A multi-modal data fault-tolerant fusion method for intelligent disassembly of decommissioned photovoltaic modules

Abstract

The application discloses a kind of multi-modal data fault-tolerant fusion methods of retired photovoltaic module intelligent disassembly, including the multi-modal data of disassembly target is synchronously collected, the health degree of the multi-modal data collected is evaluated, and finally multi-modal data is fault-tolerant fusion based on the evaluation result.This scheme introduces D-S method to the modal health degree evaluation under dynamic operation, by constructing evidence identification framework, fusing multi-feature index evidence body, the reliability of sensor data is accurately quantified, compared with traditional single index determination, the evaluation result is more suitable for the environmental change of photovoltaic power station field dynamic operation, can accurately identify the failure state of sensor data;Weight judgment is carried out based on the confidence and likelihood function interval obtained by evaluation, so that the system intelligently allocates the fusion weight of each sensor, so that low-quality data automatically reduces the "speaking right", combined with the cross-modal feature completion mechanism, feature reconstruction is completed when single mode fails, to ensure the continuity and effectiveness of the fusion process.

Description

Technical Field

[0001] This invention belongs to the field of photovoltaic module recycling technology and machine vision recognition, specifically involving a multimodal data fault-tolerant fusion method for intelligent dismantling of retired photovoltaic modules. Background Technology

[0002] As the core component of photovoltaic power generation, photovoltaic modules have seen rapid development in line with the "dual carbon" goal. With the advancement of new technologies, many older photovoltaic modules face the challenge of retirement and recycling. During the recycling process, intelligent dismantling and recycling robot systems for retired photovoltaic modules need to perform dynamic operations on-site at photovoltaic power plants. The vision recognition unit of the heavy-duty robotic arm, as the core component for precise cutting and grasping of modules, directly impacts the efficiency and accuracy of on-site dismantling operations due to its recognition accuracy and stability.

[0003] However, during the dynamic operation of the robot system on site, the sensor performance is easily affected by real-time changes in the environment. For example, when the vehicle enters a dark area, the binocular camera image is overexposed or underexposed, and the data reliability drops sharply. Dust and fog in the operation area cause the lidar point cloud to be sparse and highly discrete, and the proportion of effective points is greatly reduced. At the same time, the surface pollution, aging, high reflectivity and non-standard shape of the retired photovoltaic modules themselves further exacerbate the instability of sensor data.

[0004] While existing robot systems' visual recognition units attempt multimodal data fusion, modal health assessments often rely on single feature indicators, lacking scientific multi-evidence fusion assessment methods. This makes it difficult to accurately quantify the reliability of sensor data during dynamic operations. Furthermore, fusion methods often use fixed weights, making it difficult to dynamically allocate the "voice" of each sensor according to environmental changes. Low-quality data can easily cause serious interference to the fusion results, leading to decreased accuracy in component pose recognition or even recognition failure. This makes them unsuitable for the dynamic and complex operating conditions at photovoltaic power plants and fails to meet the high-precision cutting and grasping requirements of heavy-duty robotic arms, becoming a key technical bottleneck restricting the intelligent dismantling of retired photovoltaic modules on-site. Summary of the Invention

[0005] To address the aforementioned issues, this invention provides a multimodal data fault-tolerant fusion method for intelligent dismantling of retired photovoltaic modules.

[0006] The specific technical solution for achieving the objective of this invention is as follows:

[0007] A multimodal data fault-tolerant fusion method for intelligent dismantling of retired photovoltaic modules includes the following steps:

[0008] Step 1: Synchronously collect multimodal data of the target disassembly;

[0009] Step 2: Assess the health of the collected multimodal data;

[0010] Step 3: Based on the evaluation results, perform fault-tolerant fusion of the multimodal data.

[0011] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0012] This solution innovatively introduces the DS method into modal health assessment under dynamic operation: by constructing an evidence recognition framework and fusing multi-feature index evidence, it achieves accurate quantification of sensor data reliability. Compared with traditional single-index judgment, the assessment results are more in line with the environmental changes of dynamic operation in photovoltaic power plants and can accurately identify the failure state of sensor data. Based on the confidence level and likelihood function interval obtained from the assessment, the system makes weight judgments, allowing the system to intelligently allocate the fusion weight of each sensor, so that low-quality data automatically reduces its "voice" and avoids its interference with the fusion results from the root. Combined with the cross-modal feature completion mechanism, feature reconstruction is completed when a single mode fails, ensuring the continuity and effectiveness of the fusion process.

[0013] The present invention will be further described below with reference to specific embodiments. Attached Figure Description

[0014] Figure 1 This is a schematic diagram of the multimodal data fault-tolerant fusion method for intelligent dismantling of retired photovoltaic modules according to the present invention.

[0015] Figure 2 This is a flowchart illustrating the multimodal data fusion process in an embodiment of the present invention.

[0016] Figure 3 This is a schematic diagram of the DS evidence theory interval in an embodiment of the present invention. Detailed Implementation

[0017] Example

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. 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.

[0019] As indicated in this application and claims, unless the context clearly indicates otherwise, the words "a," "an," "an," and / or "the" do not specifically refer to the singular and may also include the plural. Generally speaking, the terms "comprising" and "including" only indicate the inclusion of explicitly identified steps and elements, which do not constitute an exclusive list, and the method or apparatus may also include other steps or elements.

[0020] Unless otherwise specifically stated, the relative arrangement, numerical expressions, and values ​​of the components and steps described in these embodiments do not limit the scope of this application. It should also be understood that, for ease of description, the dimensions of the various parts shown in the drawings are not drawn to actual scale. Techniques, methods, and devices known to those skilled in the art may not be discussed in detail, but where appropriate, such techniques, methods, and devices should be considered part of the specification. In all examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not as limitations. Therefore, other examples of exemplary embodiments may have different values. It should be noted that similar reference numerals and letters in the following drawings denote similar items; therefore, once an item is defined in one drawing, it need not be further discussed in subsequent drawings.

[0021] Combination Figure 1 A multimodal data fault-tolerant fusion method for intelligent dismantling of retired photovoltaic modules includes the following steps:

[0022] Step 1: Synchronously collect multimodal data of the target disassembly. In this embodiment, the multimodal data includes two-dimensional image data and three-dimensional point cloud data of the photovoltaic module to be disassembled.

[0023] Specifically, after the dispatch and control center issues the dismantling task, the recycling and dismantling vehicle drives the heavy-duty robotic arm to work dynamically in the target recycling area of ​​the photovoltaic power station. The synchronous trigger circuit controls the binocular camera module and the lidar module to start simultaneously, and collects two-dimensional image data and three-dimensional point cloud data of the photovoltaic module in real time. The raw data is then transmitted to the processing module of the vision recognition unit through the EtherCAT high-speed bus.

[0024] Step 2: Assess the health of the collected multimodal data:

[0025] Step 2-1: Construct a modal recognition framework. Set the recognition framework Θ={normal, abnormal, uncertain} to represent the normal and abnormal states of 2D image data and 3D point cloud data, respectively. "Normal" means that the modal data is valid and can be directly involved in fusion. "Abnormal" means that the modal data is invalid and needs to be completed by the healthy modality. "Uncertain" means that the modal data is in an intermediate state and cannot be clearly determined as normal or abnormal. Its reliability is average and it needs to be reduced in weight before participating in fusion.

[0026] Step 2-2: Determine the basic probability allocation for each modality of evidence;

[0027] In this embodiment, image clarity, number of effective feature points, and illumination uniformity are extracted as three pieces of evidence for the binocular camera, and point cloud density, effective point ratio, and point cloud dispersion are extracted as three pieces of evidence for the lidar. The extracted two-dimensional image data and three-dimensional point cloud data evidence index values ​​are normalized to the [0,1] interval.

[0028]

[0029] Where X is the original measurement value of the evidence, X min To preset a minimum value for the indicator, X max Set a maximum value for the indicator.

[0030] After normalizing the index values ​​to the [0,1] interval, the membership degree of each piece of evidence to the "normal," "abnormal," and "uncertain" propositions in the identification framework is calculated using Gaussian fuzzy membership functions. The membership function formulas for the three propositions are as follows:

[0031] 1. Normal membership degree

[0032]

[0033] 2. Abnormal membership degree

[0034]

[0035] 3. Uncertain membership degree

[0036]

[0037] Where σ is the smoothing coefficient, which is selected based on engineering experience; by normalizing the membership degree, the initial basic probability assignment function m(・) of the evidence body can be obtained. The specific formula is shown below:

[0038]

[0039] The allocation function satisfies:

[0040]

[0041] Where m(normal) represents the degree to which the evidence supports "normal", m(abnormal) represents the degree to which the evidence supports "abnormal", and m(uncertain) represents the degree to which the evidence supports "uncertain". ) represents the probability assigned to the empty set, and its value is 0, which means the probability of an impossible event is 0;

[0042] Steps 2-3, combined Figure 2 Multiple pieces of evidence are synthesized; the basic probability allocation function m of multiple pieces of evidence from the same modality is applied using the DS evidence synthesis rule. iBy performing successive pairwise synthesis, the comprehensive basic probability assignment function of this modality data is obtained:

[0043] Let the evidence to be synthesized be m1 and m2, then their basic probability distribution is as follows:

[0044] m1(A1) represents the probability that evidence body m1 supports modality normality;

[0045] m1(A2) represents the probability that evidence body m1 supports modal anomaly;

[0046] m1(A3) represents the probability that evidence body m1 supports modal uncertainty;

[0047] m2(A1) represents the probability that evidence body m2 supports modality normality;

[0048] m2(A2) represents the probability that evidence body m2 supports modal anomaly;

[0049] m2(A3) represents the probability that evidence body m2 supports modal uncertainty;

[0050] If the sum of the probabilities of support for the three possibilities for the same piece of evidence is 1, then the combined probability assignment function for evidence pieces m1 and m2 is:

[0051]

[0052]

[0053] Wherein, m12(A1), m12(A2), and m12(A3) represent the probability of supporting the propositions “normal”, “abnormal”, and “uncertain” after the evidence bodies m1 and m2 are synthesized, respectively. The contradiction coefficient K is used to characterize the degree of contradiction between the evidence bodies. The smaller the K value, the higher the consistency of the evidence and the closer the synthesis result is to the modal reality.

[0054] By eliminating conflicts between different evidence bodies through successive synthesis, the comprehensive basic probability allocation function m(·) of this modality is finally obtained.

[0055] Steps 2-4: Calculate the confidence level and likelihood function of each modality for the "normal" proposition;

[0056] Based on the comprehensive basic probability assignment function, determine the confidence function Bel(A1) and likelihood function Pl(A1) for each mode:

[0057]

[0058] Wherein, Bel(A1) is the definite support degree for the "normal" mode, Pl(A1) is the maximum possible support degree for the "normal" mode, the difference between the two Pl(A)-Bel(A) is the uncertainty degree of the "normal" mode, and the smaller the difference, the higher the reliability of the modal data;

[0059] Step 2-5, in conjunction with Figure 3 , the quantitative determination of the health degree of the modal state is realized based on the confidence interval [Bel(A1), Pl(A1)] constituted by Bel(A1) and Pl(A1):

[0060] If the differences between each of Bel(A1) and Pl(A1) and 1 are both less than the set threshold, the mode is determined to be normal, the data has high reliability, and can directly participate in fusion; if the differences between each of Bel(A1) and Pl(A1) and 0 are both less than the set threshold, the mode is determined to be abnormal, the data has no reliability, and feature completion needs to be performed; if 0<Bel(A1)<Pl(A1)<1, the mode is determined to be in a low-reliability state, normalization weighting is performed based on the confidence Bel(A1) of each mode, and the fusion weight is the quotient of Bel(A1) of the current mode and the sum of Bel(A1) of all modes (that is, the calculation process is consistent with the calculation formula of dynamic weighted fusion in Step 3), so as to weaken its interference on the fusion result.

[0061] Step 3, performing fault-tolerant fusion on the multi-modal data based on the evaluation result:

[0062] If the evaluation results of both modes are normal, the confidence values of the two modes are normalized, the fusion weights are calculated and distributed through the formula according to the confidence proportion, and dynamic weighted fusion is performed on the 2D image features and 3D point cloud features:

[0063]

[0064] Wherein, B i represents the confidence value Bel(A1) of the i-th mode, is the determined fusion weight value of the i-th modal data after weighted normalization;

[0065] If one of the modes is in an abnormal state, the following judgment is performed:

[0066] If it is determined that the 2D image data is abnormal, the 2D contour and texture features of the photovoltaic module are fitted through the features of the 3D point cloud data, to generate equivalent feature data of the binocular camera; if the 3D point cloud data is abnormal, the 3D geometry and distance features of the photovoltaic module are extracted by combining the 2D image data features with the binocular ranging principle, to generate equivalent feature data of the LiDAR; based on the generated equivalent feature data and the original feature data of the normal mode, during fusion, only the reliability function value Bel (A1) of the normal mode is used as the basis for weight distribution, and the sum of all weights is required to be 1, the normal mode is assigned a high weight, and the equivalent feature data corresponding to the abnormal mode is only assigned an extremely low weight whose difference from 0 is less than the set threshold.

[0067] If one mode is in a low-reliability state, the low-reliability mode is assigned a low fusion weight according to its Bel(A1) value, and the normal mode is assigned a high fusion weight, so as to achieve "high-reliability data leading fusion, low-reliability data assisting fusion", which not only retains the effective information of low-reliability data, but also avoids its interference on the fusion result.

[0068] Application of this embodiment under the dust working condition at the photovoltaic power station site: the LiDAR has sparse point cloud and high dispersion due to dust, after being processed by the D-S evidence theory, Bel(A1) is 0.2, Pl(A1) is 0.3, the difference between the two is 0.1, which satisfies 0<Bel(A1)<Pl(A1)<1, and it is determined to be in a low-reliability state;

[0069] The binocular camera is not affected by dust, Bel(A1) is 0.95, Pl(A1) is 0.98, the difference between both of them and 1 is less than the set threshold, and it is determined to be in a normal state;

[0070] The fusion weights are distributed according to the Bel(A1) of each mode, the weight of the binocular camera is 0.95 / (0.95+0.2)≈0.826, and the weight of the LiDAR is 0.2 / (0.95+0.2)≈0.174, which realizes the fusion led by the binocular camera and assisted by the LiDAR with a low weight, and effectively avoids the interference of low-quality point cloud data caused by dust on the fusion result.

[0071] If the photovoltaic power station enters a heavily shaded and dark area, the binocular camera image will be completely overexposed. After processing with DS evidence theory, Bel(A1) and Pl(A1) will both be close to 0, and the difference between them and 0 will be less than the set threshold, indicating an abnormal state. The lidar, unaffected by illumination, will have Bel(A1) of 0.98 and Pl(A1) of 0.99, indicating a normal state. In this case, equivalent two-dimensional image feature data of the binocular camera will be generated by fitting the three-dimensional geometric features of the lidar point cloud. This data will then be fused with the original lidar point cloud features. The weighting rule is as follows: the normal mode weight is set to its own confidence level Bel(A1) = 0.98, and the abnormal mode equivalent feature weight is set to 1 − Bel(A1) = 0.02. The sum of all weights is 1, ensuring that the normal mode dominates the fusion and guaranteeing the continuity and stability of the fusion process.

[0072] The method in this embodiment can be directly integrated into the existing visual recognition unit of the intelligent dismantling and recycling robot system for decommissioned photovoltaic modules. It can seamlessly link with the system's scheduling and control center, heavy-duty robotic arm, laser cutter, and other components. It strictly follows the core logic of the DS evidence theory and can accurately adapt to the complex working conditions of photovoltaic power station on-site dynamic operations, such as sudden changes in light intensity, dust interference, and high reflectivity / contamination of modules. Even when the sensor performance is affected by the environment, it can still output high-quality fusion feature data, providing reliable data support for subsequent operations and effectively improving the accuracy and stability of on-site dismantling operations for decommissioned photovoltaic modules.

[0073] This solution also provides a multimodal data fault-tolerant fusion system for intelligent dismantling of retired photovoltaic modules, including the following modules:

[0074] Data acquisition module: used to synchronously acquire multimodal data of target disassembly;

[0075] Evaluation module: Used to assess the health of the collected multimodal data;

[0076] Fault-tolerant fusion module: Used to perform fault-tolerant fusion of multimodal data based on the evaluation results.

[0077] This solution also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to perform the following steps:

[0078] Step 1: Synchronously collect multimodal data of the target disassembly;

[0079] Step 2: Assess the health of the collected multimodal data;

[0080] Step 3: Based on the evaluation results, perform fault-tolerant fusion of the multimodal data.

[0081] This solution also provides a computer-readable storage medium on which a computer program is stored, wherein the computer program, when executed by a processor, performs the following steps:

[0082] Step 1: Synchronously collect multimodal data of the target disassembly;

[0083] Step 2: Assess the health of the collected multimodal data;

[0084] Step 3: Based on the evaluation results, perform fault-tolerant fusion of the multimodal data.

[0085] The embodiments described above are merely one implementation method of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A multi-modal data fault-tolerant fusion method for intelligent dismantling of retired photovoltaic modules, characterized in that, comprising the following steps: Step 1: Synchronously collect multi-modal data of the disassembly target; Step 2: Evaluate the health degree of the collected multi-modal data; Step 3: Perform fault-tolerant fusion on the multi-modal data based on the evaluation result.

2. The multimodal fault-tolerant fusion identification method for intelligent dismantling of decommissioned photovoltaic modules according to claim 1, characterized in that, The multi-modal data in Step 1 includes two-dimensional image data and three-dimensional point cloud data of the photovoltaic module to be disassembled.

3. The multimodal data fault-tolerant fusion method for intelligent dismantling of decommissioned photovoltaic modules according to claim 2, characterized in that, The evaluating the health degree of the collected multi-modal data in Step 2 is specifically: Step 2-1: Construct a modal identification frame, set the identification frame Θ = {normal, abnormal, uncertain}, which respectively represent the normal and abnormal conditions of the two-dimensional image data and the three-dimensional point cloud data, where "normal" means the modal data is valid and can directly participate in fusion, "abnormal" means the modal data is invalid and feature completion needs to be completed through a healthy modal, and "uncertain" means the modal data is in an intermediate state, cannot be clearly determined as normal or abnormal, has general reliability, and needs to participate in fusion after weight reduction; Step 2-2: Determine the basic probability assignment of each modal evidence body; After normalizing the extracted evidence body index values of the two-dimensional image data and the three-dimensional point cloud data to the interval [0,1], calculate the basic probability assignment function m(·) for each evidence body to the propositions of "normal", "abnormal" and "uncertain" in the identification frame through a Gaussian fuzzy membership function, quantify the support degree of a single evidence body for the modal state, and the assignment function satisfies: ； Where m(normal) represents the degree to which the evidence supports "normal", m(abnormal) represents the degree to which the evidence supports "abnormal", and m(uncertain) represents the degree to which the evidence supports "uncertain". ) represents the probability assigned to the empty set, and its value is 0, which means the probability of an impossible event is 0; Steps 2-3: Synthesize evidence from multiple feature bodies; apply the DS evidence synthesis rule to the basic probability assignment function m of multiple evidence bodies from the same modality data. i By performing successive pairwise synthesis, the comprehensive basic probability assignment function of the modality data is obtained; Step 2-4: Calculate the confidence degree and the plausibility function of each modal for the "normal" proposition; Determine the belief function Bel(A₁) and the plausibility function Pl(A₁) of each modal according to the comprehensive basic probability assignment function: ； Wherein, Bel(A₁) is the definite support degree for the modal "normal", Pl(A₁) is the maximum possible support degree for the modal "normal", the difference Pl(A) - Bel(A) between the two is the uncertainty degree of the modal "normal", and the smaller the difference, the higher the reliability of the modal data; Step 2-5: Realize the quantitative determination of the health degree of the modal state based on the confidence interval [Bel(A₁), Pl(A₁)] composed of Bel(A₁) and Pl(A₁): If the differences between Bel(A₁) and 1 and between Pl(A₁) and 1 are both less than the set threshold, the modal is determined to be normal, the data has high reliability and can directly participate in fusion; if the differences between Bel(A₁) and 0 and between Pl(A₁) and 0 are both less than the set threshold, the modal is determined to be abnormal, the data has no reliability and feature completion needs to be completed; if 0 < Bel(A₁) < Pl(A₁) < 1, the modal is determined to be in a low-reliability state, normalized weighting is performed based on the confidence degree Bel(A₁) of each modal, the fusion weight is the quotient of the Bel(A₁) of the current modal and the sum of Bel(A₁) of all modals, so as to weaken its interference on the fusion result.

4. The multimodal data fault-tolerant fusion method for intelligent dismantling of decommissioned photovoltaic modules according to claim 3, characterized in that, The evidence bodies of the two-dimensional image data include image definition, number of effective feature points, and illumination uniformity; The evidence bodies of the three-dimensional point cloud data include point cloud density, effective point proportion, and point cloud dispersion.

5. The multimodal data fault-tolerant fusion method for intelligent dismantling of decommissioned photovoltaic modules according to claim 3, characterized in that, The process of obtaining the comprehensive basic probability assignment function of the modal data in Step 2-3 is: Assuming that the evidence bodies to be combined are m₁ and m₂, their basic probability assignments are: m1(A1) represents the probability that evidence body m1 supports modality normality; m1(A2) represents the probability that evidence body m1 supports modal anomaly; m1(A3) represents the probability that evidence body m1 supports modal uncertainty; m2(A1) represents the probability that evidence body m2 supports modality normality; m2(A2) represents the probability that evidence body m2 supports modal anomaly; m2(A3) represents the probability that evidence body m2 supports modal uncertainty; If the sum of the probabilities of support for the three possibilities for the same piece of evidence is 1, then the combined probability assignment function for evidence pieces m1 and m2 is: ； ； Wherein, m12(A1), m12(A2), and m12(A3) represent the probability of supporting the propositions "normal", "abnormal", and "uncertain" after the evidence bodies m1 and m2 are synthesized, respectively. The contradiction coefficient K is used to characterize the degree of contradiction between the evidence bodies. The smaller the K value, the higher the consistency of the evidence and the closer the synthesis result is to the modal reality. By eliminating conflicts between different evidence bodies through successive synthesis, the comprehensive basic probability allocation function m(·) of this modality is finally obtained.

6. The multimodal data fault-tolerant fusion method for intelligent dismantling of decommissioned photovoltaic modules according to claim 4, characterized in that, Step 3, based on the evaluation results, involves fault-tolerant fusion of the multimodal data, specifically as follows: If the evaluation results of both modalities are normal, the confidence values ​​of the two modalities are normalized, and the fusion weights are calculated and assigned according to the confidence ratio using a formula. The two-dimensional image features and three-dimensional point cloud features are then dynamically weighted and fused. ； Among them, B i Bel(A) represents the confidence value of the i-th mode. 1） , The fusion weight value for the determined i-th modality data; If one of its modes is in an abnormal state, the following judgment is made: If the two-dimensional image data is determined to be abnormal, the two-dimensional contour and texture features of the photovoltaic module are fitted by the features of the three-dimensional point cloud data to generate the equivalent feature data of the binocular camera; if the three-dimensional point cloud data is abnormal, the three-dimensional geometric and distance features of the photovoltaic module are extracted by combining the features of the two-dimensional image data with the binocular ranging principle to generate the equivalent feature data of the lidar; based on the generated equivalent feature data and the original feature data of the normal mode, during fusion, only the confidence function value Bel(A1) of the normal mode is used as the basis for weight allocation, and all weights are satisfied that the sum of all weights is 1, the normal mode is assigned a high weight, and the equivalent feature data corresponding to the abnormal mode is assigned only a very low weight with a difference from 0 less than a set threshold. If one of its modes is in a low reliability state, the low reliability mode is assigned a low fusion weight according to its Bel(A1) value, and the normal mode is assigned a high fusion weight, so as to achieve "high reliability data dominating fusion and low reliability data assisting fusion", which not only retains the effective information of low reliability data, but also avoids its interference with the fusion result.

7. A multimodal data fault-tolerant fusion system for intelligent dismantling of decommissioned photovoltaic modules, characterized in that, Includes the following modules: Data acquisition module: used to synchronously acquire multimodal data of target disassembly; Evaluation module: Used to assess the health of the collected multimodal data; Fault-tolerant fusion module: Used to perform fault-tolerant fusion of multimodal data based on the evaluation results.

8. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1-6.

9. A computer-storable medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1-6.

Images