Electrical equipment unmanned inspection vehicle based on multi-modal fusion and state evaluation method

Abstract

The invention provides an unmanned inspection system for electrical equipment of a transformer substation and a state evaluation method, wherein the unmanned inspection system takes an autonomous mobile inspection vehicle as a platform and integrates infrared, voiceprint, ultraviolet and visible light multi-mode perception. According to the multi-modal state evaluation method based on'equipment structure space mapping + abnormal region consistency verification + modal credibility self-adaptive correction ', health state results are in one-to-one correspondence with specific equipment structure parts, the fusion weight is automatically reduced under the condition of single-modal distortion, misjudgment caused by dislocation of multi-modal abnormal regions is eliminated, and the reliability of the multi-modal state evaluation method is improved. And the false alarm rate and the state evaluation fluctuation are obviously reduced.

Description

Technical Field

[0001] This invention relates to the field of intelligent operation and maintenance and robotic inspection technology for power systems. Specifically, it relates to an unmanned inspection system and condition assessment method for substation electrical equipment that integrates infrared, acoustic, ultraviolet and visible light multimodal sensing, using an autonomous mobile inspection vehicle as a platform. Background Technology

[0002] During long-term operation, electrical equipment such as transformers, circuit breakers, and disconnectors in substations are prone to overheating, partial discharge, insulation deterioration, and cosmetic damage due to load changes, environmental factors, or material aging.

[0003] Existing substation inspection methods mainly include manual inspection or single-sensor-assisted inspection, which have the following shortcomings: substation environments are complex and pose high-voltage risks, making it difficult for manual inspection to achieve high-frequency, full-coverage detection; relying solely on infrared, visible light, or acoustic sensors is prone to misjudgment or omission, making it difficult to accurately reflect the true health status of equipment; different modal sensors (infrared, acoustic, ultraviolet, visible light) may have spatial offsets in their observation positions of the same equipment; local environmental noise (wind noise, electromagnetic interference, reflected light) can artificially increase the probability of single-modal anomalies, thereby interfering with the fusion results.

[0004] Therefore, it is necessary to propose a comprehensive inspection system that uses unmanned inspection vehicles as the carrier, multimodal perception as the means, and state assessment as the goal. Summary of the Invention

[0005] In view of the problems existing in the prior art, the present invention provides an unmanned inspection vehicle, which includes a multimodal sensing and acquisition unit and an on-board computing and control unit. The multimodal sensing and acquisition unit includes an infrared thermal imager, an acoustic print imaging sensor, an ultraviolet imaging camera, and a high-definition visible light camera. The on-board computing and control unit preprocesses the collected data and extracts the corresponding features: infrared temperature feature vector (Fir), acoustic spectrum feature vector (Fac), ultraviolet discharge feature vector (Fuv), and appearance defect feature vector (Fvis). By inputting the above features into the multimodal fusion state evaluation model, a unified fusion feature representation is obtained: Where α1, α2, α3, and α4 represent adaptive modal attention weights, which are dynamically calculated from the signal-to-noise ratio or confidence level of each modal feature; W fusion This is the fusion transformation matrix, used to map the concatenated high-dimensional features to a unified fusion feature space dimension M; b fusion It is the bias vector; It is a nonlinear activation function to enhance the model's ability to represent complex failure modes.

[0006] Based on the fusion features, the state assessment model outputs a device health state vector.

[0007] Preferably, the infrared thermal imager is used to acquire the temperature field distribution on the surface of electrical equipment; the acoustic fingerprint sensor is used to acquire the acoustic signals generated by partial discharge; the ultraviolet imaging camera is used to detect the ultraviolet radiation characteristics generated by corona discharge; and the high-definition visible light camera is used to acquire images of the equipment's appearance.

[0008] Preferably, the infrared temperature feature vector, the acoustic spectrum feature vector, the ultraviolet discharge feature vector, and the appearance defect feature vector all carry a unified timestamp and pose information.

[0009] Preferably, the multimodal fusion includes: spatial correlation between infrared temperature anomaly regions and visible light images, correspondence between acoustic anomaly signals and equipment structural positions, and joint determination of ultraviolet discharge characteristics and equipment appearance defects.

[0010] Preferably, based on multimodal data, the on-board computing and control unit performs the following steps: S301: S302: S303: S304: S305: S306: S307: S308: S309: S300: S300: S301: S302: S303: S304: S305: S30 ...

[0011] The present invention also provides a method for assessing the condition of electrical equipment using the aforementioned inspection vehicle, the method comprising the following steps: S1: The unmanned inspection vehicle autonomously travels within the substation according to the pre-set inspection route planning information; S2: After the unmanned inspection vehicle arrives at the designated equipment area, it automatically switches to low-speed inspection mode or fixed-point parking mode. The multimodal sensing and acquisition unit jointly collects multimodal data from the same equipment under a unified time reference. S3: Based on multimodal data, the on-board computing and control unit completes data preprocessing, feature extraction, and fusion analysis in real time; The specific steps of S3 are as follows: S301: Structural space modeling of the equipment to be inspected; S302: Localization of multimodal abnormal equipment; S303: Calculation of spatial consistency constraints; S304: Adaptive correction of modal credibility; S305: Constraint fusion state evaluation.

[0012] Preferably, the infrared thermal imager is used to acquire the temperature field distribution on the surface of electrical equipment; the acoustic fingerprint sensor is used to acquire the acoustic signals generated by partial discharge; the ultraviolet imaging camera is used to detect the ultraviolet radiation characteristics generated by corona discharge; and the high-definition visible light camera is used to acquire images of the equipment's appearance.

[0013] Preferably, the infrared temperature feature vector, the acoustic spectrum feature vector, the ultraviolet discharge feature vector, and the appearance defect feature vector all carry a unified timestamp and pose information.

[0014] Preferably, the multimodal fusion includes: spatial correlation between infrared temperature anomaly regions and visible light images, correspondence between acoustic anomaly signals and equipment structural positions, and joint determination of ultraviolet discharge characteristics and equipment appearance defects.

[0015] The innovative aspects of the embodiments in this specification include: 1. This invention provides an unmanned inspection vehicle that can move autonomously or semi-autonomously in a substation environment; 2. The unmanned inspection vehicle provided by this invention can simultaneously collect multimodal status information of electrical equipment, and can perform fusion analysis on multi-source data to form equipment health status assessment results; 3. This invention is based on a multimodal state assessment method of "equipment structure spatial mapping + abnormal area consistency verification + modal credibility adaptive correction". It realizes the correspondence between health status results and specific equipment structure parts, automatically reduces its fusion weight in the case of single-modal distortion, eliminates misjudgment caused by multimodal abnormal area misalignment, and significantly reduces false alarm rate and state assessment fluctuation. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments or related technologies of this specification, the drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a structural schematic diagram of the unmanned inspection vehicle provided by the present invention; Figure 2 The flowchart shows the execution process of the on-board calculation and control unit during the acquisition process of the modal sensor provided by this invention.

[0018] Figure labels: 10, Infrared thermal imager; 20, Acoustic print imaging sensor; 30, Ultraviolet imaging camera; 40, High-definition visible light camera; 100, Unmanned inspection vehicle. Detailed Implementation

[0019] The technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] It should be noted that the terms "comprising" and "having," and any variations thereof, in the embodiments and drawings of this specification are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units is not limited to the steps or units listed, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or devices.

[0021] Example 1 like Figure 1 As shown, this invention provides an unmanned inspection vehicle capable of autonomous or semi-autonomous movement in a substation environment. The unmanned inspection vehicle 100 travels along a pre-set inspection channel in the substation. A multimodal sensor acquisition unit is fixedly installed at the front end and top of the vehicle body. The multimodal sensor acquisition unit includes an infrared thermal imager 10, an acoustic fingerprint sensor 20, an ultraviolet imaging camera 30, and a high-definition visible light camera 40. The installation angle and field of view of each sensor are mechanically calibrated according to the substation equipment layout to ensure that the target equipment is within the effective acquisition range during the inspection process.

[0022] Among them, the infrared thermal imager is used to collect the temperature field distribution on the surface of electrical equipment; the acoustic fingerprint sensor is used to collect the acoustic signals generated by partial discharge; the ultraviolet imaging camera is used to detect the ultraviolet radiation characteristics generated by corona discharge; and the high-definition visible light camera is used to collect images of the appearance of the equipment.

[0023] The unmanned inspection vehicle includes: a mobile chassis module, a sensor mounting mast, an onboard computing and control unit, a power supply system, and a communication and remote control module.

[0024] The mobile chassis module employs a four-wheel differential drive structure or a tracked drive structure, enabling it to adapt to substation environments with concrete floors, gravel roads, and slight inclines. The chassis is equipped with a motor drive unit, a reduction gear, and an independent suspension structure to ensure the stability of the inspection vehicle during operation and reduce the impact of vibration on sensor data acquisition accuracy. An integrated mounting platform is installed on top of the chassis to secure the sensor mast, onboard computing and control unit, and power supply system. The entire vehicle structure adopts a modular design, facilitating rapid deployment and maintenance in different substation environments.

[0025] The aforementioned sensors are fixed to the sensor mast or mobile chassis module via a unified mechanical mounting base, with the optical axis or acoustic center of each sensor pointing in the direction of the inspection vehicle's movement. This rigid constraint of the mechanical structure ensures that the relative positional relationship between the sensors remains stable during long-term operation, thus providing a foundation for the spatial alignment of subsequent multimodal data. Each sensor is connected to the mast via a vibration-damping structure to reduce the impact of mechanical disturbances generated during the inspection vehicle's movement on data acquisition.

[0026] The unmanned inspection vehicle proceeds sequentially to each inspection point according to a preset path or remote command, and simultaneously collects infrared thermal image data, acoustic signal data, ultraviolet imaging data, and visible light image data at each inspection point. All of the above data carries a unified timestamp and pose information to ensure the correspondence of multimodal data on the same device at the same time, and to avoid misjudgment caused by inconsistencies in time or space between multiple data sources.

[0027] The unmanned inspection vehicle also includes an onboard computing and control unit, which preprocesses the collected data and extracts corresponding features, such as the infrared temperature feature vector (F). ir ), Voiceprint spectral feature vector (F ac ), ultraviolet discharge eigenvector (F uv ) and appearance defect feature vector (F vis ).

[0028] By inputting the above features into the multimodal fusion state evaluation model, a unified fusion feature representation is obtained: Where F represents the multimodal fusion mapping function.

[0029] The multimodal fusion mapping function F is defined as a weighted concatenation and nonlinear transformation based on an attention mechanism, and its specific expression is as follows: Where α1, α2, α3, and α4 represent adaptive modal attention weights, which are dynamically calculated from the signal-to-noise ratio or confidence level of each modal feature; W fusion ∈ M×D This is the fusion transformation matrix, used to map the concatenated high-dimensional features to a unified fusion feature space dimension M; b fusion ∈ σ is the bias vector; σ() is a nonlinear activation function to enhance the model's ability to represent complex failure modes.

[0030] Based on fusion features (F fusion The state assessment model outputs a device health state vector: Where, p norm p represents the probability that the equipment is functioning correctly. heat p represents the probability of an overheating failure. pd p represents the probability of partial discharge. ins p represents the probability of insulation degradation / corona discharge. dmg This represents the probability of cosmetic damage.

[0031] The aforementioned components correspond to the confidence levels of abnormal states such as overheating, partial discharge, insulation degradation, and appearance damage. Based on preset thresholds and grading rules, the equipment status is graded and evaluated, preferably into normal, slightly abnormal, moderately abnormal, and severely abnormal categories.

[0032] The CHI status score is calculated using the following formula; CHI=100- in, These are the weighting coefficients. The severity coefficient is CHI; the maximum score is 100 points, and the lower the score, the worse the equipment's health condition.

[0033] Optionally, to avoid "average scores masking serious single-item failures", the system uses a combination of "weakest link" logic and CHI scores for final classification.

[0034] When CHI ≥ 90 and no single abnormal probability > 0.3, it means that the equipment operating parameters are within the standard range, there are no obvious defects, the normal inspection cycle is maintained, and the inspected equipment is normal. When 75≤CHI<90 or any single abnormal probability∈[0.3,0.5), it indicates the presence of a slight hot spot (temperature rise <10K) or slight external corrosion, but it does not affect the current operation. The inspection cycle needs to be shortened, and the equipment should be added to the watch list. The equipment is considered to be in a mild abnormality. When 60≤CHI<75 or any single abnormal probability∈[0.5, 0.8), it indicates the presence of obvious partial discharge signal, obvious overheating or insulator damage, and a defect work order needs to be generated. It is recommended to handle it in the most recent maintenance. The equipment is classified as a moderate abnormality. When CHI < 60 or the probability of any single abnormality is ≥ 0.8, it indicates that a strong arc discharge ultraviolet signal, critical overheating, or severe structural damage has been detected, requiring an immediate alarm to be triggered. It is recommended to apply for an emergency shutdown for repair, as the equipment is in serious abnormality.

[0035] The system automatically generates inspection results, which include at least: the abnormal equipment number and location, the type of abnormality and its corresponding modal source, the severity level of the abnormality, and recommended maintenance and repair measures. These inspection results can serve as a basis for subsequent predictive maintenance and equipment overhaul decisions.

[0036] The unmanned inspection vehicle provided by this invention reduces the risks of manual inspection, increases the inspection coverage, overcomes the limitations of single sensor inspection, reduces misjudgment and omission, and outputs results directly to serve operation and maintenance decisions, rather than just providing raw data. It is applicable to different types of substations and various electrical equipment.

[0037] Example 2 This embodiment provides a workflow for an unmanned inspection vehicle.

[0038] In actual inspection operations, the inspection vehicle autonomously travels within the substation according to pre-set inspection route planning information. The inspection route can be planned offline manually and then uploaded to the inspection vehicle, or it can be issued in real time by the remote control center as needed.

[0039] Upon reaching the designated equipment area, the inspection vehicle automatically switches to low-speed inspection mode or fixed-point parking mode to ensure sufficient stability of the multi-modal sensors during data acquisition. At this time, the mobile chassis remains stationary or moves at a constant, slow speed, and the sensor system starts synchronously. Multiple sensors jointly acquire data from the same device under a unified time reference. Through the coordinated design of mechanical motion control and sensor triggering logic, the consistency of data from different modalities in time and space is ensured, avoiding data mismatch problems caused by vehicle movement or positional shifts.

[0040] The unmanned inspection vehicle is equipped with an onboard computing and control unit for local processing of the collected multimodal data. This onboard computing and control unit is connected to various sensors via a high-speed data bus, enabling real-time data preprocessing, feature extraction, and fusion analysis at the inspection site.

[0041] Multimodal data fusion includes: spatial correlation between infrared temperature anomaly regions and visible light images, correspondence between acoustic signature anomaly signals and equipment structural locations, and joint determination of ultraviolet discharge characteristics and equipment appearance defects. Through mutual verification of multi-source data, misjudgments caused by environmental interference from a single sensor can be effectively avoided, improving the reliability of fault identification.

[0042] After completing the inspection task, the inspection vehicle uploads the status assessment results and raw data to the back-end management system via the wireless communication module, automatically generating a structured inspection report. The inspection report includes at least the equipment number, fault type, fault location, and risk level information, providing maintenance personnel with intuitive and traceable decision-making basis.

[0043] like Figure 2 As shown, during the data acquisition process by the multimodal sensor, the on-board computing and control unit performs the following steps: S301: Equipment Structural Space Modeling A structural model is pre-established for each type of equipment to be inspected: } Where: R i For key functional components of the equipment (sleeves, terminals, insulators, contacts, etc.), each component has a three-dimensional coordinate range and a functional label; S302: Multimodal anomaly equipment location A set of abnormal devices is generated from the detection results of each modality, and then projected onto a unified device coordinate system through the pose matrix of the inspection vehicle. S303: Spatial Consistency Constraint Calculation Calculate the overlap of different modal anomaly regions in the device model, construct a spatial consistency matrix, and obtain the spatial consistency mean. m .

[0044] S304: Adaptive Correction of Modal Reliability Based on historical stability coefficient H m Spatial consistency mean m Calculate modal confidence: R m =λ1˙SNR m +λ2˙ m +λ3˙H m Among them, SNR m This refers to the signal-to-noise ratio of the m-th mode; Normalization yields the fusion weights; α m = S305: Constraint Fusion Status Assessment A constrained fusion model is adopted: F* fusion =F(α) ir F ir , α ac F ac , α uv F uv , α vis F vis ) Output a health status vector and bind it to the device structure area to which the abnormal device belongs.

[0045] The aforementioned spatial mapping module, credibility assessment module, and fusion module are all deployed in the onboard computing unit of the inspection vehicle, making them suitable for upgrading and deploying existing multimodal inspection platforms.

[0046] Those skilled in the art will understand that the accompanying drawings are merely schematic diagrams of one embodiment, and the modules or processes shown in the drawings are not necessarily essential for implementing the present invention.

[0047] Those skilled in the art will understand that the modules in the apparatus of the embodiments can be distributed in the apparatus of the embodiments as described in the embodiments, or they can be located in one or more devices different from this embodiment with corresponding changes. The modules of the above embodiments can be combined into one module, or they can be further divided into multiple sub-modules.

[0048] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An unmanned inspection vehicle, characterized in that, The unmanned inspection vehicle includes a multimodal sensing and acquisition unit and an onboard computing and control unit. The multimodal sensing and acquisition unit includes an infrared thermal imager, an acoustic print imaging sensor, an ultraviolet imaging camera, and a high-definition visible light camera. The onboard computing and control unit preprocesses the collected data and extracts corresponding features: infrared temperature feature vector (F ir ), Voiceprint spectral feature vector (F ac ), ultraviolet discharge eigenvector (F uv ) and appearance defect feature vector (F vis ) By inputting the above features into the multimodal fusion state evaluation model, a unified fusion feature representation is obtained: Where α1, α2, α3, and α4 represent adaptive modal attention weights, which are dynamically calculated from the signal-to-noise ratio or confidence level of each modal feature; W fusion This is the fusion transformation matrix, used to map the concatenated high-dimensional features to a unified fusion feature space dimension M; b fusion σ is the bias vector; σ() is a nonlinear activation function to enhance the model's ability to represent complex failure modes. Based on the fusion features, the state assessment model outputs a device health state vector.

2. The inspection vehicle according to claim 1, characterized in that, The infrared thermal imager is used to collect the temperature field distribution on the surface of electrical equipment; the acoustic fingerprint sensor is used to collect acoustic signals generated by partial discharge; the ultraviolet imaging camera is used to detect the ultraviolet radiation characteristics generated by corona discharge; and the high-definition visible light camera is used to collect images of the equipment's appearance.

3. The inspection vehicle according to claim 1, characterized in that, The infrared temperature feature vector, the acoustic spectrum feature vector, the ultraviolet discharge feature vector, and the appearance defect feature vector all carry a unified timestamp and pose information.

4. The inspection vehicle according to claim 1, characterized in that, The multimodal fusion includes: spatial correlation between infrared temperature anomaly regions and visible light images, correspondence between acoustic anomaly signals and equipment structural locations, and joint determination of ultraviolet discharge characteristics and equipment appearance defects.

5. The inspection vehicle according to claim 1, characterized in that, Based on multimodal data, the vehicle-mounted computing and control unit performs the following steps: S301: Structural space modeling of the equipment to be inspected; S302: Localization of multimodal abnormal equipment; S303: Spatial consistency constraint measurement.

6. A method for assessing the condition of electrical equipment using any one of the inspection vehicles described in claims 1-5, characterized in that: The method includes the following steps: S1: The unmanned inspection vehicle autonomously travels within the substation according to the pre-set inspection route planning information; S2: After the unmanned inspection vehicle arrives at the designated equipment area, it automatically switches to low-speed inspection mode or fixed-point parking mode. The multimodal sensing and acquisition unit jointly collects multimodal data from the same equipment under a unified time reference. S3: Based on multimodal data, the on-board computing and control unit completes data preprocessing, feature extraction, and fusion analysis in real time; The specific steps of S3 are as follows: S301: Structural space modeling of the equipment to be inspected; S302: Localization of multimodal abnormal equipment; S303: Calculation of spatial consistency constraints; S304: Adaptive correction of modal credibility; S305: Constraint fusion state evaluation.

7. The method according to claim 6, characterized in that, The infrared thermal imager is used to collect the temperature field distribution on the surface of electrical equipment; the acoustic fingerprint sensor is used to collect acoustic signals generated by partial discharge; the ultraviolet imaging camera is used to detect the ultraviolet radiation characteristics generated by corona discharge; and the high-definition visible light camera is used to collect images of the equipment's appearance.

8. The method according to claim 6, characterized in that, The infrared temperature feature vector, the acoustic spectrum feature vector, the ultraviolet discharge feature vector, and the appearance defect feature vector all carry a unified timestamp and pose information.

9. The method according to claim 6, characterized in that, The multimodal fusion includes: spatial correlation between infrared temperature anomaly regions and visible light images, correspondence between acoustic anomaly signals and equipment structural locations, and joint determination of ultraviolet discharge characteristics and equipment appearance defects.

Images