PMSM fault diagnosis method and system based on image color features

Abstract

The application discloses a PMSM fault diagnosis method and system based on image color features, and the method comprises the following steps: acquiring original vibration signals of a motor to be tested under different states and various working conditions; performing empirical mode decomposition on the original vibration signals to obtain different level signal components by solving different level intrinsic mode functions; converting the original vibration signals into symmetric dot matrix diagrams of three RGB channels according to the different level signal components; improving DenseNet to learn the shape distribution and color features of images by fusing scSE modules, and constructing a classification network model; and inputting the symmetric dot matrix diagrams of the three RGB channels into the classification network model to output fault diagnosis results. The application has high diagnosis accuracy, non-invasive and high robustness, can realize rapid diagnosis of turn-to-turn short circuit faults of a permanent magnet synchronous motor, and has good generalization ability and anti-interference ability, thereby ensuring faster fault diagnosis and more comprehensive and accurate fault feature extraction.

Description

Technical Field

[0001] This invention belongs to the field of motor technology and relates to a PMSM fault diagnosis method and system based on image color features. Background Technology

[0002] Permanent magnet synchronous motors (PMSMs) are widely used in new energy vehicles due to their advantages such as light weight, reliable operation, low noise, and high efficiency. However, due to inherent manufacturing defects in automotive PMSMs and the effects of wear, deformation, and corrosion during vehicle operation, the performance of automotive PMSMs gradually deteriorates as component performance worsens, potentially triggering safety hazards and even causing shutdowns and significant economic losses. Periodic intermittent operation and frequent starts can easily lead to overheating of the motor windings, resulting in insulation failure. Common fault types in automotive PMSMs include winding short circuits, demagnetization faults, and mechanical faults. Among these, inter-turn short circuit faults, uneven demagnetization faults, and eccentricity faults share similar fault characteristics, making it difficult to distinguish between them in current research. Because the acceleration signals of inter-turn short circuit faults, uneven demagnetization faults, and eccentricity faults have certain similarities and overlaps, distinguishing them using a single sensor's time-domain signal is challenging. Therefore, accurately diagnosing faults with small differences in single sensor signals is of significant research value.

[0003] In recent years, Symmetrical Point Maps (SDPs), as an intuitive signal representation method, have been widely used in fault diagnosis. Unlike traditional methods, SDPs can easily transform the original signal into a mirror-symmetric snowflake image with relatively low computational cost. SDPs identify and differentiate signals by analyzing the amplitude and frequency differences in the image. Therefore, SDPs are considered an effective and reliable fault diagnosis method. However, the cases studied in these SDP applications all involve different fault types with significant differences in the time domain signals. When the differences in the fault's time domain signals are small, the converted SDPs have high similarity, easily leading to misdiagnosis. Information from a single time domain or time-frequency domain dimension cannot fully reflect the unique information of the fault; it may be necessary to combine information from both dimensions to compensate for the limitations of a single dimension and improve the distinguishability of images from different types of motors.

[0004] Deep learning methods have made significant progress in fault diagnosis, with the advantage of automatically learning discriminative features without relying on specialized knowledge. However, most current deep learning methods can only operate under specific working conditions and cannot meet practical needs. To address this issue, many intelligent fault diagnosis methods for non-stationary conditions have emerged in recent years. Although these methods have certain advantages, they still have the following drawbacks due to the strict requirements of traditional time-frequency methods on the working environment: (1) These algorithms are only applicable to situations with small speed changes. (2) When the speed is transient, the algorithm performance will drop significantly. Summary of the Invention

[0005] The purpose of this invention is to address the problems in the prior art by providing a PMSM fault diagnosis method and system based on image color features. This method and system have high diagnostic accuracy, are non-invasive and highly robust, and can quickly diagnose inter-turn short-circuit faults in permanent magnet synchronous motors. Furthermore, it has good generalization ability and anti-interference capability.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] Firstly, a PMSM fault diagnosis method based on image color features is provided, including:

[0008] Acquire the original vibration signals of the motor under test under different states and multiple operating conditions;

[0009] The original vibration signal is subjected to empirical mode decomposition, and the eigenmode functions at different levels are solved to obtain signal components at different levels.

[0010] The original vibration signal is converted into a symmetrical dot matrix diagram with three RGB channels according to the signal components of different levels.

[0011] By integrating the scSE module, DenseNet learns the shape distribution and color features of images, and a classification network model is constructed.

[0012] Input the symmetrical dot matrix image of the three RGB channels into the classification network model and output the fault diagnosis result.

[0013] As a preferred embodiment, acquiring the original vibration signals of the motor under test under different states and multiple operating conditions includes:

[0014] Construct a fault simulation test bench and pre-set inter-turn short circuit fault, uneven demagnetization fault, eccentricity fault and normal motor;

[0015] Acceleration signal segments were collected for each motor under the following operating conditions:

[0016] Nine steady-state operating conditions: speeds of 1000 r / min, 1500 r / min, and 2000 r / min, and loads of no load, half load, and rated load, respectively; and six transient operating conditions: acceleration and deceleration conditions with loads of no load, half load, and rated load, respectively.

[0017] Each acceleration signal segment is divided into a data subset, which is then divided into a training set, a validation set, and a test set.

[0018] As a preferred embodiment, the step of performing empirical mode decomposition on the original vibration signal and solving for the eigenmode functions at different levels to obtain signal components at different levels includes:

[0019] The preprocessed original vibration signal is subjected to empirical mode decomposition to solve for the eigenmode functions at different levels.

[0020] Index all signals from the first three levels of empirical mode decomposition after decomposition, and solve for the maximum value z of the absolute value of all sampling points.

[0021] As a preferred embodiment, the process of converting the original vibration signal into a symmetrical dot matrix image with three RGB channels according to different levels of signal components includes:

[0022] The time-domain characteristics of the original vibration signal are represented by the shape and distribution of the image, and the time-frequency domain characteristics are represented by the RGB color characteristics of the image. The signal components at different levels are converted into RGB color characteristics and displayed on the image. The magnitudes of the first three signal components are selected to correspond to the three RGB channels respectively, so as to characterize the unique fault characteristics of the signal from different time scales.

[0023] As a preferred embodiment, the process of converting the original vibration signal into a symmetrical dot matrix image with three RGB channels according to different levels of signal components includes:

[0024] For a time series signal x, x i It is the value of the i-th sampling point of the time series signal x, x i+t It is adjacent time ω sq The sampling point values ​​after an interval t; when the time domain point x i Transforming to polar coordinate space S(r(i),θ(i),φ(i)), it can be expressed as:

[0025]

[0026] In the formula, x min and x max Let θ represent the minimum and maximum values ​​of the time-domain signal, respectively, with respect to the angle θ of the nth mirror-symmetric plane. nAs the axis of symmetry of polar coordinates, a fan-shaped lobe image is formed; θ(i) and Φ(i) represent the clockwise and counterclockwise rotation angles along the nth mirror symmetry plane in polar coordinates, respectively; as the amplification factor ζ and the time delay factor t change, each sampling point of the time-domain signal is intuitively displayed in the polar coordinate image, presenting different images;

[0027] Using the maximum absolute value z of all sampling points, solve for the RGB three-channel feature sequence T. R (i)T G (i)T B (i) is:

[0028]

[0029] Among them, IMF1 i For the first-level empirical mode decomposition component, corresponding to the i-th value, IMF1 i+t For the first-level empirical mode decomposition component, corresponding to the (i+t)th value;

[0030] The RGB three-channel feature sequences are mapped to the range of 0-255 by scaling according to the following formula:

[0031]

[0032] Where Round() is the rounding function; k = R, G, B represent the image color channels, and the R, G, and B channels correspond to the first three levels of intrinsic mode components, respectively; T k (i) represents the RGB three-channel feature sequence;

[0033] Display the color features corresponding to all points on the symmetrical bitmap.

[0034] As a preferred approach, the step of improving the shape distribution and color features of DenseNet learning images by integrating the scSE module and constructing a classification network model involves extracting the channel and spatial importance of the input feature maps, adding them together, and then strengthening the feature sub-maps that simultaneously possess important channel and spatial features, thereby prompting the network to learn more meaningful feature information.

[0035] As a preferred embodiment, the method of improving the shape distribution and color features of DenseNet learned images by fusing the scSE module to construct a classification network model includes:

[0036] Input feature map P = [p1, p2, ..., p N ], where the channel L and W are the height and width of the feature map, respectively;

[0037] Global spatial features are embedded into vector v through a global pooling layer, where... C represents the number of channels;

[0038] The obtained vector v is passed through a fully connected layer with weights ω1 and ω2, and then sequentially processed by the ReLU activation function and Sigmoid normalization to obtain the i-th channel p. i The degree of importance of features in, The value is

[0039] The number of channels is C and the weight is v. sq A 1×1 convolutional block compresses the input feature map P into channels, outputting a feature map fq = ω with 1 channel and size L×W. sq *P;

[0040] The obtained feature map fq is normalized by Sigmod to obtain the spatial information importance degree σ(fq(i,j)) of each spatial location (i,j) in the feature map, thereby enhancing the feature of important spatial locations;

[0041] right Add σ(fq(i,j)) to obtain the feature subgraph P′.

[0042] Secondly, a PMSM fault diagnosis system based on image color features is provided, including:

[0043] The raw vibration signal acquisition module is used to acquire the raw vibration signals of the motor under test under different states and multiple operating conditions;

[0044] The Empirical Mode Decomposition (EMD) module is used to perform EMD on the original vibration signal and solve for the eigenmode functions at different levels to obtain signal components at different levels.

[0045] The symmetrical dot matrix conversion module is used to convert the original vibration signal into a symmetrical dot matrix with three RGB channels according to different levels of signal components.

[0046] The classification network model building module is used to improve the shape distribution and color features of DenseNet images by fusing the scSE module, and to build a classification network model.

[0047] The output module is used to input the symmetrical dot matrix image of the three RGB channels into the classification network model and output the fault diagnosis results.

[0048] Thirdly, an electronic device is provided, comprising: a memory storing at least one instruction; and a processor executing the instruction stored in the memory to implement the PMSM fault diagnosis method based on image color features.

[0049] Fourthly, a computer-readable storage medium is provided, the computer-readable storage medium storing a computer program, which, when executed by a processor, implements the PMSM fault diagnosis method based on image color features.

[0050] Compared with the prior art, the present invention has at least the following beneficial effects:

[0051] To improve the ability to distinguish similar fault types, this invention proposes a signal-to-image conversion method based on Empirical Mode Decomposition-Symmetric Point Pattern (EMD-SDP). This method displays the time-frequency domain features of EMD in color on the SDP image, reducing the inter-class similarity of SDP images generated by similar faults. Furthermore, this invention proposes a classification network model based on an improved DenseNet convolutional neural network fused with the scSE attention mechanism. This enhances the network's ability to learn important channels and spatial characteristics in the feature map, optimizing the fault diagnosis effect. Experimental results on different types of PMSMs under various operating conditions verify that the proposed method outperforms other benchmark methods. The proposed method exhibits high diagnostic accuracy under constant speed and load conditions and a certain generalization ability under variable speed and constant load conditions. It overcomes the limitation of low accuracy in diagnosing faults with small signal differences using a single sensor. This provides theoretical guidance and practical reference for reducing maintenance costs and improving service quality of complex motor systems, ensuring faster fault diagnosis and more comprehensive and accurate fault feature extraction. Attached Figure Description

[0052] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0053] Figure 1 This is a flowchart of the PMSM fault diagnosis method based on image color features according to an embodiment of the present invention;

[0054] Figure 2 This is a schematic diagram of the improved SDP conversion process for EMD-RGB features according to an embodiment of the present invention;

[0055] Figure 3 This is a schematic diagram of the DenseNet network structure integrating the scSE module in an embodiment of the present invention;

[0056] Figure 4 These are EMD-SDP images of different loads at a rotation speed of 1000 r / min, according to an embodiment of the present invention.

[0057] Figure 5 These are EMD-SDP images at different speeds under rated load according to an embodiment of the present invention;

[0058] Figure 6 These are EMD-SDP images of a normal motor during no-load acceleration and deceleration in an embodiment of the present invention.

[0059] Figure 7 These are EMD-SDP images of a motor under half-load acceleration and deceleration during an inter-turn short circuit according to an embodiment of the present invention;

[0060] Figure 8 These are EMD-SDP images of the eccentric motor under rated load acceleration and deceleration according to an embodiment of the present invention;

[0061] Figure 9 This embodiment of the invention obtains a confusion matrix diagram of a steady-state dataset using deep learning methods;

[0062] Figure 10 This is a bar chart showing the diagnostic accuracy of different methods in constant load acceleration and deceleration according to embodiments of the present invention. Detailed Implementation

[0063] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, those skilled in the art can obtain other embodiments without creative effort.

[0064] This invention proposes a PMSM fault diagnosis method based on image color features. This method ensures faster fault diagnosis and more comprehensive and accurate fault feature extraction. It has good performance under both steady-state and variable-speed transient conditions. The overall performance of the proposed method is better than other methods, and it can be effectively used for fault diagnosis when speed changes.

[0065] First, it should be noted that increasing the number of sampling points may result in redundant interference information, which may affect image recognition. In the field of fault diagnosis, a higher sampling rate is usually used in order to obtain as much information as possible. Therefore, the fault data in the following embodiments of the present invention were all measured at 10 kHz.

[0066] Example 1

[0067] Please see Figure 1 A PMSM fault diagnosis method based on image color features includes the following steps:

[0068] S1: The experiment obtained vibration signals of motors in different states under various operating conditions, including four motor states: healthy state, inter-turn short circuit fault, local demagnetization fault and eccentric fault, as well as various speed and load conditions, and vibration acceleration signals of different motors under constant load variable speed conditions.

[0069] S2: Perform EMD on the preprocessed vibration signal to solve the intrinsic mode functions at different levels. This can characterize the unique fault features of the signal from different time scales. It is suitable for analyzing nonlinear and non-stationary signal sequences and has a high signal-to-noise ratio.

[0070] S3: The original vibration signal is converted into an RGB-SDP image. The time-domain characteristics of the original signal are represented by the shape and distribution of the image, and the time-frequency domain characteristics are represented by the RGB color characteristics of the image. Different levels of IMF are converted into RGB color features and displayed on the image. The sizes of the first three IMF layers correspond to the three RGB channels respectively, allowing for the characterization of the signal's unique fault characteristics at different time scales. Figure 2 As shown.

[0071] S4: By integrating the scSE module, DenseNet learns the shape distribution and color features of images, constructs a classification network model, improves the method to enhance the network's feature extraction ability, enhances the network's attention to important image features through a weighted strategy, and enhances the backbone network's ability to represent important spatial and channel features, thus better adapting to various complex scenarios.

[0072] S5: Perform steps S2 and S3 on the vibration signal of the motor under test, and use step S4 to input the image into the constructed classification model to diagnose motor faults.

[0073] In one possible implementation, step S1 specifically includes:

[0074] S11: A PMSM fault simulation test bench was constructed, with pre-set inter-turn short circuit faults, uneven demagnetization faults, eccentricity faults, and normal motors.

[0075] S12: Collect acceleration signal segments for each motor under various operating conditions: 9 steady-state conditions (speeds of 1000 r / min, 1500 r / min, and 2000 r / min, and loads of no-load, half-load, and rated load), and 6 transient conditions (acceleration / deceleration conditions with no-load, half-load, and rated load). Figures 4 to 8 As shown;

[0076] S13: Divide each acceleration signal segment into 400 data subsets, and divide all data subsets into three parts: training set, validation set, and test set;

[0077] In one possible implementation, step S2 specifically includes:

[0078] S21: Perform EMD on the preprocessed vibration signal to solve for the intrinsic mode functions at different levels;

[0079] S22: Index all signals of the first three layers of EMD after decomposition, and solve for the maximum value z of the absolute value of all sampling points.

[0080] In one possible implementation, step S3 specifically includes:

[0081] S31: For a time series signal x, x i It is the value of the i-th sampling point of signal x, x i+t It is adjacent time ω sq The sampling point values ​​after an interval t. According to the principle of SDP, when the time domain point x... i Transforming to polar coordinate space S(r(i),θ(i),φ(i)), it can be expressed as:

[0082]

[0083] In the formula, x min and x max Let θ represent the minimum and maximum values ​​of the time-domain signal, respectively, with respect to the angle θ of the nth mirror-symmetric plane. n As the axis of symmetry in polar coordinates, a fan-shaped lobe image is formed. θ(i) and Φ(i) represent the clockwise and counterclockwise rotation angles along the nth mirror symmetry plane in polar coordinates, respectively. With the changes in the amplification factor ζ and the time delay factor t, each sampling point of the time-domain signal can be intuitively displayed in the polar coordinate image, presenting different images.

[0084] S32: Because x is used in the conversion process of SDP images i and x i+t The values ​​of two sampling points, the IMF after EMD transformation i and IMF i+t Correspondingly, a three-layer EMD is performed on all acquired signals, and all decomposed signals are indexed. The maximum absolute value z of all sampling points is calculated, and the RGB three-channel feature sequence T is solved. R (i)T G (i)T B (i) is:

[0085]

[0086] Among them, IMF1 i For the i-th value corresponding to the first-level empirical mode decomposition component, IMF1 i+t This represents the (i+t)th value corresponding to the first-level empirical mode decomposition component.

[0087] S33: After scaling down by a certain proportion, the RGB three-channel feature sequence is mapped to the range of 0-255. The specific processing method is as follows:

[0088]

[0089] Where Round() is the rounding function; k = R, G, B represent the image color channels, and the R, G, and B channels correspond to the first three levels of intrinsic mode components, respectively; T k (i) represents the RGB three-channel feature sequence.

[0090] S34: Display the color features corresponding to all points on the SDP image.

[0091] In one possible implementation, the scSE module in step S4 extracts the channel and spatial importance of the input feature map separately, and then adds them together. The resulting feature sub-map with high importance (i.e., possessing both important channel and spatial features) is then more strongly stimulated, prompting the network to learn more meaningful feature information. For example... Figure 3 As shown, the specific steps for this module to extract and optimize features are as follows:

[0092] S41: Input feature map P = [p1, p2, ..., p N (Among them, the channel) L and W are the height and width of the feature map, respectively.

[0093] S42: Global spatial features are embedded into vector v through a global pooling layer (where... (C is the number of channels)

[0094] S43: Pass the obtained vector v through a fully connected layer with weights ω1 and ω2, and then pass it through the ReLU activation function (δ(·)) and Sigmoid normalization (σ(·)) to obtain the i-th channel p. i The degree of importance of features in The value is

[0095] S44: The number of channels is C, and the weight is ω. sq A 1×1 convolutional block compresses the input feature map P into channels, outputting a feature map with 1 channel and size L×W, fq=ω. sq *P;

[0096] S45: Normalize the obtained feature map fq using Sigmod (σ(·)) to obtain the spatial information importance of each spatial location (i,j) in the feature map (σ(fq(i,j))), in order to enhance the features of important spatial locations.

[0097] S46: Yes Adding σ(fq(i,j)) to obtain the feature subgraph P′;

[0098] In one possible implementation, step S5 specifically includes:

[0099] S51: Perform EMD on the vibration signal to be measured to solve for the intrinsic mode functions at different levels;

[0100] S52: Convert the vibration signal to be measured into an SDP image, and convert different levels of IMF into color features to display on the SDP image;

[0101] S53: Input the SDP image improved by converting the vibration signal to be tested into EMD-RGB features, input the DenseNet classification network model improved by fusing scSE attention mechanism, and output the diagnostic result of the signal to be tested.

[0102] This invention obtains vibration signals of motors under various operating conditions in different states through experiments; performs EMD on the preprocessed vibration signals to solve for intrinsic mode functions (IMFs) at different levels; converts the original vibration signals into RGB-SDP images, where the time-domain features of the original signals are represented by the shape and distribution of the image, and the time-frequency domain features are represented by the RGB color features of the image. IMFs at different levels are converted into RGB color features and displayed on the image. The sizes of the first three IMF layers correspond to the three RGB channels, allowing for the characterization of unique fault features of the signal at different time scales. Combining time-domain and time-frequency domain features provides a more comprehensive characterization of the unique features of the fault signal, facilitating further feature extraction and model construction. By integrating the scSE module, the DenseNet learns the shape distribution and color features of the image, constructing a classification network model. This improved method enhances the network's feature extraction capabilities, and a weighted strategy increases the network's focus on important image features, enhancing the backbone network's ability to represent important spatial and channel features, thus better adapting to various complex scenarios. Steps S2 and S3 are executed on the vibration signal of the motor under test, and the classification model constructed in step S4 is used to diagnose motor faults. This invention presents a PMSM fault diagnosis method based on image color features, ensuring faster fault diagnosis and more comprehensive and accurate fault feature extraction. It exhibits good performance under both steady-state and variable-speed transient conditions, and the overall performance of the proposed method surpasses other methods, effectively enabling fault diagnosis during speed variations. This method overcomes the limitation of low accuracy in diagnosing faults with small signal differences using a single sensor, providing theoretical guidance and practical reference for reducing maintenance costs and improving service quality of complex motor systems.

[0103] Example 2

[0104] Based on the same inventive concept as the PMSM fault diagnosis method based on image color features in Embodiment 1 above, the present invention also provides a PMSM fault diagnosis system based on image color features, the system comprising:

[0105] The raw vibration signal acquisition module is used to acquire the raw vibration signals of the motor under test under different states and multiple operating conditions;

[0106] The Empirical Mode Decomposition (EMD) module is used to perform EMD on the original vibration signal and solve for the eigenmode functions at different levels to obtain signal components at different levels.

[0107] The symmetrical dot matrix conversion module is used to convert the original vibration signal into a symmetrical dot matrix with three RGB channels according to different levels of signal components.

[0108] The classification network model building module is used to improve the shape distribution and color features of DenseNet images by fusing the scSE module, and to build a classification network model.

[0109] The output module is used to input the symmetrical dot matrix image of the three RGB channels into the classification network model and output the fault diagnosis results.

[0110] Example 3

[0111] Based on the same inventive concept as the PMSM fault diagnosis method based on image color features in Embodiment 1, the present invention also provides an electronic device, including: a memory storing at least one instruction; and a processor executing the instructions stored in the memory to implement the PMSM fault diagnosis method based on image color features as described in Embodiment 1, as follows:

[0112] The device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor, when executing the program, performs the following steps:

[0113] Vibration signals of PMSMs under various speed and load conditions were acquired when they were in a healthy state, had inter-turn short circuit faults, local demagnetization faults, and eccentric faults.

[0114] The preprocessed vibration signal was subjected to EMD to solve for the intrinsic mode functions at different levels.

[0115] The original vibration signal is converted into an SDP image, and different levels of IMF are converted into color features and displayed on the SDP image.

[0116] Improve the DenseNet learning image dataset by incorporating the scSE attention mechanism to build a classification network model;

[0117] The vibration signal of the motor under test is processed through steps S2 and S3. Then, the image is input into the constructed classification model in step S4 to diagnose the motor fault.

[0118] Example 4

[0119] Based on the same inventive concept as the PMSM fault diagnosis method based on image color features in Embodiment 1, the present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the PMSM fault diagnosis method based on image color features described in Embodiment 1.

[0120] The experiment obtained vibration signals of motors in different states under various operating conditions, including four motor states: healthy state, inter-turn short circuit fault, local demagnetization fault, and eccentric fault, as well as various speed and load conditions.

[0121] The preprocessed vibration signal was subjected to EMD to solve for the intrinsic mode functions at different levels.

[0122] The original vibration signal is converted into an SDP image, and different levels of IMF are converted into color features and displayed on the SDP image.

[0123] Improve the DenseNet learning image dataset by incorporating the scSE attention mechanism to build a classification network model;

[0124] The vibration signal of the motor under test is processed through steps S2 and S3. Then, the image is input into the constructed classification model in step S4 to diagnose the motor fault.

[0125] In this embodiment of the invention, vibration data is collected using a PMSM fault testing platform to experimentally verify the aforementioned related theories. The motor fault testing platform mainly consists of the motor under test, vibration sensors, a positioning platform, a digital signal acquisition device, an encoder, a DC power supply, and a computer. The motor parameters are shown in Table 1.

[0126] Table 1 shows the motor parameters.

[0127] 750w 4 380V 1.35A 50Hz 3000rpm

[0128] Faulty motors mainly include motors with inter-turn short circuit faults, motors with uneven demagnetization faults, and motors with eccentric faults.

[0129] like Figure 9 and Figure 10 As shown in the summary, this invention addresses the problem that traditional diagnostic methods cannot meet practical needs when the operating conditions of PMSMs change, and proposes a PMSM fault diagnosis method based on image color features. This fault diagnosis method exhibits high accuracy under constant speed and load conditions and good generalization ability under variable speed and constant load conditions. Furthermore, the proposed diagnostic method is non-invasive and can be widely applied to the condition monitoring and diagnosis of industrial motors. This method overcomes the limitation of low accuracy in diagnosing faults with small differences in signals from a single sensor, providing theoretical guidance and practical reference for reducing maintenance costs and improving service quality of complex motor systems.

[0130] This invention proposes an improved SDP time-series signal image conversion method based on EMD-RGB features, where the magnitudes of the first three IMF layers correspond to the R, G, and B channel intensities of a color image, respectively. The time-domain features of the original signal are represented by the image's shape and distribution, while the time-frequency domain features are represented by the image's RGB color features. Compared to grayscale images, this method preserves complete original time-domain signal information while also reflecting time-frequency domain signal characteristics, increasing the differences between image classes and enabling the differentiation of fault types with small time-domain signal differences. Combining time-domain and time-frequency domain features allows for a more comprehensive characterization of the unique features of fault signals, facilitating further feature extraction and model construction.

[0131] This invention embeds the scSE attention module into the DenseNet network to improve the network's feature extraction performance. By employing a weighted strategy, it enhances the network's attention to important image features, strengthens the backbone network's ability to represent important spatial and channel features, and effectively alleviates the network's poor performance in extracting complex high-level features. By calculating the importance of channel and spatial features in the feature map, it enhances the network's ability to learn important channel and spatial characteristics in the feature map, optimizes fault diagnosis performance, and improves the diagnostic model's ability to extract image distribution and color features.

[0132] This invention constructs a PMSM fault simulation test bench to test different types of PMSMs under constant speed and constant load conditions and variable speed and constant load conditions, thus validating the proposed method. The method uses experimental data from various speeds and loads under constant speed and constant load conditions for training to construct a classification model. Fault diagnosis results show that the proposed method has the highest diagnostic accuracy. This method is also applicable to constant load acceleration and constant load deceleration conditions of motors, and can be effectively used for fault diagnosis under speed changes, exhibiting certain generalization performance.

[0133] The diagnostic approach proposed in this invention is applicable to other time-series imaging methods, exhibiting good versatility and application results. Using a single dimension of the signal for fault diagnosis is not ideal; utilizing features from multiple dimensions of the signal can significantly improve diagnostic accuracy. Convolutional neural networks reduce the complexity of the network model through three strategies: local receptive field, weight sharing, and downsampling, while also possessing invariance to translation, rotation, and scaling.

[0134] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0135] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0136] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0137] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0138] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A PMSM fault diagnosis method based on image color features, characterized in that, include: Acquire the original vibration signals of the motor under test under different states and multiple operating conditions; The original vibration signal is subjected to empirical mode decomposition, and the eigenmode functions at different levels are solved to obtain signal components at different levels. The original vibration signal is converted into a symmetrical dot matrix diagram with three RGB channels according to the signal components at different levels. By integrating the scSE module, DenseNet learns the shape distribution and color features of images, and a classification network model is constructed. Input the symmetrical dot matrix image of the three RGB channels into the classification network model and output the fault diagnosis result; The process of converting the original vibration signal into a symmetrical dot matrix image with three RGB channels according to different levels of signal components includes: The time-domain characteristics of the original vibration signal are represented by the shape and distribution of the image, and the time-frequency domain characteristics are represented by the RGB color characteristics of the image. The signal components at different levels are converted into RGB color characteristics and displayed on the image. The magnitudes of the first three signal components are selected to correspond to the three RGB channels respectively, so as to characterize the unique fault characteristics of the signal from different time scales. The process of converting the original vibration signal into a symmetrical dot matrix image with three RGB channels according to different levels of signal components includes: For time series signals , It is a time series signal The i Values ​​of each sampling point Adjacent times The sampling point values ​​after an interval t; when the time domain point Transform to polar coordinate space , is represented as: In the formula, and Let represent the minimum and maximum values ​​of the time-domain signal, respectively, with respect to the angle of the nth mirror-symmetric plane. As the axis of symmetry of polar coordinates, a fan-shaped lobe image is formed; and These represent the clockwise and counterclockwise rotation angles along the nth mirror symmetry plane in polar coordinates, respectively; with the magnification factor... With the change of the time delay factor t, each sampling point of the time-domain signal is intuitively displayed in the polar coordinate image, presenting different images; Use the maximum absolute value of all sample points Solve the RGB three-channel feature sequence for: in, For the first layer of empirical mode decomposition components, corresponding to the first layer... i A number, For the first layer of empirical mode decomposition components, corresponding to the first layer... i + t A number; The RGB three-channel feature sequences are mapped to the range of 0-255 by scaling according to the following formula: Where Round() is the rounding function; k=R,G,B represent the image color channels, and the R, G and B channels correspond to the first three levels of intrinsic mode components, respectively; Represents the RGB three-channel feature sequence; Display the color features corresponding to all points on the symmetrical bitmap; The step of improving the shape distribution and color features of DenseNet learning images by fusing the scSE module and constructing a classification network model involves extracting the channel and spatial importance of the input feature maps, adding them together, and then strengthening the feature sub-maps that have both important channel and important spatial features, so as to enable the network to learn more meaningful feature information. The method of improving the shape distribution and color features of DenseNet images by fusing the scSE module and constructing a classification network model includes: Input feature map Among them, the channel , L and W These are the height and width of the feature map, respectively. Global spatial features are embedded into the vector through a global pooling layer. v In, among them, C is the number of channels; The obtained vector v is weighted as follows: and The fully connected layer is processed sequentially by ReLU activation and Sigmoid normalization to obtain the i-th channel. The degree of importance of features ,in, The value is ; With channel number C and weight of A 1×1 convolutional block compresses the input feature map P into channels, outputting a feature map with 1 channel and a size of L×W. ; The obtained feature map After Sigmoid normalization, each spatial location in the feature map is obtained. Spatial information importance Enhance the understanding of important spatial location features; right and Perform the addition operation to obtain the feature subgraph. .

2. The PMSM fault diagnosis method based on image color features according to claim 1, characterized in that, The acquisition of the original vibration signals of the motor under test under different states and multiple operating conditions includes: Construct a fault simulation test bench and pre-set inter-turn short circuit fault, uneven demagnetization fault, eccentricity fault and normal motor; Acceleration signal segments were collected for each motor under the following operating conditions: Nine steady-state operating conditions: speeds of 1000 r / min, 1500 r / min, and 2000 r / min, and loads of no load, half load, and rated load, respectively; and six transient operating conditions: acceleration and deceleration conditions with loads of no load, half load, and rated load, respectively. Each acceleration signal segment is divided into a data subset, which is then divided into a training set, a validation set, and a test set.

3. The PMSM fault diagnosis method based on image color features according to claim 1, characterized in that, The step of performing empirical mode decomposition on the original vibration signal and solving for the eigenmode functions at different levels to obtain signal components at different levels includes: The preprocessed original vibration signal is subjected to empirical mode decomposition to solve for the eigenmode functions at different levels. Index all signals from the first three levels of empirical mode decomposition after decomposition, and solve for the maximum absolute value of all sampling points. .

4. A PMSM fault diagnosis system based on image color features, characterized in that, Performing the PMSM fault diagnosis method based on image color features as described in any one of claims 1 to 3 includes: The raw vibration signal acquisition module is used to acquire the raw vibration signals of the motor under test under different states and multiple operating conditions; The Empirical Mode Decomposition (EMD) module is used to perform EMD on the original vibration signal and solve for the eigenmode functions at different levels to obtain signal components at different levels. The symmetrical dot matrix conversion module is used to convert the original vibration signal into a symmetrical dot matrix with three RGB channels according to different levels of signal components. The classification network model building module is used to improve the shape distribution and color features of DenseNet images by fusing the scSE module, and to build a classification network model. The output module is used to input the symmetrical dot matrix image of the three RGB channels into the classification network model and output the fault diagnosis results.

5. An electronic device, characterized in that, include: Memory, storing at least one instruction; The processor executes the instructions stored in the memory to implement the PMSM fault diagnosis method based on image color features as described in any one of claims 1 to 3.

6. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the PMSM fault diagnosis method based on image color features as described in any one of claims 1 to 3.

Images