Non-contact multi-parameter elbow joint fault diagnosis method and device

Abstract

The embodiment of the application discloses a non-contact multi-parameter elbow joint fault diagnosis method and device; multi-parameter original signals of an area where an elbow joint is located in a same time reference and a same physical microenvironment are collected; the multi-parameter original signals are denoised, zero offset corrected and filtered to obtain preprocessed multi-parameter signals; a space-time coupling compensation matrix is applied to compensate the preprocessed multi-parameter signals; feature extraction is performed on the compensated multi-parameter signals to obtain a feature vector, and a response time lag between channels of each parameter signal is calculated; a weighted time distance between a current observation stage and typical modes of each fault stage is calculated, and the weighted time distance is mapped to a preset fault evolution chain to calculate posterior probabilities of the typical modes of each fault stage through the fault evolution chain; according to a comparison result of the posterior probabilities and a preset fault threshold, a fault diagnosis result of the area where the elbow joint is located and a confidence degree corresponding to the fault diagnosis result are output.

Description

Technical Field

[0001] This application relates to the field of power system state perception and intelligent diagnosis technology, specifically to a non-contact multi-parameter elbow joint fault diagnosis method and device. Background Technology

[0002] Ring main units (RMS) are compact high-voltage power distribution equipment suitable for ring network power supply and terminal power supply scenarios in urban and rural power distribution networks. They integrate switching and protection components, enabling power distribution, control, and rapid fault isolation. They are characterized by their small size, convenient operation, and high reliability, ensuring the stable and efficient operation of the power distribution network. With the development of power distribution systems towards higher density and greater enclosure, RMS are widely used in urban medium-voltage power grids. Their internal structure typically uses dry air insulation, and they are compact with limited space. The cable compartment contains key electrical connection components such as elbow joints, load switches, and surge arresters, making it the area most prone to partial discharge and thermal degradation. Under long-term operation, impurities in the insulation layer, moisture penetration, and concentrated interfacial electric fields can lead to the accumulation of micro-discharges, thermal stress buildup, and chemical decomposition, ultimately evolving into a combustion and explosion accident.

[0003] Existing technologies for monitoring the cable elbow joint area of ​​ring main units mostly employ multiple discrete sensors for independent detection, such as temperature and humidity sensors, gas sensors, ultraviolet detection, or infrared thermometry. Although these technologies can reflect the local state to a certain extent, they have the following problems: (1) The sensors are scattered and large in size, making it impossible to arrange them in the limited space of the cable elbow joint area, and the difference in the installation position of each sensor leads to spatial asynchrony; (2) The sensor data acquisition cycle is inconsistent, lacking a unified time reference, and it is impossible to form a multidimensional feature relationship under the same spatiotemporal framework; (3) When predicting faults, it relies on single threshold judgment or independent parameter trend analysis, which cannot reveal the causality and evolution law between multimodal physical processes, resulting in low accuracy of early fault warning.

[0004] Therefore, how to accurately detect faults in the cable elbow joint area at a local spatial scale is a problem that urgently needs to be solved. Summary of the Invention

[0005] This application provides a non-contact, multi-parameter elbow joint fault diagnosis method and apparatus, which can accurately detect faults in the elbow joint area of ​​cables at a local spatial scale.

[0006] This application provides a non-contact, multi-parameter elbow joint fault diagnosis method, including:

[0007] Using a pre-configured non-contact multi-parameter probe, multi-parameter raw signals of the elbow joint area are acquired at the same time reference and in the same physical microenvironment.

[0008] The original multi-parameter signal is denoised, zero-biased corrected, and filtered to obtain a preprocessed multi-parameter signal;

[0009] The preprocessed multi-parameter signal is compensated by applying a spatiotemporal coupling compensation matrix to obtain a compensated multi-parameter signal.

[0010] Feature extraction is performed on the compensated multi-parameter signal to obtain a feature vector, and the response time delay between the channels of each parameter signal is calculated;

[0011] Based on the feature vector and the response delay, the weighted time distance between the current observation stage and the typical modes of each fault stage is calculated, and the weighted time distance is mapped to a preset fault evolution chain, so as to calculate the posterior probability of the typical modes of each fault stage through the fault evolution chain.

[0012] Based on the comparison between the posterior probability and the preset fault threshold, the fault diagnosis result of the area where the elbow joint is located and the confidence level corresponding to the fault diagnosis result are output.

[0013] The spatiotemporal coupling compensation matrix includes a dynamic compensation matrix and a hysteresis matrix. The application of the spatiotemporal coupling compensation matrix to compensate the preprocessed multi-parameter signal yields a compensated multi-parameter signal, including:

[0014] The preprocessed multi-parameter signal is compensated using the dynamic compensation matrix and the hysteresis matrix to obtain the compensated multi-parameter signal.

[0015] The compensated multi-parameter signal is represented as follows:

[0016]

[0017] in, For the compensated multi-parameter signal at the nth sampling point, For dynamic compensation matrix, This represents the relative humidity signal at the nth sampling point. For the temperature signal of the nth sampling point, This is the preprocessed multi-parameter signal for the nth sampling point. The size of the maximum hysteresis window. Let be the hysteresis matrix of the k-th hysteresis window. This is the preprocessed multi-parameter signal of the nk-th sampling point;

[0018] The dynamic compensation matrix is ​​obtained by adding a humidity-dependent offset matrix and a temperature-dependent offset matrix to a 6th-order identity matrix. The dynamic compensation matrix is ​​used to correct the signal deviation caused by temperature and humidity cross-sensitivity.

[0019] The hysteresis matrix adopts an exponential decay form and is used to characterize the impact of airflow delay and sensor response time on historical signals.

[0020] The step of extracting features from the compensated multi-parameter signal to obtain a feature vector includes:

[0021] Extract the instantaneous values ​​of each compensated parameter signal from its channels;

[0022] Calculate the first-order difference of each channel of the compensated parameter signal;

[0023] Calculate the slope of the sliding window for each compensated parameter signal channel;

[0024] Count the instantaneous pulses of the ultraviolet signal channel;

[0025] The spectral energy of the smoke signal is calculated using short-time Fourier transform.

[0026] Calculate the short-window Pearson correlation coefficient between any two channels of the compensated parametric signals;

[0027] The feature vector is obtained based on the channel instantaneous value, the channel first-order difference, the channel sliding window slope, the instantaneous pulse count, the spectral energy, and the short-window Pearson correlation coefficient.

[0028] The step of calculating the weighted time distance between the current observation stage and the typical modes of each fault stage based on the feature vector and the response delay includes:

[0029] Multiple typical fault stage patterns are preset, and each typical fault stage pattern corresponds to a set of typical feature trajectories and a typical time delay.

[0030] For the current observation phase, calculate the weighted time distance of the current observation phase relative to the typical mode of each fault phase. The weighted time distance is represented as follows:

[0031]

[0032] in, This represents the weighted time distance between the nth sampling point in the current observation phase and the typical mode of the kth fault phase. For the feature vector, Let m be the weight of the m-th feature. For the m-th feature observation value of the n-th sampling point, This is the m-th typical feature trajectory of the typical mode of the k-th fault stage. This represents the m-th typical time delay of the typical mode of the k-th fault stage. This represents the set of typical feature vectors for the typical mode of the k-th failure stage.

[0033] The process of mapping the weighted time distance to a preset fault evolution chain, and calculating the posterior probability of typical modes at each fault stage through the fault evolution chain, includes:

[0034] The weighted time distance is mapped to a preset fault evolution chain to construct a likelihood function, which represents the probability of observing the corresponding weighted time distance when given a typical pattern of a certain fault stage.

[0035] The posterior probability is calculated using the likelihood function, and the posterior probability is expressed as follows:

[0036]

[0037] in, Represents the posterior probability. Represents the likelihood function. This represents the prior probability of the typical mode for the k-th failure stage. Index representing typical failure phases. This represents the total number of typical failure phases. This represents the weighted time distance of the current observation phase relative to the typical pattern of the k-th failure phase. For the first Weighted time distance observed under typical fault stage modes The conditional probability, For the first Prior probabilities of typical failure phase patterns.

[0038] Accordingly, embodiments of this application also provide a non-contact multi-parameter elbow joint fault diagnosis device, comprising:

[0039] A non-contact multi-parameter probe is used to acquire multi-parameter raw signals in the area where the elbow joint is located, under the same time reference and the same physical microenvironment.

[0040] The diagnostic unit is used to denoise, perform zero-bias correction, and filter on the original multi-parameter signal to obtain a preprocessed multi-parameter signal; apply a spatiotemporal coupling compensation matrix to compensate the preprocessed multi-parameter signal to obtain a compensated multi-parameter signal; extract features from the compensated multi-parameter signal to obtain feature vectors, and calculate the response delay between the channels of each parameter signal; based on the feature vectors and the response delays, calculate the weighted time distance between the current observation stage and the typical modes of each fault stage, and map the weighted time distance to a preset fault evolution chain to calculate the posterior probability of the typical modes of each fault stage through the fault evolution chain; and output the fault diagnosis result of the elbow joint area and the confidence level corresponding to the fault diagnosis result based on the comparison result of the posterior probability and the preset fault threshold.

[0041] The non-contact multi-parameter probe adopts a three-dimensional stacked modular structure. From top to bottom, the non-contact multi-parameter probe includes an optical sensing module, a gas detection module, an environmental parameter monitoring module, and an electromagnetic shielding heat dissipation module.

[0042] The optical sensing module includes a sapphire optical window, an ultraviolet detector, and a laser scattering optical component.

[0043] The surface of the sapphire optical window is coated with a double-layer interference film;

[0044] The ultraviolet detector and the laser scattering optical component share the sapphire optical window to form a coaxial optical path.

[0045] The gas detection module includes:

[0046] The microfluidic cavity has its inlet located on the side wall of the non-contact multi-parameter probe and its outlet located below the non-contact multi-parameter probe. The microfluidic cavity includes a pulse delay cavity structure, which is used to form a brief dead zone so that the optical sensing module, gas detection module, environmental parameter monitoring module and electromagnetic shielding heat dissipation module generate a time offset for the same sample.

[0047] Metal-oxide-semiconductor gas-sensitive arrays are used to detect volatile organic compounds and volatile gaseous products.

[0048] Electrochemical unit used for detecting carbon monoxide.

[0049] The environmental parameter monitoring module includes:

[0050] A thin-film thermistor, wherein the thin-film thermistor is thermally coupled to the bottom of the non-contact multi-parameter probe via a thermal bridge;

[0051] Thin-film capacitive humidity sensor, used to detect ambient humidity;

[0052] A resistance temperature detector is used to assist the thin-film thermistor in detecting ambient temperature.

[0053] The electromagnetic shielding heat dissipation module includes:

[0054] Magnetic ring shielding is used to shield high-frequency electromagnetic interference generated by electric arc or partial discharge;

[0055] A micron-sized conductive carbon film is coated on the inner wall of the non-contact multi-parameter probe. The micron-sized conductive carbon film is used to absorb local charges to reduce the transient impact of discharge pulses on gas signals and optical signals.

[0056] In this embodiment, firstly, a pre-configured non-contact multi-parameter probe is used to acquire raw multi-parameter signals of the elbow joint area within the same time reference and physical microenvironment. Then, the raw multi-parameter signals are denoised, zero-biased corrected, and filtered to obtain pre-processed multi-parameter signals. Next, a spatiotemporal coupling compensation matrix is ​​applied to compensate the pre-processed multi-parameter signals to obtain compensated multi-parameter signals. Then, feature extraction is performed on the compensated multi-parameter signals to obtain feature vectors, and the response time delays between the channels of each parameter signal are calculated. Then, based on the feature vectors and the response time delays, a weighted time distance is calculated between the current observation stage and the typical modes of each fault stage, and this weighted time distance is mapped to a preset fault evolution chain to calculate the posterior probability of the typical modes of each fault stage through the fault evolution chain. Finally, based on the comparison between the posterior probability and a preset fault threshold, the fault diagnosis result of the elbow joint area and the corresponding confidence level are output.

[0057] This solution utilizes pre-configured non-contact multi-parameter probes to acquire raw multi-parameter signals within the same timeframe and physical micro-environment. This integrates multi-parameter sensing units (thermal, humidity, electrical, optical, gas, and flame sensors) into a millimeter-level space, physically eliminating the spatial asynchrony issues caused by varying installation locations in traditional distributed sensors. This ensures that multi-parameter data originates from the same monitoring scenario, laying a data foundation for accurate diagnosis. Pre-processing steps, including denoising, zero-bias correction, and filtering, effectively remove interference noise from the raw signals, improving signal purity. A spatiotemporal coupling compensation matrix is ​​then applied to compensate for the pre-processed signals, correcting signal deviations caused by temperature and humidity cross-sensitivity, airflow delay, and sensor response lag. This achieves signal self-calibration under dynamic environments, ensuring the accuracy of multi-parameter signals. Accuracy and consistency are achieved by extracting multi-dimensional feature vectors containing instantaneous values, first-order differences, and sliding window slopes, and calculating the response delay of each channel. This comprehensively captures the signal change patterns and causal relationships between parameters during the fault evolution process, overcoming the limitations of traditional single-feature analysis. Based on the feature vectors and response delays, a weighted time distance is calculated, mapped to a preset fault evolution chain, and the posterior probability is calculated. The fault evolution chain model is used to reconstruct the progressive fault process of the cable terminal from thermal degradation to discharge breakdown. A statistical probability model replaces the single threshold judgment, improving the accuracy of early fault identification. By comparing the posterior probability with the preset threshold, the diagnostic results and confidence levels are output, reducing false alarms and missed alarms. Ultimately, accurate detection and graded diagnosis of elbow joint area faults at a local spatial scale are achieved. Attached Figure Description

[0058] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0059] Figure 1 This is a flowchart illustrating the non-contact multi-parameter elbow joint fault diagnosis method provided in the embodiments of this application.

[0060] Figure 2 This is a schematic diagram of the non-contact multi-parameter elbow joint fault diagnosis device provided in the embodiments of this application. Detailed Implementation

[0061] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0062] This application provides a non-contact, multi-parameter elbow joint fault diagnosis method and apparatus. Specifically, this non-contact, multi-parameter elbow joint fault diagnosis apparatus can be integrated into an electronic device, such as a terminal or server.

[0063] The server can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud pre-built databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, network acceleration services (Content Delivery Network, CDN), as well as big data and artificial intelligence platforms.

[0064] The terminal can be a smartphone, tablet, laptop, desktop computer, smart speaker, smartwatch, smart TV, in-vehicle terminal, etc., but is not limited to these. The terminal and the server can be connected directly or indirectly through wired or wireless communication, which is not limited herein.

[0065] The following sections provide detailed descriptions of each example. It should be noted that the order in which the embodiments are described is not intended to limit the preferred order of the embodiments.

[0066] Figure 1 This is a flowchart illustrating the non-contact, multi-parameter elbow joint fault diagnosis method provided in this application embodiment. Figure 1 As shown, the specific process of this non-contact multi-parameter elbow joint fault diagnosis method is as follows:

[0067] 101. Using a pre-configured non-contact multi-parameter probe, acquire multi-parameter raw signals of the elbow joint area in the same time reference and the same physical microenvironment.

[0068] Non-contact multi-parameter probes are a type of detection device that does not directly contact the object being measured and can simultaneously acquire two or more physical or chemical parameters. Their core advantage lies in avoiding interference from contact measurements on the operation of the measured equipment. They are also adaptable to confined, complex, or high-risk installation environments and are widely used in equipment condition monitoring in fields such as power and chemical industries. Specifically, in this embodiment, the non-contact multi-parameter probe has an integrated three-dimensional stacked modular structure, with an overall near-cylindrical or flat cylindrical shape, an outer diameter not exceeding 25mm, and a length not exceeding 60mm. The probe's interior integrates, from top to bottom, an optical sensing module, a gas detection module, an environmental parameter monitoring module, and an electromagnetic shielding and heat dissipation module. It can simultaneously acquire six types of multi-parameter signals: heat, humidity, electricity, light, gas, and flame. Furthermore, all sensing units share the same microfluidic cavity, ensuring spatial and temporal consistency of the signals.

[0069] In this context, "same time reference" means that all parameter acquisition is based on a unified clock signal, avoiding data misalignment in the time dimension; "same physical microenvironment" means that each sensing unit is in the same local monitoring scenario, ensuring that the acquired multiple parameters can reflect the state of the same area. These two are the fundamental prerequisites for multi-parameter collaborative analysis, avoiding data correlation distortion caused by spatiotemporal differences. Specifically, in this embodiment, the same time reference is achieved through a high-stability crystal oscillator clock built into the probe, and all sensing channels complete data acquisition under the synchronization of this clock; the same physical microenvironment is achieved through the probe's common cavity gas-optic flow path design, where optical sensing, gas detection, and other modules share the same microfluidic cavity and monitoring area, eliminating the spatial asynchrony problem of traditional distributed sensors at the physical level.

[0070] Among them, multi-parameter raw signals refer to raw physical or chemical signals that are directly acquired through various sensing elements without processing. They typically include signal types related to equipment status, such as heat (temperature), humidity, electricity (electric field / discharge), light (ultraviolet / visible light), gas (gas concentration), and flame (flame / infrared). These are the basic data for subsequent signal processing and diagnostic analysis.

[0071] 102. The original multi-parameter signal is denoised, zero-biased corrected, and filtered to obtain the preprocessed multi-parameter signal.

[0072] Denoising refers to removing useless components introduced by electromagnetic interference, environmental noise, etc. from the signal; zero-bias correction refers to correcting the zero-point offset error of the sensing element itself to ensure the accuracy of the signal reference; filtering refers to retaining the effective frequency band components related to the monitoring target in the signal and suppressing irrelevant frequency band interference. The combined effect of these three is the key to improving signal quality and ensuring the accuracy of subsequent analysis. Specifically, in this embodiment, the acquired 6-dimensional multi-parameter raw signal is subjected to this preprocessing operation. Targeted denoising algorithms (such as wavelet denoising) are used to eliminate noise caused by high-frequency electromagnetic interference in the ring main unit. Zero-bias correction is used to eliminate sensor baseline drift. Filtering algorithms (such as low-pass filtering) are used to retain fault-related signal characteristics, ultimately obtaining the preprocessed multi-parameter signal.

[0073] 103. Apply the spatiotemporal coupling compensation matrix to compensate the preprocessed multi-parameter signal to obtain the compensated multi-parameter signal.

[0074] Among them, the spatiotemporal coupling compensation matrix is ​​a type of mathematical model used to correct spatiotemporal correlation errors in multi-parameter signals. It is mainly used to address signal distortion caused by temperature and humidity cross-sensitivity, signal transmission delay, and sensor response lag in multi-sensor systems. It achieves dynamic calibration of signals through matrix operations and is widely used in the field of multimodal sensor data fusion.

[0075] 104. Perform feature extraction on the compensated multi-parameter signal to obtain feature vectors, and calculate the response time delay between the channels of each parameter signal.

[0076] Feature extraction refers to the process of extracting key information that reflects changes in the target state from the preprocessed signal and transforming the high-dimensional original signal into a low-dimensional feature vector. The extracted features need to have discriminability and stability. Common features include instantaneous values, differences, slopes, pulse counts, and spectral energy. It is the core link of pattern recognition and fault diagnosis.

[0077] Among them, response delay refers to the difference in response time between different sensing channels to the same excitation signal. It originates from factors such as differences in signal transmission paths and different response speeds of the sensors themselves. It is a key spatiotemporal feature that needs to be quantified in multi-parameter collaborative analysis and can be calculated by methods such as cross-correlation functions.

[0078] 105. Based on the feature vector and the response delay, calculate the weighted time distance between the current observation stage and the typical modes of each fault stage, and map the weighted time distance to a preset fault evolution chain, so as to calculate the posterior probability of the typical modes of each fault stage through the fault evolution chain.

[0079] Weighted temporal distance is a type of metric used to measure the difference between currently observed features and typical pattern features. By assigning weights to different features, it highlights the influence of key features and combines time dimension offset correction to achieve accurate comparison of dynamic patterns. It is widely used in the field of pattern recognition of time series data.

[0080] The fault evolution chain is a progressive evolution path model based on the physical mechanism of equipment faults, describing the fault's progression from its inception to its outbreak. It includes multiple ordered fault stages and typical characteristics of each stage, overcoming the limitations of traditional single-threshold judgments and reflecting the causal relationship and temporal pattern of fault evolution. Specifically, in this embodiment, the preset fault evolution chain is constructed based on the physical mechanism of cable terminal faults, covering ordered fault stages such as local overheating, insulation oxidation, decomposition gas generation, micro-discharge, smoke generation, and arc breakdown. Each stage pre-stores a set of typical feature trajectories and typical time delays, providing a model basis for posterior probability calculation.

[0081] Among them, posterior probability is the conditional probability of an event occurring after the acquisition of observation data, calculated based on Bayes' theorem. It can integrate the correlation between prior information and observation data, quantify the credibility of the event occurrence, and is widely used in the fields of statistical decision-making and fault diagnosis.

[0082] 106. Based on the comparison between the posterior probability and the preset fault threshold, output the fault diagnosis result of the area where the elbow joint is located and the confidence level corresponding to the fault diagnosis result.

[0083] Among them, the preset fault threshold is a critical value used to judge the fault status, which is calibrated according to actual application needs and historical data. It is usually divided into early warning threshold and alarm threshold. It is a key parameter for realizing fault classification decision-making and needs to balance diagnostic sensitivity and false alarm rate.

[0084] The fault diagnosis result is a qualitative judgment of the equipment status (e.g., normal, early warning of a certain stage, fault alarm of a certain stage), while the confidence level is a quantitative representation of the reliability of the diagnosis result. The combination of the two can provide users with a clear and reliable basis for decision-making, avoiding the uncertainty of a single qualitative judgment. Specifically, in this embodiment, the diagnosis result is output based on the comparison between the posterior probability and the preset fault threshold, including "normal / low confidence", "early warning", "fault alarm", etc. The confidence level is directly determined by the posterior probability, realizing the synchronous output of the diagnosis result and reliability.

[0085] As can be seen from the above, the non-contact multi-parameter elbow joint fault diagnosis method in this application uses a pre-configured non-contact multi-parameter probe to collect multi-parameter raw signals in the same time reference and the same physical micro-environment. This integrates multi-parameter sensing units such as heat, humidity, electricity, light, gas, and flame into a millimeter-level space, eliminating the spatial asynchrony problem caused by the different installation positions of traditional distributed sensors at the physical level. This ensures that the multi-parameter data originates from the same monitoring scene, laying a data foundation for accurate diagnosis. Through preprocessing steps of denoising, zero-bias correction, and filtering, interference noise in the raw signal is effectively removed, improving signal purity. Then, a spatiotemporal coupling compensation matrix is ​​applied to compensate the preprocessed signal, correcting signal deviations caused by temperature and humidity cross-sensitivity, airflow delay, and sensor response lag, achieving signal self-calibration in dynamic environments. The system ensures the accuracy and consistency of multi-parameter signals. By extracting multi-dimensional feature vectors including instantaneous values, first-order differences, and sliding window slopes, and calculating the response delay of each channel, it comprehensively captures the signal change patterns and causal relationships between parameters during the fault evolution process, breaking through the limitations of traditional single-feature analysis. Based on the feature vectors and response delays, a weighted time distance is calculated, mapped to a preset fault evolution chain, and the posterior probability is calculated. The fault evolution chain model is used to reconstruct the progressive fault process of the cable terminal from thermal degradation to discharge breakdown. The statistical probability model replaces the single threshold judgment, improving the accuracy of early fault identification. By comparing the posterior probability with the preset threshold, the diagnostic results and confidence levels are output, reducing false alarms and missed alarms. Ultimately, it achieves accurate detection and graded diagnosis of elbow joint area faults at a local spatial scale.

[0086] In some embodiments, the spatiotemporal coupling compensation matrix includes a dynamic compensation matrix and a hysteresis matrix. The application of the spatiotemporal coupling compensation matrix to compensate the preprocessed multi-parameter signal to obtain a compensated multi-parameter signal includes:

[0087] The preprocessed multi-parameter signal is compensated using the dynamic compensation matrix and the hysteresis matrix to obtain the compensated multi-parameter signal.

[0088] The compensated multi-parameter signal is represented as follows:

[0089]

[0090] in, For the compensated multi-parameter signal at the nth sampling point, For dynamic compensation matrix, This represents the relative humidity signal at the nth sampling point. For the temperature signal of the nth sampling point, This is the preprocessed multi-parameter signal for the nth sampling point. The size of the maximum hysteresis window. Let be the hysteresis matrix of the k-th hysteresis window. This is the preprocessed multi-parameter signal of the nk-th sampling point;

[0091] The dynamic compensation matrix is ​​obtained by adding a humidity-dependent offset matrix and a temperature-dependent offset matrix to a 6th-order identity matrix. The dynamic compensation matrix is ​​used to correct the signal deviation caused by temperature and humidity cross-sensitivity.

[0092] The hysteresis matrix adopts an exponential decay form and is used to characterize the impact of airflow delay and sensor response time on historical signals.

[0093] The dynamic compensation matrix is ​​a matrix-type tool used in multi-parameter sensor signal processing to correct signal deviations caused by environmental interference (such as temperature and humidity changes) in real time. Its core is to dynamically adjust the matrix coefficients to offset the cross-sensitivity effects of environmental factors on different sensor channels, ensuring that the signal accurately reflects the state of the measured object. It is widely used in multimodal data calibration scenarios. Specifically, in this embodiment, the matrix is ​​6×6 in dimension, dynamically generated based on the relative humidity and temperature signals of the current sampling point. It is constructed by adding a humidity-dependent offset matrix and a temperature-dependent offset matrix to a 6th-order identity matrix, specifically used to correct the deviations caused by the cross-sensitivity of temperature and humidity in multi-parameter signals, ensuring the accuracy of the signal under fluctuating temperature and humidity conditions.

[0094] The humidity-dependent offset matrix is ​​a matrix that quantifies the degree of interference of humidity changes on the signals of each sensing channel. It includes a humidity cross-sensitivity coefficient, the magnitude of which is related to the degree of humidity deviation from the reference value. It is used to specifically counteract the influence of humidity on signals of different parameters.

[0095] The temperature-dependent offset matrix is ​​used to quantify the interference of temperature fluctuations on the signals of each sensing channel. It includes temperature cross-sensitivity coefficients and, by combining it with the temperature scaling function, can dynamically adjust the correction strength to offset the systematic errors caused by temperature changes to multi-parameter signals.

[0096] The 6th-order identity matrix is ​​a 6×6 matrix with diagonal elements of 1 and all other elements of 0. In matrix operations, it serves as the benchmark for signal correction, ensuring that the signal retains its original characteristics and does not generate additional distortion when there is no environmental interference. It is the basic building block of the dynamic compensation matrix. Specifically, in this embodiment, this matrix is ​​the baseline portion of the dynamic compensation matrix, providing an initial benchmark for each sensing channel (heat, humidity, electricity, light, gas, flame), ensuring that multi-parameter signals are not over-corrected when there is no significant interference from temperature and humidity.

[0097] Hysteresis matrices are matrix tools used in time-series signal processing to characterize the influence of historical signals on current signals. By quantifying the historical effects caused by signal transmission delays and sensor response delays, they can compensate for the delay of the current signal. They are commonly found in multi-sensor systems with dynamic delays.

[0098] The size of the maximum hysteresis window is the threshold for the number of historical signal sampling points set in the timing compensation, which determines the range of historical data participating in the delay compensation. It needs to be calibrated according to the system delay characteristics (such as airflow transmission time and sensor response speed) to ensure that all historical samples that have a significant impact on the current signal are covered.

[0099] Among them, the hysteresis matrix of the k-th hysteresis window is the hysteresis matrix corresponding to the hysteresis window index k. The coefficients of the hysteresis matrix decrease as k increases, reflecting the rule that the more distant the historical signal, the less influence it has on the current signal.

[0100] Among them, the preprocessed multi-parameter signal of the nk-th sampling point is the preprocessed multi-parameter data of the k-th time before the n-th current sampling point in the time sequence sampling. It contains parameter information such as heat, humidity, electricity, light, gas, and flame at that time and is the specific source of historical signals in hysteresis compensation.

[0101] In this embodiment, a dynamic compensation matrix composed of a 6th-order identity matrix, a humidity-dependent offset matrix, and a temperature-dependent offset matrix can accurately correct the deviations caused by the cross-sensitivity of temperature and humidity to multi-parameter signals based on real-time temperature and humidity signals, avoiding signal distortion caused by environmental fluctuations. By adopting a hysteresis matrix in the form of exponential decay, the influence of historical signals caused by airflow delay and sensor response time can be quantified, and the coefficient decaying according to an exponential law can accurately match the strength changes of the actual delay effect, ensuring that historical delays are effectively offset. The synergistic effect of the dynamic compensation matrix and the hysteresis matrix enables the compensated multi-parameter signals to eliminate environmental interference and correct timing delays, significantly improving the accuracy and consistency of the signals. This provides a high-quality data foundation for subsequent feature extraction, time delay calculation, and fault evolution chain matching, thereby solving the problem of low diagnostic accuracy caused by environmental interference and delay effects in traditional multi-parameter monitoring, and indirectly improving the accuracy and reliability of elbow joint fault detection.

[0102] In some implementations, the step of extracting features from the compensated multi-parameter signal to obtain a feature vector includes:

[0103] Extract the instantaneous values ​​of each compensated parameter signal from its channels;

[0104] Calculate the first-order difference of each channel of the compensated parameter signal;

[0105] Calculate the slope of the sliding window for each compensated parameter signal channel;

[0106] Count the instantaneous pulses of the ultraviolet signal channel;

[0107] The spectral energy of the smoke signal is calculated using short-time Fourier transform.

[0108] Calculate the short-window Pearson correlation coefficient between any two channels of the compensated parametric signals;

[0109] The feature vector is obtained based on the channel instantaneous value, the channel first-order difference, the channel sliding window slope, the instantaneous pulse count, the spectral energy, and the short-window Pearson correlation coefficient.

[0110] Among them, the instantaneous value of a channel refers to the real-time measurement value output by a single sensing channel at a specific sampling moment. It directly reflects the instantaneous state of the corresponding physical or chemical parameters at that moment and is the most basic original feature in signal feature extraction, which can provide a real-time state benchmark for subsequent analysis.

[0111] The first-order difference refers to the difference between signals from the same channel at two adjacent sampling times. The mathematical expression is the subtraction of signal values ​​at adjacent times. It is used to quantify the rate of change of a signal in a very short time and can quickly capture the initial stage of sudden changes or gradual trends in parameters. It is one of the core features reflecting the dynamic changes of signals.

[0112] The sliding window slope refers to the average value of the first-order difference within a continuous sampling point window of a set size W. It is used to smooth out interference caused by instantaneous fluctuations and reflect the overall trend of signal change over a period of time. The window size W needs to be calibrated according to the signal characteristics and monitoring requirements to balance response speed and stability.

[0113] Among them, the instantaneous pulse count of the ultraviolet signal channel refers to the number of ultraviolet pulses detected by the ultraviolet detector per unit time (such as per second). It is used to quantify the intensity and frequency of ultraviolet radiation and is a key feature for identifying ultraviolet radiation phenomena such as partial discharge and electric arc. It is widely used in the field of electrical equipment discharge monitoring.

[0114] The Short-Time Fourier Transform (SFT) is a signal processing method that decomposes a time-domain signal into a time-frequency domain. By performing a Fourier transform on the signal through a sliding time window, it can reflect both the time-domain changes of the signal and capture its frequency-domain characteristics. It is suitable for the analysis of non-stationary signals and is a core means of extracting frequency-domain information from signals. Specifically, in this embodiment, this method is used to process the smoke signal detected by the laser scattering component, transforming the time-domain waveform of the smoke signal into a time-frequency distribution, providing a basis for subsequent calculation of spectral energy.

[0115] The spectral energy of the smoke signal refers to the total energy of the smoke signal within a specific frequency range after undergoing a short-time Fourier transform. It reflects the intensity distribution of the smoke signal in the frequency domain. Smoke with different particle sizes and concentrations corresponds to different spectral energy characteristics, making it a key frequency domain feature for identifying the presence and severity of smoke. Specifically, in this embodiment, this feature is the energy value calculated from the smoke laser scattering signal after a short-time Fourier transform. It is used to quantify the concentration and particle characteristics of the smoke, indirectly reflecting the number of solid particles generated by insulation decomposition.

[0116] The short-window Pearson correlation coefficient is a statistical indicator that measures the degree of linear correlation between two variables within a set short time window. Its value ranges from -1 to 1. The closer the absolute value is to 1, the stronger the correlation, and the closer it is to 0, the weaker the correlation. It is used to reveal the intrinsic relationship between different parameters.

[0117] In this embodiment, by extracting the instantaneous values ​​of each channel, the real-time status of each parameter during the fault evolution process can be accurately captured, providing basic data support for diagnosis. By calculating the first-order difference and the slope of the sliding window, the instantaneous rate of change of the parameter and the overall trend of change over a period of time are reflected, respectively, avoiding the limitation that a single instantaneous value cannot reflect dynamic changes. By statistically counting the instantaneous pulses of the ultraviolet signal channel, the partial discharge intensity can be directly correlated, providing targeted characteristics for discharge-related faults. By calculating the spectral energy of the smoke signal through short-time Fourier transform, the frequency domain characteristics of smoke particles can be effectively captured. This approach improves the accuracy of smoke detection. By calculating the short-window Pearson correlation coefficient, it reveals the intrinsic relationships between different parameters (such as the correlation between overheating and gas production), aligning with the causal logic of fault evolution. Ultimately, it integrates multi-dimensional features into a feature vector, comprehensively covering the static state, dynamic changes, specific fault signals, and parameter correlation characteristics of the parameters. This overcomes the limitations of traditional single-feature analysis and provides a rich, comprehensive, and distinctive feature foundation for subsequent pattern matching and posterior probability calculation based on the fault evolution chain, significantly improving the accuracy and early identification capability of elbow joint fault diagnosis.

[0118] In some implementations, calculating the weighted time distance between the current observation stage and typical modes of each fault stage based on the feature vector and the response delay includes:

[0119] Multiple typical fault stage patterns are preset, and each typical fault stage pattern corresponds to a set of typical feature trajectories and a typical time delay.

[0120] For the current observation phase, calculate the weighted time distance of the current observation phase relative to the typical mode of each fault phase. The weighted time distance is represented as follows:

[0121]

[0122] in, This represents the weighted time distance between the nth sampling point in the current observation phase and the typical mode of the kth fault phase. For the feature vector, Let m be the weight of the m-th feature. For the m-th feature observation value of the n-th sampling point, This is the m-th typical feature trajectory of the typical mode of the k-th fault stage. This represents the m-th typical time delay of the typical mode of the k-th fault stage. This represents the set of typical feature vectors for the typical mode of the k-th failure stage.

[0123] Among them, the typical fault stage mode is a standard state mode corresponding to each stage of fault development, based on the physical evolution mechanism of equipment faults. Each mode contains the core characteristic performance of each monitoring parameter under that stage, and serves as the benchmark for fault mode matching. It is widely used in the field of time-series fault diagnosis. Specifically, in this embodiment, the typical fault stage mode is based on the preset cable terminal fault evolution chain, covering six ordered stages: local overheating (F1), insulation oxidation (F2), decomposition gas generation (F3), micro-discharge (F4), smoke generation (F5), and arc breakdown (F6). Each mode corresponds to a unique set of typical characteristic trajectories and typical time delays, providing a standard reference for fault matching in the current observation stage.

[0124] Among them, the typical feature trajectory set is a set of standard time-series curves showing the changes of each monitoring feature over time under a typical mode of a certain fault stage. It reflects the dynamic evolution of features under the fault stage and is trained by a large number of laboratory-induced fault samples and on-site historical fault data, thus possessing stability and recognizability.

[0125] Among them, typical time delay is the inherent response time difference between various features under a typical mode of a certain fault stage. It reflects the temporal characteristics of the causal relationship of parameters under the fault stage. It is determined by the physical mechanism of the fault and is a key parameter for correcting the effects of time delay and achieving accurate mode matching.

[0126] The weight of the m-th feature is the importance coefficient of the m-th feature in the feature vector in pattern matching, with a value range of [0,1]. The sum of all feature weights is 1. It is obtained by training through optimization algorithms (such as cross-entropy minimization and maximum a posteriori likelihood) and is used to highlight features that contribute highly to fault stage identification and suppress the interference of secondary features.

[0127] Among them, the m-th typical feature trajectory of the typical pattern of the k-th fault stage is the standard time series change curve of the m-th feature under the fault stage. It is the ideal representation of the feature in the fault stage and is used to make accurate comparison with the currently observed features.

[0128] Among them, the m-th typical time delay of the typical mode of the k-th fault stage is the standard response delay of the m-th feature relative to the baseline feature under the fault stage. It is used to correct the misalignment between the current observed feature and the typical feature trajectory in the time dimension and ensure the spatiotemporal consistency of the comparison.

[0129] Among them, the typical feature vector set of the typical pattern of the kth fault stage is a set of standard feature vectors of all features at each sampling time under the fault stage. It is a complete time-series record of the feature state of the fault stage, providing comprehensive standard data support for pattern matching.

[0130] In this embodiment, by pre-setting multiple typical fault stage patterns based on the fault evolution chain and configuring a set of typical feature trajectories and typical time delays that conform to the laws of physical evolution for each pattern, a precise standard benchmark is provided for fault matching in the current observation stage, solving the problem of lack of temporal benchmarks in traditional diagnosis. By introducing feature weights, the dominant role of key features in fault stage identification is highlighted, the interference of secondary features is suppressed, and the discriminative power of distance calculation is improved. By incorporating typical time delays into the weighted time distance formula, the comparison error caused by temporal misalignment between features is corrected, ensuring the consistency between the current observation and the typical patterns in the time dimension. Finally, the weighted time distance is used to quantify the degree of difference between the current observation and the typical patterns of each fault stage, providing a reliable difference measurement basis for subsequently mapping the differences to the fault evolution chain and calculating the posterior probability. This effectively breaks through the limitation that traditional single threshold judgment cannot reflect the temporal evolution law of faults, significantly improving the accuracy and reliability of elbow joint fault stage identification and providing key support for early fault warning.

[0131] In some implementations, the weighted time distance is mapped to a preset fault evolution chain to calculate the posterior probability of typical modes at each fault stage through the fault evolution chain, including:

[0132] The weighted time distance is mapped to a preset fault evolution chain to construct a likelihood function, which represents the probability of observing the corresponding weighted time distance when given a typical pattern of a certain fault stage.

[0133] The posterior probability is calculated using the likelihood function, and the posterior probability is expressed as follows:

[0134]

[0135] in, Represents the posterior probability. Represents the likelihood function. This represents the prior probability of the typical mode for the k-th failure stage. Index representing typical failure phases. This represents the total number of typical failure phases. This represents the weighted time distance of the current observation phase relative to the typical pattern of the k-th failure phase. For the first Weighted time distance observed under typical fault stage modes The conditional probability, For the first Prior probabilities of typical failure phase patterns.

[0136] The likelihood function is a function in statistics used to characterize the probability of observing a certain result given certain model parameters (or conditions). Its core is to quantify the degree of matching between observed data and model parameters. It is widely used in parameter estimation and decision-making in fields such as probability statistics and pattern recognition.

[0137] Among them, prior probability refers to the probability of an event occurring based on historical experience, statistical data, or the frequency of failures, before any observation data is obtained. It is the initial basis for Bayesian probability inference and is used to incorporate prior knowledge to improve the rationality of the inference.

[0138] Here, index j is an identifying parameter used to traverse all typical modes of failure stages. Its value ranges from 1 to K (K is the total number). Its core function is to sequentially refer to each failure stage, realizing the calculation and comparison of all modes one by one.

[0139] The total number (K) of typical fault stage patterns is the total number of typical fault stage patterns included in the preset fault evolution chain. It is determined by the physical evolution mechanism of equipment faults and needs to fully cover the complete process from fault initiation to outbreak, ensuring that no critical stages are omitted. Specifically, in this embodiment, the value of K is consistent with the stage division of the fault evolution chain, corresponding to 6 stages: local overheating, insulation oxidation, decomposition gas generation, micro-discharge, smoke generation, and arc breakdown, i.e., K=6, ensuring that all possible fault evolution stages are included in the posterior probability calculation.

[0140] In this embodiment, by mapping weighted time distance to the fault evolution chain and constructing a likelihood function, the degree of difference between observations and typical patterns is transformed into a quantifiable probability value, solving the problem that traditional difference measures cannot be directly used for decision-making. By introducing prior probability, empirical information on the historical fault occurrence patterns of ring main unit elbow joints is incorporated, making probabilistic inference not only dependent on current observation data but also based on historical statistics, thus improving the rationality of the inference. By combining the likelihood function with prior probability using Bayes' formula, the posterior probability of each fault stage is calculated, quantifying the credibility of the current observation for each fault stage, and providing accurate probabilistic basis for subsequent threshold-based fault diagnosis. This process breaks through the limitations of traditional single threshold judgment or independent parameter analysis, achieving deep integration of observation data, historical experience, and fault evolution patterns through probabilistic statistical methods, significantly improving the scientificity and reliability of fault stage identification, and laying a core foundation for outputting high-confidence diagnostic results.

[0141] The method described in the above embodiments will be further described in detail below.

[0142] In this embodiment, the non-contact multi-parameter elbow joint fault diagnosis device will be specifically integrated into an electronic device, with the electronic device being a terminal, for illustration.

[0143] A non-contact, multi-parameter elbow joint fault diagnosis method, the specific process of which is as follows:

[0144] The probe synchronously samples to obtain the raw multi-channel data Xraw(t) (under the same time reference);

[0145] Preprocessing (denoising, zero bias correction, filtering) → X(t);

[0146] Apply the spatiotemporal coupling compensation matrix (STCM) → X'(t) (normalized value after compensation);

[0147] Feature extraction (trend slope, derivative, instantaneous pulse count, band energy, etc.) → feature vector Φ(t);

[0148] E-delay / delay estimation (calculate τi between channels);

[0149] Calculate the weighted time distance D (a measure of multivariate dynamic offset);

[0150] Map D to the Fault Evolution Chain (FEC) and compute the posterior probability P(Fk|D,Φ);

[0151] Output the judgment (stage label) and confidence level, and trigger a warning / alarm if necessary.

[0152] Assume the original multi-parameter vector obtained by the system sampling is:

[0153]

[0154] in:

[0155] : The 6-dimensional original multimodal signal column vector at time t;

[0156] Heat (temperature) represents the local temperature detected by the probe at time t, in °C.

[0157] : relative humidity, representing the relative humidity of the air in the monitored area at time t, in %RH;

[0158] Electricity represents the intensity of the local electric field or discharge signal at time t, which can be the electric field amplitude (V / m) or the instantaneous voltage of the arc discharge.

[0159] : Light, representing the intensity of ultraviolet / visible radiation or pulse count at time t, with units of pulse / s or mV;

[0160] : gas, representing the concentration of insulating decomposition gas (such as CO, VOC, etc.) or the output value of the gas sensor at time t, in ppm;

[0161] Flame: indicates the intensity of the flame or infrared radiation signal at time t, in mV or radiation power (W / m²).

[0162] The vector For a moment The six-modal raw signal input, synchronously acquired by the probe, is used for subsequent spatiotemporal compensation and fault evolution analysis.

[0163] Define the compensated vector for:

[0164]

[0165] in, Let be the compensated multi-parameter signal vector at time t. For dynamic compensation matrix, This indicates that instantaneous environmental compensation is performed on the signal at the current moment. For time delay variables, It is a hysteresis kernel function. for The preprocessed multi-parameter signal vector at time step, This represents the weighted cumulative effect on historical multimodal signals.

[0166] The temporal memory effect is represented by two parts: an instantaneous multiplier matrix term plus a convolutional hysteresis term. For ease of implementation and patent disclosure, we present a discretized implementation (sampling interval Δt):

[0167]

[0168] in, This represents the 6-dimensional multimodal signal column vector after compensation at the nth sampling point;

[0169] It is a 6×6 dynamic compensation matrix;

[0170] This represents the original 6-dimensional multimodal signal column vector at the nth sampling point;

[0171] Indicates the maximum hysteresis window size;

[0172] Let be the 6×6 hysteresis matrix for the k-th hysteresis window;

[0173] This represents the original 6-dimensional multimodal signal column vector of the nk-th sampling point.

[0174] Set M as the baseline identity matrix plus a small temperature and humidity dependent offset matrix:

[0175]

[0176] in: It is a 6×6 dynamic compensation matrix;

[0177] It is a 6th-order identity matrix;

[0178] It is a humidity scaling function;

[0179] It is a temperature scaling function;

[0180] This is a 6×6 humidity cross-sensitivity coefficient matrix;

[0181] It is a 6×6 temperature cross-sensitivity coefficient matrix;

[0182] This represents the actual relative humidity.

[0183] For reference relative humidity;

[0184] This refers to the actual temperature.

[0185] This is a reference temperature.

[0186] and This is a scaling function that represents the intensity of the effect when relative humidity / temperature deviates from the reference value; it can be in linear or saturated form.

[0187]

[0188] in, Represents the humidity scaling function; Indicates the intensity coefficient of the effect of humidity; Indicates the humidity saturation coefficient; This indicates the actual relative humidity; This indicates the reference relative humidity.

[0189] Specific JH elements This indicates the influence coefficient (cross sensitivity) of the original channel j on channel i when humidity changes. For example, increased humidity can cause VOC sensor baseline drift, so the corresponding negative / positive values ​​are filled in the corresponding row and column of M.

[0190] K[k] is used to describe the historical effects of airflow delay and sensor response time, such as the delay caused by the sample in the microfluidic cavity lingering before entering different sensors. Each element of K[k] is obtained through system identification / calibration and can be approximated using an exponential decay form:

[0191]

[0192] in, This represents the 6×6 hysteresis matrix for the k-th hysteresis window; Indicates the global scaling parameter; Indicates the attenuation coefficient; Indicates the lagging window index; This represents a 6×6 directional coefficient matrix.

[0193] The process of constructing M and K requires a calibration method:

[0194] Static baseline measurement: Under a clean atmosphere, take 5–7 sampling points for temperature T and humidity H respectively (covering the expected working range, such as 0–80℃, 10–95%RH), record the steady-state response of X[n], and obtain the baseline X0(Hi,Tj).

[0195] Cross-sensitivity test: At a fixed temperature, humidity was varied and VOC / CO samples of known concentration were injected. The baseline drift between channels was recorded, and the JH elements were fitted using the least squares method (i.e., the sensitivity coefficient was solved for each pair of i,j).

[0196] Dynamic delay identification: Inject step gas concentration pulses, record the response curves of different channels, and estimate the K[k] series through system identification.

[0197] Verification and iteration: Apply the obtained M and K to the validation set (different temperatures and humidity, different gas spectra) to evaluate the stability and consistency of the corrected signal. If the error exceeds the set limit, repeat the calibration and introduce nonlinear terms (such as quadratic correction) or saturation functions.

[0198] Online adaptive calibration: During runtime, track regression or Kalman filtering is used to periodically fine-tune J and K to adapt to sensor drift or contamination.

[0199] Traditional threshold methods trigger alarms based solely on a single parameter exceeding a set value, neglecting the sequential causal relationships between different physical parameters. Cable terminal faults are often a gradual evolutionary process, with physical mechanisms including localized overheating, insulation oxidation, decomposition and gas generation, micro-discharge, smoke generation, and ultimately, arc breakdown. This invention utilizes the temporal coupling characteristics of multimodal data to construct a fault chain model, replacing single-point judgments with time-series pattern matching, thereby achieving hierarchical diagnosis of the entire fault process.

[0200] This module mainly extracts multiple features from the compensated time series X'[n] defined in the second module to form a feature vector Φ[n]. Commonly used features include, but are not limited to:

[0201] Instantaneous value: Current value of each channel

[0202] First-order difference (rate):

[0203] Sliding window slope:

[0204] Instantaneous pulse counting: (Unit: pulse / s)

[0205] Spectral energy: (Calculated via short-time Fourier transform)

[0206] Cross-channel correlation coefficient: (Short window related to Pearson)

[0207] Overall eigenvector writing:

[0208]

[0209] Index i corresponds to different channels.

[0210] in, This represents the feature vector of the nth sampling point; This represents the instantaneous value of the i-th channel after compensation; This represents the first-order difference after compensation for the i-th channel; This represents the slope of the sliding window after compensation for the i-th channel; Indicates the ultraviolet pulse count; This represents the energy spectrum of the smoke. Let represent the short-window Pearson correlation coefficient between channel i and channel j.

[0211] The time delay τi,j (i.e., the response delay of channel j to channel i) is calculated using the short-time cross-correlation function:

[0212]

[0213] in, This represents the response time delay of the j-th channel to the i-th channel; Indicates the candidate value for time delay; Indicates the candidate value for the maximum time delay; This represents the function for calculating the Pearson correlation coefficient. This represents the signal after compensation for the i-th channel; This represents the signal after compensation for the j-th channel.

[0214] In the implementation, τ is calculated for a certain target parameter (such as VOC) and other parameters to obtain the offset used for subsequent D calculations.

[0215] Let D represent the "dynamic offset" between the current observation and the typical pattern of each fault stage. There are K fault stages (F1………FK), and each stage obtains a set of typical feature trajectories during training / calibration. For the current time n, the weighted time distance relative to stage k is defined as:

[0216]

[0217] Indicator set This indicates the selected feature (such as VOC slope, CO rise rate, UV pulse count, smoke energy, etc.).

[0218] in, This represents the weighted time distance between the nth sampling point and the kth fault stage; Indicates the fault stage index; Indicates feature index; Represents the selected feature set; This represents the weight of the m-th feature; This represents the m-th feature observation value of the n-th sampling point; This represents the m-th typical characteristic trajectory of the k-th fault stage; This represents the typical time delay of the m-th feature under the k-th fault stage; This represents the set of typical feature vectors for the k-th fault stage.

[0219] If only a single time point comparison is used, the above formula is the weighted absolute difference; in practice, a sliding window integral form can be used to increase robustness:

[0220]

[0221] in, Indicates the size of the sliding window; Indicates the index within the sliding window; This represents the time decay weight.

[0222] With the value of Dk, the subsequent judgment consists of two steps: probability mapping and threshold decision. Treating Dk as the observation distance under the conditions of this stage, we construct the likelihood function:

[0223]

[0224] in, Let F_k represent the likelihood probability of observing a distance D_k given the k-th fault stage F_k; This represents the k-th fault stage; Let D_k represent the variance of D_k under the k-th failure stage; Let D_k represent the mean value of D_k under the k-th fault stage; This represents the weighted time distance relative to the k-th fault stage; This represents an exponential function.

[0225] Calculate the posterior probability using Bayes' theorem:

[0226]

[0227] in, Let F_k represent the posterior probability of the k-th fault stage occurring given an observation distance D_k; Represents the likelihood probability; This represents the prior probability of the k-th failure stage; Indicates the fault stage index; Indicates the total number of failure stages; This represents the weighted time distance relative to the k-th fault stage.

[0228] Decision threshold and confidence output:

[0229] If the posterior of k at a certain stage (For example, 0.9), then this stage alarm will be triggered and "extremely high confidence" will be output. This is the alarm probability threshold. The alarm probability threshold can be used to determine whether a fault has occurred. When a fault is determined to have occurred, a fault alarm can be issued.

[0230] like (For example, 0.7–0.9) will trigger an early warning, prompting manual inspection. The warning probability threshold is used to identify potential fault risks and provide early warnings before faults occur. If potential fault risks are not identified and detected, they will lead to faults during subsequent use. Therefore, when a potential fault risk is identified through the warning probability threshold, a manual inspection prompt is triggered to achieve early intervention in the fault.

[0231] If all posterior values ​​are low, continue observation or output "normal / low confidence".

[0232] In addition, the posterior time can be smoothed across multiple time points (e.g., by exponential weighted averaging) to avoid false alarms caused by sudden noise.

[0233] The weights wm and the typical trajectory f{(k)} can be learned from the training set. Common methods include:

[0234] Collect a large number of fault evolution samples (laboratory-induced and field-based historical faults) and label the data according to stages.

[0235] For each stage, the characteristic mean trajectory and time delay τm{(k)} are calculated (based on cross correlation).

[0236] Use an optimization objective (such as cross-entropy or maximum a posteriori likelihood) to solve for wm, so that the classification / posterior probabilities are optimal on the validation set. L1 / L2 regularization can also be used to control the sparsity of the weights.

[0237] A confidence score is calculated for each diagnostic result. This score is based on a Bayesian statistical model and is determined by feature matching degree, signal correlation, and environmental stability. When multiple parameters simultaneously conform to the fault evolution trend, the confidence score automatically increases, thereby reducing false alarms and missed alarms.

[0238] As shown above, by using a pre-configured non-contact multi-parameter probe, raw multi-parameter signals are collected within the same time reference and physical micro-environment. This integrates multi-parameter sensing units such as heat, humidity, electricity, light, gas, and flame within a millimeter-level space, eliminating the spatial asynchrony problem caused by the different installation positions of traditional distributed sensors at the physical level. This ensures that multi-parameter data originates from the same monitoring scenario, laying a data foundation for accurate diagnosis. Through preprocessing steps such as noise reduction, zero-bias correction, and filtering, interference noise in the raw signal is effectively eliminated, improving signal purity. Then, a spatiotemporal coupling compensation matrix is ​​applied to compensate for the preprocessed signal, correcting signal deviations caused by temperature and humidity cross-sensitivity, airflow delay, and sensor response lag. This achieves signal self-calibration under dynamic environments, ensuring the accuracy of multi-parameter signals. The system achieves high accuracy and consistency by extracting multi-dimensional feature vectors containing instantaneous values, first-order differences, and sliding window slopes, and calculating the response delay of each channel. This comprehensively captures the signal change patterns and causal relationships between parameters during the fault evolution process, overcoming the limitations of traditional single-feature analysis. Based on the feature vectors and response delays, a weighted time distance is calculated, mapped to a preset fault evolution chain, and the posterior probability is calculated. The fault evolution chain model is used to reconstruct the progressive fault process of the cable terminal from thermal degradation to discharge breakdown. A statistical probability model replaces the single threshold judgment, improving the accuracy of early fault identification. By comparing the posterior probability with the preset threshold, the diagnostic results and confidence levels are output, reducing false alarms and missed alarms. Ultimately, this achieves accurate detection and graded diagnosis of elbow joint area faults at a local spatial scale.

[0239] To better implement the above methods, this application also provides a non-contact multi-parameter elbow joint fault diagnosis device. This non-contact multi-parameter elbow joint fault diagnosis device can be integrated into an electronic device, such as a terminal or server. The terminal can be a mobile phone, tablet computer, smart Bluetooth device, laptop computer, or personal computer; the server can be a single server or a server cluster composed of multiple servers.

[0240] Figure 2 This is a schematic diagram of the non-contact multi-parameter elbow joint fault diagnosis device provided in an embodiment of this application. Figure 2 As shown, the non-contact multi-parameter elbow joint fault diagnosis device may include a non-contact multi-parameter probe 201 and a diagnosis unit 202, as follows:

[0241] (1) Non-contact multi-parameter probe 201;

[0242] The non-contact multi-parameter probe 201 is used to acquire multi-parameter raw signals in the area where the elbow joint is located, under the same time reference and the same physical microenvironment.

[0243] (2) Diagnostic unit 202;

[0244] The diagnostic unit 202 is used to denoise, perform zero-bias correction, and filter on the original multi-parameter signal to obtain a preprocessed multi-parameter signal; apply a spatiotemporal coupling compensation matrix to compensate the preprocessed multi-parameter signal to obtain a compensated multi-parameter signal; extract features from the compensated multi-parameter signal to obtain feature vectors, and calculate the response delay between the channels of each parameter signal; calculate the weighted time distance between the current observation stage and the typical modes of each fault stage based on the feature vectors and the response delays, and map the weighted time distance to a preset fault evolution chain to calculate the posterior probability of the typical modes of each fault stage through the fault evolution chain; and output the fault diagnosis result of the elbow joint area and the confidence level corresponding to the fault diagnosis result based on the comparison result of the posterior probability and the preset fault threshold.

[0245] In some embodiments, the non-contact multi-parameter probe adopts a three-dimensional stacked modular structure, and the non-contact multi-parameter probe includes, from top to bottom, an optical sensing module, a gas detection module, an environmental parameter monitoring module, and an electromagnetic shielding heat dissipation module.

[0246] The optical sensing module includes a sapphire optical window, an ultraviolet detector, and a laser scattering optical component.

[0247] The surface of the sapphire optical window is coated with a double-layer interference film;

[0248] The ultraviolet detector and the laser scattering optical component share the sapphire optical window to form a coaxial optical path.

[0249] The modular structure of three-dimensional stacking is a design that integrates components with different functions by stacking them in a three-dimensional spatial order. By utilizing vertical space instead of horizontal tiling, the overall volume can be significantly reduced, while achieving close collaboration among modules. It is a core structural design method for miniaturized and highly integrated devices, and is widely used in precision device fields such as MEMS and sensor arrays. Specifically, in this embodiment, the structure is manifested as an optical sensing module, a gas detection module, an environmental parameter monitoring module, and an electromagnetic shielding and heat dissipation module stacked sequentially from top to bottom inside a non-contact multi-parameter probe. Modular integration is achieved using MEMS technology, ensuring that the probe's outer diameter does not exceed 25mm and its length does not exceed 60mm, achieving multi-parameter collaborative detection in an extremely small space.

[0250] The optical sensing module is a functional module used to detect optical signals (such as ultraviolet, visible light, infrared, and scattered light). It typically integrates an optical window, a photodetector, and optical conduction / focusing components. It converts photophysical signals into electrical signals and is a core component for optical parameter monitoring, widely used in scenarios such as discharge detection and smoke detection. Specifically, in this embodiment, this module is specifically used to detect ultraviolet light generated by partial discharge from elbow joints and scattered light from smoke particles generated by insulation decomposition. Its core components include a sapphire optical window, a solar-blind UVC ultraviolet detector, and a laser scattering optical component. These three components work together to achieve high signal-to-noise ratio optical signal acquisition.

[0251] The gas detection module is a functional module used to identify gas components and detect gas concentrations. It typically integrates gas-sensitive elements and gas flow channels, converting gas chemical signals into electrical signals. It is a key component for monitoring gas parameters and is widely used in scenarios such as equipment insulation degradation and fault-induced gas generation. Specifically, in this embodiment, this module shares the same microfluidic cavity with the optical sensing module, and incorporates a metal-oxide-semiconductor gas-sensitive array and an electrochemical unit, enabling rapid response to VOCs, etc. It can accurately detect volatile products and harmful gases such as CO, enabling precise detection of gases that decompose insulation.

[0252] The environmental parameter monitoring module is a functional module used to detect key physical parameters (such as temperature and humidity) in the equipment's operating environment. It integrates corresponding environmental sensors and provides environmental background data for core parameter monitoring, correcting measurement deviations caused by environmental interference. It is an auxiliary module for improving monitoring accuracy. Specifically, in this embodiment, the module includes a thin-film thermistor and a thin-film capacitive humidity sensor, coupled to a copper-based heat sink at the bottom of the probe via a thermal bridge, accurately detecting the temperature and humidity of the monitored area and providing temperature and humidity parameter support for the spatiotemporal coupling compensation matrix.

[0253] Among them, the electromagnetic shielding and heat dissipation module is a composite module that combines electromagnetic interference shielding and heat dissipation functions. It typically integrates conductive / magnetic shielding materials and heat dissipation structures, which can block the influence of external electromagnetic interference on internal components and simultaneously dissipate internally generated heat, ensuring the stable operation of the module in complex environments. It is widely used in equipment in high-voltage and strong electromagnetic interference scenarios. Specifically, in this embodiment, the module uses a copper substrate as the main body, with thermally conductive through holes and a magnetic ring shield embedded inside. The inner wall of the cavity is coated with a micron-level conductive carbonized film, which can shield high-frequency electromagnetic interference generated by electric arcs or partial discharges and dissipate excess heat inside the probe, ensuring the stable operation of each sensing module.

[0254] The sapphire optical window is an optically transparent component made of sapphire material. Sapphire has high light transmittance (especially in the ultraviolet band), high temperature resistance, wear resistance, and corrosion resistance, providing a stable channel for optical signal transmission while protecting internal optical components. It is widely used in optical monitoring equipment in harsh environments. Specifically, in this embodiment, the window is approximately 0.6 mm thick, located on the top layer of the probe, providing a common optical channel for the ultraviolet detector and laser scattering components. It directly contacts the internal environment of the ring main unit, requiring a balance between light transmittance and contamination resistance.

[0255] Among them, the ultraviolet (UV) detector is a device specifically designed to detect UV light signals. It converts UV light intensity into electrical signals and can be classified according to the detection wavelength into solar-blind (responding only to deep UV light) and near-UV types. Solar-blind detectors avoid interference from UV light in sunlight and are suitable for partial discharge detection in electrical equipment. Specifically, in this embodiment, the detector is a solar-blind UVC (deep ultraviolet) detector, specifically designed to respond to the deep UV light generated by partial discharge at the elbow joint, and is not affected by ambient visible light, ensuring the specificity of discharge signal detection.

[0256] The laser scattering optical component is an optical part used to emit laser light and receive scattered light. It typically integrates a laser emitter, collimating lens, scattered light receiving lens, and optical damping cavity. By detecting the scattered light signal from the laser beam, the presence and concentration of particles can be determined, and it is widely used in particle detection scenarios such as smoke and dust. Specifically, in this embodiment, the component includes a miniature collimating lens and an optical damping cavity. The emitted laser beam, after collimation, enters the optical damping cavity and interacts with smoke particles in the microfluidic cavity to generate scattered light. By receiving the scattered light, high signal-to-noise ratio detection of insulating decomposed solid particles is achieved.

[0257] Among them, the double-layer interference film is an optical thin film (such as oxide or nitride) deposited on the surface of an optical component using a vacuum deposition process. It utilizes the principle of light interference to enhance the transmission of light in a specific wavelength band and block light in other wavelength bands. It is a key technology for optimizing the spectral characteristics of optical windows and is widely used in optical filtering, anti-reflection, and other applications. Specifically, in this embodiment, the thin film is... The dual-layer structure, coated on the surface of the sapphire optical window, has the core function of blocking visible light and enhancing the transmittance of ultraviolet light and laser scattering wavelengths, while also improving the anti-fouling and self-cleaning properties of the window surface.

[0258] Sharing a sapphire optical window to form a coaxial optical path allows multiple optical detection components to share the same optical window and position the optical paths of each component on the same axis (or parallel coaxially). This reduces the number of optical windows and the space occupied by the optical path, while ensuring consistency in the detection areas of different optical parameters and improving the synergy of multiple optical parameters. It is a common design approach for miniaturized optical modules. Specifically, in this embodiment, the ultraviolet detector and the laser scattering optical component share the same sapphire optical window, and the optical path is designed to be coaxial. This ensures that the detection areas of ultraviolet light detection and laser scattering detection highly overlap, guaranteeing spatial consistency between the optical signal and the subsequent gas signal.

[0259] In some embodiments, the gas detection module includes:

[0260] The microfluidic cavity has its inlet located on the side wall of the non-contact multi-parameter probe and its outlet located below the non-contact multi-parameter probe. The microfluidic cavity includes a pulse delay cavity structure, which is used to form a brief dead zone so that the optical sensing module, gas detection module, environmental parameter monitoring module and electromagnetic shielding heat dissipation module generate a time offset for the same sample.

[0261] Metal-oxide-semiconductor gas-sensitive arrays are used to detect volatile organic compounds and volatile gaseous products.

[0262] Electrochemical unit used for detecting carbon monoxide.

[0263] The microfluidic cavity is a miniaturized fluid containment and transport cavity, typically composed of microchannels, reaction zones, and other structures. Its small size (generally below cubic centimeters) allows for precise control of the flow, residence, and distribution of fluid samples. It is the core structure in miniature gas sensing devices that enables effective sample contact with the sensing element and is widely used in portable and embedded gas detection modules. Specifically, in this embodiment, the microfluidic cavity is the core cavity of the gas detection module, approximately 4 mm high, with a total volume of 1–2 cm³. The inlet is located on the probe sidewall, and the outlet is located below. The channel direction follows the natural thermal convection direction inside the ring main unit. All gas-sensitive elements (MOS gas-sensitive array, electrochemical unit) are integrated within the cavity, forming an intersecting layout with the optical path of the optical sensing module to ensure that the gas sample and the optical signal detection area coincide. Furthermore, the cavity includes a pulse delay cavity structure, combining sample transmission, sensing detection, and time feature construction functions.

[0264] The pulse delay cavity structure is a dead zone structure formed by localized protrusions or depressions within a microfluidic cavity. This allows the fluid sample to briefly reside in this area, resulting in a slight difference in the time it takes for sensing elements at different locations to contact the same sample. This creates a controllable time offset, which is not a measurement error but a characteristic signal used for time-series analysis. It is widely applied in micro-sensing systems that require tracing fluid transport paths or performing time-series correlation analysis. Specifically, in this embodiment, the structure creates a brief dead zone within the microfluidic cavity, causing a measurable time offset in the contact time of sensing elements from different modules, such as the optical sensing module and the gas detection module, with the same sample. This time offset is used in subsequent algorithms to determine the temporal relationship of gas sources, enabling time-series-based fault tracing analysis.

[0265] Among them, the metal-oxide-semiconductor (MOS) gas-sensitive array is an array-type sensing structure composed of multiple MOS gas-sensitive elements of different materials or structures. Each element has different sensitivities to different volatile gases. By combining the overall response modes of the array, the gas type can be identified and the concentration quantified. It features fast response speed, low cost, and small size, and is widely used in multi-component gas detection scenarios. Specifically, in this embodiment, the array adopts a MEMS packaging form, and the chip spacing and exposed area have been calibrated and optimized. It is specifically designed for rapid response to volatile organic compounds, hydrogen, and other volatile products generated by insulation degradation, while taking into account detection sensitivity, repeatability, and stability.

[0266] The electrochemical unit is a gas detection element designed based on the principle of electrochemical redox reactions. It typically consists of a working electrode, a reference electrode, an auxiliary electrode, and an electrolyte. It exhibits strong selectivity and detection accuracy for specific gases, enabling quantitative detection of low-concentration harmful gases. This overcomes the shortcomings of metal oxide semiconductor gas sensors in terms of selectivity for certain gases, and it is widely used in high-precision gas detection systems. Specifically, in this embodiment, the electrochemical unit is an independently configured high-precision detection component specifically designed for the precise quantitative detection of carbon monoxide generated during insulation decomposition. It complements the MOS gas-sensitive array, enhancing the comprehensiveness and reliability of fault gas detection.

[0267] In some embodiments, the environmental parameter monitoring module includes:

[0268] A thin-film thermistor, wherein the thin-film thermistor is thermally coupled to the bottom of the non-contact multi-parameter probe via a thermal bridge;

[0269] Thin-film capacitive humidity sensor, used to detect ambient humidity;

[0270] A resistance temperature detector is used to assist the thin-film thermistor in detecting ambient temperature.

[0271] Thin-film thermistors are temperature-sensing elements fabricated on an insulating substrate using thin-film deposition technology. They are characterized by small size, low heat capacity, fast response speed, and wide temperature measurement range. They can quickly sense temperature changes and convert them into resistance changes, and are widely used in miniaturized, high-precision temperature monitoring scenarios. Specifically, in this embodiment, the thin-film thermistor is the core temperature detection element of the environmental parameter monitoring module. It achieves thermal coupling with the copper-based heat sink at the bottom of the probe through a thermal bridge, ensuring that the measured temperature accurately reflects the actual thermal state of the cable joint. At the same time, the air isolation layer design avoids interference from temperature drift of the gas sensing cavity.

[0272] A thermal bridge, in this context, refers to a short thermal resistance path structure built between two thermal components. Typically made of high thermal conductivity materials (such as copper or aluminum), it can rapidly transfer heat, achieving thermal synchronization between the two components. Furthermore, the direction of heat conduction can be controlled through structural design, making it widely used in micro-sensing systems requiring precise thermal coupling. Specifically, in this embodiment, the thermal bridge is a short thermal resistance structure made of a thin copper sheet. One end is connected to a thin-film thermistor, and the other end is connected to a copper-based heat sink at the bottom of the probe. This allows the thermistor to quickly sense temperature changes at the cable connector. Simultaneously, an air isolation layer is placed below the thermal bridge to block heat conduction to the gas sensing cavity, preventing temperature detection from interfering with gas measurement.

[0273] Bottom thermal coupling refers to achieving a tight thermal connection between the temperature sensing element and a high thermal conductivity component (such as a metal substrate) at the bottom of the device through a specific structure. This creates a stable heat transfer path between the temperature sensing element and the object being measured, ensuring that the measured temperature matches the actual temperature of the object. This is a key design feature for improving the accuracy of temperature measurements. Specifically, in this embodiment, bottom thermal coupling is achieved by connecting a thin-film thermistor to a copper-based heat sink at the bottom of the probe via a thermal bridge. The copper-based heat sink then undergoes thermal coupling with the ring network cabinet through a path with limited thermal resistance, ultimately achieving temperature synchronization between the temperature sensing element and the cable connector, thus ensuring the accuracy of temperature detection.

[0274] Among them, resistance temperature detectors are temperature sensing elements designed based on the physical property that the resistance value of a conductor or semiconductor changes with temperature. They are typically made of materials such as platinum and nickel, and feature high measurement accuracy, good stability, and excellent linearity. They can provide accurate temperature data over a wide temperature range and are often used as auxiliary temperature sensing elements in conjunction with other thermistors to improve the reliability and accuracy of the temperature measurement system. Specifically, in this embodiment, the resistance temperature detector is a platinum resistance type, working in conjunction with a thin-film thermistor, specifically for assisting in the detection of ambient temperature. Data fusion corrects the measurement deviation of the thin-film thermistor, further improving the accuracy and stability of temperature detection and avoiding detection failure caused by the malfunction of a single temperature sensing element.

[0275] In some embodiments, the electromagnetic shielding heat dissipation module includes:

[0276] Magnetic ring shielding is used to shield high-frequency electromagnetic interference generated by electric arc or partial discharge;

[0277] A micron-sized conductive carbon film is coated on the inner wall of the non-contact multi-parameter probe. The micron-sized conductive carbon film is used to absorb local charges to reduce the transient impact of discharge pulses on gas signals and optical signals.

[0278] The magnetic ring shield is a ring-shaped electromagnetic shielding element made of soft magnetic material. Based on the principles of electromagnetic induction and hysteresis loss, it can strongly absorb and attenuate high-frequency electromagnetic waves, blocking their penetration and propagation. It is a core component in electronic equipment for suppressing high-frequency electromagnetic interference and is widely used in precision sensing systems in strong electromagnetic environments. Specifically, in this embodiment, the magnetic ring shield is embedded inside the copper substrate of the electromagnetic shielding heat dissipation module, spaced apart from the thermally conductive vias. It is specifically designed to shield against high-frequency electromagnetic interference generated by electric arcs or partial discharges within the ring main unit, preventing interference signals from intruding into the circuit systems of optical sensing, gas detection, and other modules, thus ensuring the stability of electrical signal transmission.

[0279] Among them, the micron-level conductive carbon film is a carbon-based conductive thin film with a thickness on the order of micrometers (typically 1-10 μm). It has excellent conductivity, charge adsorption capacity, and chemical stability. It can quickly capture and dissipate locally accumulated static electricity or transient charge, reducing the impact of electromagnetic pulses on sensitive elements. It is often used for internal electromagnetic protection of precision sensing equipment, making up for the shortcomings of traditional metal shielding in suppressing transient pulses. Specifically, in this embodiment, the film is uniformly coated on the inner wall of the cavity of the non-contact multi-parameter probe with a thickness of about 3 μm. It forms a dual protection of active adsorption and passive shielding with the magnetic ring shield, specifically absorbing the transient charge generated by partial discharge, weakening the transient impact of discharge pulses on components such as gas sensors and ultraviolet detectors, and avoiding signal spike distortion.

[0280] Furthermore, embodiments of this application also provide an electronic device, including a processor and a memory, wherein the memory stores an application program, and the processor is used to run the application program in the memory to execute the non-contact multi-parameter elbow joint fault diagnosis method provided in embodiments of this application.

[0281] In practice, each of the above units can be implemented as an independent entity or can be arbitrarily combined to be implemented as the same or several entities. For the specific implementation of each of the above units, please refer to the previous method embodiments, which will not be repeated here.

[0282] As can be seen from the above, the non-contact multi-parameter elbow joint fault diagnosis device in this application embodiment acquires multi-parameter raw signals in the same time reference and the same physical micro-environment through a pre-configured non-contact multi-parameter probe. This integrates multi-parameter sensing units such as heat, humidity, electricity, light, gas, and flame into a millimeter-level space, eliminating the spatial asynchrony problem caused by the different installation positions of traditional distributed sensors at the physical level. This ensures that the multi-parameter data originates from the same monitoring scene, laying a data foundation for accurate diagnosis. Through preprocessing steps of denoising, zero-bias correction, and filtering, interference noise in the raw signal is effectively removed, improving signal purity. Then, a spatiotemporal coupling compensation matrix is ​​applied to compensate the preprocessed signal, correcting signal deviations caused by temperature and humidity cross-sensitivity, airflow delay, and sensor response lag, achieving signal self-regulation in dynamic environments. Calibration ensures the accuracy and consistency of multi-parameter signals. By extracting multi-dimensional feature vectors including instantaneous values, first-order differences, and sliding window slopes, and calculating the response delay of each channel, the signal change patterns and causal relationships between parameters during the fault evolution process are comprehensively captured, overcoming the limitations of traditional single-feature analysis. Based on the feature vectors and response delays, a weighted time distance is calculated, mapped to a preset fault evolution chain, and the posterior probability is calculated. The fault evolution chain model is used to reconstruct the progressive fault process of the cable terminal from thermal degradation to discharge breakdown. The statistical probability model replaces the single threshold judgment, improving the accuracy of early fault identification. By comparing the posterior probability with the preset threshold, the diagnostic results and confidence levels are output, reducing false alarms and missed alarms. Ultimately, accurate detection and graded diagnosis of elbow joint area faults at the local spatial scale are achieved.

[0283] The foregoing has provided a detailed description of a non-contact, multi-parameter elbow joint fault diagnosis method and apparatus provided in the embodiments of this application. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A non-contact, multi-parameter elbow joint fault diagnosis method, characterized in that, include: Using a pre-configured non-contact multi-parameter probe, multi-parameter raw signals are collected in the area where the elbow joint is located, under the same time reference and in the same physical microenvironment. The signal types of the multi-parameter raw signals include heat, humidity, electricity, light, gas, and flame. The original multi-parameter signal is denoised, zero-biased corrected, and filtered to obtain a preprocessed multi-parameter signal; The preprocessed multi-parameter signal is compensated by applying a spatiotemporal coupling compensation matrix to obtain a compensated multi-parameter signal. The spatiotemporal coupling compensation matrix is ​​used to dynamically calibrate the signal through matrix operations to address signal distortion caused by temperature and humidity cross-sensitivity, signal transmission delay, and sensor response lag. Feature extraction is performed on the compensated multi-parameter signal to obtain a feature vector, and the response time delay between the channels of each parameter signal is calculated; Based on the feature vector and the response delay, the weighted time distance between the current observation stage and the typical modes of each fault stage is calculated, and the weighted time distance is mapped to a preset fault evolution chain, so as to calculate the posterior probability of the typical modes of each fault stage through the fault evolution chain. Based on the comparison between the posterior probability and the preset fault threshold, the fault diagnosis result of the area where the elbow joint is located and the confidence level corresponding to the fault diagnosis result are output.

2. The non-contact multi-parameter elbow joint fault diagnosis method according to claim 1, characterized in that, The spatiotemporal coupling compensation matrix includes a dynamic compensation matrix and a hysteresis matrix. The application of the spatiotemporal coupling compensation matrix to compensate the preprocessed multi-parameter signal yields a compensated multi-parameter signal, including: The preprocessed multi-parameter signal is compensated using the dynamic compensation matrix and the hysteresis matrix to obtain the compensated multi-parameter signal. The compensated multi-parameter signal is represented as follows: in, For the compensated multi-parameter signal at the nth sampling point, For dynamic compensation matrix, This represents the relative humidity signal at the nth sampling point. For the temperature signal of the nth sampling point, This is the preprocessed multi-parameter signal for the nth sampling point. The size of the maximum hysteresis window. Let be the hysteresis matrix of the k-th hysteresis window. This is the preprocessed multi-parameter signal of the nk-th sampling point; The dynamic compensation matrix is ​​obtained by adding a humidity-dependent offset matrix and a temperature-dependent offset matrix to a 6th-order identity matrix. The dynamic compensation matrix is ​​used to correct the signal deviation caused by temperature and humidity cross-sensitivity. The hysteresis matrix adopts an exponential decay form and is used to characterize the impact of airflow delay and sensor response time on historical signals.

3. The non-contact multi-parameter elbow joint fault diagnosis method according to claim 1, characterized in that, The step of extracting features from the compensated multi-parameter signal to obtain a feature vector includes: Extract the instantaneous values ​​of each compensated parameter signal from its channels; Calculate the first-order difference of each channel of the compensated parameter signal; Calculate the slope of the sliding window for each compensated parameter signal channel; Count the instantaneous pulses of the ultraviolet signal channel; The spectral energy of the smoke signal is calculated using short-time Fourier transform. Calculate the short-window Pearson correlation coefficient between any two channels of the compensated parametric signals; The feature vector is obtained based on the channel instantaneous value, the channel first-order difference, the channel sliding window slope, the instantaneous pulse count, the spectral energy, and the short-window Pearson correlation coefficient.

4. The non-contact multi-parameter elbow joint fault diagnosis method according to claim 1, characterized in that, The step of calculating the weighted time distance between the current observation stage and the typical modes of each fault stage based on the feature vector and the response delay includes: Multiple typical fault stage patterns are preset, and each typical fault stage pattern corresponds to a set of typical feature trajectories and a typical time delay. For the current observation phase, calculate the weighted time distance of the current observation phase relative to the typical mode of each fault phase. The weighted time distance is represented as follows: in, This represents the weighted time distance between the nth sampling point in the current observation phase and the typical mode of the kth fault phase. For the feature vector, Let m be the weight of the m-th feature. For the m-th feature observation value of the n-th sampling point, This is the m-th typical feature trajectory of the typical mode of the k-th fault stage. This represents the m-th typical time delay of the typical mode of the k-th fault stage. This represents the set of typical feature vectors for the typical mode of the k-th failure stage.

5. The non-contact multi-parameter elbow joint fault diagnosis method according to claim 4, characterized in that, The weighted time distance is mapped to a preset fault evolution chain to calculate the posterior probability of typical modes at each fault stage through the fault evolution chain, including: The weighted time distance is mapped to a preset fault evolution chain to construct a likelihood function, which represents the probability of observing the corresponding weighted time distance when given a typical pattern of a certain fault stage. The posterior probability is calculated using the likelihood function, and the posterior probability is expressed as follows: in, Represents the posterior probability. Represents the likelihood function. This represents the prior probability of the typical mode for the k-th failure stage. Index representing typical failure phases. This represents the total number of typical failure phases. This represents the weighted time distance of the current observation phase relative to the typical pattern of the k-th failure phase. For the first Weighted time distance observed under typical fault stage modes The conditional probability, For the first Prior probabilities of typical failure phase patterns.

6. A non-contact, multi-parameter elbow joint fault diagnosis device, characterized in that, include: A non-contact multi-parameter probe is used to collect multi-parameter raw signals in the area where the elbow joint is located, under the same time reference and the same physical microenvironment. The signal types of the multi-parameter raw signals include heat, humidity, electricity, light, gas, and flame. The diagnostic unit is used to perform denoising, zero-bias correction, and filtering on the original multi-parameter signal to obtain a preprocessed multi-parameter signal. The preprocessed multi-parameter signal is compensated by applying a spatiotemporal coupling compensation matrix to obtain a compensated multi-parameter signal. The spatiotemporal coupling compensation matrix is ​​used to dynamically calibrate the signal through matrix operations to address signal distortion caused by temperature and humidity cross-sensitivity, signal transmission delay, and sensor response lag. Feature extraction is performed on the compensated multi-parameter signal to obtain feature vectors, and the response delay between the channels of each parameter signal is calculated. Based on the feature vectors and the response delays, the weighted time distance between the current observation stage and the typical modes of each fault stage is calculated, and the weighted time distance is mapped to a preset fault evolution chain to calculate the posterior probability of the typical modes of each fault stage through the fault evolution chain. According to the comparison result between the posterior probability and the preset fault threshold, the fault diagnosis result of the elbow joint area and the confidence level corresponding to the fault diagnosis result are output.

7. The non-contact multi-parameter elbow joint fault diagnosis device according to claim 6, characterized in that: The non-contact multi-parameter probe adopts a three-dimensional stacked modular structure. From top to bottom, the non-contact multi-parameter probe includes an optical sensing module, a gas detection module, an environmental parameter monitoring module, and an electromagnetic shielding heat dissipation module. The optical sensing module includes a sapphire optical window, an ultraviolet detector, and a laser scattering optical component. The surface of the sapphire optical window is coated with a double-layer interference film; The ultraviolet detector and the laser scattering optical component share the sapphire optical window to form a coaxial optical path.

8. The non-contact multi-parameter elbow joint fault diagnosis device according to claim 7, characterized in that, The gas detection module includes: The microfluidic cavity has its inlet located on the side wall of the non-contact multi-parameter probe and its outlet located below the non-contact multi-parameter probe. The microfluidic cavity includes a pulse delay cavity structure, which is used to form a brief dead zone so that the optical sensing module, gas detection module, environmental parameter monitoring module and electromagnetic shielding heat dissipation module generate a time offset for the same sample. Metal-oxide-semiconductor gas-sensitive arrays are used to detect volatile organic compounds and volatile gaseous products. Electrochemical unit used for detecting carbon monoxide.

9. The non-contact multi-parameter elbow joint fault diagnosis device according to claim 7, characterized in that, The environmental parameter monitoring module includes: A thin-film thermistor, wherein the thin-film thermistor is thermally coupled to the bottom of the non-contact multi-parameter probe via a thermal bridge; Thin-film capacitive humidity sensor, used to detect ambient humidity; A resistance temperature detector is used to assist the thin-film thermistor in detecting ambient temperature.

10. The non-contact multi-parameter elbow joint fault diagnosis device according to claim 7, characterized in that, The electromagnetic shielding heat dissipation module includes: Magnetic ring shielding is used to shield high-frequency electromagnetic interference generated by electric arcs or partial discharges. A micron-sized conductive carbon film is coated on the inner wall of the non-contact multi-parameter probe. The micron-sized conductive carbon film is used to absorb local charges to reduce the transient impact of discharge pulses on gas signals and optical signals.

Images