CN-122017489-A - Ultrahigh voltage cable insulation detection method and ultrahigh voltage cable maintenance method

CN122017489ACN 122017489 ACN122017489 ACN 122017489ACN-122017489-A

Abstract

The present invention discloses a method for detecting insulation of ultra-high voltage cables and a method for maintaining ultra-high voltage cables, which relates to the field of cross-linked polyethylene cable manufacturing technology. By in-situ and non-contact terahertz time-domain spectroscopy three-dimensional full field scanning of the produced cables, the time-domain waveforms of each spatial point of the cable insulation layer in the cable are obtained, and a four-dimensional database containing spatial coordinates and corresponding dielectric spectra is calculated and generated. The present invention comprehensively perceives the microstructure of insulation through non-destructive terahertz scanning, and combines manufacturing data with molecular dynamics mechanism, using hybrid AI The model achieves a leap from micro defect localization to long-term performance prediction, enabling in-situ and quantitative evaluation and life prediction of insulation reliability of cables after production is completed, and forming a closed loop to guide process optimization, ultimately achieving a fundamental transformation from passive inspection to active predictive quality control.

Inventors

ZHANG WENHUA
LIU CHANGRONG
QIU LIANG
WANG YANG
LIU PING
LUO ZHIQUN

Assignees

江西电缆有限责任公司

Dates

Publication Date: 20260512
Application Date: 20260205

Claims (10)

1. The insulation detection method for the ultra-high voltage cable is characterized by comprising the following steps of: s100, carrying out in-situ non-contact terahertz time-domain spectrum three-dimensional full-field scanning on the produced cable, obtaining time-domain waveforms of all space points of a cable insulating layer in the cable, and calculating and generating a four-dimensional database containing space coordinates and corresponding dielectric spectrums; S200, synchronously acquiring key process parameters of the whole cable production process, digitally associating the key process parameters with the corresponding cable segment space coordinates in the S100, and generating associated process parameters; s300, constructing a mechanistic knowledge base of a cross-linked polyethylene micro defect structure under different process deviations and an evolution rule under electric-thermal stress based on molecular dynamics simulation, and extracting key mechanism characteristic parameters; s400, constructing a mechanism-data hybrid driving artificial intelligent model; Taking the dielectric spectrum of the S100, the associated technological parameters of the S200 and the key mechanism characteristic parameters matched from the S300 as mixed input, and taking the microscopic defect type, the severity level and the long-term performance prediction index as output, training a mechanism-data mixed driving artificial intelligent model; S500, inputting the dielectric spectrum and related process parameters of the cable to be tested into a trained mechanism-data hybrid driving artificial intelligent model to obtain three-dimensional defect distribution and long-term performance prediction results of the cable insulating layer, and generating a visual quality assessment report according to the three-dimensional defect distribution and long-term performance prediction results; And S600, generating a production process optimization feedback suggestion according to the correlation analysis of the three-dimensional defect distribution and the correlation process parameters in the visual quality evaluation report.
2. The method for detecting insulation of an ultra-high voltage cable according to claim 1, wherein the step S100 specifically comprises: s110, enabling the produced cable to axially pass through an annular terahertz time-domain spectrum scanning device at a constant speed; s120, controlling a transmitting and receiving unit of the terahertz time-domain spectrum scanning device to rotate at a constant speed along the circumferential direction of the cable and synchronously move along the axial direction of the cable, so that terahertz pulse light beams scan the cable insulation layer point by point in a spiral track until the whole section of cable is covered; S130, recording transmitted or reflected terahertz pulse time-domain waveforms at each scanning point, and separating out main pulse signals representing the body characteristics of the insulating layer; S140, carrying out Fourier transformation on the main pulse signal of each scanning point to obtain a frequency domain spectrum of the corresponding scanning point, and calculating a complex dielectric constant real part epsilon '(omega) and an imaginary part epsilon' (omega) of the corresponding scanning point under different frequencies to form a dielectric spectrum of the corresponding scanning point; And S150, storing the dielectric spectrum of each scanning point in association with space coordinates, wherein the space coordinates comprise axial positions, radial depths and circumferential angles, so as to construct the four-dimensional database.
3. The method for detecting insulation of an ultra-high voltage cable according to claim 2, wherein the step S200 specifically comprises: s210, synchronously collecting a predefined key process parameter set from a production control system in real time from the beginning to the end of production of a cable section to be evaluated; S220, arranging and integrating the key process parameter sets according to the production time sequence, and associating the key process parameter sets with the unique production identifier of the cable section to generate a structured process parameter digital model; s230, calculating the space axial position of the insulating layer corresponding to each time point of the cable section on the production line according to the traction speed history record of the cable section; and S240, establishing a mapping relation between a parameter set of each time point in the process parameter digital model and all scanning points in the corresponding axial positions in the four-dimensional database constructed in the step S150 according to the corresponding relation obtained by calculation in the step S230, completing the digital association, and generating associated process parameters.
4. The method for detecting insulation of an ultra-high voltage cable according to claim 3, wherein the step S300 specifically comprises: S310, establishing an atomic-level molecular dynamics initial model of crosslinked polyethylene, and setting density, crosslinking degree and chain length distribution which are consistent with the cable insulating material by using a COMPASS or similar force field; S320, in the initial model, different preset process defects are introduced by changing simulation conditions, and a plurality of micro defect structure models are constructed; the preset process defects comprise insufficient local crosslinking degree, embedding of specific impurity atoms or clusters and gradient distribution of crystallinity caused by a non-uniform temperature field; S330, applying constant electric stress and thermal stress corresponding to actual operation conditions of the cable to each micro-defect structure model in a simulation environment, performing long-term molecular dynamics simulation, and recording the evolution process of the microstructure of the micro-defect structure model along with simulation time; S340, extracting key mechanism characteristic parameters representing defect dynamic behaviors from the evolution process, wherein the key mechanism characteristic parameters comprise the size and distribution change rate of free volume holes, anisotropic parameters of molecular chain segment mobility and the evolution trend of trap charge density and energy distribution; S350, carrying out association storage on the initial condition of each simulation case, the corresponding micro-defect structure model and the key mechanism characteristic parameters evolved by the micro-defect structure model, and constructing the mechanistic knowledge base.
5. The method for detecting insulation of an ultra-high voltage cable according to claim 4, wherein the step S400 specifically comprises: S410, constructing a mechanism-data hybrid driving artificial intelligent model with a multi-mode input branch, wherein the mechanism-data hybrid driving artificial intelligent model comprises: A first input branch for receiving normalized dielectric spectral data from a specific spatial point of S100; A second input branch for receiving a vectorized complete set of key process parameters associated with the spatial point from S200; a third input branch for receiving key mechanism characteristic parameters from S300 that best match the set of key process parameters, the key mechanism characteristic parameters being embedded as physical constraints; S420, preparing a training data set, wherein the data set takes S100, S200 and S300 data of a historical production cable as input characteristics, and takes actual defect types, positions and performance degradation data obtained in an accelerated aging experiment or long-term operation of a corresponding cable section as a supervision tag; S430, training the mechanism-data hybrid drive artificial intelligent model, wherein the total loss function of the mechanism-data hybrid drive artificial intelligent model consists of three parts, namely main task loss for defect classification and positioning, regression task loss for long-term performance prediction, and mechanism consistency regularization loss of a special punishment mechanism-data hybrid drive artificial intelligent model prediction result against a physical rule revealed by the key mechanism characteristic parameter of S300; S440, after training is completed, curing mechanism-data hybrid driving artificial intelligent model parameters, so that the mechanism-data hybrid driving artificial intelligent model has the functions of synchronously outputting microscopic defect classification and grade corresponding to the current space point, and performance degradation curve and residual electric life predicted value under the preset running condition of the point according to the input multi-mode data.
6. The method for detecting insulation of an ultra-high voltage cable according to claim 5, wherein the step S500 specifically comprises: S510, synchronously inputting four-dimensional database data acquired in S100 and associated technological parameters of space-time association acquired and completed in S200 of a cable to be tested into the mechanism-data hybrid driving artificial intelligent model subjected to S400 training and solidification; S520, traversing all scanning points by the mechanism-data hybrid driving artificial intelligent model according to a space coordinate sequence, and outputting microscopic defect classification and grade labels of each scanning point in parallel and a corresponding predicted residual life value array, wherein the predicted residual life value array represents failure probability under different operation years; S530, generating a three-dimensional defect type distribution cloud picture and a defect grade distribution cloud picture of the cable insulation layer by adopting a three-dimensional interpolation algorithm based on microscopic defect classification and grade labels of all scanning points; S540, based on the predicted remaining life value arrays of all the scanning points, extracting the failure probability under the same operation period threshold value, generating a three-dimensional thermodynamic diagram of the failure probability of the cable insulating layer at nodes with different service times; S550, carrying out comprehensive quality grading on the cable according to a preset quality grading rule by combining the three-dimensional defect distribution cloud chart and the failure probability thermodynamic diagram, wherein the quality grading rule is used for dividing the cable into at least four grades of an excellent grade product, a qualified product, a reworked product and a disqualified product at least according to the overall defect density and the minimum predicted residual life value; S560, automatically generating a visual quality assessment report containing the three-dimensional distribution cloud chart, the thermodynamic diagram, the quality grade, the specific defect position coordinate list and the maintenance advice.
7. The method for detecting insulation of an ultra-high voltage cable according to claim 6, wherein the step S600 specifically comprises: s610, analyzing the visual quality assessment report generated in the step S500, and extracting the spatial position information, the type and the grade of all the defects marked in the visual quality assessment report; s620, mapping each defect position back to the corresponding key process parameters which are effective when the position insulating layer is produced according to the mapping relation established in S200; S630, comparing the key process parameter set of all the defect points with the key process parameter set of the defect-free or superior level points by adopting statistical correlation analysis or machine learning attribution algorithm, and identifying at least one key process parameter and deviation direction thereof which are obviously related to the specific defect type and/or the low predicted life; s640, based on the identification result of S630, generating a process optimization suggestion list comprising specific parameter adjustment quantity, adjustment priority and adjusted expected quality improvement degree according to each key process parameter and combining the physical allowable range and corresponding defect evolution knowledge in the S300 mechanistic knowledge base; and S650, pushing the process optimization suggestion list to a corresponding execution unit of a production control system in real time in the form of a structured instruction, and guiding the closed-loop control of the production process of the next batch or the current online cable.
8. The method according to claim 5, wherein in step S410, the key mechanism characteristic parameters received by the third input branch are structurally fused into the forward propagation of the hybrid neural network model by at least one of: The key mechanism characteristic parameters are input into a special auxiliary decoder network, and the auxiliary decoder network tries to reconstruct the key mechanism characteristic parameters from the mechanism-data hybrid driving artificial intelligent model intermediate layer hidden characteristics, so that a main network is forced to learn the characteristic representation consistent with the physical rule; The key mechanism characteristic parameters are used for constructing a priori physical rule sub-network, and the output of the physical rule sub-network and the middle characteristics of the main network are subjected to weighted fusion or gating so as to selectively strengthen signal paths conforming to the physical rule in different reasoning stages; And thirdly, the key mechanism characteristic parameters are used as a group of ideal characteristic vectors for initializing or applying bias to a part of weight matrix of a mechanism-data hybrid driving artificial intelligent model core layer, so that the mechanism-data hybrid driving artificial intelligent model has reasoning tendency conforming to a physical rule at the beginning of training.
9. The method according to claim 5, wherein in the step S410, the feature vector Fd obtained by encoding the data of the first and second input branches is fused with the key mechanism feature parameter vector Fm from the third input branch by a mechanism guiding fusion module to generate a hybrid feature vector F hybrid ; The mechanism directs the fusion module to: S411, inputting the key mechanism characteristic parameter vector F m into a parameterized network to generate a group of attention weight vectors W a or a characteristic transformation matrix M m ; S412, carrying out weighted screening or projective transformation on the characteristic vector F d by utilizing the W a or the M m so that characteristic dimensions related to the current physical mechanism in the output characteristic F hybrid are enhanced; the total loss function L total in step S430 is specifically: L total =α(λL cls +(1 λ)L reg_life )+β Lmech +γL reg The method comprises the steps of obtaining a model prediction model, wherein L cls is cross entropy loss of a defect classification task, L reg_life is smooth L1 loss of a life prediction task, L mech is mechanism consistency regularization loss, calculating deviation between an intermediate or final result depending on the model prediction and a physical relationship implied by a mechanism characteristic parameter Fm, L reg is weight regularization loss, and alpha, beta, gamma and lambda are preset positive weight coefficients.
10. An ultra-high voltage cable maintenance method, characterized in that the cable segment is maintained after the current cable segment is detected by the ultra-high voltage cable insulation detection method according to any one of claims 1 to 9.

Description

Ultrahigh voltage cable insulation detection method and ultrahigh voltage cable maintenance method Technical Field The invention relates to the technical field of manufacturing of crosslinked polyethylene cables, in particular to an ultrahigh voltage cable insulation detection method and an ultrahigh voltage cable maintenance method. Background Crosslinked polyethylene (XLPE) cables have become a core equipment in the field of ultra-high voltage transmission due to their excellent electrical and mechanical properties. The long-term operational reliability of a cable is fundamentally dependent on the quality of its insulation, so that after production is completed, the insulation must be subjected to strict quality checks. At present, an insulation detection method commonly adopted in the industry is mainly based on electrical performance test, such as partial discharge detection, power frequency withstand voltage test or ultra-low frequency dielectric loss test. These methods play an important role in engineering by applying high voltages to evaluate the electrical strength of insulation or to detect macroscopic defects. However, these conventional methods have a fundamental limitation in that they are essentially a "post-hoc" based "destructive stress" pass-through test. The detection signal reflects the transient behavior of the insulation at the test voltage or the response to macroscopic defects (e.g., air gaps, impurities) that have progressed to a certain scale. They cannot nondestructively and directly sense the microstructure states (such as molecular chain crosslinking uniformity, micron-sized micropore distribution, early impurity dispersion and the like) which determine the long-term aging performance of the insulation, and further cannot evaluate the evolution trend of the microstructures under the combined action of an electric field and a thermal field for decades. Therefore, one of the outstanding technical problems faced in the prior art is how to detect the existing microscopic insulation defects and quantitatively evaluate and predict the long-term reliability (or residual electric life) of the cable under the ultra-high voltage operation condition under the conditions that the cable is not destroyed and can be scanned comprehensively after the cable is produced. Disclosure of Invention The invention aims to overcome the defects of the prior art and provides a technical scheme for solving the problems. In order to achieve the above purpose, the present invention provides the following technical solutions: the insulation detection method of the ultra-high voltage cable comprises the following steps: s100, carrying out in-situ non-contact terahertz time-domain spectrum three-dimensional full-field scanning on the produced cable, obtaining time-domain waveforms of all space points of a cable insulating layer in the cable, and calculating and generating a four-dimensional database containing space coordinates and corresponding dielectric spectrums; S200, synchronously acquiring key process parameters of the whole cable production process, digitally associating the key process parameters with the corresponding cable segment space coordinates in the S100, and generating associated process parameters; s300, constructing a mechanistic knowledge base of a cross-linked polyethylene micro defect structure under different process deviations and an evolution rule under electric-thermal stress based on molecular dynamics simulation, and extracting key mechanism characteristic parameters; s400, constructing a mechanism-data hybrid driving artificial intelligent model; Taking the dielectric spectrum of the S100, the associated technological parameters of the S200 and the key mechanism characteristic parameters matched from the S300 as mixed input, and taking the microscopic defect type, the severity level and the long-term performance prediction index as output, training a mechanism-data mixed driving artificial intelligent model; S500, inputting the dielectric spectrum and related process parameters of the cable to be tested into a trained mechanism-data hybrid driving artificial intelligent model to obtain three-dimensional defect distribution and long-term performance prediction results of the cable insulating layer, and generating a visual quality assessment report according to the three-dimensional defect distribution and long-term performance prediction results; And S600, generating a production process optimization feedback suggestion according to the correlation analysis of the three-dimensional defect distribution and the correlation process parameters in the visual quality evaluation report. As a further scheme of the invention, the step S100 specifically comprises the following steps: s110, enabling the produced cable to axially pass through an annular terahertz time-domain spectrum scanning device at a constant speed; s120, controlling a transmitting and receiving unit of the terahertz time-do