Explainable output and evidence binding method, device, equipment, storage medium and program product for generator diagnosis

Abstract

This disclosure relates to a method, apparatus, device, storage medium, and program product for interpretable output and evidence binding in generator diagnosis. The method includes: obtaining diagnostic results, including abnormal location information and feature weight information, from inspection images of a generator based on an artificial intelligence model, and converting these results into engineering semantic text descriptions; extracting features of the region of interest of the artificial intelligence model in the inspection images based on the feature weight information, and generating a visualization layer representing the decision-making basis of the artificial intelligence model; constructing a structured evidence object based on the inspection images, visualization layer, engineering semantic text description, physical space coordinate information, and algorithm metadata; and parsing the evidence object corresponding to the traceability request in response to a traceability request for the diagnostic results, and displaying it on an interactive interface. This achieves transparency and visualization of the artificial intelligence diagnostic process, enhances the trust level of human-machine collaboration, establishes a complete closed-loop evidence chain, and provides data support for the maintenance of power equipment.

Description

Technical Field

[0001] This disclosure relates to the field of industrial artificial intelligence technology, and in particular to a method, apparatus, device, storage medium, and program product for binding interpretable output and evidence in generator diagnostics. Background Technology

[0002] This section is intended to provide background or context for embodiments of this disclosure. The description herein is not intended to imply that it is prior art simply because it is included in this section.

[0003] In the operation and maintenance of large generators, computer vision-based artificial intelligence diagnostic technology is gradually replacing traditional manual visual inspection methods. In existing technologies, visual AI solutions typically only output a classification result and its corresponding confidence score, such as determining the presence of cracks in an inspection image and providing a confidence score.

[0004] However, the above-mentioned existing technical solutions have the following drawbacks: The lack of interpretability in the output of artificial intelligence models has led to a crisis of trust. Existing object detection or classification networks typically only output defect category labels and confidence scores, lacking intuitive visual attention indicators and logical reasoning. Users cannot determine whether high confidence scores are based on real defect textures or artifacts or noise in the background, making diagnostic results difficult to accept.

[0005] The disconnect between diagnostic conclusions and supporting evidence is a significant problem. In the traditional maintenance report generation process, the final defect list is often separated from the massive amount of original inspection image data. When it is necessary to review a historical defect, it is difficult to quickly retrieve the original image corresponding to the conclusion, the model version used at the time, and the parameter thresholds, making traceability extremely difficult. Summary of the Invention

[0006] In order to at least partially address one of the technical problems in the related art, this disclosure provides a method, apparatus, device, storage medium, and program product for binding interpretable output and evidence for generator diagnostics.

[0007] To achieve the above objectives, a first aspect of the exemplary embodiments of this disclosure provides a method for binding interpretable output and evidence in generator diagnostics, comprising: Obtain the inspection image corresponding to the generator, perform anomaly diagnosis on the inspection image based on the artificial intelligence model, and obtain the diagnosis result. The diagnosis result includes anomaly location information and feature weight information. Convert the diagnosis result into an engineering semantic text description. Based on the feature weight information, the features of the region of interest of the artificial intelligence model in the inspection image are extracted, and based on the features of the region of interest, a visualization layer representing the decision basis of the artificial intelligence model is generated. Based on the inspection images, the visualization layers, the engineering semantic text descriptions, physical space coordinate information, and algorithm metadata, a structured evidence object is constructed. The physical space coordinate information is used to identify the physical location of the abnormal location information on the generator. The algorithm metadata includes the model version information and diagnosis time information of the artificial intelligence model. In response to a traceability request for a diagnostic result, the evidence object corresponding to the traceability request is parsed, and the inspection image, the visualization layer, the engineering semantic text description, the physical space coordinate information, and the algorithm metadata are displayed on the interactive interface.

[0008] In some exemplary embodiments, extracting the region of interest features of the inspection image by the artificial intelligence model includes: The gradient-weighted class activation mapping algorithm is used to determine the importance weight of the feature map output by the last convolutional layer in the artificial intelligence model to the anomaly category. Based on the importance weights, the feature map regions that contribute the most to the classification decision are extracted as the regions of interest.

[0009] In some exemplary embodiments, determining the importance weight of the feature map output by the last convolutional layer in the artificial intelligence model to the anomaly category includes: For each feature map channel output by the last convolutional layer, determine the gradient of the anomaly category prediction score with respect to each pixel in each feature map channel; Global average pooling is performed on the gradient values ​​corresponding to each feature map channel to obtain the importance weight of each feature map channel for the anomaly category.

[0010] In some exemplary embodiments, extracting the feature map region that contributes the most to the classification decision as the feature of the region of interest based on the importance weight includes: Using the importance weights corresponding to each feature map channel as coefficients, a weighted sum is performed on all feature map channels output by the last convolutional layer; The weighted summation result is processed to retain positive contributions through activation, thereby removing features that have a negative impact on classification decisions.

[0011] In some exemplary embodiments, generating a visualization layer representing the decision-making basis of the artificial intelligence model includes: A heatmap is generated based on the features of the region of interest; Adjust the size of the heat map to be the same as the inspection image; The resized heatmap is overlaid on the inspection image to form the visualization layer.

[0012] In some exemplary embodiments, the step of overlaying the resized heatmap onto the inspection image includes: The heat map is mapped to a preset color image; The color image is overlaid on the inspection image as a semi-transparent overlay.

[0013] Based on the same inventive concept, a second aspect of the exemplary embodiments of this disclosure provides an interpretable output and evidence binding device for generator diagnostics, comprising: The diagnostic result determination module is configured to acquire the inspection image corresponding to the generator, perform anomaly diagnosis on the inspection image based on an artificial intelligence model, obtain a diagnostic result, the diagnostic result includes anomaly location information and feature weight information, and convert the diagnostic result into an engineering semantic text description. The visualization generation module is configured to extract the region of interest features of the artificial intelligence model on the inspection image based on the feature weight information, and generate a visualization layer representing the decision basis of the artificial intelligence model based on the region of interest features. The evidence object construction module is configured to construct a structured evidence object based on the inspection image, the visualization layer, the engineering semantic text description, physical space coordinate information, and algorithm metadata. The physical space coordinate information is used to identify the physical location of the abnormal location information on the generator, and the algorithm metadata includes the model version information and diagnosis time information of the artificial intelligence model. The traceability interaction module is configured to respond to a traceability request for a diagnostic result, parse the evidence object corresponding to the traceability request, and display the inspection image, the visualization layer, the engineering semantic text description, the physical space coordinate information, and the algorithm metadata on the interactive interface.

[0014] Based on the same inventive concept, a third aspect of the exemplary embodiments of this disclosure provides an electronic device, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the method as described in the first aspect.

[0015] Based on the same inventive concept, a fourth aspect of the exemplary embodiments of this disclosure provides a computer-readable storage medium having a computer program / instructions stored thereon, which, when executed by a processor, implement the steps of the method as described in the first aspect.

[0016] Based on the same inventive concept, a fifth aspect of the exemplary embodiments of this disclosure provides a computer program product, including a computer program / instructions that, when executed by a processor, implement the steps of the method as described in the first aspect.

[0017] The technical solution provided in this disclosure has the following advantages compared with the prior art: It achieves transparency and visualization in the AI ​​diagnostic process. By extracting features of the region of interest based on feature weight information and generating a visualization layer, the internal decision-making basis of the AI ​​model is presented in an intuitive form such as a heatmap. Users can determine whether the model is focused on the correct defect texture, effectively identify false alarms caused by background oil stains or reflections, and transform the black-box model into an understandable and verifiable transparent tool, thereby improving the trust level of human-computer collaboration.

[0018] A complete, auditable, closed-loop evidence chain was established. By constructing structured evidence objects that include inspection images, visualization layers, engineering semantic text descriptions, physical space coordinate information, and algorithm metadata, diagnostic conclusions are strongly bound to their background and encapsulated in an tamper-proof manner, providing data support for the maintenance of critical power equipment. Attached Figure Description

[0019] The accompanying drawings, which are incorporated in and form a part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure.

[0020] To more clearly illustrate the technical solutions in the embodiments of this disclosure or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 A schematic flowchart of a generator diagnostics method for binding interpretable outputs with evidence, provided as an exemplary embodiment of this disclosure; Figure 2 A schematic diagram of a generator diagnostic interpretable output and evidence binding device provided as an exemplary embodiment of the present disclosure; Figure 3 A schematic diagram of the structure of an electronic device provided as an exemplary embodiment of the present disclosure. Detailed Implementation

[0022] It is understood that the data involved in this technical solution (including but not limited to the data itself, the acquisition or use of the data) shall comply with the requirements of relevant laws, regulations and related provisions.

[0023] Embodiments of this disclosure will now be described in more detail with reference to the accompanying drawings. While some embodiments of this disclosure are shown in the drawings, it should be understood that this disclosure can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of this disclosure. It should be understood that the accompanying drawings and embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of protection of this disclosure.

[0024] It should be understood that the steps described in the method embodiments of this disclosure may be performed in different orders and / or in parallel. Furthermore, the method embodiments may include additional steps and / or omit the steps shown. The scope of this disclosure is not limited in this respect.

[0025] The term "comprising" and its variations as used herein are open-ended inclusions, meaning "including but not limited to". The term "based on" means "at least partially based on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments". Definitions of other terms will be given in the description below.

[0026] It should be noted that the concepts of "first" and "second" mentioned in this disclosure are used only to distinguish different devices, modules or units, and are not used to limit the order of functions performed by these devices, modules or units or their interdependencies.

[0027] It should be noted that the terms "a" and "a plurality of" used in this disclosure are illustrative rather than restrictive, and those skilled in the art should understand that, unless otherwise expressly indicated in the context, they should be understood as "one or more".

[0028] The names of messages or information exchanged between multiple devices in the embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of such messages or information.

[0029] To better understand the above-mentioned objectives, features, and advantages of this disclosure, the solutions disclosed herein will be further described below. It should be noted that, unless otherwise specified, the embodiments and features described herein can be combined with each other.

[0030] Numerous specific details are set forth in the following description in order to provide a full understanding of this disclosure, but this disclosure may also be implemented in other ways different from those described herein; obviously, the embodiments in the specification are only some, and not all, of the embodiments of this disclosure.

[0031] To address the aforementioned issues, this disclosure provides a method for binding interpretable output and evidence in generator diagnostics. The method will be described below with reference to specific embodiments.

[0032] Figure 1 This is a flowchart illustrating a method for binding interpretable outputs and evidence in generator diagnostics, provided in an embodiment of this disclosure. This method can be executed by a generator diagnostic interpretable output and evidence binding device, which can be implemented in software and / or hardware, and is generally integrated into an electronic device. Figure 1 As shown, the method includes: Step 110: Obtain the inspection image corresponding to the generator, perform anomaly diagnosis on the inspection image based on the artificial intelligence model, and obtain the diagnosis result. The diagnosis result includes anomaly location information and feature weight information. Convert the diagnosis result into an engineering semantic text description.

[0033] Among them, the inspection image corresponding to the generator is obtained: the inspection image can be a visible light image obtained by taking pictures of the generator stator or rotor surface by an endoscope, industrial camera or other image acquisition equipment.

[0034] After acquiring the inspection image, it is input into a pre-trained artificial intelligence (AI) model for anomaly diagnosis. This AI model can be a deep convolutional neural network model for object detection, capable of locating and classifying potential defect areas in the inspection image. After processing the inspection image, the AI ​​model outputs a corresponding diagnostic result. This diagnostic result contains at least two types of key information: first, anomaly location information, indicating the spatial coordinate range of the defect area identified by the AI ​​model in the inspection image, represented, for example, by a bounding box or pixel-level mask; second, feature weight information, reflecting the contribution of each feature channel within the AI ​​model to the final anomaly classification, and is a quantified value output from the intermediate layer of the AI ​​model.

[0035] Specifically, the numerical labels directly output by the artificial intelligence model in the diagnostic results are converted into engineering semantic text descriptions. A preset feature semantic mapping relationship can be maintained, which maps different AI model output category identifiers to standardized engineering terms and corresponding causal analysis or treatment suggestions. By querying this mapping relationship, the abstract diagnostic results can be transformed into natural language text that is readable and understandable to users, containing defect names, possible causes, and suggested measures, thus completing the conversion process from numerical model output to engineering semantic text descriptions.

[0036] As an example, this mapping table associates different anomaly category identifiers output by the AI ​​model with standardized defect category names and corresponding engineering handling suggestions. For instance, when the AI ​​model outputs an anomaly category identifier of a certain value for a specific inspection image area, querying the mapping table will retrieve the corresponding defect category name as "anti-halo layer corrosion," along with the associated engineering suggestion text, such as a suggestion to check the slot potential and retest partial discharge. Through this mapping query mechanism, the originally abstract classification numbers are transformed into engineering semantic text descriptions that maintenance personnel can directly read and understand.

[0037] For some AI models with multi-attribute output capabilities, their diagnostic results may include quantitative descriptions of defect geometry or size attributes in addition to anomaly categories. The process of converting diagnostic results into engineering semantic text descriptions also includes natural language generation of attribute vectors. For example, besides outputting anomaly categories, the AI ​​model outputs an attribute vector containing numerical values ​​for defect length, defect area, and shape descriptors. The text generation module combines these values ​​and descriptors from the attribute vector to generate a complete natural language description, such as: "A linear defect was identified, approximately 15 millimeters long, distributed in a linear pattern, suspected to be a mechanical scratch." This results in richer and more specific engineering semantic text descriptions, providing users with more comprehensive decision-making information.

[0038] Step 120: Based on the feature weight information, extract the features of the region of interest of the artificial intelligence model in the inspection image, and generate a visualization layer representing the decision basis of the artificial intelligence model based on the features of the region of interest.

[0039] Based on the feature weight information obtained from the diagnostic results, the internal decision-making logic of the artificial intelligence model is further analyzed and visualized. Specifically, based on the feature weight information, the AI ​​model extracts the region of interest features for the inspection image. These region of interest features refer to the feature representations of image regions that play a crucial supporting role in the AI ​​model's decision-making when determining the existence of specific anomalies in the inspection image. During image processing, the AI ​​model's deep convolutional layers learn different levels of visual features. The feature maps of convolutional layers closer to the output already possess strong semantic information, reflecting the differences in the model's level of attention to different parts of the image. By retrospectively analyzing the feature weight information, the pixels or regions of greatest interest to the model can be located from these feature maps and extracted as the region of interest features.

[0040] Specifically, a visualization layer representing the decision-making basis of the artificial intelligence model is generated based on the extracted features of the region of interest. This visualization layer can be generated in ways including, but not limited to, generating a heatmap based on the intensity distribution of the features of the region of interest, or forming visual annotations on the inspection image in the form of contour outlining or brightness mapping. The core function of this visualization layer is to transform the feature weights and activation distributions originally hidden within the artificial intelligence model into visual elements that users can intuitively understand. This allows the image areas on which the model relies to make anomaly diagnosis conclusions to be highlighted or emphasized, thus providing clear visual guidance for subsequent manual review and result acceptance.

[0041] In some exemplary embodiments, extracting the region of interest features of the inspection image by the artificial intelligence model includes: The gradient-weighted class activation mapping algorithm is used to determine the importance weight of the feature map output by the last convolutional layer in the artificial intelligence model to the anomaly category. Based on the importance weights, the feature map regions that contribute the most to the classification decision are extracted as the regions of interest.

[0042] The basic idea of ​​the gradient-weighted class activation mapping algorithm is to use the gradient information generated by the artificial intelligence model during forward inference to measure the sensitivity of the feature map output by the last convolutional layer to a specific anomaly category.

[0043] Specifically, the feature map output by the last convolutional layer in the AI ​​model is determined. This feature map consists of multiple feature map channels, each representing the response strength of a different filter to the input inspection image in a certain abstract semantic dimension. For this feature map, its importance weight for the anomaly category is calculated. The importance weight reflects the relative contribution of each feature map channel to the model's final anomaly classification, and its calculation depends on the gradient information of the model's predicted score with respect to the pixel values ​​of the feature map channels. After obtaining the importance weights corresponding to each feature map channel, the feature map region with the highest contribution to the classification decision can be extracted from the feature map based on these weights, and this region is used as the region of interest. Through this method, the spatial range in which the AI ​​model's visual attention is most concentrated when making anomaly diagnosis decisions can be accurately located, providing a reliable basis for subsequently generating interpretable visualization layers.

[0044] In some exemplary embodiments, determining the importance weight of the feature map output by the last convolutional layer in the artificial intelligence model to the anomaly category includes: For each feature map channel output by the last convolutional layer, determine the gradient of the anomaly category prediction score with respect to each pixel in each feature map channel; Global average pooling is performed on the gradient values ​​corresponding to each feature map channel to obtain the importance weight of each feature map channel for the anomaly category.

[0045] In this process, the AI ​​model outputs a predicted score for a specific anomaly category during anomaly diagnosis. This predicted score reflects the likelihood that the model believes the inspected image contains that type of anomaly. For each feature map channel output by the last convolutional layer, the gradient of the anomaly category predicted score relative to each pixel value in that feature map channel is calculated. This gradient represents the degree to which a small change in a pixel on the feature map channel will cause a corresponding change in the anomaly category predicted score; therefore, this gradient is directly related to the influence of each pixel in the feature map channel on the classification decision. After calculating the gradient values ​​corresponding to all pixels in each feature map channel, global average pooling is performed on these gradient values. Global average pooling involves summing the gradient values ​​at all spatial locations within the same feature map channel and then dividing by the total number of spatial locations to obtain an overall representative value for that feature map channel. This value is the importance weight of that feature map channel for the anomaly category. A larger importance weight indicates a greater contribution of the corresponding feature map channel to the model's anomaly category judgment, and vice versa.

[0046] In some exemplary embodiments, extracting the feature map region that contributes the most to the classification decision as the feature of the region of interest based on the importance weight includes: Using the importance weights corresponding to each feature map channel as coefficients, a weighted sum is performed on all feature map channels output by the last convolutional layer; The weighted summation result is processed to retain positive contributions through activation, thereby removing features that have a negative impact on classification decisions.

[0047] Specifically, using the importance weights corresponding to each feature map channel as coefficients, a weighted summation is performed on all feature map channels output by the last convolutional layer: the pixel value at each spatial location within each feature map channel is multiplied by the importance weight of its corresponding channel, and then the weighted summation is accumulated pixel-by-pixel at the corresponding spatial locations to obtain a two-dimensional weighted summation result image with the same spatial size as each feature map channel. In this weighted summation result image, regions with larger values ​​indicate a higher degree of dependence of the artificial intelligence model on that region when making the current anomaly category judgment.

[0048] The weighted summation result undergoes activation processing to retain positive contributions. This activation process removes features that negatively impact classification decisions, retaining only those that positively contribute to improving anomaly prediction scores while setting negative values ​​that might suppress prediction scores to zero. The feature map regions obtained after this processing constitute the regions of interest that contribute most to classification decisions. These regions of interest spatially correspond highly to the key textures or structural regions in the inspection images that actually trigger anomaly alarms in the AI ​​model.

[0049] In some exemplary embodiments, generating a visualization layer representing the decision-making basis of the artificial intelligence model includes: A heatmap is generated based on the features of the region of interest; Adjust the size of the heat map to be the same as the inspection image; The resized heatmap is overlaid on the inspection image to form the visualization layer.

[0050] Specifically, a heatmap is generated based on the features of the region of interest: since the features of the region of interest are themselves a two-dimensional numerical matrix, the magnitude of the value at each spatial location represents the degree of attention the artificial intelligence model pays to that location, so this numerical matrix can be directly mapped to a heatmap. Different colors or brightness levels are used in the heatmap to represent the degree of attention; for example, high-attention areas correspond to warm colors, and low-attention areas correspond to cool colors.

[0051] Since the features of the region of interest typically originate from deep convolutional layers of an artificial intelligence model, their spatial resolution is lower than that of the original inspection image. Therefore, the size of the heatmap needs to be adjusted to be enlarged to the same size as the inspection image. This size adjustment can be achieved using an interpolation algorithm to ensure that the heatmap corresponds point-by-point to the pixel grid of the inspection image.

[0052] The heatmap, after being resized, is overlaid on the inspection image to form the visualization layer. The resulting visualization layer combines all the texture details of the original inspection image with the model attention distribution information conveyed by the heatmap, allowing users to observe the defects themselves and the focus of the artificial intelligence model on them in the same view.

[0053] In some exemplary embodiments, the step of overlaying the resized heatmap onto the inspection image includes: The heat map is mapped to a preset color image; The color image is overlaid on the inspection image as a semi-transparent overlay.

[0054] The heatmap is mapped to a preset color image. This mapping process follows a pre-defined color mapping scheme. For example, areas with higher values ​​in the heatmap can be mapped to red or yellow, while areas with lower values ​​can be mapped to blue or black, thus forming a color image with clear visual hierarchy. This color image is essentially a visual encoding of the numerical distribution of features in the area of ​​interest.

[0055] The color image is overlaid on the inspection image as a semi-transparent overlay: the semi-transparent overlay means that the color image and the inspection image are not completely covered when overlaid, but rather a transparency parameter between completely opaque and completely transparent is set so that the original content of the inspection image can be recognized by the naked eye through the color layer.

[0056] By adjusting the transparency, users can balance the discernibility between the original image details and the model's attention indicators according to their actual needs. The resulting visualization layer retains the true texture information of the inspection image while intuitively displaying the image areas on which the artificial intelligence model bases its decisions, thus achieving a clear and adjustable visual interpretation of the model's decision-making basis.

[0057] In the process of generating a visualization layer based on the features of the region of interest, this disclosure may further include the step of generating a close-up defect image: After obtaining the features of the region of interest, connected regions composed of pixels whose values ​​exceed a preset activation threshold can be selected according to the numerical values ​​of each spatial location in the features of the region of interest. For each connected region, its minimum bounding rectangle is automatically calculated, and a preset number of pixels are extended outward from the boundary of this rectangle to retain necessary background texture information. Subsequently, based on the coordinate range of the extended rectangle, the corresponding local image region is cropped from the original inspection image, and the cropped image is the close-up defect image. This close-up defect image focuses on the core region and its surrounding context in which the artificial intelligence model determines anomalies, facilitating high-resolution detail review by the user. The generated close-up defect image can be used as one of the evidentiary elements and encapsulated together with other information into subsequent structured evidence objects.

[0058] Step 130: Based on the inspection image, the visualization layer, the engineering semantic text description, the physical space coordinate information, and the algorithm metadata, construct a structured evidence object, wherein the physical space coordinate information is used to identify the physical location of the abnormal location information on the generator, and the algorithm metadata includes the model version information and diagnosis time information of the artificial intelligence model.

[0059] Building upon the previously generated inspection images, visualization layers, and engineering semantic text descriptions, this step further constructs a standardized data unit to integrate the diagnostic conclusions with all supporting evidence. The core output of this step is a structured evidence object, which aggregates and binds various related information originally scattered across different modules or systems according to a pre-defined data organization structure, thereby achieving complete traceability of the diagnostic conclusions. The information elements incorporated when constructing this structured evidence object include the following aspects.

[0060] The inspection images are included as the original visual material for this diagnosis and are considered as evidence. These inspection images can be unmodified original image files output from the original acquisition device, or storage paths and hash digests that uniquely identify the original image file, ensuring accurate identification of the original image data during subsequent tracing without replacement or alteration.

[0061] The visualization layer is incorporated into the evidence as key interpretative evidence. This visualization layer is the output of step 120; it exists as an image overlaid with heatmap annotations or other forms of visual enhancement, intuitively revealing the image regions that the artificial intelligence model focuses on when making diagnostic decisions, providing a visual basis for users to understand the model's behavior.

[0062] The engineering semantic text description is incorporated into the evidence object. This text description is a natural language expression converted from the numerical diagnostic results output by the artificial intelligence model in step 110. It includes the defect category name, possible cause analysis, and suggested remedial measures, so that the diagnostic conclusions can be directly read and understood by personnel who do not have professional knowledge of artificial intelligence models.

[0063] Physical spatial coordinate information is incorporated into the evidence object. This physical spatial coordinate information is used to identify the physical location of the abnormal location information on the generator, and its content may include, but is not limited to, some or all of the five-dimensional positioning parameters such as unit number, component name, stator slot number, axial distance, and circumferential angle. By binding physical spatial coordinate information to the evidence object, each diagnostic conclusion can be correlated with a precise physical location on the generator's physical structure, making it easy for on-site users to directly locate the defect based on the report's guidance.

[0064] Algorithm metadata is incorporated into the evidence object. This algorithm metadata includes model version information and diagnosis time information of the artificial intelligence model. The model version information records the version identifier of the artificial intelligence model used to generate the current diagnostic conclusion. This identifier can be associated with the dataset version, network structure configuration, and weight file used during model training, ensuring accurate knowledge of the algorithm's state at the time when it is necessary to reproduce the diagnostic process or compare results across versions. The diagnosis time information records the moment the diagnostic inference occurred, providing a basis for constructing an audit trajectory over time. Furthermore, the algorithm metadata may also include configuration information such as the confidence threshold and post-processing rule parameters used during inference.

[0065] The aforementioned information elements are organized according to a pre-defined field hierarchy within the structured evidence object. For example, they can be described using an extensible data exchange format, forming an independent, self-contained data container that is easy to transmit and parse. By calculating hash digests of the core fields within this evidence object, it can be further ensured that the diagnostic conclusions and supporting evidence carried by the evidence object have not been illegally modified throughout its entire lifecycle, from generation and storage to retrieval and review. This step lays a solid data foundation for subsequent one-click traceability from the final diagnostic report entries to the original evidence scenario.

[0066] As an example, this structured evidence object can be defined using a scalable data exchange format, containing multiple hierarchically organized fields. The first field is a unique identifier used for unique indexing of the diagnostic evidence. The second field is the set of original evidence information, which may further include subfields for the original image storage path and the original image hash value, and may also include a subfield for the storage path of the defect close-up image. The third field is the set of interpretability output information, which may further include subfields for the storage path of the visualization layer, the engineering semantic text description, and the attention score or confidence level. The fourth field is the set of physical space context information, which may further include subfields for the stator slot number, axial distance, and component name. The fifth field is the set of algorithm state snapshot information, which may further include subfields for the model version number, diagnostic time information, and the confidence threshold used for inference. These fields and their nested relationships together constitute a complete, self-describing structured evidence object, ensuring the clarity and stability of the relationships between the various evidence elements.

[0067] Step 140: In response to the traceability request for the diagnostic results, parse the evidence object corresponding to the traceability request, and display the inspection image, the visualization layer, the engineering semantic text description, the physical space coordinate information, and the algorithm metadata on the interactive interface.

[0068] This includes a diagnostic conclusion traceability and review function provided to users or reviewers, based on the structured evidence objects already constructed and stored in the system. When a user browses the list of diagnostic reports or defect items through the front-end interactive interface and initiates a traceability request for a specific diagnostic result, the system responds to the request by performing the corresponding evidence object parsing and rendering operations.

[0069] First, based on the diagnostic result item selected by the user, a unique identifier for the structured evidence object associated with that item is obtained, and the corresponding complete evidence object is retrieved from the data storage location based on this identifier. During the retrieval process, the integrity of the evidence object can be verified simultaneously, for example, by comparing whether the pre-stored hash digest value in the evidence object is consistent with the currently recalculated hash value, to confirm that its content has not been tampered with or replaced since the evidence object was generated.

[0070] Subsequently, the information contained in the parsed evidence object is presented in a graphical interface. Within the main view area of ​​the interactive interface, the inspection image is loaded and displayed; this inspection image is the raw image data used by the artificial intelligence model for diagnosis. To facilitate quick location of defects, corresponding location markers or bounding boxes can be overlaid on the inspection image based on the abnormal location information recorded in the evidence object.

[0071] Simultaneously, a visualization layer is overlaid on the inspection image. This visualization layer, presented as a semi-transparent heatmap or other visual annotations, represents the image layer upon which the artificial intelligence model makes decisions. The interactive interface provides a transparency adjustment control, allowing users to dynamically adjust the transparency parameters of the visualization layer. By adjusting the transparency, users can smoothly switch between fully revealing the texture of the original inspection image and highlighting the areas of interest for the model, thereby autonomously verifying whether the artificial intelligence model is focusing its attention on the correct areas related to defect features.

[0072] In addition, the information display area of ​​the interactive interface also simultaneously presents the engineering semantic text description, the physical space coordinate information, and the algorithm metadata. The engineering semantic text description explains the meaning of the diagnostic conclusion, the defect name, and possible handling suggestions to the user in natural language. The physical space coordinate information precisely indicates the corresponding location of the defect on the generator's physical structure, such as the stator slot number and axial distance, providing clear location guidance for on-site users. The algorithm metadata includes the model version information and diagnosis time information of the artificial intelligence model, and optionally displays the confidence threshold parameters used during inference. When it is detected that the version or parameter configuration of the currently running artificial intelligence model is inconsistent with the algorithm metadata recorded in the evidence object, the interactive interface can further generate a warning prompt to remind the user that the current environment has changed, which may affect the reproduction and verification of the diagnostic conclusion. Through the above interface display method, one-click traceability from the diagnostic conclusion item to the complete evidence scenario is realized, enabling users to obtain all the supporting information needed to make a review judgment in a unified view, effectively improving the transparency and credibility of the diagnostic results.

[0073] When displaying inspection images on the interactive interface, to further assist users in quickly locating the corresponding position of defects on the generator's physical structure, a corresponding positioning cursor can be drawn on the inspection image based on the physical space coordinate information recorded in the evidence object. Specifically, parameters such as axial distance, circumferential angle, or slot number related to the abnormal location are parsed from the physical space coordinate information. Using a pre-calibrated mapping relationship between the image pixel coordinate system and the generator's physical space coordinate system, the corresponding indicated position of the defect area on the inspection image is calculated. Subsequently, a visual positioning cursor, such as a crosshair or arrow, is rendered at this indicated position. This positioning cursor intuitively indicates to the user the approximate location of the currently examined defect within the overall generator structure, further improving the accuracy and efficiency of diagnostic tracing.

[0074] It should be noted that the method of this disclosure embodiment can be executed by a single device, such as a computer or server. The method of this embodiment can also be applied to a distributed scenario, where multiple devices cooperate to complete the task. In such a distributed scenario, one of these devices may execute only one or more steps of the method of this disclosure embodiment, and the multiple devices will interact with each other to complete the method described.

[0075] It should be noted that the above description describes some embodiments of this disclosure. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in a different order than that shown in the above embodiments and still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0076] To achieve the above embodiments, this disclosure also proposes an interpretable output and evidence binding device for generator diagnostics.

[0077] Figure 2 This is a schematic diagram of a generator diagnostic interpretable output and evidence binding device provided in an embodiment of this disclosure. This device can be implemented by software and / or hardware, and is generally integrated into an electronic device. Figure 2 As shown, the device includes: The diagnostic result determination module 210 is configured to acquire the inspection image corresponding to the generator, perform anomaly diagnosis on the inspection image based on an artificial intelligence model, obtain a diagnostic result, the diagnostic result including anomaly location information and feature weight information, and convert the diagnostic result into an engineering semantic text description. The visualization generation module 220 is configured to extract the features of the region of interest of the artificial intelligence model on the inspection image based on the feature weight information, and generate a visualization layer representing the decision basis of the artificial intelligence model based on the features of the region of interest. The evidence object construction module 230 is configured to construct a structured evidence object based on the inspection image, the visualization layer, the engineering semantic text description, physical space coordinate information and algorithm metadata. The physical space coordinate information is used to identify the physical location of the abnormal location information on the generator, and the algorithm metadata includes the model version information and diagnosis time information of the artificial intelligence model. The traceability interaction module 240 is configured to respond to a traceability request for a diagnostic result, parse the evidence object corresponding to the traceability request, and display the inspection image, the visualization layer, the engineering semantic text description, the physical space coordinate information, and the algorithm metadata on the interactive interface.

[0078] In some exemplary embodiments, the visualization generation module 220 is configured to: The gradient-weighted class activation mapping algorithm is used to determine the importance weight of the feature map output by the last convolutional layer in the artificial intelligence model to the anomaly category. Based on the importance weights, the feature map regions that contribute the most to the classification decision are extracted as the regions of interest.

[0079] In some exemplary embodiments, the visualization generation module 220 is configured to: For each feature map channel output by the last convolutional layer, determine the gradient of the anomaly category prediction score with respect to each pixel in each feature map channel; Global average pooling is performed on the gradient values ​​corresponding to each feature map channel to obtain the importance weight of each feature map channel for the anomaly category.

[0080] In some exemplary embodiments, the visualization generation module 220 is configured to: Using the importance weights corresponding to each feature map channel as coefficients, a weighted sum is performed on all feature map channels output by the last convolutional layer; The weighted summation result is processed to retain positive contributions through activation, thereby removing features that have a negative impact on classification decisions.

[0081] In some exemplary embodiments, the visualization generation module 220 is configured to: A heatmap is generated based on the features of the region of interest; Adjust the size of the heat map to be the same as the inspection image; The resized heatmap is overlaid on the inspection image to form the visualization layer.

[0082] In some exemplary embodiments, the visualization generation module 220 is configured to: The heat map is mapped to a preset color image; The color image is overlaid on the inspection image as a semi-transparent overlay.

[0083] For ease of description, the above apparatus is described in terms of its functions, divided into various modules. Of course, in implementing this disclosure, the functions of each module can be implemented in one or more software and / or hardware.

[0084] The generator diagnostic interpretable output and evidence binding device provided in this disclosure can execute the generator diagnostic interpretable output and evidence binding method provided in any embodiment of this disclosure, and has the corresponding functional modules and beneficial effects of the execution method.

[0085] To implement the above embodiments, this disclosure also proposes a computer program product, including a computer program / instruction, which, when executed by a processor, implements the interpretable output and evidence binding method for generator diagnosis in the above embodiments.

[0086] Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this disclosure.

[0087] The following is a detailed reference. Figure 3 The diagram illustrates a structural schematic suitable for implementing the electronic device in the embodiments of this disclosure. The electronic device in the embodiments of this disclosure may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (personal digital assistants), PADs (tablet computers), PMPs (portable multimedia players), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. Figure 3 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments disclosed herein.

[0088] like Figure 3 As shown, the electronic device may include a processor (e.g., a central processing unit, a graphics processing unit, etc.) 301, which can perform various appropriate actions and processes according to a program stored in read-only memory (ROM) 302 or a program loaded from memory 308 into random access memory (RAM) 303. The RAM 303 also stores various programs and data required for the operation of the electronic device. The processor 301, ROM 302, and RAM 303 are interconnected via a bus 304. An input / output (I / O) interface 305 is also connected to the bus 304.

[0089] Typically, the following devices can be connected to I / O interface 305: input devices 306 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 307 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; memory devices 308 including, for example, magnetic tapes, hard disks, etc.; and communication devices 309. Communication device 309 allows electronic devices to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 3 Electronic devices with various devices are shown, but it should be understood that it is not required to implement or have all of the devices shown. More or fewer devices may be implemented or have alternatively.

[0090] In particular, according to embodiments of this disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this disclosure include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication device 309, or installed from memory 308, or installed from ROM 302. When the computer program is executed by processor 301, it performs the functions defined in the generator diagnostic interpretable output and evidence binding method of embodiments of this disclosure.

[0091] It should be noted that the computer-readable medium described in this disclosure can be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this disclosure, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this disclosure, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wires, optical fibers, RF (radio frequency), etc., or any suitable combination thereof.

[0092] In some implementations, clients and servers can communicate using any currently known or future-developed network protocol such as HTTP (Hypertext Transfer Protocol) and can interconnect with digital data communication (e.g., communication networks) of any form or medium. Examples of communication networks include local area networks (“LANs”), wide area networks (“WANs”), the Internet (e.g., the Internet of Things), and peer-to-peer networks (e.g., ad hoc peer-to-peer networks), as well as any currently known or future-developed networks.

[0093] The aforementioned computer-readable medium may be included in the aforementioned electronic device; or it may exist independently and not assembled into the electronic device.

[0094] The aforementioned computer-readable medium carries one or more programs, which, when executed by the electronic device, cause the electronic device to perform the aforementioned generator diagnostic interpretable output and evidence binding method.

[0095] Electronic devices can be programmed with computer program code in one or more programming languages ​​or a combination thereof to perform the operations of this disclosure. These programming languages ​​include, but are not limited to, object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0096] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0097] The units described in the embodiments of this disclosure can be implemented in software or hardware. The names of the units are not, in some cases, intended to limit the specific unit.

[0098] The functions described above in this document can be performed at least in part by one or more hardware logic components. For example, exemplary types of hardware logic components that can be used, without limitation, include: field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), system-on-a-chip (SoCs), complex programmable logic devices (CPLDs), and so on.

[0099] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0100] The above description is merely a preferred embodiment of this disclosure and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of this disclosure is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features disclosed in this disclosure that have similar functions.

[0101] Furthermore, while the operations are described in a specific order, this should not be construed as requiring these operations to be performed in the specific order shown or in a sequential order. In certain environments, multitasking and parallel processing may be advantageous. Similarly, while several specific implementation details are included in the above discussion, these should not be construed as limiting the scope of this disclosure. Certain features described in the context of individual embodiments may also be implemented in combination in a single embodiment. Conversely, various features described in the context of a single embodiment may also be implemented individually or in any suitable sub-combination in multiple embodiments.

[0102] Although the subject matter has been described using language specific to structural features and / or methodological logic, it should be understood that the subject matter defined in the appended claims is not necessarily limited to the specific features or actions described above. Rather, the specific features and actions described above are merely illustrative examples of implementing the claims.

Claims

1. A method for binding interpretable output and evidence in generator diagnostics, characterized in that, include: Obtain the inspection image corresponding to the generator, perform anomaly diagnosis on the inspection image based on the artificial intelligence model, and obtain the diagnosis result. The diagnosis result includes anomaly location information and feature weight information. Convert the diagnosis result into an engineering semantic text description. Based on the feature weight information, the features of the region of interest of the artificial intelligence model in the inspection image are extracted, and based on the features of the region of interest, a visualization layer representing the decision basis of the artificial intelligence model is generated. Based on the inspection images, the visualization layers, the engineering semantic text descriptions, physical space coordinate information, and algorithm metadata, a structured evidence object is constructed. The physical space coordinate information is used to identify the physical location of the abnormal location information on the generator. The algorithm metadata includes the model version information and diagnosis time information of the artificial intelligence model. In response to a traceability request for a diagnostic result, the evidence object corresponding to the traceability request is parsed, and the inspection image, the visualization layer, the engineering semantic text description, the physical space coordinate information, and the algorithm metadata are displayed on the interactive interface.

2. The method according to claim 1, characterized in that, The step of extracting the region of interest features of the inspection image by the artificial intelligence model includes: The gradient-weighted class activation mapping algorithm is used to determine the importance weight of the feature map output by the last convolutional layer in the artificial intelligence model to the anomaly category. Based on the importance weights, the feature map regions that contribute the most to the classification decision are extracted as the regions of interest.

3. The method of claim 2, wherein, Determining the importance weight of the feature map output by the last convolutional layer in the artificial intelligence model to the anomaly category includes: For each feature map channel output by the last convolutional layer, determine the gradient of the anomaly category prediction score with respect to each pixel in each feature map channel; Global average pooling is performed on the gradient values ​​corresponding to each feature map channel to obtain the importance weight of each feature map channel for the anomaly category.

4. The method of claim 2, wherein, The step of extracting the feature map region that contributes the most to the classification decision as the feature of the region of interest based on the importance weight includes: Using the importance weights corresponding to each feature map channel as coefficients, a weighted sum is performed on all feature map channels output by the last convolutional layer; The weighted summation result is processed to retain positive contributions through activation, thereby removing features that have a negative impact on classification decisions.

5. The method of claim 1, wherein, The generation of the visualization layer representing the decision-making basis of the artificial intelligence model includes: A heatmap is generated based on the features of the region of interest; Adjust the size of the heat map to be the same as the inspection image; The resized heatmap is overlaid on the inspection image to form the visualization layer.

6. The method of claim 5, wherein, The step of overlaying the resized heatmap onto the inspection image includes: The heat map is mapped to a preset color image; The color image is overlaid on the inspection image as a semi-transparent overlay.

7. An apparatus for binding the interpretable output of a generator diagnosis with evidence, characterized in that include: The diagnostic result determination module is configured to acquire the inspection image corresponding to the generator, perform anomaly diagnosis on the inspection image based on an artificial intelligence model, obtain a diagnostic result, the diagnostic result includes anomaly location information and feature weight information, and convert the diagnostic result into an engineering semantic text description. The visualization generation module is configured to extract the region of interest features of the artificial intelligence model on the inspection image based on the feature weight information, and generate a visualization layer representing the decision basis of the artificial intelligence model based on the region of interest features. The evidence object construction module is configured to construct a structured evidence object based on the inspection image, the visualization layer, the engineering semantic text description, physical space coordinate information, and algorithm metadata. The physical space coordinate information is used to identify the physical location of the abnormal location information on the generator, and the algorithm metadata includes the model version information and diagnosis time information of the artificial intelligence model. The traceability interaction module is configured to respond to a traceability request for a diagnostic result, parse the evidence object corresponding to the traceability request, and display the inspection image, the visualization layer, the engineering semantic text description, the physical space coordinate information, and the algorithm metadata on the interactive interface.

8. An electronic device, comprising: The method includes a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the method according to any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, It stores a computer program / instruction thereon, which, when executed by a processor, implements the steps of the method according to any one of claims 1 to 6.

10. A computer program product, characterised in that, Includes a computer program / instructions that, when executed by a processor, implement the steps of the method according to any one of claims 1 to 6.

Images