading response information corresponding to each sampling time are organized in the order of sampling time to obtain a sequence of correction results and a sequence of loading response information that correspond one-to-one with the sampling time, forming the target output vector sequence;

[0045] Read the reservoir state vector corresponding to each sampling time from the current improved DESN model in the edge computing node, and obtain the current output vector sequence based on the adaptive output weight calculation module;

[0046] The difference between the target output vector sequence and the current output vector sequence is calculated to obtain the difference vector. The output weight increment vector is then calculated using the difference vector, the preset learning rate parameter, and the reservoir state vector at the corresponding sampling time to obtain the updated improved DESN model.

[0047] The beneficial effects of this invention are:

[0048] This invention achieves a unified measurement effect of high real-time performance, high reliability, and high accuracy, which is difficult to achieve with existing technologies, by introducing an edge computing architecture, a dynamic feature extraction mechanism, an improved DESN error compensation model, and a multi-source measurement data fusion method into the plasticity meter measurement system. Addressing the technical limitations of traditional plasticity meters, which rely on centralized processing, suffer from high latency, weak error compensation capabilities, and an inability to handle material nonlinear responses, this invention moves the preprocessing of measurement data, feature extraction, error correction, and data consistency verification to edge computing nodes. This enables the system to maintain stable real-time processing capabilities under high-speed loading and complex operating conditions, significantly reducing the impact of data transmission latency accumulation and network instability on measurement performance. Simultaneously, by utilizing the instantaneous strain rate, material plasticity coefficient, and nonlinear response characteristics constructed through dynamic feature extraction, this invention can more accurately describe the transient behavior of materials during the plastic stage, providing high-quality feature inputs for error compensation.

[0049] Regarding error compensation and temperature drift correction, this invention employs an improved DESN model. Through input feature encoding, a reservoir state evolution mechanism approximated by the Koopman operator, and adaptive output weight calculation, the model can accurately characterize nonlinear errors and temperature drift in high-dimensional dynamic states, improving the stability and robustness of measurement results. Simultaneously, this invention introduces a multi-dimensional consistency verification method based on feature similarity to effectively eliminate inconsistent measurement data caused by noise interference or transient anomalies, avoiding the influence of outliers on parameter calculations and thus improving overall data quality. In a multi-channel measurement environment, this invention further generates continuous, smooth, and physically consistent strain change curves by constructing a channel correlation matrix and a multi-source fusion mechanism based on feature structures, providing a reliable foundation for high-precision calculation of plastic deformation rate and strain energy parameters.

[0050] Furthermore, this invention synchronizes the fused measurement results to the host computer in real time via edge nodes for visualization, generates correction results based on measurement deviation indicators, and then converts these correction results into loading adjustment commands, feeding them back to the loading control unit. This achieves dynamic closed-loop adjustment of the measurement and loading processes. Through continuous online updates of the improved DESN model, this invention can continuously self-optimize as the test progresses and the environment changes, improving the system's adaptability to different materials, loading paths, and environmental conditions. In summary, this invention not only overcomes the shortcomings of traditional plasticity testers, such as poor real-time performance, limited error compensation capabilities, weak data consistency, and inability to dynamically adjust, but also enhances the intelligence of the measurement process, achieving rapid, precise, and stable measurement of plasticity parameters under highly dynamic conditions. It has significant engineering application value and promotional significance. Attached Figure Description

[0051] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0052] Figure 1 This is an overall flowchart of a rapid plasticity meter precision measurement method based on edge computing proposed in this invention;

[0053] Figure 2 This is a schematic diagram of the real-time characteristic value of a rapid plasticity meter precision measurement method based on edge computing proposed in this invention;

[0054] Figure 3 This is a schematic diagram of the module structure of an improved DESN model for a rapid plasticity meter precision measurement method based on edge computing proposed in this invention. Detailed Implementation

[0055] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0056] refer to Figure 1-3 A rapid plasticity meter precision measurement method based on edge computing includes the following steps:

[0057] A precision measurement system for a plasticity meter is built to collect measurement data and transmit it to an edge computing node, where the measurement data is preprocessed.

[0058] Dynamic feature extraction and parameter identification are performed on the preprocessed measurement data to form real-time measurement feature values;

[0059] Real-time feature values ​​are input into the improved DESN model for error compensation and temperature drift correction, generating a corrected feature vector.

[0060] The modified feature vector is compared with the historical measurement samples, and the data consistency test method is used to obtain the effective measurement dataset.

[0061] Within an edge computing node, multi-source measurement data fusion is performed on the effective measurement dataset to obtain the fused result dataset.

[0062] The fusion processing result dataset is synchronized to the host computer display and storage module through the high-speed communication interface of the edge node. Visual results are generated on the host computer, and the measurement deviation index is calculated and the correction result is output.

[0063] On the host computer, the correction result is converted into a loading adjustment command according to the set sampling period and control logic and fed back to the loading control unit. Loading response information is generated based on the execution status of the loading control unit and the returned measurement data.

[0064] Based on the correction results and loading response information, the improved DESN model in the edge computing node is updated to obtain the updated improved DESN model.

[0065] In this embodiment, the plasticity meter precision measurement system includes a plasticity meter sensing unit, an edge computing node, a signal acquisition module, a data processing module, an error correction module, and a host computer display and control module. The preprocessing includes noise filtering, outlier detection, and signal smoothing.

[0066] In this embodiment, the formation of the real-time measurement feature value specifically includes:

[0067] Based on the preprocessed measurement data, the instantaneous strain rate is calculated and constructed into an instantaneous strain rate sequence in chronological order. The calculation process involves extracting the strain measurement value corresponding to each sampling time from the preprocessed measurement data, selecting two adjacent sampling times in sequence, subtracting the strain measurement value of the previous sampling time from the strain measurement value of the current sampling time, and dividing by the time interval between the current sampling time and the previous sampling time to obtain the instantaneous strain rate.

[0068] Within a preset plastic deformation range, a plastic measurement window containing a predetermined number of continuous sampling times is selected from the preprocessed measurement data. For each sampling time within the plastic measurement window, a plastic response coefficient sequence is calculated. The plastic response coefficient sequence is obtained by reading the load data and plastic-related data corresponding to each sampling time within the plastic measurement window, and calculating them according to a preset relationship between the load data and plastic-related data. The arithmetic mean of the plastic response coefficient sequence is then taken as the material plasticity coefficient.

[0069] At the selected sampling time, the absolute value of the difference between the actual strain measurement value and the strain measurement value of the corresponding linear reference is calculated, and the arithmetic mean of the absolute values ​​of the difference corresponding to all sampling times is taken as the nonlinear response characteristic.

[0070] The instantaneous strain rate sequence, material plasticity coefficient, and nonlinear response characteristics are combined to form real-time characteristic values.

[0071] In this embodiment, the generation of the modified feature vector specifically includes:

[0072] Real-time feature values ​​are input into the improved DESN model for error compensation and temperature drift correction. The improved DESN model includes an input feature encoding module, a deep reservoir dynamic state generation module, and an adaptive output weight calculation module. The input feature encoding module normalizes the real-time feature values ​​and uses multi-scale wavelet decomposition to form an input encoding vector. The deep reservoir dynamic state generation module introduces a state evolution mechanism approximating the Koopman operator to generate the reservoir state vector at the current moment. The adaptive output weight calculation module calculates the error compensation amount and temperature drift correction amount based on the reservoir state vector at the current moment.

[0073] In the input feature encoding module, the real-time feature values ​​are normalized and multi-scale wavelet decomposition is used to obtain feature components at different scales to form the input encoding vector.

[0074] In the deep reserve pool dynamic state generation module, based on the input encoding vector, a state evolution mechanism approximating the Koopman operator is introduced to generate the reserve pool state vector at the current moment. The generation process is to map the input encoding vector to a high-dimensional feature space to obtain a high-dimensional feature vector, and use the approximate Koopman operator to perform linear evolution processing on the high-dimensional feature vector, and map the evolved high-dimensional feature vector to the reserve pool state vector at the current moment.

[0075] In the adaptive output weight calculation module, based on the current state vector of the reservoir, error compensation amount and temperature drift correction amount are generated. The generation process is to multiply each component in the reservoir state vector with the preset output weight parameters one by one and sum all the products. According to the preset component mapping relationship, it is divided into error compensation amount and temperature drift correction amount.

[0076] The error compensation amount and the temperature drift correction amount are concatenated to form a correction feature vector.

[0077] In this embodiment, obtaining the effective measurement dataset specifically includes:

[0078] The modified feature vectors are organized according to the sampling time to establish a sequence of modified feature vectors, and the corresponding historical measurement sample feature vectors are matched for each modified feature vector.

[0079] For each modified feature vector in the modified feature vector sequence, a similarity calculation is performed with the corresponding historical measurement sample feature vector. The similarity calculation process involves successively calculating the difference between the values ​​of the modified feature vector and the historical measurement sample feature vector at each corresponding component, squaring each component difference, summing the differences, and taking the square root.

[0080] Based on the data consistency verification method, the similarity value is compared with a preset similarity threshold. When the similarity value is less than the similarity threshold, the measurement point data at the corresponding sampling time is determined to be measurement point data that meets the consistency requirements. When the similarity value is greater than the similarity threshold, the measurement point data at the corresponding sampling time is determined to be measurement point data that does not meet the consistency requirements and is removed, thus forming a valid measurement dataset.

[0081] In this embodiment, obtaining the fusion processing result dataset specifically includes:

[0082] Within the edge computing node, the effective measurement dataset is organized according to the measurement channel and sampling time to obtain multi-source measurement data, which includes strain measurement data, voltage measurement data, temperature measurement data, and load measurement data.

[0083] Using the sampling time as an index, the measurements of each measurement channel in the multi-source measurement data at the same sampling time are combined into a multi-dimensional measurement feature vector, and a channel correlation matrix is ​​constructed. The construction process involves calculating the numerical differences between the multi-dimensional measurement feature vectors at each sampling time and filling them into the corresponding positions of the matrix according to the correspondence between channels. By traversing the combination relationships of all measurement channels, a channel correlation matrix is ​​formed.

[0084] Based on the channel correlation matrix, a channel correlation-driven feature aggregation operation is performed on the multi-source measurement data to obtain a fused measurement feature vector, which is then used as the fused measurement data corresponding to the sampling time.

[0085] The fused measurement data are combined and stored in the order of sampling time to form a fused processing result dataset.

[0086] In this embodiment, the output of the correction result specifically includes:

[0087] The fusion processing result dataset is synchronized to the host computer display and storage module through the high-speed communication interface of the edge node. Based on the fusion processing result dataset, the measurement data corresponding to each sampling time is called, and the visualization results are generated according to the time order and the category of the measured physical quantity.

[0088] Based on the measurement data in the fusion processing result dataset, the measurement deviation index is calculated. The calculation process is to take the difference between the measurement data in the fusion processing result dataset and the corresponding reference measurement data, square the difference successively, accumulate them, and divide by the number of sampling times to perform a square root operation to obtain the measurement deviation index.

[0089] Based on the measurement deviation index and the preset correction rules, a correction result is generated. The correction result is obtained by adjusting the direction and magnitude of the fused measurement data in terms of numerical values ​​according to the measurement deviation index. Then, the fused measurement data corresponding to each sampling time is numerically corrected one by one according to the preset correction rules. The corrected fused measurement data of each sampling time is arranged and combined according to the sampling time sequence to obtain the result.

[0090] In this embodiment, the generation of the loading response information specifically includes:

[0091] On the host computer, based on the set sampling period and control logic, at each sampling moment, a loading adjustment command is generated from the corresponding correction result according to the preset correction rule, forming a loading adjustment command sequence.

[0092] The loading adjustment command sequence is sent to the loading control unit through the control interface, and the loading control unit is controlled to perform loading adjustment operation according to the command sequence at each sampling time.

[0093] The host computer receives the execution status information and returned measurement data from the loading control unit, associates and binds the execution status information and the returned measurement data to form a combined data entry containing the loading execution status and measurement results, thus forming loading response information.

[0094] In this embodiment, obtaining the updated and improved DESN model specifically includes:

[0095] The correction results and loading response information corresponding to each sampling time are organized in the order of sampling time to obtain a sequence of correction results and a sequence of loading response information that correspond one-to-one with the sampling time, forming the target output vector sequence;

[0096] Read the reservoir state vector corresponding to each sampling time from the current improved DESN model in the edge computing node, and obtain the current output vector sequence based on the adaptive output weight calculation module;

[0097] The difference between the target output vector sequence and the current output vector sequence is calculated to obtain the difference vector. The output weight increment vector is then calculated using the difference vector, the preset learning rate parameter, and the reservoir state vector at the corresponding sampling time to obtain the updated improved DESN model.

[0098] Example 1:

[0099] In a metal materials laboratory, to verify the plastic deformation characteristics of newly developed high-strength steel under high-speed tensile conditions, researchers built a measurement system comprising a plasticity gauge sensing unit, a loading control unit, edge computing nodes, and a host computer. In this scenario, the experiment required real-time capture of strain, voltage, and temperature changes of the sample during the plastic stage at a high sampling frequency of 200Hz. It also required calculation of the material's instantaneous strain rate, plasticity coefficient, and nonlinear response, while simultaneously correcting for measurement deviations caused by temperature drift in real time to ensure high-precision plasticity parameter measurement capabilities during high-speed loading.

[0100] Traditional testing methods suffer from latency exceeding 50ms due to data uploading to a host computer for processing. Furthermore, high measurement noise and poor consistency between channels often lead to jitter in strain curves and shifts in the peak position of instantaneous strain rate. This invention, however, migrates preprocessing, dynamic feature extraction, error compensation, data consistency verification, and multi-source fusion to edge computing nodes, significantly shortening the processing path in the measurement chain and achieving real-time response capabilities for high-frequency measurements.

[0101] During the specific testing process, after the sample is installed, the plasticity gauge sensing unit begins to synchronously record strain, voltage, and temperature signals. The acquired measurement data is transmitted to the edge computing node in real time, where the node immediately performs preprocessing steps such as filtering, baseline calibration, and time synchronization. Subsequently, the system automatically extracts real-time measured feature values ​​such as instantaneous strain rate, material plasticity coefficient, and nonlinear response characteristics from the preprocessed data, and inputs them into the improved DESN model for error compensation and temperature drift correction, generating a corrected feature vector.

[0102] To ensure data reliability, the feature vector is compared with historical measurement samples in the same feature space for similarity. If the feature vector at a certain sampling moment deviates too much from the historical feature distribution, the system will automatically mark and remove the abnormal measurement point to avoid affecting subsequent curve generation and parameter calculation. In this way, even if a channel is occasionally subjected to impact or electromagnetic interference during the test, causing signal abnormalities, it will not affect the stability of the entire measurement curve. Subsequently, the edge computing nodes perform multi-source fusion on the remaining data, making the curve smoother and more physically consistent, and calculating key parameters such as plastic deformation rate and strain energy.

[0103] The entire data processing results are pushed to the host computer in real time and displayed as curves, allowing testers to observe the plastic response behavior of the material in real time. Simultaneously, the host computer generates correction results based on measurement deviation indicators and dynamically adjusts the loading process through the loading control unit to maintain stability and high precision during testing. After the test, the system uses the correction results and loading response information to update and improve the DESN model, enabling the model to further adapt to environmental and material properties in subsequent tests and improve the accuracy of continuous measurements.

[0104] The following are typical data comparison results obtained from a high-speed tensile test, reflecting the improvements of the present invention in terms of strain measurement accuracy, data stability, and strain rate identification capability.

[0105] Table 1. Comparison of measurement performance data between the traditional method and the method of this invention in high-speed tensile testing.

[0106] Measurement indicators Traditional method measurement value Measurement values ​​of this invention degree of improvement Measurement delay (ms) 51.7 7.3 ↓85.9% Peak noise amplitude (με) 34.2 8.1 ↓76.3% Curve jitter index (%) 12.5 3.4 ↓72.8% Strain energy calculation error (%) 9.2 2.1 ↓77.2% Deviation caused by temperature drift (με) 27.8 4.5 ↓83.8% Channel consistency correlation coefficient 0.82 0.96 ↑17.1% Instantaneous strain rate peak identification error (%) 6.8 1.4 ↓79.4% Abnormal measurement point rejection rate (%) 2.3 0.4 More stable Curve smoothness (%) after multi-source fusion 83.6 95.1 ↑13.7% Stability measurement under high dynamic loading (rating) 3.1 / 5 4.8 / 5 ↑Significant

[0107] As shown in Table 1, in high-speed tensile tests, the method of this invention exhibits significant advantages over traditional centralized measurement methods in all key indicators. The measurement latency is significantly reduced from 51.7 ms in the traditional method to 7.3 ms, thanks to edge computing nodes undertaking the main computational tasks, which greatly shortens the measurement link and eliminates reliance on the centralized processing capabilities of remote servers. Regarding noise suppression and curve stability, this invention reduces the peak noise amplitude by 76.3% and the curve jitter index by 72.8% through dynamic feature extraction, error compensation, and multi-source data fusion, reflecting the effectiveness of data filtering and consistency enhancement mechanisms.

[0108] Key parameters of the material's plastic stage, such as strain energy and instantaneous strain rate, have also been identified with higher accuracy. The strain energy calculation error has decreased from 9.2% to 2.1%, and the peak instantaneous strain rate identification error has been reduced by nearly 80%, indicating that the strain change curve processed by this invention is more continuous and realistic, accurately reflecting the physical behavior of the material in the plastic stage. Simultaneously, the method of this invention has a significant effect on correcting temperature drift, reducing the measurement deviation caused by temperature drift by more than 80%, ensuring the stability of long-term measurements.

[0109] The multi-channel correlation improved from 0.82 to 0.96, indicating that the proposed method, through similarity consistency testing and fusion strategies, exhibits significant robustness and reliability in multi-channel measurement environments. Furthermore, the reduced outlier removal rate signifies greater system stability, while the smoothness of the fusion curve increased to 95.1%, further enhancing the accuracy of parameter calculations. In summary, the proposed method successfully addresses the problems of insufficient real-time performance, severe error accumulation, ineffective compensation for nonlinear response, and weak consistency of multi-source data inherent in traditional methods, achieving rapid measurement capabilities for plasticity meters with high dynamics, high precision, and high reliability.

Claims

1. A fast plastic gauge precision measurement method based on edge computing, characterized in that, Includes the following steps: A precision measurement system for a plasticity meter is built to collect measurement data and transmit it to an edge computing node, where the measurement data is preprocessed. Dynamic feature extraction and parameter identification are performed on the preprocessed measurement data to form real-time measurement feature values; Real-time feature values ​​are input into the improved DESN model for error compensation and temperature drift correction, generating a corrected feature vector. The modified feature vector is compared with the historical measurement samples, and the data consistency test method is used to obtain the effective measurement dataset. Within an edge computing node, multi-source measurement data fusion is performed on the effective measurement dataset to obtain the fused result dataset. The fusion processing result dataset is synchronized to the host computer display and storage module through the high-speed communication interface of the edge node. Visual results are generated on the host computer, and the measurement deviation index is calculated and the correction result is output. On the host computer, the correction result is converted into a loading adjustment command according to the set sampling period and control logic and fed back to the loading control unit. Loading response information is generated based on the execution status of the loading control unit and the returned measurement data. Based on the correction results and loading response information, the improved DESN model in the edge computing node is updated to obtain the updated improved DESN model.

2. The fast plastometer precision measurement method based on edge computing according to claim 1, characterized in that, The plasticity gauge precision measurement system includes a plasticity gauge sensing unit, an edge computing node, a signal acquisition module, a data processing module, an error correction module, and a host computer display and control module. The preprocessing includes noise filtering, outlier detection, and signal smoothing.

3. The fast plastometer precision measurement method based on edge computing according to claim 1, characterized in that, The formation of the real-time measurement feature values ​​specifically includes: Based on the preprocessed measurement data, the instantaneous strain rate is calculated and constructed into an instantaneous strain rate sequence in chronological order. The calculation process involves extracting the strain measurement value corresponding to each sampling time from the preprocessed measurement data, selecting two adjacent sampling times in sequence, subtracting the strain measurement value of the previous sampling time from the strain measurement value of the current sampling time, and dividing by the time interval between the current sampling time and the previous sampling time to obtain the instantaneous strain rate. Within a preset plastic deformation range, a plastic measurement window containing a predetermined number of continuous sampling times is selected from the preprocessed measurement data. For each sampling time within the plastic measurement window, a plastic response coefficient sequence is calculated. The plastic response coefficient sequence is obtained by reading the load data and plastic-related data corresponding to each sampling time within the plastic measurement window, and calculating them according to a preset relationship between the load data and plastic-related data. The arithmetic mean of the plastic response coefficient sequence is then taken as the material plasticity coefficient. At the selected sampling time, the absolute value of the difference between the actual strain measurement value and the strain measurement value of the corresponding linear reference is calculated, and the arithmetic mean of the absolute values ​​of the difference corresponding to all sampling times is taken as the nonlinear response characteristic. The instantaneous strain rate sequence, material plasticity coefficient, and nonlinear response characteristics are combined to form real-time characteristic values.

4. The fast plastometer precision measurement method based on edge computing according to claim 1, characterized in that, The generation of the modified feature vector specifically includes: Real-time feature values ​​are input into the improved DESN model for error compensation and temperature drift correction. The improved DESN model includes an input feature encoding module, a deep reservoir dynamic state generation module, and an adaptive output weight calculation module. The input feature encoding module normalizes the real-time feature values ​​and uses multi-scale wavelet decomposition to form an input encoding vector. The deep reservoir dynamic state generation module introduces a state evolution mechanism approximating the Koopman operator to generate the reservoir state vector at the current moment. The adaptive output weight calculation module calculates the error compensation amount and temperature drift correction amount based on the reservoir state vector at the current moment. In the input feature encoding module, the real-time feature values ​​are normalized and multi-scale wavelet decomposition is used to obtain feature components at different scales to form the input encoding vector. In the deep reserve pool dynamic state generation module, based on the input encoding vector, a state evolution mechanism approximating the Koopman operator is introduced to generate the reserve pool state vector at the current moment. The generation process is to map the input encoding vector to a high-dimensional feature space to obtain a high-dimensional feature vector, and use the approximate Koopman operator to perform linear evolution processing on the high-dimensional feature vector, and map the evolved high-dimensional feature vector to the reserve pool state vector at the current moment. In the adaptive output weight calculation module, based on the current state vector of the reservoir, error compensation amount and temperature drift correction amount are generated. The generation process is to multiply each component in the reservoir state vector with the preset output weight parameters one by one and sum all the products. According to the preset component mapping relationship, it is divided into error compensation amount and temperature drift correction amount. The error compensation amount and the temperature drift correction amount are concatenated to form a correction feature vector.

5. The fast plastometer precision measurement method based on edge computing according to claim 1, characterized in that, Obtaining the effective measurement dataset specifically includes: The modified feature vectors are organized according to the sampling time to establish a sequence of modified feature vectors, and the corresponding historical measurement sample feature vectors are matched for each modified feature vector. For each modified feature vector in the modified feature vector sequence, a similarity calculation is performed with the corresponding historical measurement sample feature vector. The similarity calculation process involves successively calculating the difference between the values ​​of the modified feature vector and the historical measurement sample feature vector at each corresponding component, squaring each component difference, summing the differences, and taking the square root. Based on the data consistency verification method, the similarity value is compared with a preset similarity threshold. When the similarity value is less than the similarity threshold, the measurement point data at the corresponding sampling time is determined to be measurement point data that meets the consistency requirements. When the similarity value is greater than the similarity threshold, the measurement point data at the corresponding sampling time is determined to be measurement point data that does not meet the consistency requirements and is removed, thus forming a valid measurement dataset.

6. The fast plastometer precision measurement method based on edge computing according to claim 1, characterized in that, The specific steps involved in obtaining the fusion processing result dataset are as follows: Within the edge computing node, the effective measurement dataset is organized according to the measurement channel and sampling time to obtain multi-source measurement data, which includes strain measurement data, voltage measurement data, temperature measurement data, and load measurement data. Using the sampling time as an index, the measurements of each measurement channel in the multi-source measurement data at the same sampling time are combined into a multi-dimensional measurement feature vector, and a channel correlation matrix is ​​constructed. The construction process involves calculating the numerical differences between the multi-dimensional measurement feature vectors at each sampling time and filling them into the corresponding positions of the matrix according to the correspondence between channels. By traversing the combination relationships of all measurement channels, a channel correlation matrix is ​​formed. Based on the channel correlation matrix, a channel correlation-driven feature aggregation operation is performed on the multi-source measurement data to obtain a fused measurement feature vector, which is then used as the fused measurement data corresponding to the sampling time. The fused measurement data are combined and stored in the order of sampling time to form a fused processing result dataset.

7. The method for rapid plasticity measurement based on edge computing according to claim 1, characterized in that, The output of the correction result specifically includes: The fusion processing result dataset is synchronized to the host computer display and storage module through the high-speed communication interface of the edge node. Based on the fusion processing result dataset, the measurement data corresponding to each sampling time is called, and the visualization results are generated according to the time order and the category of the measured physical quantity. Based on the measurement data in the fusion processing result dataset, the measurement deviation index is calculated. The calculation process is to take the difference between the measurement data in the fusion processing result dataset and the corresponding reference measurement data, square the difference successively, accumulate them, and divide by the number of sampling times to perform a square root operation to obtain the measurement deviation index. Based on the measurement deviation index and the preset correction rules, a correction result is generated. The correction result is obtained by adjusting the direction and magnitude of the fused measurement data in terms of numerical values ​​according to the measurement deviation index. Then, the fused measurement data corresponding to each sampling time is numerically corrected one by one according to the preset correction rules. The corrected fused measurement data of each sampling time is arranged and combined according to the sampling time sequence to obtain the result.

8. The method for rapid plasticity measurement based on edge computing according to claim 1, characterized in that, The generation of the loading response information specifically includes: On the host computer, based on the set sampling period and control logic, at each sampling moment, a loading adjustment command is generated from the corresponding correction result according to the preset correction rule, forming a loading adjustment command sequence. The loading adjustment command sequence is sent to the loading control unit through the control interface, and the loading control unit is controlled to perform loading adjustment operation according to the command sequence at each sampling time. The host computer receives the execution status information and returned measurement data from the loading control unit, associates and binds the execution status information and the returned measurement data to form a combined data entry containing the loading execution status and measurement results, thus forming loading response information.

9. The method for rapid plasticity measurement based on edge computing according to claim 1, characterized in that, The updated and improved DESN model is obtained specifically through: The correction results and loading response information corresponding to each sampling time are organized in the order of sampling time to obtain a sequence of correction results and a sequence of loading response information that correspond one-to-one with the sampling time, forming the target output vector sequence; Read the reservoir state vector corresponding to each sampling time from the current improved DESN model in the edge computing node, and obtain the current output vector sequence based on the adaptive output weight calculation module; The difference between the target output vector sequence and the current output vector sequence is calculated to obtain the difference vector. The output weight increment vector is then calculated using the difference vector, the preset learning rate parameter, and the reservoir state vector at the corresponding sampling time to obtain the updated improved DESN model.

Images