Multi-parameter nondestructive testing method and system for quality of low-voltage cable joint in electric power engineering

Abstract

The invention discloses a multi-parameter nondestructive testing method and a multi-parameter nondestructive testing system for the quality of a low-voltage cable joint in electric power engineering. The multi-parameter nondestructive testing method comprises the following steps: arranging a multi-modal sensor, acquiring multi-domain fusion data, and preprocessing to generate a multi-source time sequence data set; executing phase extraction, constructing four types of phase trajectories, and generating a parameter phase space deep fingerprint; perturbation excitation is applied under loop operation, transient changes are collected, and multi-parameter transient data are generated; performing transition event identification, extracting four types of jump features, and constructing an energy domain response feature set; combining the three-domain features, constructing a coupling manifold embedded structure, and generating high-dimensional tensor representation; and executing tensor contraction, outputting a state vector, and generating a joint quality detection result. According to the method, by constructing multi-parameter deep fingerprints and introducing perturbation excitation analysis and multi-domain coupling manifold modeling, high-sensitivity nondestructive testing and accurate state evaluation of early hidden defects of the low-voltage cable joint are achieved.

Description

Technical Field

[0001] This invention relates to the field of power engineering testing technology, and in particular to a multi-parameter non-destructive testing method and system for the quality of low-voltage cable joints in power engineering. Background Technology

[0002] Current condition monitoring of low-voltage cable joints primarily relies on single-parameter detection methods, such as temperature rise detection, current fluctuation detection, partial discharge detection, or magnetic field leakage detection. These methods largely depend on changes in a single physical parameter, judging the joint's condition through abnormal temperature, increased resistance, enhanced current harmonics, or the appearance of partial discharge signals. However, low-voltage cable joints often do not exhibit obvious single-parameter anomalies in the early, latent degradation stages. Minor contact defects, microcrack formation, or changes in the oxidation interface all exhibit weak responses, making it difficult for traditional single-parameter detection to capture the key characteristics of early defects. Existing detection methods typically employ fixed thresholds or static judgment approaches, which are ill-suited to the differentiated condition characteristics of individual joints, leading to early anomalies often being overlooked.

[0003] In recent years, multi-parameter detection methods have been gradually introduced into cable condition monitoring, improving diagnostic reliability by simultaneously acquiring temperature rise, current, magnetic field, and acoustic information. However, existing multi-parameter technologies still mainly rely on parameter amplitudes, trend changes, or simple feature splicing, lacking in-depth exploration of the inherent coupling relationships between multiple parameters and failing to reveal deep anomaly patterns using the correlation characteristics between signals from multiple physical domains. Furthermore, existing technologies primarily employ passive monitoring modes during operation, making it difficult to actively stimulate metastable joint defects, thus hindering the identification of latent problems. Traditional data fusion methods are mostly linear superpositions or shallow models, unable to construct a high-dimensional state feature space that reflects the complex mechanisms within the joint.

[0004] Existing technologies generally suffer from problems such as coarse detection granularity, insufficient sensitivity, and inability to accurately capture multi-parameter correlation anomalies. In particular, they lack the ability to effectively locate early latent defects, which often leads to joint degradation being discovered only after it has developed to a significant stage.

[0005] Therefore, how to provide a multi-parameter non-destructive testing method and system for the quality of low-voltage cable joints in power engineering is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] One objective of this invention is to propose a multi-parameter non-destructive testing method and system for low-voltage cable joint quality in power engineering. This invention utilizes multi-modal sensing detection, deep phase space fingerprint construction, perturbation excitation transient analysis, and multi-domain coupled manifold embedding technology to perform full-process acquisition, phase correlation analysis, transient transition identification, and multi-domain fusion modeling of the cable joint's operating state, achieving a comprehensive evaluation of the thermal, electrical, magnetic, and acoustic characteristics of the low-voltage cable joint. By constructing a parametric phase space deep fingerprint, this invention can characterize the stable modes of the joint's multi-parameter operating characteristics; by introducing small-amplitude multi-frequency current perturbations and analyzing transient transition behavior, this invention can actively excite the weak response characteristics of early latent defects; by establishing a multi-domain coupled manifold embedding structure, this invention achieves deep fusion and state-space mapping of topological, energy, and physical domain characteristics. Compared with existing technologies, this invention has advantages such as high sensitivity, strong multi-domain sensing capability, ability to identify early latent defects, and strong adaptability, providing a more reliable technical means to improve the operational safety of low-voltage cable joints.

[0007] A multi-parameter non-destructive testing method for the quality of low-voltage cable joints in power engineering, according to an embodiment of the present invention, includes:

[0008] Multimodal sensors are deployed at low-voltage cable joints to collect multi-domain fusion data. The multi-domain fusion data is preprocessed to generate multi-source time-series datasets.

[0009] Phase extraction is performed on multi-source time series datasets to construct temperature rise phase trajectories, current phase trajectories, magnetic field phase trajectories, and acoustic phase trajectories, forming a set of multi-parameter phase trajectories, and constructing a deep spatial fingerprint of the parametric phase of low-voltage cable joints;

[0010] Under the condition that the circuit to which the connector belongs remains in operation, a small-amplitude multi-frequency current perturbation excitation is applied to the circuit. During the perturbation excitation, transient change data of temperature rise parameter, electrical parameter, magnetic field parameter and acoustic parameter are collected to generate multi-parameter transient data.

[0011] Transition event identification is performed on multi-parameter transient data, and transition features are extracted from transient changes in temperature rise, current, magnetic field and acoustic modes to construct energy domain response features;

[0012] By combining deep fingerprints of the parametric phase space, energy domain response features, and physical domain features, a three-domain multi-parameter input of topological domain features, energy domain features, and physical domain features is constructed, and a multi-domain coupled manifold embedding structure is built to generate a high-dimensional tensor representation.

[0013] Tensor contraction operation is performed on the high-dimensional tensor representation to output the state vector. Based on the state vector, the quality level, defect type and degradation trend of the low-voltage cable joint are determined, and the detection results of the low-voltage cable joint are generated as the operation monitoring output.

[0014] Optionally, the multimodal sensor includes a temperature rise sensor, a current sensor, a magnetic field sensor, and an ultrasonic sensor.

[0015] Optionally, the multi-domain fusion data includes temperature rise data, current data, magnetic field data, and acoustic data used to characterize the operating status of low-voltage cable joints. The preprocessing of the multi-domain fusion data includes performing time synchronization, noise filtering, and normalization on the temperature rise data, current data, magnetic field data, and acoustic data.

[0016] Optionally, the process of forming a set of multi-parameter phase trajectories to construct a deep spatial fingerprint of the parametric phase of the low-voltage cable joint includes:

[0017] Temperature rise time series data, current time series data, magnetic field time series data and acoustic time series data corresponding to the same low-voltage cable joint are extracted from the multi-source time series dataset, maintaining a one-to-one correspondence between the various types of time series data in the sampling time;

[0018] Phase extraction processing was performed on the temperature rise time series data, current time series data, magnetic field time series data, and acoustic time series data respectively to obtain the time-varying temperature rise phase sequence, current phase sequence, magnetic field phase sequence, and acoustic phase sequence.

[0019] At each sampling moment, the corresponding temperature rise phase value, current phase value, magnetic field phase value and acoustic phase value are combined to form a multi-parameter phase feature vector. All multi-parameter phase feature vectors are arranged according to the sampling time order to construct a set of multi-parameter phase trajectories covering the detection time process.

[0020] The value ranges of the temperature rise phase sequence, current phase sequence, magnetic field phase sequence and acoustic phase sequence are divided into several non-overlapping phase intervals according to preset rules. The multi-parameter phase feature vector is mapped to the multi-dimensional phase region. The number of times the multi-parameter phase trajectory stays in each multi-dimensional phase region and the transfer relationship between different multi-dimensional phase regions are recorded to form the multi-dimensional phase region access distribution and multi-dimensional phase region transfer sequence.

[0021] Based on the multidimensional phase region access distribution and multidimensional phase region transfer sequence, feature indicators are extracted to characterize the stability, phase coupling mode and phase evolution path of low-voltage cable joints in multi-parameter phase space. The feature indicators are combined into a fixed-length fingerprint description vector in a preset order, and the fingerprint description vector is used as the deep fingerprint of the parametric phase space of the low-voltage cable joint.

[0022] Optionally, the generation of multi-parameter transient data includes:

[0023] When the circuit to which the low-voltage cable joint belongs is in normal operation, obtain the rated current parameters and current operating current parameters of the circuit, and determine the upper limit of the current disturbance amplitude and the set of current disturbance frequencies at multiple frequency points;

[0024] Without changing the normal operating conditions of the circuit, a small-amplitude current perturbation excitation signal composed of multiple frequency components is injected into the circuit according to the preset upper limit of current perturbation amplitude and the set of multiple frequency points of current perturbation frequency, so as to keep the circuit running continuously within the preset perturbation excitation duration.

[0025] Within a preset time window before, during, and after the injection of the perturbation excitation signal, the temperature rise parameters, electrical parameters, magnetic field parameters, and acoustic parameters at the low-voltage cable joint are synchronously collected at a transient sampling frequency higher than the sampling frequency of the multi-source time-series dataset to form transient sampling data.

[0026] The collected transient data on temperature rise, electrical transient data, magnetic field transient data, and acoustic transient data are truncated and aligned according to the perturbation excitation trigger time, and transient change data segments within the corresponding perturbation excitation time period are extracted.

[0027] The transient data segments of temperature rise, electrical transient data, magnetic field transient data, and acoustic transient data are combined in chronological order to form multi-parameter transient data.

[0028] Optionally, the construction of the energy domain response characteristics includes:

[0029] Based on the perturbation excitation trigger time, a transient analysis time window including the pre-trigger period, the trigger period, and the post-trigger period is determined. Temperature rise transient data, current transient data, magnetic field transient data, and acoustic transient data are acquired within the time window, maintaining a one-to-one correspondence between the four types of transient data at the sampling time.

[0030] For the four types of transient data, the difference between adjacent sampling points is calculated and the difference is calculated again to obtain the derived sequence. The amplitude is normalized and the baseline is calibrated based on the statistics of the period before the trigger. Fixed-length median filtering and short window smoothing are then performed on the derived sequence.

[0031] Candidate transition event detection is performed on the calibrated and smoothed derived sequence. An amplitude threshold, a minimum number of sustained samples, and start-end identification rules are set. Intervals that satisfy the amplitude threshold and the minimum number of sustained samples are marked as single-domain candidate transition events.

[0032] Perform multi-domain consistency judgment on single-domain candidate transition events, set multi-domain consistency time window and cross-domain phase order constraint, require candidate transition events of temperature rise parameter and electrical parameter to exist at the same time window and satisfy the preset order constraint, mark the event that satisfies the multi-domain consistency judgment as confirmed transition event, and generate event attribute set for each confirmed transition event.

[0033] Transition mode vectors are generated by encoding confirmed transition events in a fixed order. The transition mode vectors consist of parameter type combination encoding, time overlap level encoding, peak amplitude level encoding, continuous sample number level encoding, and sequence identifier encoding. Corresponding transition mode vectors are generated for four types of parameters: temperature rise, current, magnetic field, and acoustics. The four types of transition mode vectors are connected in order to form energy domain response characteristics.

[0034] Optionally, the construction of the multi-domain coupled manifold embedding structure to generate a high-dimensional tensor representation includes:

[0035] The deep fingerprint of the parametric phase space, energy domain response features, and physical domain features are obtained and combined according to the preset field order to form a three-domain multi-parameter input sample set containing topological domain features, energy domain features, and physical domain features.

[0036] Based on the three-domain multi-parameter input sample set, a multi-domain coupled manifold embedding structure is constructed. The multi-domain coupled manifold embedding structure includes intra-domain manifold construction units, cross-domain manifold alignment units, and coupled manifold generation units. The intra-domain manifold construction units process the topological domain features, energy domain features, and physical domain features respectively, determine the neighborhood relationship between the feature samples of each domain, and construct the corresponding intra-domain manifold subspace.

[0037] By using the cross-domain manifold alignment unit, samples that correspond one-to-one with the time index or operating condition are selected from the three-domain multi-parameter input sample set as cross-domain anchor points. Based on the distribution relationship of the cross-domain anchor points in the manifold subspaces of each domain, scale normalization, coordinate translation and local neighborhood structure adjustment are performed on the topological domain manifold subspace, energy domain manifold subspace and physical domain manifold subspace.

[0038] By using coupled manifold generation units, cross-domain connections are established between the aligned topological domain manifold subspace, energy domain manifold subspace, and physical domain manifold subspace. The neighborhood connections and cross-domain connections of the manifold subspaces within each domain are jointly modeled to form a joint manifold structure containing information from the three types of manifold subspaces and cross-domain coupling information. The joint manifold structure is used as the core representation of the multi-domain coupled manifold embedding structure.

[0039] Based on the multi-domain coupled manifold embedding structure, the embedding coordinates of the three-domain multi-parameter input sample set are solved. The embedding coordinates of each sample in the joint manifold structure are arranged and organized according to a preset dimensional order to generate a high-dimensional tensor representation.

[0040] Optionally, the detection results of the generated low-voltage cable joints are used as operational monitoring outputs, including:

[0041] The high-dimensional tensor representation is indexed and organized according to the joint sample dimension, time dimension, domain dimension and feature dimension to form a high-dimensional tensor sample set with a single low-voltage cable joint as the basic unit.

[0042] For each low-voltage cable joint in the high-dimensional tensor sample set, based on the corresponding energy domain response characteristics and deep fingerprint of the parametric phase space, the shrinkage priority and shrinkage weight of the topological domain, energy domain and physical domain in the tensor shrinkage operation are determined;

[0043] In the time dimension, the first stage of tensor contraction operation is performed on the high-dimensional tensor representation. According to the preset sliding time window and anomaly sensitive window length, the tensor elements in each time window are aggregated and calculated to compress the high-dimensional tensor representation of the same low-voltage cable joint in the entire detection time process into a time-contracted tensor representation that reflects the time evolution characteristics.

[0044] Combining the shrinkage priority and shrinkage weight, the second-stage tensor shrinkage operation is performed on the domain dimension and feature dimension. The tensor components related to the topological domain, the tensor components related to the energy domain, and the tensor components related to the physical domain are grouped, aggregated, and weighted to generate a state vector.

[0045] Each component in the state vector is compared with the preset set of quality level judgment thresholds, defect type judgment rules, and degradation trend judgment rules to determine the corresponding quality level identifier, defect type identifier, and degradation trend identifier. The state vector, quality level identifier, defect type identifier, and degradation trend identifier are then combined according to the joint number and detection time to generate a low-voltage cable joint detection result record as the operation monitoring output.

[0046] A multi-parameter non-destructive testing system for the quality of low-voltage cable joints in power engineering, according to an embodiment of the present invention, includes the following modules:

[0047] The multi-domain data processing module is used to deploy multimodal sensors, collect multi-domain fusion data, perform preprocessing, and generate multi-source time-series datasets.

[0048] The phase fingerprint construction module is used to extract phase from multi-source time series datasets, generate a set of multi-parameter phase trajectories, and construct a deep spatial fingerprint of the parametric phase.

[0049] The perturbation transient acquisition module is used to apply small-amplitude multi-frequency current perturbation excitation and acquire transient change data of various physical parameters to generate multi-parameter transient data.

[0050] The transition feature extraction module is used to identify transition events in multi-parameter transient data, extract the transition features of each parameter, and form energy domain response features.

[0051] The multi-domain manifold embedding module is used to combine deep fingerprints of parametric phase space, energy domain response features and physical domain features to generate multi-domain coupled manifold embedding structures.

[0052] The state output module is used to generate state vectors based on the multi-domain coupled manifold embedding structure and output the quality level, defect type and degradation trend.

[0053] The beneficial effects of this invention are:

[0054] This invention constructs a deep fingerprint of parametric phase space, enabling a refined characterization of the multi-parameter operating behavior of low-voltage cable joints, significantly expanding the traditional detection methods that rely on single-parameter amplitude changes. Compared to the limitations of existing technologies that can only respond to overt anomalies, this invention unifies the phase evolution relationships of multiple physical domains such as temperature rise, current, magnetic field, and acoustics into a single fingerprint structure, thereby capturing weak, implicit, and evolving phase coupling characteristics during joint operation. This deep fingerprint characterization method effectively improves the stable identification capability under individual joint differences, allowing for the accurate recording of subtle phase shifts that occur during mild contact defects, the initial stage of interface oxidation, and the formation of local microcracks, providing a reliable foundation for subsequent dynamic analysis.

[0055] This invention applies small-amplitude, multi-frequency current perturbation excitation under normal operating conditions, combined with transient transition event recognition technology. This actively excites and amplifies latent defects, which are imperceptible by traditional passive monitoring, into observable features. By analyzing the transient changes in four parameters—temperature rise, current, magnetic field, and acoustics—this invention can effectively identify subtle structural changes occurring at the material interface in the metastable region, thus exposing the sensitive characteristics of the joint in the early stages of degradation. This active excitation-based detection method significantly enhances the system's response to early-stage defects, enabling the invention to make judgments in the early stages of defect formation and effectively improving the foresight and reliability of the detection results.

[0056] This invention constructs a multi-domain coupled manifold embedding structure that organically integrates topological, energy, and physical domain features in a high-dimensional space, achieving deep coupling of multi-parameter information and overall state mapping. Compared to traditional linear fusion or shallow models, this invention can extract more representative comprehensive feature expressions from multi-domain interaction relationships, forming a more accurate representation of the overall operating state of the joint. Through analysis of the high-dimensional manifold representation, the state vector output by this invention can accurately reflect the joint quality level, defect type, and degradation trend, providing a clear basis for operation and maintenance decisions. The overall effect of this invention is to improve the early defect identification capability of low-voltage cable joints, enhance the accuracy of multi-parameter comprehensive judgment, and effectively improve the safety and stability of power system operation. Attached Figure Description

[0057] 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:

[0058] Figure 1 This is a flowchart of the multi-parameter non-destructive testing method for the quality of low-voltage cable joints in power engineering proposed in this invention.

[0059] Figure 2 This is a schematic diagram of the multi-parameter non-destructive testing system for the quality of low-voltage cable joints in power engineering proposed in this invention. Detailed Implementation

[0060] 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.

[0061] refer to Figure 1 Multi-parameter non-destructive testing methods for the quality of low-voltage cable joints in power engineering include:

[0062] Multimodal sensors are deployed at low-voltage cable joints to collect multi-domain fusion data. The multi-domain fusion data is preprocessed to generate multi-source time-series datasets.

[0063] Phase extraction is performed on multi-source time series datasets to construct temperature rise phase trajectories, current phase trajectories, magnetic field phase trajectories, and acoustic phase trajectories, forming a set of multi-parameter phase trajectories, and constructing a deep spatial fingerprint of the parametric phase of low-voltage cable joints;

[0064] Under the condition that the circuit to which the connector belongs remains in operation, a small-amplitude multi-frequency current perturbation excitation is applied to the circuit. During the perturbation excitation, transient change data of temperature rise parameter, electrical parameter, magnetic field parameter and acoustic parameter are collected to generate multi-parameter transient data.

[0065] Transition event identification is performed on multi-parameter transient data, and transition features are extracted from transient changes in temperature rise, current, magnetic field and acoustic modes to construct energy domain response features;

[0066] By combining deep fingerprints of the parametric phase space, energy domain response features, and physical domain features, a three-domain multi-parameter input of topological domain features, energy domain features, and physical domain features is constructed, and a multi-domain coupled manifold embedding structure is built to generate a high-dimensional tensor representation.

[0067] Tensor contraction operation is performed on the high-dimensional tensor representation to output the state vector. Based on the state vector, the quality level, defect type and degradation trend of the low-voltage cable joint are determined, and the detection results of the low-voltage cable joint are generated as the operation monitoring output.

[0068] In this embodiment, the multimodal sensor includes a temperature rise sensor, a current sensor, a magnetic field sensor, and an ultrasonic sensor.

[0069] In this embodiment, the multi-domain fusion data includes temperature rise data, current data, magnetic field data, and acoustic data used to characterize the operating status of low-voltage cable joints. The preprocessing of the multi-domain fusion data includes performing time synchronization, noise filtering, and normalization on the temperature rise data, current data, magnetic field data, and acoustic data.

[0070] In this embodiment, the step of forming a set of multi-parameter phase trajectories to construct a deep spatial fingerprint of the parametric phase of a low-voltage cable joint includes:

[0071] Temperature rise time series data, current time series data, magnetic field time series data and acoustic time series data corresponding to the same low-voltage cable joint are extracted from the multi-source time series dataset, maintaining a one-to-one correspondence between the various types of time series data in the sampling time;

[0072] Phase extraction processing is performed on the temperature rise time series data, current time series data, magnetic field time series data, and acoustic time series data respectively to obtain time-varying phase sequences of temperature rise, current, magnetic field, and acoustic data. Specifically, the phase extraction processing is performed on the temperature rise time series data, current time series data, magnetic field time series data, and acoustic time series data respectively as follows:

[0073] Phase extraction processing is performed on the temperature rise time series data. Based on the periodic or quasi-periodic variation characteristics of temperature rise fluctuations in the sampling time series, the continuous temperature rise data is divided into several adjacent temperature rise change segments. The positional relationship between the rising segment, the stable segment and the falling segment is extracted in each segment, and the temperature rise phase sequence is constructed based on the internal variation pattern of the segment.

[0074] Phase extraction processing is performed on the current time series data. Based on the continuous periodic variation law of the current waveform under the operating power grid frequency, the relative phase position of each sampling point is determined according to the zero crossing position, waveform peak position and waveform valley position, and a current phase sequence that changes continuously along time is generated.

[0075] Phase extraction processing is performed on magnetic field time series data and acoustic time series data. By identifying significant fluctuation segments, peak and trough positions and local pulsation changes that occur in the sampling time, each sampling point is mapped to its phase position in the magnetic field change mode or acoustic pulsation mode, generating magnetic field phase sequence and acoustic phase sequence respectively.

[0076] At each sampling moment, the corresponding temperature rise phase value, current phase value, magnetic field phase value and acoustic phase value are combined to form a multi-parameter phase feature vector. All multi-parameter phase feature vectors are arranged according to the sampling time order to construct a set of multi-parameter phase trajectories covering the detection time process.

[0077] The value ranges of the temperature rise phase sequence, current phase sequence, magnetic field phase sequence and acoustic phase sequence are divided into several non-overlapping phase intervals according to preset rules. The multi-parameter phase feature vector is mapped to the multi-dimensional phase region. The number of times the multi-parameter phase trajectory stays in each multi-dimensional phase region and the transfer relationship between different multi-dimensional phase regions are recorded to form the multi-dimensional phase region access distribution and multi-dimensional phase region transfer sequence.

[0078] Based on the multidimensional phase region access distribution and multidimensional phase region transfer sequence, feature indicators are extracted to characterize the stability, phase coupling mode, and phase evolution path of low-voltage cable joints in the multi-parameter phase space. The feature indicators are combined into a fixed-length fingerprint description vector in a preset order. The fingerprint description vector is used as the deep fingerprint of the low-voltage cable joint in the parametric phase space. The preset order is as follows: multidimensional phase region access frequency feature, adjacent multidimensional phase region transfer probability feature, multi-parameter phase coupling strength feature, multi-parameter phase evolution direction feature, and multi-parameter phase change stability feature.

[0079] In this embodiment, the generation of multi-parameter transient data includes:

[0080] When the circuit to which the low-voltage cable joint belongs is in normal operation, obtain the rated current parameters and current operating current parameters of the circuit, and determine the upper limit of the current disturbance amplitude and the set of current disturbance frequencies at multiple frequency points;

[0081] Without altering the normal operating conditions of the circuit, a small-amplitude current perturbation excitation signal composed of multiple frequency components is injected into the circuit according to the preset upper limit of the current perturbation amplitude and the set of multiple frequency points for the current perturbation. The circuit continues to operate continuously for the preset duration of the perturbation excitation. Specifically:

[0082] The preset upper limit for current disturbance amplitude is 3%–5% of the rated current;

[0083] The preset duration of the perturbation excitation is 0.3 seconds to 1.0 seconds;

[0084] Within a preset time window before, during, and after the injection of the perturbation excitation signal, the temperature rise parameter, electrical parameter, magnetic field parameter, and acoustic parameter at the low-voltage cable joint are synchronously collected at a transient sampling frequency higher than the sampling frequency of the multi-source time series dataset to form transient sampling data. The transient sampling frequency higher than the sampling frequency of the multi-source time series dataset means that the sampling frequency of the multi-source time series dataset is 100Hz to 1kHz, and the transient sampling frequency is 5 to 10 times the sampling frequency of the multi-source time series dataset. The preset time window is from 50 milliseconds before the perturbation excitation trigger moment to 150 milliseconds after the perturbation excitation ends.

[0085] The collected transient data on temperature rise, electrical transient data, magnetic field transient data, and acoustic transient data are truncated and aligned according to the perturbation excitation trigger time, and transient change data segments within the corresponding perturbation excitation time period are extracted.

[0086] The transient data segments of temperature rise, electrical transient data, magnetic field transient data, and acoustic transient data are combined in chronological order to form multi-parameter transient data.

[0087] In this embodiment, constructing the energy domain response characteristics includes:

[0088] Based on the perturbation excitation trigger time, a transient analysis time window including the pre-trigger period, the trigger period, and the post-trigger period is determined. Temperature rise transient data, current transient data, magnetic field transient data, and acoustic transient data are acquired within the time window, maintaining a one-to-one correspondence between the four types of transient data at the sampling time.

[0089] For each of the four types of transient data, the difference between adjacent sampling points and the difference between the two points are calculated again to obtain the derived sequence. Based on the statistics of the period before the trigger, the derived sequence is subjected to amplitude normalization and baseline calibration. Fixed-length median filtering and short-window smoothing are then applied to the derived sequence. Specifically, the calculation of the difference between adjacent sampling points and the difference between the two points are performed on each of the four types of transient data.

[0090] The transient temperature rise data is processed by interpolating adjacent sampling points to calculate the change between adjacent sampling points. Based on this, the continuous change is interpolated again to obtain the derived sequence that reflects the rate and degree of transient temperature rise.

[0091] The transient current data is processed by interpolation between adjacent sampling points to extract the current change between every two adjacent sampling points. The interpolation is then performed again on the sequence of changes to form an derived sequence that reflects the strength of the transient current change and the characteristics of the change acceleration.

[0092] The magnetic field transient data and acoustic transient data are processed by the difference between adjacent sampling points and the difference is recalculated to transform their respective continuous change patterns into derived sequences that highlight local rapid changes.

[0093] Candidate transition event detection is performed on the calibrated and smoothed derived sequence. An amplitude threshold, a minimum number of sustained samples, and start / end identification rules are set. Intervals satisfying the amplitude threshold and the minimum number of sustained samples are marked as single-domain candidate transition events. Specifically, the candidate transition event detection for the calibrated and smoothed derived sequence involves:

[0094] The amplitude changes of the temperature rise derived sequence after calibration and smoothing are checked point by point. When the amplitude in a certain continuous interval exceeds the preset amplitude threshold and remains no less than the preset minimum continuous sample number, the continuous interval is marked as a single-domain candidate transition event of temperature rise.

[0095] The calibrated and smoothed current-derived sequences are tested according to the same amplitude threshold and minimum number of sustained samples. Continuous intervals that meet the threshold and duration conditions are identified and marked as candidate current single-domain transition events.

[0096] Amplitude threshold detection and continuous sample number detection were performed on the calibrated and smoothed magnetic field derived sequence and acoustic derived sequence, respectively. The continuous intervals that met the detection conditions were marked as magnetic field single-domain candidate transition events and acoustic single-domain candidate transition events, respectively.

[0097] The start and end recognition rule is that in the exported sequence, when the amplitude first continuously exceeds the preset amplitude threshold, the sampling point is marked as the start point of the transition event, and when the amplitude continuously falls back to below the preset amplitude threshold and remains no less than the preset number of convergence samples, the sampling point is marked as the end point of the transition event.

[0098] Multi-domain consistency determination is performed on single-domain candidate transition events. A multi-domain consistency time window and cross-domain phase order constraints are set. Within the same time window, candidate transition events for temperature rise parameters and electrical parameters are required to exist simultaneously and satisfy the preset order constraints. Events that satisfy the multi-domain consistency determination are marked as confirmed transition events. An event attribute set is generated for each confirmed transition event. The preset order constraints require that within the same multi-domain consistency time window, the occurrence time of magnetic field candidate transition events is earlier than the occurrence time of current candidate transition events, the occurrence time of current candidate transition events is earlier than the occurrence time of temperature rise candidate transition events, and the occurrence time of temperature rise candidate transition events is earlier than the occurrence time of acoustic candidate transition events. The two types of physical parameters refer to:

[0099] Transition mode vectors are generated by encoding confirmed transition events in a fixed order. The transition mode vectors consist of parameter type combination encoding, time overlap level encoding, peak amplitude level encoding, continuous sample number level encoding, and sequence identifier encoding. Corresponding transition mode vectors are generated for four types of parameters: temperature rise, current, magnetic field, and acoustics. The four types of transition mode vectors are connected in order to form energy domain response characteristics.

[0100] In this embodiment, the construction of the multi-domain coupled manifold embedding structure to generate a high-dimensional tensor representation includes:

[0101] The deep fingerprint of the parameter phase space, the energy domain response features, and the physical domain features are obtained and combined according to the preset field order to form a three-domain multi-parameter input sample set containing topological domain features, energy domain features, and physical domain features. The preset field order is as follows: first, the topological domain feature fields corresponding to the deep fingerprint of the parameter phase space are arranged; second, the energy domain response feature fields are arranged; and finally, the physical domain feature fields are arranged.

[0102] Based on the three-domain multi-parameter input sample set, a multi-domain coupled manifold embedding structure is constructed. The multi-domain coupled manifold embedding structure includes intra-domain manifold construction units, cross-domain manifold alignment units, and coupled manifold generation units. The intra-domain manifold construction units process the topological domain features, energy domain features, and physical domain features respectively, determine the neighborhood relationship between the feature samples of each domain, and construct the corresponding intra-domain manifold subspace.

[0103] Through the cross-domain manifold alignment unit, samples that correspond one-to-one with the time index or operating condition are selected from the three-domain multi-parameter input sample set as cross-domain anchor points. Based on the distribution relationship of the cross-domain anchor points in the manifold subspaces of each domain, scale normalization, coordinate translation, and local neighborhood structure adjustment are performed on the topological domain manifold subspace, energy domain manifold subspace, and physical domain manifold subspace. Specifically, the scale normalization, coordinate translation, and local neighborhood structure adjustment are performed on the topological domain manifold subspace, energy domain manifold subspace, and physical domain manifold subspace.

[0104] Scale normalization is performed on the topological domain manifold subspace, energy domain manifold subspace, and physical domain manifold subspace respectively. By scaling the value range of cross-domain anchor point samples in each manifold subspace to a uniform numerical range, the three types of manifold subspaces are kept consistent under the same scale framework.

[0105] Coordinate translation is performed on the topological domain manifold subspace, the energy domain manifold subspace, and the physical domain manifold subspace respectively. By using the center position of the cross-domain anchor point in the three types of manifold subspaces, the overall coordinate distribution of each manifold subspace is simultaneously translated to a unified reference coordinate position.

[0106] Local neighborhood structure adjustment processing is performed on the topological domain manifold subspace, energy domain manifold subspace, and physical domain manifold subspace respectively. By rearranging the sample connection relationships in the neighborhood of cross-domain anchor points, the three manifold subspaces have a consistent neighborhood connection structure in the local neighborhood.

[0107] By using coupled manifold generation units, cross-domain connections are established between the aligned topological domain manifold subspace, energy domain manifold subspace, and physical domain manifold subspace. The neighborhood connections and cross-domain connections of the manifold subspaces within each domain are jointly modeled to form a joint manifold structure containing information from the three types of manifold subspaces and cross-domain coupling information. The joint manifold structure is used as the core representation of the multi-domain coupled manifold embedding structure.

[0108] Based on a multi-domain coupled manifold embedding structure, embedding coordinates are solved for a three-domain multi-parameter input sample set. The embedding coordinates of each sample in the joint manifold structure are arranged and organized according to a preset dimensional order to generate a high-dimensional tensor representation. Specifically, the embedding coordinates are solved for the three-domain multi-parameter input sample set.

[0109] For a set of three-domain multi-parameter input samples, a neighborhood search is performed in the embedded structure of a multi-domain coupled manifold. By determining the set of neighborhood samples of each sample in the joint manifold structure, neighborhood constraints are provided for solving the embedded coordinates.

[0110] Based on the neighborhood constraints, joint manifold coordinates are obtained for each sample. By maintaining the relative positional relationship between samples within the neighborhood, each sample is mapped to the embedding coordinate position in the multi-domain coupled manifold embedding structure.

[0111] The obtained embedded coordinates are arranged in a preset dimensional order, and the embedded coordinates of the same connector are organized in the order of samples to form a high-dimensional tensor representation.

[0112] The preset dimensional order is as follows: first, the topological domain embedded coordinate dimension is arranged; second, the energy domain embedded coordinate dimension is arranged; and finally, the physical domain embedded coordinate dimension is arranged.

[0113] In this embodiment, the detection result of generating the low-voltage cable joint is used as the operation monitoring output, including:

[0114] The high-dimensional tensor representation is indexed and organized according to the joint sample dimension, time dimension, domain dimension and feature dimension to form a high-dimensional tensor sample set with a single low-voltage cable joint as the basic unit.

[0115] For each low-voltage cable joint in the high-dimensional tensor sample set, based on the corresponding energy domain response characteristics and deep fingerprint of the parametric phase space, the shrinkage priority and shrinkage weight of the topological domain, energy domain and physical domain in the tensor shrinkage operation are determined;

[0116] The first stage of tensor contraction operation is performed on the high-dimensional tensor representation in the time dimension. According to the preset sliding time window and anomaly sensitive window length, the tensor elements in each time window are aggregated and calculated to compress the high-dimensional tensor representation of the same low-voltage cable joint in the entire detection time process into a time-contracted tensor representation that reflects the time evolution characteristics. The preset sliding time window length is 20 milliseconds to 50 milliseconds, and the preset anomaly sensitive window length is 5 milliseconds to 15 milliseconds.

[0117] Combining the shrinkage priority and shrinkage weight, the second-stage tensor shrinkage operation is performed on the domain dimension and feature dimension. The tensor components related to the topological domain, the tensor components related to the energy domain, and the tensor components related to the physical domain are grouped, aggregated, and weighted to generate a state vector.

[0118] Each component in the state vector is compared with the preset set of quality level judgment thresholds, defect type judgment rules, and degradation trend judgment rules to determine the corresponding quality level identifier, defect type identifier, and degradation trend identifier. The state vector, quality level identifier, defect type identifier, and degradation trend identifier are then combined according to the joint number and detection time to generate a low-voltage cable joint detection result record as the operation monitoring output.

[0119] refer to Figure 2 A multi-parameter non-destructive testing system for the quality of low-voltage cable joints in power engineering includes the following modules:

[0120] The multi-domain data processing module is used to deploy multimodal sensors, collect multi-domain fusion data, perform preprocessing, and generate multi-source time-series datasets.

[0121] The phase fingerprint construction module is used to extract phase from multi-source time series datasets, generate a set of multi-parameter phase trajectories, and construct a deep spatial fingerprint of the parametric phase.

[0122] The perturbation transient acquisition module is used to apply small-amplitude multi-frequency current perturbation excitation and acquire transient change data of various physical parameters to generate multi-parameter transient data.

[0123] The transition feature extraction module is used to identify transition events in multi-parameter transient data, extract the transition features of each parameter, and form energy domain response features.

[0124] The multi-domain manifold embedding module is used to combine deep fingerprints of parametric phase space, energy domain response features and physical domain features to generate multi-domain coupled manifold embedding structures.

[0125] The state output module is used to generate state vectors based on the multi-domain coupled manifold embedding structure and output the quality level, defect type and degradation trend.

[0126] Example 1:

[0127] To verify the feasibility of this invention in practice, it was applied to a 10kV distribution substation containing 47 low-voltage cable joints that had been in operation for 6 to 10 years. Some of these joints had exhibited slight signs of intermittent overheating and abnormal noises in the past two years, but these did not reach traditional alarm thresholds. This invention underwent 10 consecutive days of actual testing in this substation to verify its ability to identify early-stage latent defects in low-voltage cable joints, as well as its stability and multi-domain fusion capabilities under complex field conditions.

[0128] A 0.4kV branch line at the test site contained eight key monitoring joints. Three of these joints exhibited subtle anomalies such as short-term increases in operating temperature, weak acoustic pulsations during nighttime load fluctuations, and localized magnetic field disturbances. However, traditional methods could not confirm whether these were actual potential hazards. This invention utilizes multi-modal sensors to collect temperature rise data, current data, magnetic field data, and acoustic data, and processes this data using the multi-domain coupling analysis method of this invention.

[0129] In field applications, multimodal sensors are first installed on the outer surface and adjacent locations of each connector, including infrared temperature nodes, current transformers, Hall flux density sensors, and high-sensitivity ultrasonic probes. During operation, the system automatically collects temperature rise, current, magnetic field, and acoustic signals, and performs preprocessing to obtain stable multi-source time-series data. Subsequently, using the phase space deep fingerprint construction method of this invention, these four types of data are converted into phase trajectories, forming the phase space distribution, region access map, and region transfer sequence for each connector in the background system. Compared to traditional amplitude data, the deep fingerprint of this invention can characterize the unique "phase morphology" of the connector's operational behavior, thus accurately capturing potential offset trends even when the connector appears normal.

[0130] Under permissible load conditions, this invention applies small-amplitude, multi-frequency current perturbation excitation to eight connectors during low-load nighttime periods. The transient response behavior of each connector in the thermal, electrical, magnetic, and acoustic domains is observed through this multi-frequency perturbation. In the acquired transient responses, three connectors exhibited sudden, micro-amplitude jumps in different parameters, all with amplitudes smaller than traditional thresholds. However, this invention successfully captured these subtle changes through a transition event identification mechanism. System analysis revealed that these jumps exhibit synchronous characteristics across different physical domains, indicating the possibility of initial interface degradation or poor contact within the connectors. Such phenomena cannot be identified using traditional detection methods because traditional detection does not actively excite perturbations and cannot analyze cross-domain synchronous transitions.

[0131] After constructing the energy domain response features, this invention further inputs the deep phase fingerprint, transition event features, and physical domain features into the multi-domain coupled manifold embedding structure. After embedding, a high-dimensional tensor representation is formed, and a state vector is finally generated through tensor contraction analysis. By analyzing the quality level and defect type of the state vector, this invention identifies three joints as having a state level of "slight degradation" or "moderate degradation," with the most typical joint numbered LM-04.

[0132] To verify the advantages of this invention, the system simultaneously recorded the phase fingerprint changes and energy domain transition characteristics of each connector over 10 days. The results showed that the LM-04 connector exhibited decreased phase stability in three parameters: temperature rise, current, and magnetic field, indicating a difference in its operational behavior compared to the normal group. During perturbation excitation, the LM-04 connector experienced 11 cross-domain synchronous transition events, with a cross-domain consistency rate of 82.6%, significantly higher than the 5%–11% range of healthy connectors. In cases where traditional methods cannot provide a clear diagnosis, this invention can identify potential problems in advance. Actual disassembly and inspection by maintenance personnel confirmed the presence of mild oxidation and localized microcracks on the metal surface of the internal crimping area of ​​the LM-04 connector, completely consistent with the detection results of this invention.

[0133] Table 1. Statistical Table of Multi-domain Parameter Test Results for Low-voltage Cable Joints

[0134] Connector Number Average daily temperature rise (°C) Phase stability offset index (%) Number of perturbation transition events Cross-domain consistency (%) Status level determination On-site dismantling and inspection conclusions LM-01 3.2 1.8 0 4.5 normal normal LM-02 3.5 2.1 1 7.2 normal normal LM-03 4.8 4.3 3 10.1 normal normal LM-04 6.1 14.7 11 82.6 Moderate degradation Oxidation + Microcracks LM-05 3.4 2.4 2 6.8 normal normal LM-06 5.2 8.6 7 35.4 Mild degradation Minor poor contact LM-07 3.1 3.2 1 9.3 normal normal LM-08 4.7 7.9 5 22.8 Mild degradation Loose crimping of connector

[0135] As can be seen from the data in Table 1, this invention can accurately distinguish between healthy joints and joints with latent degradation in real power operation scenarios. The daily average temperature rise, phase shift, number of transition events, and cross-domain consistency of LM-01, LM-02, LM-03, LM-05, and LM-07 are all at low levels, indicating stable operation. No synchronous anomalies were observed among various multi-domain parameters. The final judgment results and on-site inspections were both normal. This invention accurately identifies healthy joints without generating false alarms, verifying the reliability of this method in normal joint scenarios.

[0136] The data show that the behavioral characteristics of LM-04, LM-06, and LM-08 differ significantly from those of healthy connectors. In particular, LM-04 exhibits significantly higher phase stability shift index, perturbation transition event frequency, and cross-domain consistency, reflecting anomalous synchronous changes with strong correlations across multiple physical domains. This invention combines deep phase fingerprinting with perturbation transient transition characteristics, enabling the identification of early weak signals that are difficult to capture using traditional methods. LM-04 was ultimately determined to be moderately degraded, and on-site disassembly and inspection confirmed the presence of oxidation and microcracks internally, perfectly consistent with the detection results. This demonstrates that this invention has extremely high sensitivity to structural defects and can detect latent defects much earlier than traditional monitoring methods.

[0137] Although LM-06 and LM-08 did not show significant abnormal temperature rise, their phase shift, number of transition events, and cross-domain consistency were significantly higher than the normal range. Based on this, the present invention classifies them as mild degradation. On-site inspection confirmed the presence of slight contact defects and loose crimping, respectively. These results demonstrate that the present invention achieves deep fusion analysis of the thermal, electrical, magnetic, and acoustic domains through a multi-domain coupled manifold embedding structure, revealing hidden degradation patterns that traditional single-parameter monitoring cannot detect. Overall, the present invention has significant advantages in early defect identification, anomaly relationship mining, and state level classification, providing higher reliability and forward-looking support for cable joint maintenance.

[0138] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A multi-parameter non-destructive testing method for the quality of low-voltage cable joints in power engineering, characterized in that, include: Multimodal sensors are deployed at low-voltage cable joints to collect multi-domain fusion data. The multi-domain fusion data is preprocessed to generate multi-source time-series datasets. Phase extraction is performed on multi-source time series datasets to construct temperature rise phase trajectories, current phase trajectories, magnetic field phase trajectories, and acoustic phase trajectories, forming a set of multi-parameter phase trajectories, and constructing a deep spatial fingerprint of the parametric phase of low-voltage cable joints; Under the condition that the circuit to which the connector belongs remains in operation, a small-amplitude multi-frequency current perturbation excitation is applied to the circuit. During the perturbation excitation, transient change data of temperature rise parameter, electrical parameter, magnetic field parameter and acoustic parameter are collected to generate multi-parameter transient data. Transition event identification is performed on multi-parameter transient data, and transition features are extracted from transient changes in temperature rise, current, magnetic field and acoustic modes to construct energy domain response features; By combining deep fingerprints of the parametric phase space, energy domain response features, and physical domain features, a three-domain multi-parameter input of topological domain features, energy domain features, and physical domain features is constructed, and a multi-domain coupled manifold embedding structure is built to generate a high-dimensional tensor representation. Tensor contraction operation is performed on the high-dimensional tensor representation to output the state vector. Based on the state vector, the quality level, defect type and degradation trend of the low-voltage cable joint are determined, and the detection results of the low-voltage cable joint are generated as the operation monitoring output.

2. The multi-parameter non-destructive testing method for the quality of low-voltage cable joints in power engineering according to claim 1, characterized in that, The multimodal sensor includes a temperature rise sensor, a current sensor, a magnetic field sensor, and an ultrasonic sensor.

3. The multi-parameter non-destructive testing method for the quality of low-voltage cable joints in power engineering according to claim 1, characterized in that, The multi-domain fusion data includes temperature rise data, current data, magnetic field data, and acoustic data used to characterize the operating status of low-voltage cable joints. The preprocessing of the multi-domain fusion data includes performing time synchronization, noise filtering, and normalization on the temperature rise data, current data, magnetic field data, and acoustic data.

4. The multi-parameter non-destructive testing method for the quality of low-voltage cable joints in power engineering according to claim 1, characterized in that, The process of forming a set of multi-parameter phase trajectories to construct a deep spatial fingerprint of the parametric phase of a low-voltage cable joint includes: Temperature rise time series data, current time series data, magnetic field time series data and acoustic time series data corresponding to the same low-voltage cable joint are extracted from the multi-source time series dataset, maintaining a one-to-one correspondence between the various types of time series data in the sampling time; Phase extraction processing was performed on the temperature rise time series data, current time series data, magnetic field time series data, and acoustic time series data respectively to obtain the time-varying temperature rise phase sequence, current phase sequence, magnetic field phase sequence, and acoustic phase sequence. At each sampling moment, the corresponding temperature rise phase value, current phase value, magnetic field phase value and acoustic phase value are combined to form a multi-parameter phase feature vector. All multi-parameter phase feature vectors are arranged according to the sampling time order to construct a set of multi-parameter phase trajectories covering the detection time process. The value ranges of the temperature rise phase sequence, current phase sequence, magnetic field phase sequence and acoustic phase sequence are divided into several non-overlapping phase intervals according to preset rules. The multi-parameter phase feature vector is mapped to the multi-dimensional phase region. The number of times the multi-parameter phase trajectory stays in each multi-dimensional phase region and the transfer relationship between different multi-dimensional phase regions are recorded to form the multi-dimensional phase region access distribution and multi-dimensional phase region transfer sequence. Based on the multidimensional phase region access distribution and multidimensional phase region transfer sequence, feature indicators are extracted to characterize the stability, phase coupling mode and phase evolution path of low-voltage cable joints in multi-parameter phase space. The feature indicators are combined into a fixed-length fingerprint description vector in a preset order, and the fingerprint description vector is used as the deep fingerprint of the parametric phase space of the low-voltage cable joint.

5. The multi-parameter non-destructive testing method for the quality of low-voltage cable joints in power engineering according to claim 1, characterized in that, The generation of multi-parameter transient data includes: When the circuit to which the low-voltage cable joint belongs is in normal operation, obtain the rated current parameters and current operating current parameters of the circuit, and determine the upper limit of the current disturbance amplitude and the set of current disturbance frequencies at multiple frequency points; Without changing the normal operating conditions of the circuit, a small-amplitude current perturbation excitation signal composed of multiple frequency components is injected into the circuit according to the preset upper limit of current perturbation amplitude and the set of multiple frequency points of current perturbation frequency, so as to keep the circuit running continuously within the preset perturbation excitation duration. Within a preset time window before, during, and after the injection of the perturbation excitation signal, the temperature rise parameters, electrical parameters, magnetic field parameters, and acoustic parameters at the low-voltage cable joint are synchronously collected at a transient sampling frequency higher than the sampling frequency of the multi-source time-series dataset to form transient sampling data. The collected transient data on temperature rise, electrical transient data, magnetic field transient data, and acoustic transient data are truncated and aligned according to the perturbation excitation trigger time, and transient change data segments within the corresponding perturbation excitation time period are extracted. The transient data segments of temperature rise, electrical transient data, magnetic field transient data, and acoustic transient data are combined in chronological order to form multi-parameter transient data.

6. The multi-parameter non-destructive testing method for the quality of low-voltage cable joints in power engineering according to claim 1, characterized in that, The constructed energy domain response characteristics include: Based on the perturbation excitation trigger time, a transient analysis time window including the pre-trigger period, the trigger period, and the post-trigger period is determined. Temperature rise transient data, current transient data, magnetic field transient data, and acoustic transient data are acquired within the time window, maintaining a one-to-one correspondence between the four types of transient data at the sampling time. For the four types of transient data, the difference between adjacent sampling points is calculated and the difference is calculated again to obtain the derived sequence. The amplitude is normalized and the baseline is calibrated based on the statistics of the period before the trigger. Fixed-length median filtering and short window smoothing are then performed on the derived sequence. Candidate transition event detection is performed on the calibrated and smoothed derived sequence. An amplitude threshold, a minimum number of sustained samples, and start-end identification rules are set. Intervals that satisfy the amplitude threshold and the minimum number of sustained samples are marked as single-domain candidate transition events. Perform multi-domain consistency judgment on single-domain candidate transition events, set multi-domain consistency time window and cross-domain phase order constraint, require candidate transition events of temperature rise parameter and electrical parameter to exist at the same time window and satisfy the preset order constraint, mark the event that satisfies the multi-domain consistency judgment as confirmed transition event, and generate event attribute set for each confirmed transition event. Transition mode vectors are generated by encoding confirmed transition events in a fixed order. The transition mode vectors consist of parameter type combination encoding, time overlap level encoding, peak amplitude level encoding, continuous sample number level encoding, and sequence identifier encoding. Corresponding transition mode vectors are generated for four types of parameters: temperature rise, current, magnetic field, and acoustics. The four types of transition mode vectors are connected in order to form energy domain response characteristics.

7. The multi-parameter non-destructive testing method for the quality of low-voltage cable joints in power engineering according to claim 1, characterized in that, The construction of the multi-domain coupled manifold embedding structure to generate high-dimensional tensor representations includes: The deep fingerprint of the parametric phase space, energy domain response features, and physical domain features are obtained and combined according to the preset field order to form a three-domain multi-parameter input sample set containing topological domain features, energy domain features, and physical domain features. Based on the three-domain multi-parameter input sample set, a multi-domain coupled manifold embedding structure is constructed. The multi-domain coupled manifold embedding structure includes intra-domain manifold construction units, cross-domain manifold alignment units, and coupled manifold generation units. The intra-domain manifold construction units process the topological domain features, energy domain features, and physical domain features respectively, determine the neighborhood relationship between the feature samples of each domain, and construct the corresponding intra-domain manifold subspace. By using the cross-domain manifold alignment unit, samples that correspond one-to-one with the time index or operating condition are selected from the three-domain multi-parameter input sample set as cross-domain anchor points. Based on the distribution relationship of the cross-domain anchor points in the manifold subspaces of each domain, scale normalization, coordinate translation and local neighborhood structure adjustment are performed on the topological domain manifold subspace, energy domain manifold subspace and physical domain manifold subspace. By using coupled manifold generation units, cross-domain connections are established between the aligned topological domain manifold subspace, energy domain manifold subspace, and physical domain manifold subspace. The neighborhood connections and cross-domain connections of the manifold subspaces within each domain are jointly modeled to form a joint manifold structure containing information from the three types of manifold subspaces and cross-domain coupling information. The joint manifold structure is used as the core representation of the multi-domain coupled manifold embedding structure. Based on the multi-domain coupled manifold embedding structure, the embedding coordinates of the three-domain multi-parameter input sample set are solved. The embedding coordinates of each sample in the joint manifold structure are arranged and organized according to a preset dimensional order to generate a high-dimensional tensor representation.

8. The multi-parameter non-destructive testing method for the quality of low-voltage cable joints in power engineering according to claim 1, characterized in that, The detection results of the generated low-voltage cable joints are used as operational monitoring outputs, including: The high-dimensional tensor representation is indexed and organized according to the joint sample dimension, time dimension, domain dimension and feature dimension to form a high-dimensional tensor sample set with a single low-voltage cable joint as the basic unit. For each low-voltage cable joint in the high-dimensional tensor sample set, based on the corresponding energy domain response characteristics and deep fingerprint of the parametric phase space, the shrinkage priority and shrinkage weight of the topological domain, energy domain and physical domain in the tensor shrinkage operation are determined; In the time dimension, the first stage of tensor contraction operation is performed on the high-dimensional tensor representation. According to the preset sliding time window and anomaly sensitive window length, the tensor elements in each time window are aggregated and calculated to compress the high-dimensional tensor representation of the same low-voltage cable joint in the entire detection time process into a time-contracted tensor representation that reflects the time evolution characteristics. Combining the shrinkage priority and shrinkage weight, the second-stage tensor shrinkage operation is performed on the domain dimension and feature dimension. The tensor components related to the topological domain, the tensor components related to the energy domain, and the tensor components related to the physical domain are grouped, aggregated, and weighted to generate a state vector. Each component in the state vector is compared with the preset set of quality level judgment thresholds, defect type judgment rules, and degradation trend judgment rules to determine the corresponding quality level identifier, defect type identifier, and degradation trend identifier. The state vector, quality level identifier, defect type identifier, and degradation trend identifier are then combined according to the joint number and detection time to generate a low-voltage cable joint detection result record as the operation monitoring output.

9. A multi-parameter non-destructive testing system for the quality of low-voltage cable joints in power engineering, comprising the multi-parameter non-destructive testing method for the quality of low-voltage cable joints in power engineering as described in any one of claims 1 to 8, characterized in that, Includes the following modules: The multi-domain data processing module is used to deploy multimodal sensors, collect multi-domain fusion data, perform preprocessing, and generate multi-source time-series datasets. The phase fingerprint construction module is used to extract phase from multi-source time series datasets, generate a set of multi-parameter phase trajectories, and construct a deep spatial fingerprint of the parametric phase. The perturbation transient acquisition module is used to apply small-amplitude multi-frequency current perturbation excitation and acquire transient change data of various physical parameters to generate multi-parameter transient data. The transition feature extraction module is used to identify transition events in multi-parameter transient data, extract the transition features of each parameter, and form energy domain response features. The multi-domain manifold embedding module is used to combine deep fingerprints of parametric phase space, energy domain response features and physical domain features to generate multi-domain coupled manifold embedding structures. The state output module is used to generate state vectors based on the multi-domain coupled manifold embedding structure and output the quality level, defect type and degradation trend.

Images