Induction motor fault identification method and device, electronic equipment, computer readable storage medium and computer program product

Abstract

This application discloses a method, apparatus, electronic device, computer-readable storage medium, and computer program product for induction motor fault identification. The method includes the following steps: acquiring infrared thermal images of the induction motor under multiple operating conditions; preprocessing the acquired infrared thermal images to obtain real samples; and constructing a diffusion generation model jointly constrained by component region conditions and fault state conditions. The component region conditions characterize the component region category to which the infrared thermal image was acquired, and the fault state conditions characterize the fault state category of the induction motor when the infrared thermal image was acquired. This ensures consistency between the component region conditions and the fault state conditions between the real samples and the diffusion-enhanced samples, increases the number of enhanced samples consistent with the real thermal mechanism, and avoids affecting the model's generalization ability.

Description

Technical Field

[0001] This invention generally relates to the field of induction motors, specifically to the field of infrared thermal imaging detection and intelligent fault diagnosis technology, and particularly to an induction motor fault identification method, device, electronic device, computer-readable storage medium, and computer program product. Background Technology

[0002] Infrared thermal imaging technology can acquire the surface temperature distribution of induction motors without stopping the machine or contacting the equipment, offering advantages such as high detection efficiency, wide coverage, and suitability for online inspection. Different faults in induction motors can cause changes in frictional heating, electromagnetic losses, ventilation and heat dissipation paths, and local thermal coupling relationships, resulting in different hot spot distributions, degrees of thermal asymmetry, and changes in thermal gradients in the fan area, bearing area, and junction box area.

[0003] While existing deep learning-based infrared fault diagnosis methods can automatically extract thermal features, in industrial scenarios, infrared samples of induction motors typically suffer from limited sample size, significant load variations, strong environmental temperature disturbances, and inconsistent surface emissivity. Conventional geometric enhancement and photometric enhancement can only alter the appearance of the image and are insufficient to simulate the reasonable changes in hotspot intensity, thermal gradient, and temperature rise diffusion range of the induction motor fault thermal field. This can easily lead to inconsistencies between the enhanced samples and the actual thermal mechanisms, thereby affecting the model's generalization ability. Summary of the Invention

[0004] In view of the above-mentioned defects or deficiencies in the prior art, it is desirable to provide an induction motor fault identification method, device, electronic device, computer-readable storage medium and computer program product to increase the number of enhanced samples that are consistent with the real thermal mechanism.

[0005] In a first aspect, the induction motor fault identification method provided in the embodiments of this application includes the following steps: Acquire infrared thermal images of induction motors under multiple operating conditions; The acquired infrared thermal images are preprocessed to obtain real samples; A diffusion generation model is constructed under the joint constraints of component region conditions and fault state conditions. The component region conditions are used to characterize the component region category to which the infrared thermal image acquisition location belongs, and the fault state conditions are used to characterize the fault state category of the induction motor when the infrared thermal image is acquired. Generative enhancement of real samples is performed using a diffusion generation model to obtain diffusion-enhanced samples. The diffusion-enhanced samples and their corresponding real samples maintain the same component region conditions and fault state conditions, and the diffusion-enhanced samples and their corresponding real samples satisfy the preset thermal distribution difference constraint. The diffusion-enhanced samples are mixed with real samples to form training data, which is used to train the infrared thermal image classification model to obtain the infrared fault identification model of the induction motor. The infrared thermal image of the induction motor to be identified is input into the infrared fault identification model of the induction motor to obtain the component area category and fault state category of the induction motor to be identified.

[0006] Secondly, the infrared fault identification device for induction motors provided in the embodiments of this application includes: The acquisition module is used to acquire infrared thermal images of the induction motor under multiple operating conditions; The preprocessing module is used to preprocess the acquired infrared thermal images to obtain real samples; The construction module is used to build a diffusion generation model subject to joint constraints of component region conditions and fault state conditions. The component region conditions are used to characterize the component region category to which the infrared thermal image acquisition location belongs, and the fault state conditions are used to characterize the fault state category of the induction motor when the infrared thermal image is acquired. The diffusion enhancement module is used to perform generative enhancement on real samples using a diffusion generation model to obtain diffusion-enhanced samples. The diffusion-enhanced samples and their corresponding real samples maintain the same component region conditions and fault state conditions, and the diffusion-enhanced samples and their corresponding real samples satisfy the preset thermal distribution difference constraints. The training module is used to mix diffusion-enhanced samples with real samples to form training data, train the infrared thermal image classification model, and obtain the infrared fault identification model of induction motor. An infrared fault identification model for induction motors is used to process the infrared thermal image of the induction motor to be identified, and to obtain the component area category and fault state category of the induction motor to be identified.

[0007] Thirdly, the electronic device provided in the embodiments of this application includes a memory and a processor. The memory stores a computer program, and the processor executes computer execution instructions stored in the memory, causing the processor to perform the first aspect and / or various possible implementations of the first aspect as described above.

[0008] Fourthly, the computer-readable storage medium provided in the embodiments of this application stores a computer program thereon, and the computer execution instructions are executed by a processor to implement the first aspect and / or various possible implementations of the first aspect as described above.

[0009] Fifthly, the computer program product provided in the embodiments of this application includes a computer program that, when executed by a processor, implements the first aspect and / or various possible implementations of the first aspect.

[0010] According to the induction motor fault identification method, apparatus, electronic device, computer-readable storage medium, and computer program product provided in the embodiments of this application, the real samples are generatively enhanced by a diffusion generation model to obtain diffusion-enhanced samples. The diffusion generation model is jointly constrained by component region conditions and fault state conditions, which ensures that the component region conditions and fault state conditions are consistent between the real samples and the diffusion-enhanced samples. This increases the number of enhanced samples that are consistent with the real thermal mechanism and avoids affecting the model's generalization ability. Attached Figure Description

[0011] Other features, objects, and advantages of this application will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings: Figure 1 This is a flowchart illustrating an embodiment of an induction motor fault identification method according to the present invention. Figure 2 The training accuracy curve of an induction motor fault identification method according to an embodiment of the present invention; Figure 3 The training loss curve of an induction motor fault identification method according to an embodiment of the present invention; Figure 4 This is a confusion matrix of an induction motor fault identification method according to an embodiment of the present invention; Figure 5 This is a schematic diagram of the structure of a computer-readable storage medium according to another embodiment of the present invention. Detailed Implementation

[0012] The present application will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and not intended to limit it. Furthermore, it should be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings.

[0013] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0014] In one embodiment of this application, reference is made to Figure 1 A method for identifying faults in an induction motor includes the following steps: Step S100: Acquire infrared thermal images of the induction motor under multiple operating conditions.

[0015] In one possible implementation, multiple operating conditions include normal operating conditions, rotor eccentricity fault conditions, bearing damage fault conditions, and rotor bar breakage fault conditions, and the infrared thermal image acquisition area includes the fan observation area, the bearing observation area, and the junction box observation area.

[0016] Specifically, the fan observation area, bearing observation area, and junction box observation area are the external surface observation areas of the induction motor that have the ability to characterize the differences in thermal response under different operating conditions. The fan area mainly characterizes the thermal field characteristics related to end heat dissipation, the bearing area mainly characterizes the thermal field characteristics related to frictional heating of the support part, and the junction box area mainly characterizes the external thermal field characteristics related to winding leads and electromagnetic thermal coupling.

[0017] Alternatively, infrared thermal images can be acquired using a fixed infrared thermal imager or a handheld infrared thermal imager.

[0018] Specifically, a fixed or handheld infrared thermal imager is used to collect infrared radiation signals from the induction motor, and the collected temperature data is converted into a pseudo-color infrared thermal image for subsequent visualization and model training. The pseudo-color infrared thermal image refers to a color thermal image obtained by converting infrared temperature or heat intensity information according to a preset color mapping rule.

[0019] Optionally, during the acquisition of infrared thermal images, the parameters of the acquisition equipment, ambient temperature, and operating conditions are recorded. These parameters are then used to calibrate and compensate the infrared temperature measurement results, eliminating the influence of differences in shooting parameters, environmental interference, and operating load on temperature distribution. This ensures the consistency, accuracy, and comparability of the infrared thermal image data, providing a standardized and high-quality data foundation for subsequent model training and fault diagnosis.

[0020] Step S200: Preprocess the acquired infrared thermal image to obtain a real sample.

[0021] Optionally, preprocessing includes region correspondence correction, size unification, and normalization.

[0022] Specifically, when preprocessing infrared thermal images, the images are first adjusted to a uniform input size, and then standardized using normalization parameters that match the pre-trained classification backbone network.

[0023] Optionally, preprocessing includes one or more of the following: thermal contrast adjustment, random flipping, random small-angle rotation, and noise perturbation of the infrared thermal image, wherein the monotonic relationship of the infrared temperature mapping in the infrared thermal image remains unchanged. This can improve the model's adaptability to changes in imaging viewpoint and thermal contrast, and enhance training robustness.

[0024] Specifically, maintaining the monotonic relationship of infrared temperature mapping in infrared thermal images involves: after region registration of the infrared thermal images before and after enhancement, calculating the temperature order correlation coefficient within the same component region; if the temperature order correlation coefficient is not less than a preset correlation coefficient threshold, the monotonic relationship of infrared temperature mapping is considered unchanged. For pseudo-color infrared thermal images, the hue perturbation amplitude, brightness perturbation amplitude, and contrast perturbation amplitude are also limited to corresponding preset ranges. In other words, to maintain the monotonic relationship of infrared temperature mapping in infrared thermal images, after thermal contrast adjustment, random small-angle rotation, noise perturbation, or pseudo-color perturbation, region registration is performed on the infrared thermal images before and after enhancement, and the monotonic relationship of infrared temperature mapping is determined within the same component region. The same component region can be a fan region, a bearing region, or a junction box region.

[0025] For infrared thermal images with the original temperature matrix, the temperature order correlation coefficient is calculated based on the original temperature matrix before and after enhancement. For pseudo-color infrared thermal images, the temperature order correlation coefficient is calculated based on the grayscale thermal image before pseudo-color mapping or the thermal intensity image obtained through color inverse mapping. The temperature order correlation coefficient is used to characterize whether the relative temperature order of high-temperature and low-temperature regions remains consistent before and after enhancement.

[0026] When the temperature order correlation coefficient is not less than a preset correlation coefficient threshold, the enhanced infrared thermal image is determined to maintain the monotonic relationship of the infrared temperature mapping. For example, the preset correlation coefficient threshold can be between 0.90 and 0.98, preferably 0.95. The temperature order correlation coefficient can be either the Spearman rank correlation coefficient or the Kendall rank correlation coefficient.

[0027] For pseudocolor infrared thermal images, the amplitudes of hue perturbation, luminance perturbation, and contrast perturbation are further limited to avoid reversing the color representation order in high-temperature and low-temperature regions. For example, the hue perturbation amplitude does not exceed 10 degrees, the luminance perturbation amplitude does not exceed 10% of the original luminance range, and the contrast perturbation amplitude does not exceed 15% of the original contrast. When the pseudocolor mapping rule is known, the color representation order is rechecked according to the original pseudocolor mapping rule after perturbation; when the pseudocolor mapping rule is unknown, enhancement operations are preferentially performed on the original temperature matrix or grayscale thermal image before pseudocolor mapping.

[0028] If the enhanced infrared thermal image does not meet the preset correlation coefficient threshold, or if the pseudo-color perturbation exceeds the corresponding preset range, the enhanced image is discarded, or the preprocessing enhancement operation is re-executed. This method allows for a quantitative determination of whether the monotonic relationship of the infrared temperature mapping remains unchanged, preventing the enhancement operation from causing a reversal of the temperature semantics between high-temperature and low-temperature regions.

[0029] Step S300: Construct a diffusion generation model constrained by component region conditions and fault state conditions. The component region conditions are used to characterize the component region category to which the infrared thermal image acquisition location belongs, and the fault state conditions are used to characterize the fault state category of the induction motor when the infrared thermal image is acquired.

[0030] Optionally, an infrared thermal image classification model can be used as the classification backbone network. The backbone network is responsible for extracting global thermal distribution features and local hotspot texture features from the infrared thermal image. After the backbone network, a random deactivation layer and a fully connected classification layer are connected to output the probability of each fault category. The backbone network parameters can be frozen or only the later layers can be fine-tuned to avoid overfitting due to too many model parameters under small sample conditions.

[0031] Step S400: Generative enhancement of the real sample is performed using a diffusion generation model to obtain a diffusion-enhanced sample. The diffusion-enhanced sample and its corresponding real sample maintain the same component region conditions and fault state conditions, and the diffusion-enhanced sample and its corresponding real sample satisfy the preset thermal distribution difference constraint.

[0032] Optionally, the diffusion generation model employs a dual-condition generation method to augment the infrared thermal image.

[0033] Specifically, random noise, diffusion time steps, component region conditions, and fault state conditions are input into the diffusion generation model to generate infrared thermal images of induction motors of corresponding categories. This model can precisely constrain the component location and fault semantic category of the infrared thermal images, enabling targeted generation of infrared thermal images for specified components and fault types. While ensuring the physical rationality and category consistency of the generated samples, it significantly improves sample diversity and data availability, effectively alleviating the problems of scarce infrared fault samples and class imbalance, and providing high-quality training data with well-structured, accurately labeled, and diverse distributions for subsequent classification models.

[0034] Optionally, the diffusion generation model uses an image-to-image enhancement approach to augment the infrared thermal image.

[0035] Specifically, the diffusion generation model is fed with a real infrared thermal image of an induction motor, control noise intensity, and corresponding component region conditions and fault state conditions. Semantic-preserving thermal field enhancement is performed on the real image, outputting an enhanced image. Controlled noise is added to the real infrared thermal image, followed by conditional denoising, outputting an enhanced image that is semantically consistent with the original image but with slightly different thermal distribution. This achieves controllable and physically reasonable thermal field enhancement while fully preserving the semantics and structure of the real image. The sample fidelity is high, the distribution is realistic and reliable, and it can effectively improve dataset quality and model robustness.

[0036] For example, thermal distribution difference does not refer to pixel-level irregular differences caused by random noise, but rather to measurable changes in the infrared thermal field in terms of hot spot intensity, thermal gradient distribution, temperature rise diffusion range, or local background thermal texture, provided that the component area category and fault state category remain consistent.

[0037] Specifically, when determining whether the diffusion-enhanced sample has a reasonable thermal distribution difference, the diffusion-enhanced sample and its corresponding real sample are first registered to the same component region, and the similarity of thermal field structure, temperature order correlation, main hot spot center offset, hot spot intensity difference, thermal gradient distribution difference and temperature rise diffusion range difference are judged within the component region.

[0038] When the thermal field structure similarity between the diffusion-enhanced sample and its corresponding real sample is greater than the first threshold, the temperature order correlation is greater than the second threshold, and the offset of the main hot spot center is less than the third threshold, and at least one of the hot spot intensity difference, thermal gradient distribution difference, or temperature rise diffusion range difference is greater than the fourth threshold and less than the fifth threshold, it is determined that the diffusion-enhanced sample and its corresponding real sample satisfy the preset thermal distribution difference constraint.

[0039] Among them, the first to third thresholds are used to ensure that the diffusion-enhanced samples do not deviate from the original component region semantics and fault state semantics; the fourth threshold is used to exclude near-duplicate samples or samples with weak noise disturbances; and the fifth threshold is used to exclude abnormally generated samples with excessive thermal field changes or fault semantic distortion.

[0040] Diffusion-enhanced samples that do not meet the preset thermal distribution difference constraints are removed, or their sampling probability in the training batch is reduced. In this way, the diffusion-enhanced samples entering the training phase maintain both the semantic consistency of the component region and the semantic consistency of the fault state, and have measurable thermal field diversity, thus distinguishing them from conventional enhancement methods such as simple random noise perturbation, pruning, and flipping.

[0041] Step S500: Mix the diffusion-enhanced samples with real samples to form training data, train the infrared thermal image classification model, and obtain the infrared fault identification model for induction motors.

[0042] For example, both the diffusion-enhanced samples obtained by the diffusion generation model through dual-condition generation of infrared thermal images and the diffusion-enhanced samples obtained by the diffusion generation model through image-to-image enhancement of infrared thermal images can be mixed with real samples to form training data.

[0043] The proportion of augmented samples can be set based on the validation set performance. In practice, multiple candidate proportions can be pre-defined for comparison, including but not limited to 10%, 20%, 30%, 40%, and 50%. These candidate proportions are trained under the same number of training epochs and parameters, and the optimal proportion is selected based on the classification accuracy, recall, F1 score, or loss convergence on the validation set. A lower proportion of augmented samples is used in the early stages of training to ensure the model learns the true hot distribution features first. The proportion of augmented samples is gradually increased in the later stages of training to enhance sample diversity. When augmented samples cause a decline in validation set performance or increased class confusion, their proportion is reduced or the introduction of corresponding augmented samples is stopped.

[0044] When training the infrared image classification model, a batch training data is constructed by mixing real samples and diffusion-enhanced samples to expand the coverage of similar samples in the thermal distribution space and improve the separability between different fault categories.

[0045] Optionally, when training the infrared thermal image classification model, a cross-entropy loss function can be used to constrain the consistency between the predicted class and the true label, and an adaptive moment estimation optimizer can be used for parameter updates. A fixed random seed can be maintained, and consistency can be ensured in image partitioning, batch training order, and optimization parameter settings to guarantee reproducibility of results.

[0046] After training, an infrared fault identification model for induction motors with multiple components and operating conditions is obtained.

[0047] Step S600: Input the infrared thermal image of the induction motor to be identified into the induction motor infrared fault identification model to obtain the component area category and fault state category of the induction motor to be identified.

[0048] Specifically, during the inference phase, the infrared thermal image of the induction motor to be identified is input into the trained induction motor infrared fault identification model. The model outputs the predicted probabilities for each category and provides the final fault state category. If the input image comes from any of the fan area, bearing area, or junction box area, the induction motor infrared fault identification model can simultaneously determine the component area category and fault state category corresponding to the infrared thermal image of the induction motor to be identified. In practical deployment, this result can be used for equipment inspection and early warning, condition assessment, maintenance decision-making, and fault trend analysis.

[0049] In one possible implementation, a diffusion generation model constrained by both component region conditions and fault state conditions is established, including: encoding the component region conditions and fault state conditions separately to obtain component region condition codes and fault state condition codes; cross-fusion the component region condition codes and fault state condition codes to obtain dual-condition fusion features; and inputting the dual-condition fusion features as conditional information into the diffusion generation model so that the diffusion generation model is simultaneously constrained by both component region conditions and fault state conditions during the reverse denoising process.

[0050] Specifically, the cross-fusion is achieved using a cross-attention mechanism. The component region condition code is used as the first conditional feature, and the fault state condition code is used as the second conditional feature. An attention weight is calculated between the first and second conditional features, and the fault state condition code or component region condition code is weighted and adjusted according to this weight to obtain a dual-conditional fusion feature. This dual-conditional fusion feature simultaneously contains component region thermal field structure information and fault state thermal anomaly information, which is used to constrain the reverse denoising process of the diffusion generation model.

[0051] Dual-condition fusion features can be used as supplementary inputs to the diffusion time step conditions, conditional inputs to the intermediate layer of the inverse denoising network, conditional normalization parameters, or attention weights, and injected into the inverse denoising network of the diffusion generation model, so that the diffusion generation model can perceive the component region to which the infrared thermal image belongs and the corresponding fault state at each diffusion time step.

[0052] It should be noted that the cross-fusion in this application differs from directly merging component region conditions and fault state conditions into twelve single labels. The twelve single labels only represent the final combined category, such as "fan-normal" or "bearing-bearing damage," which essentially compresses two semantic dimensions into a single category number. In contrast, this application retains the component region condition encoding and the fault state condition encoding separately, and calculates the association weight between the two through a cross-attention mechanism, thereby displaying the interaction between the modeled component region thermal field structure and the fault state thermal anomaly.

[0053] Specifically, infrared thermal images are labeled using a dual-label method of "component area - fault status," forming at least twelve categories of labels. The first label in the dual-label method indicates the component area to which the infrared thermal image was acquired, while the second label indicates the overall operating status of the induction motor at the time the infrared thermal image was acquired.

[0054] The twelve classification labels include: Fan - Normal, Fan - Rotor Eccentricity, Fan - Bearing Damage, Fan - Rotor Broken Bar, Bearing - Normal, Bearing - Rotor Eccentricity, Bearing - Bearing Damage, Bearing - Rotor Broken Bar, Junction Box - Normal, Junction Box - Rotor Eccentricity, Junction Box - Bearing Damage, and Junction Box - Rotor Broken Bar. For example, "Fan - Rotor Eccentricity" indicates that the infrared thermal image was acquired from the fan area, and the induction motor as a whole was in a rotor eccentric state at the time of acquisition. The accuracy of each fault category under the diffusion enhancement neural network is shown in Table 1.

[0055] Table 1: Accuracy of each fault category under the diffusion-enhanced neural network In one possible implementation, the method further includes: using a diffusion generation model to learn the thermal field distribution characteristics of different component regions of the induction motor and under different fault states; and using the forward noise addition process and the reverse noise reduction process of the diffusion generation model to establish a mapping from the noise space to the infrared thermal image space.

[0056] Specifically, the diffusion generation model is trained using a category-constrained approach. During the forward noise addition process, noise is progressively added to the real infrared thermal image; during the backward denoising process, a neural network is used to learn and progressively denoise, enabling the diffusion generation model to recover an infrared thermal image with target fault semantics under given category conditions. Because the generative enhancement process can introduce random but reasonable changes in thermal texture, the generated result can maintain the component region conditions and fault state conditions unchanged, while altering the intensity of hot spots, thermal gradient distribution, temperature rise diffusion range, and local background texture.

[0057] In one possible implementation, the infrared thermal image classification model uses a deep neural network as the backbone network for feature extraction, and the classification head of the infrared thermal image classification model includes at least one randomly deactivated layer and a fully connected layer.

[0058] Specifically, the infrared image classification model uses a convolutional neural network or other deep neural network as the backbone network for feature extraction.

[0059] Specifically, the classification head includes at least one randomly deactivated layer and a fully connected layer to output the predicted probabilities of each category. The fully connected layer efficiently maps the high-dimensional features extracted from the backbone to the category space, completing feature fusion and classification decisions, and outputting category probabilities. The randomly deactivated layer randomly masks some neurons during training, reducing neuronal co-adaptation, effectively suppressing overfitting, and improving the model's generalization ability and robustness.

[0060] The induction motor fault identification method of this application is applicable to non-contact infrared fault detection of induction motors under conditions of load fluctuation, ambient temperature change, surface emissivity difference and imaging angle change, and is used to improve the accuracy, robustness and generalization ability of infrared fault identification of induction motors.

[0061] The following is a complete and detailed explanation of the entire process of the induction motor fault identification method.

[0062] The induction motor fault identification method of this embodiment includes the following steps: S1, Infrared Image Acquisition and Category Labeling. Acquire infrared images of the induction motor in the fan area, bearing area, and junction box area. The operating status includes at least the normal state, rotor eccentricity state, bearing damage state, and rotor bar breakage state.

[0063] For example, suppose the original sample set is: in Indicates the first 1 infrared image, This represents the corresponding category label. The formula indicates that a supervised learning dataset is constructed using "image-label" sample pairs, providing the input basis for subsequent diffusion augmentation and classification training.

[0064] S2, Image Preprocessing. The input image is uniformly adjusted to a preset size and normalized.

[0065] For example, the normalization expression is: in and These represent the mean vector and standard deviation vector used in the pre-trained network, respectively. This formula maps the pixel distribution under different imaging conditions to a uniform numerical range, thereby reducing the impact of brightness scale differences on network training and improving model convergence stability.

[0066] S3, Conditional Diffusion Modeling. Forward noise addition is performed on the real infrared image to obtain noise samples at different time steps.

[0067] For example, its expression is: This can be further written as: in Indicates the first The noise intensity of each diffusion step. This is the cumulative fidelity factor. This refers to random noise that follows a standard Gaussian distribution. The meaning of the above formula is that by gradually injecting noise into the real infrared image, a diffusion trajectory from the clear heatmap to the noise space is established, enabling the model to learn the statistical distribution of various fault thermal textures.

[0068] S4, Diffusion-enhanced sample generation. Based on class conditions. and noisy images Train the denoising network to predict noise.

[0069] For example, its objective can be written as: This formula represents the parameter as The denoising network at a given diffusion step With category conditions During the process, noise components are estimated, thereby gradually recovering an infrared image that satisfies the semantic constraints of the target category in the reverse process. This mechanism can be used to generate enhanced samples that are semantically consistent with real samples but have reasonable differences in local hotspot distribution and thermal texture. .

[0070] S5, training the classification network with a mixture of real and diffused samples. Real and diffused enhanced samples are input together into the infrared image classification model to form a training set. Let the output of the classification network be the first... The class prediction probability is: in Output the category value for the corresponding category. This represents the total number of categories. The formula indicates that by mapping the network output to the probabilities of each category using a normalized exponential function, the model can simultaneously perform component identification and fault state identification.

[0071] Furthermore, the classification training loss function can be expressed as: in This is a one-hot encoding of the true label. The formula indicates that cross-entropy is used to measure the difference between the predicted distribution and the true distribution. When the model assigns a higher probability to the true class, the loss value decreases, thus driving the network to learn more stable discriminative features.

[0072] During the parameter update phase, the model parameters iterate as follows: in Indicates the learning rate. This represents the loss function for the current batch. This formula reflects the basic process by which the optimizer updates model parameters based on the gradient of the loss function; in practical implementations, an adaptive moment estimation optimizer can be used to improve convergence speed and training stability.

[0073] S6, Fault Identification Output. After the infrared image to be tested is input into the trained network, the output category prediction result is given: This formula indicates that the category corresponding to the highest predicted probability is taken as the final diagnostic label, thereby obtaining the component area to which the induction motor belongs and its fault status; at the same time, it can also output the probability of each category to assist in completing early warning analysis, status assessment and maintenance decision-making.

[0074] Figure 2 The figure shows the training accuracy curves for the infrared thermal image classification model. In the initial training phase (0-10 epochs), both the training accuracy (blue) and validation accuracy (orange) rapidly increased from a low initial level to over 95%, indicating high learning efficiency and rapid convergence. In the later training phase (10-50 epochs), both curves rose steadily and eventually stabilized above 98%. The training accuracy was slightly higher than the validation accuracy, but the difference was minimal, showing no significant overfitting. This demonstrates the model's strong generalization ability and robustness on unseen data. The later curves showed no significant oscillations, indicating a smooth training process, sufficient parameter optimization, and the achievement of ideal classification accuracy.

[0075] Figure 3 The figure shows the training loss curves for training the infrared thermal image classification model. In the initial training phase (0-10 epochs), both the training and validation losses rapidly decreased from high levels (1.8 / 1.2). The validation loss stabilized around 10 epochs, while the training loss continued to decrease slowly until it approached zero, indicating that the model had fully learned the features of the training data. The validation loss (orange) stabilized at around 0.2 after 10 epochs, without showing an upward trend, and the difference between it and the training loss remained within a reasonable range, ruling out overfitting and verifying the model's generalization ability. The loss curves showed no abnormal oscillations, indicating that the optimizer parameters were set reasonably, the model training process was stable, and it eventually converged to near the global optimum.

[0076] Figure 4The confusion matrix of the induction motor infrared fault identification model is shown. The values ​​on the diagonal (correctly classified) lines are significantly higher than those on the off-diagonal (misclassified) lines, indicating accurate identification of the vast majority of categories. The accuracy of the normal category (junction box / bearing / fan - normal) is close to 100% (almost all 356 / 355 / 356 samples are correct). The accuracy of identifying various fault states (rotor eccentricity, rotor bar breakage, bearing damage) is above 95%, with only a very small number of cross-category misclassifications. These misclassifications mostly occur between different faults within the same component, with no serious semantic errors. The confusion matrix clearly demonstrates the excellent performance of the induction motor infrared fault identification model in the induction motor infrared fault classification task, accurately distinguishing the thermal field characteristics of different components and different fault states, meeting the accuracy requirements of industrial fault diagnosis.

[0077] Based on the same inventive concept, this application also provides an infrared fault identification device for induction motors to implement the aforementioned induction motor fault identification method. The solution provided by this infrared fault identification device is similar to the solution described in the aforementioned induction motor fault identification method. Therefore, the specific limitations in one or more device embodiments provided below can be found in the limitations of the induction motor fault identification method described above, and will not be repeated here.

[0078] In another embodiment of this application, an infrared fault identification device for an induction motor includes: an acquisition module for acquiring infrared thermal images of the induction motor under multiple operating conditions; a preprocessing module for preprocessing the acquired infrared thermal images to obtain real samples; a construction module for constructing a diffusion generation model constrained by component region conditions and fault state conditions, wherein the component region conditions characterize the component region category to which the infrared thermal image acquisition location belongs, and the fault state conditions characterize the fault state category of the induction motor when the infrared thermal image is acquired; a diffusion enhancement module for generatively enhancing the real samples using the diffusion generation model to obtain diffusion-enhanced samples, wherein the diffusion-enhanced samples and their corresponding real samples maintain consistency in component region conditions and fault state conditions, and the diffusion-enhanced samples and their corresponding real samples satisfy a preset thermal distribution difference constraint; a training module for mixing the diffusion-enhanced samples and real samples to form training data, training the infrared thermal image classification model, and obtaining an infrared fault identification model for the induction motor; and an infrared fault identification model for processing the infrared thermal image of the induction motor to be identified to obtain the component region category and fault state category of the induction motor to be identified.

[0079] In another embodiment of this application, an electronic device includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps of the above-described induction motor fault identification method.

[0080] In another embodiment of this application, a computer-readable storage medium is provided, on which a computer program is stored, which, when executed by a processor, implements the steps of the above-described induction motor fault identification method.

[0081] In another embodiment of this application, a computer program product includes a computer program that, when executed by a processor, implements the steps of the above-described induction motor fault identification method.

[0082] refer to Figure 5 The computer system includes a central processing unit (CPU) 501, which can perform various appropriate actions and processes based on programs stored in read-only memory (ROM) 502 or programs loaded from storage into random access memory (RAM) 503. RAM 503 also stores various programs and data required for system operation. CPU 501, ROM 502, and RAM 503 are interconnected via bus 504. Input / output (I / O) interface 505 is also connected to bus 504.

[0083] The following components are connected to I / O interface 505: an input section 506 including a keyboard, mouse, etc.; an output section including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and speakers, etc.; a storage section 508 including a hard disk, etc.; and a communication section 509 including a network interface card such as a LAN card, modem, etc. The communication section 509 performs communication processing via a network such as the Internet. A drive is also connected to I / O interface 505 as needed. A removable medium 511, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on drive 510 as needed so that computer programs read from it can be installed into storage section 508 as needed.

[0084] In particular, according to embodiments of the present invention, the above-described reference process Figure 1 The described process can be implemented as a computer software program. For example, one embodiment of the invention includes a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network via a communication component, and / or installed from a removable medium. When the computer program is executed by a central processing unit (CPU) 501, it performs the functions defined in the system of this application.

[0085] It should be noted that the computer-readable medium shown in this invention 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 invention, 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 invention, 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. Computer-readable signal media can also be any computer-readable medium other than computer-readable storage media, 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: wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.

[0086] 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 the present invention. 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 a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may 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.

[0087] The units described in the embodiments of the present invention can be implemented in software or hardware, and can also be housed in a processor. The names of these units are not necessarily limiting of the unit itself. The described units or modules can also be housed in a processor; for example, a processor may be described as including an acquisition module, a preprocessing module, a diffusion enhancement module, a training module, and an infrared fault identification model for an induction motor. The names of these units or modules are not necessarily limiting of the unit or module itself; for example, the acquisition module may also be described as "an acquisition module for acquiring infrared thermal images of an induction motor under multiple operating conditions."

[0088] In another aspect, this application also provides a computer-readable medium, which may be included in the electronic device described in the above embodiments; or it may exist independently and not assembled into the electronic device. The computer-readable medium carries one or more programs, which, when executed by the electronic device, cause the electronic device to implement the induction motor fault identification method as described in the above embodiments.

[0089] For example, the electronic device can achieve the following: Figure 1 The steps are as follows: Step S100, acquiring infrared thermal images of the induction motor under multiple operating conditions; Step S200, preprocessing the acquired infrared thermal images to obtain real samples; Step S300, constructing a diffusion generation model constrained by component region conditions and fault state conditions, wherein the component region conditions characterize the component region category to which the infrared thermal image acquisition location belongs, and the fault state conditions characterize the fault state category of the induction motor when the infrared thermal image is acquired; Step S400, using the diffusion generation model to perform generative enhancement on the real samples to obtain diffusion-enhanced samples, wherein the diffusion-enhanced samples and their corresponding real samples maintain consistency in component region conditions and fault state conditions, and the diffusion-enhanced samples and their corresponding real samples satisfy a preset thermal distribution difference constraint; Step S500, mixing the diffusion-enhanced samples and real samples to form training data, training the infrared thermal image classification model, and obtaining an infrared fault identification model for the induction motor; Step S600, inputting the infrared thermal image of the induction motor to be identified into the infrared fault identification model for the induction motor to obtain the component region category and fault state category of the induction motor to be identified.

[0090] It should be noted that although several modules or units for the device used to perform actions have been mentioned in the detailed description above, this division is not mandatory. In fact, according to embodiments of this disclosure, the features and functions of two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided and embodied by multiple modules or units.

[0091] Furthermore, although the steps of the method in this disclosure are described in a specific order in the accompanying drawings, this does not require or imply that the steps must be performed in that specific order, or that all the steps shown must be performed to achieve the desired result. Additionally or alternatively, certain steps may be omitted, multiple steps may be combined into one step, and / or a step may be broken down into multiple steps.

[0092] Through the above description of the embodiments, those skilled in the art will readily understand that the exemplary embodiments described herein can be implemented by software or by combining software with necessary hardware.

[0093] The above description is merely a preferred embodiment of this application and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in this application 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 inventive concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features with similar functions disclosed in this application.

Claims

1. A method for identifying faults in an induction motor, characterized in that, Includes the following steps: Acquire infrared thermal images of induction motors under multiple operating conditions; The acquired infrared thermal images are preprocessed to obtain real samples; A diffusion generation model is constructed under the joint constraints of component region conditions and fault state conditions. The component region conditions are used to characterize the component region category to which the infrared thermal image acquisition location belongs, and the fault state conditions are used to characterize the fault state category of the induction motor when the infrared thermal image is acquired. Generative enhancement of the real sample is performed using a diffusion generation model to obtain a diffusion-enhanced sample. The diffusion-enhanced sample and the real sample corresponding to it maintain the same component region conditions and fault state conditions, and the diffusion-enhanced sample and the real sample corresponding to it satisfy a preset thermal distribution difference constraint. The diffusion-enhanced samples are mixed with real samples to form training data, which is used to train the infrared thermal image classification model to obtain the infrared fault identification model of the induction motor. The infrared thermal image of the induction motor to be identified is input into the infrared fault identification model of the induction motor to obtain the component area category and fault state category of the induction motor to be identified.

2. The method according to claim 1, characterized in that, A diffusion generation model constrained by both component region conditions and fault state conditions is established, including: The component area conditions are encoded to obtain the component area condition code, and the fault state conditions are encoded to obtain the fault state condition code; The component region condition code and the fault state condition code are cross-fused to obtain a dual-condition fusion feature; The dual-condition fusion features are input as conditional information into the inverse denoising network of the diffusion generation model to establish a diffusion generation model jointly constrained by component region conditions and fault state conditions.

3. The method according to claim 1, characterized in that, Also includes: The thermal field distribution characteristics of different component regions of an induction motor under different fault conditions are learned using a diffusion generation model. By utilizing the forward noise addition process and the reverse noise reduction process of the diffusion generation model, a mapping from the noise space to the infrared thermal image space is established.

4. The method according to claim 1, characterized in that, The preprocessing includes region correspondence correction, size unification, and normalization; and The infrared thermal image is subjected to one or more of the following: thermal contrast adjustment, random flipping, random small-angle rotation, and noise perturbation, wherein the monotonic relationship of the infrared temperature mapping in the infrared thermal image is maintained unchanged.

5. The method according to claim 1, characterized in that, The diffusion-enhanced sample and its corresponding real sample satisfy a preset thermal distribution difference constraint, including: Register the diffusion-enhanced sample and its corresponding real sample to the same component region; Within the component area, the similarity of the thermal field structure, the correlation of temperature order, the offset of the main hot spot center, the difference in hot spot intensity, the difference in thermal gradient distribution, and the difference in temperature rise diffusion range are determined.

6. The method according to claim 1, characterized in that, The infrared thermal image classification model uses a deep neural network as the backbone network for feature extraction. The classification head of the infrared thermal image classification model includes at least one randomly deactivated layer and a fully connected layer.

7. An infrared fault identification device for an induction motor, characterized in that, include: The acquisition module is used to acquire infrared thermal images of the induction motor under multiple operating conditions; The preprocessing module is used to preprocess the acquired infrared thermal image to obtain a real sample; The construction module is used to build a diffusion generation model subject to joint constraints of component region conditions and fault state conditions. The component region conditions are used to characterize the component region category to which the infrared thermal image acquisition location belongs, and the fault state conditions are used to characterize the fault state category of the induction motor when the infrared thermal image is acquired. The diffusion enhancement module is used to perform generative enhancement on the real sample using a diffusion generation model to obtain a diffusion-enhanced sample. The diffusion-enhanced sample and the real sample corresponding to it maintain the same component region conditions and fault state conditions, and the diffusion-enhanced sample and the real sample corresponding to it satisfy a preset thermal distribution difference constraint. The training module is used to mix diffusion-enhanced samples with real samples to form training data, train the infrared thermal image classification model, and obtain the infrared fault identification model of induction motor. An infrared fault identification model for induction motors is used to process the infrared thermal image of the induction motor to be identified, and to obtain the component area category and fault state category of the induction motor to be identified.

8. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.

9. A computer-readable storage 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 to 6.

10. A computer program product, comprising a computer program, 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 to 6.

Images