Apparatus and method for detecting faults in electric vehicle drive motor and reducer, and diagnostic system therefor

Abstract

The present invention relates to a device for detecting defects occurring in a drive motor and a reduction gear of an electric vehicle, comprising: a preprocessing unit that converts at least one signal detected from the motor and the reduction gear into a spectrogram; a feature fusion unit that fuses features extracted from at least one converted spectrogram and outputs fused features; and a defect detection unit that extracts a first feature related to the type of defect from the fused features through a first branch of a pre-prepared model, inputs the first feature into a classification module of the model to output the type of defect, extracts a second feature related to the location of the defect from the fused feature through a second branch of the model, inputs the feature formed by merging the first feature and the second feature into a segmentation module of the model to output a spectrogram that visualizes the type and location of the defect in the time-frequency domain.

Description

Technology Field

[0001] The present invention relates to an apparatus and method for detecting defects in an electric vehicle drive motor and a reduction gear, and a diagnostic system for the same. Background Technology

[0003] Recently, as the depletion of fossil fuels and air pollution caused by overuse have become serious issues, research and development regarding the use of renewable energy and eco-friendly transportation methods are actively underway. Among eco-friendly transportation options, electric vehicles emit no air pollutants during operation and do not cause noise pollution, leading automakers worldwide to rush to launch them into the market.

[0004] Electric motors play the role of converting electrical energy into mechanical energy, while reduction gears are devices that convert the rotational force of the electric motor into appropriate speed and power during the process of transmitting it to the axle. To enhance the performance and efficiency of electric vehicles, technology that monitors the condition of these motors and reduction gears in real time and detects defects early is crucial. This is because neglecting defects in the electric motors and reduction gears of electric vehicles can lead to safety issues resulting from degraded vehicle performance, potentially threatening the safety of passengers and pedestrians, and causing economic losses due to product degradation.

[0005] Meanwhile, existing non-invasive methods for detecting defects in the aforementioned electric motors and reduction gears utilize current or vibration signal processing techniques. However, these methods often detect defect signals only after mechanical defects have occurred, making early detection difficult. Furthermore, they are vulnerable to complex analyses that consider abnormal signals and internal / external interference, which makes them difficult to apply to actual operating vehicles.

[0006] Therefore, research is needed on defect detection methods that can detect defects in electric motors and reduction gears of electric vehicles at an early stage and are less sensitive to external environments. Prior art literature

[0008] Korean Patent Publication No. 10-2023-0007781 The problem to be solved

[0009] The present invention has been devised to solve the above-mentioned problems, and the objective of the present invention is to provide an electric vehicle drive motor and reduction gear defect detection device and method, and a diagnostic system for the same. means of solving the problem

[0011] A device for detecting defects occurring in a drive motor and a reduction gear of an electric vehicle according to an embodiment of the present invention for achieving the above objective comprises: a preprocessing unit that converts at least one signal detected from the motor and the reduction gear into a spectrogram; a feature fusion unit that fuses features extracted from at least one converted spectrogram and outputs fused features; and a defect detection unit that extracts a first feature related to the type of defect from the fused features through a first branch of a pre-prepared model, inputs the first feature into a classification module of the model to output the type of defect, extracts a second feature related to the location of the defect from the fused feature through a second branch of the model, inputs the feature formed by merging the first feature and the second feature into a segmentation module of the model to output a spectrogram that visualizes the type and location of the defect in the time-frequency domain.

[0012] A method for detecting defects occurring in a drive motor and a reduction gear of an electric vehicle according to an embodiment of the present invention for achieving the above objective comprises: a step in which a preprocessing unit converts at least one signal detected from the motor and the reduction gear into a spectrogram; a step in which a feature fusion unit fuses features extracted from the converted at least one spectrogram and outputs fused features; and a step in which a defect detection unit extracts a first feature related to the type of defect from the fused features through a first branch of a pre-prepared model, inputs the first feature into a classification module of the model to output the type of defect, extracts a second feature related to the location of the defect from the fused feature through a second branch of the model, inputs the feature formed by merging the first feature and the second feature into a segmentation module of the model to output a spectrogram that visualizes the type and location of the defect in the time-frequency domain.

[0013] A fault diagnosis system for an electric vehicle drive motor and a reduction gear according to an embodiment of the present invention for achieving the above objective comprises: a sensor unit for detecting at least one signal generated from the motor and the reduction gear; and a processing unit for converting the detected at least one signal into a spectrogram, fusing features extracted from the converted at least one spectrogram to output a fused feature, extracting a first feature related to the type of the fault from the fused feature through a first branch of a pre-prepared model, inputting the first feature into a classification module of the model to output the type of the fault, extracting a second feature related to the location of the fault from the fused feature through a second branch of the model, and inputting a feature formed by merging the first feature and the second feature into a segmentation module of the model to output a spectrogram that visualizes the type and location of the fault in the time-frequency domain. Effects of the invention

[0015] According to one aspect of the present invention described above, by providing an electric vehicle drive motor and reduction gear defect detection device and method, and a diagnostic system for the same, the type and location of the defect can be diagnosed simultaneously, thereby increasing the accuracy of defect diagnosis and shortening the time for problem solving.

[0016] In addition, by visualizing the type and location of defects as spectrograms in the time-frequency domain, users can more intuitively understand the causes and effects of defects and identify them more easily. Brief explanation of the drawing

[0018] FIG. 1 is a schematic diagram illustrating an electric vehicle drive motor and reduction gear fault diagnosis system according to an embodiment of the present invention. FIG. 2 is a diagram for explaining the operation of the processing unit illustrated in FIG. 1, FIG. 3 is a drawing showing the detailed configuration of the interactive concentration module illustrated in FIG. 2. FIG. 4 is a drawing showing the detailed configuration of the ASPP illustrated in FIG. 2. FIG. 5 is a diagram showing an example of frequency boundary information obtained from the spectrogram of the correct data. And, FIG. 6 is a flowchart showing the operation of an electric vehicle defect detection device according to an embodiment of the present invention. Specific details for implementing the invention

[0019] The following detailed description of the invention refers to the accompanying drawings, which illustrate specific embodiments in which the invention may be practiced. These embodiments are described in sufficient detail to enable those skilled in the art to practice the invention. It should be understood that various embodiments of the invention are different but need not be mutually exclusive. For example, specific shapes, structures, and characteristics described herein may be implemented in other embodiments without departing from the spirit and scope of the invention in relation to one embodiment. It should also be understood that the location or arrangement of individual components within each disclosed embodiment may be changed without departing from the spirit and scope of the invention. Accordingly, the following detailed description is not intended to be limiting, and the scope of the invention is limited only by the appended claims, including all equivalents to those claimed therein, provided appropriately described. Similar reference numerals in the drawings refer to the same or similar functions across various aspects.

[0020] The components according to the present invention are defined by functional distinction rather than physical distinction, and can be defined by the functions each performs. Each component may be implemented as hardware or as program code and processing units that perform each function, and the functions of two or more components may be included and implemented in a single component. Therefore, it should be noted that the names assigned to the components in the following embodiments are not intended to physically distinguish each component but are assigned to imply the representative function performed by each component, and that the technical concept of the present invention is not limited by the names of the components.

[0021] Preferred embodiments of the present invention will be described in more detail below with reference to the drawings.

[0022] FIG. 1 is a schematic diagram illustrating an electric vehicle drive motor and reduction gear fault diagnosis system according to an embodiment of the present invention, FIG. 2 is a diagram for explaining the operation of a processing unit illustrated in FIG. 1, FIG. 3 is a diagram showing the detailed configuration of a bidirectional concentration module illustrated in FIG. 2, FIG. 4 is a diagram showing the detailed configuration of an ASPP illustrated in FIG. 2, and FIG. 5 is a diagram showing an example of frequency boundary information obtained from a spectrogram of ground truth data.

[0023] The described defect diagnosis system includes a sensor unit (200) that detects a signal generated from the powertrain (100) of an electric vehicle and a processing unit (300) that detects a defect in the electric vehicle by processing the detected signal according to the method proposed in the present invention. The processing unit (300) is composed of a preprocessing unit (310), a feature fusion unit (320), and a defect detection unit (330). In the embodiment of the present invention, the case where the processing unit (300) is mounted inside the electric vehicle is described as an example, but it is understood that the processing unit (300) may also be provided remotely.

[0024] The above powertrain (100) is a core system of an electric vehicle that converts power and transmits power to the wheels of the vehicle, and may include a drive motor (110) and a reducer (120). The sensor unit (200) detects at least one signal generated from the motor (110) and the reducer (120). That is, the sensor unit (200) detects a current signal (210) and a vibration signal (220) generated from the motor (110), and detects a vibration signal (230) generated from the reducer (120).

[0025] The preprocessing unit (310) applies a Short-Time Fourier Transform (STFT) method to at least one signal detected by the sensor unit (200) to convert it into a spectrogram that represents the frequency components according to the temporal change of the signal. Here, the STFT method refers to a method of dividing the signal into short time intervals and applying a Fourier Transform to each interval to analyze the frequency components over time.

[0026] The feature fusion unit (320) extracts features from at least one converted spectrogram and fuses the extracted features to output fused features.

[0027] The defect detection unit (330) extracts semantic information, namely a first feature related to the type of defect occurring in the motor (110) and the reduction gear (120), from the fused feature through the first branch (331) of the pre-prepared model, and inputs the first feature into the classification module (333) of the model to output the type of defect. At this time, the defect detection unit (330) applies Global Average Pooling (GAP) to reduce the feature map size before inputting the first feature into the classification module (333). Here, the types of defects may include, for example, motor eccentricity defects, reduction gear defects, motor demagnetization defects, etc.

[0028] Additionally, the defect detection unit (330) extracts spatially detailed information, namely a second feature related to the location of the defect, from the fused feature through the second branch (332) of the model, and inputs the feature formed by the merger of the first feature and the second feature into the segmentation module (334) of the model to output a spectrogram that visualizes the type and location of the defect in the time-frequency domain.

[0029] At this time, before merging the first feature and the second feature, the defect detection unit (330) enhances the contextual representation of the first feature through the ASPP (Atrous Spatial Pyramid Pooling) module (336) configured as in FIG. 4, upscales the first feature with enhanced contextual representation to restore the feature map size to its original resolution, and then merges it with the second feature.

[0030] Thus, the defect detection unit (330) has the advantage of being able to extract defect components quickly and accurately with a small number of parameters by extracting feature information related to the type of defect and feature information related to the location of the defect through the first branch (331) and the second branch (332), respectively. In addition, the defect detection unit (330) applies a multi-output regression method to simultaneously train the classification module (333) and the segmentation module (334) into a single model.

[0031] Additionally, the defect detection unit (330) further includes a bilateral attention module (BAM) (337), the detailed configuration of which is shown in FIG. 3. The BAM (337) learns the dependency between time and frequency by applying an attention mechanism to a feature map in which a first feature extracted through a first branch (331) and a second feature extracted through a second branch (332) are merged. Then, the BAM (337) summarizes the feature map in terms of time and frequency and highlights useful areas in the summarized feature map that can help with defect detection.

[0032] Below, the operation of this BAM (337) will be explained in more detail through mathematical formulas.

[0033] X, which is the i-th stage output of the first branch (331) and the second branch (332). d ∈ R H x W x C , X s ∈ cast , The process of converting can be represented by the following mathematical formulas 1 to 3. In addition, by redistributing the feature map obtained through this process to the model, each branch can emphasize areas useful for defect detection in the feature map and suppress unnecessary components.

[0034]

[0035]

[0036]

[0037] Here F 1x1 represents the sequential operation of 1x1 convolution and batch normalization, and F 3x3 represents the sequential operation of 3x3 convolution and batch normalization, g represents the GELU function, and conv1x1 represents a 1x1 convolution operation, and U represents upscaling that restores resolution by applying bilinear interpolation.

[0038] That is, BAM (337) is the feature map X of the first feature extracted through the first branch (331) as shown in Equation 1. s Compress the channel dimension of and the feature map X of the second feature extracted through the second branch (332). d Reduce the resolution of to match the dimensions of the two feature maps. That is, X s channel dimension 2 i C is X d Since it is relatively larger than the channel dimension C, the channel dimension is compressed, and X d The resolution H x W is X s resolution 2 -i H x 2 -i Since it is relatively higher than W, the resolution is reduced. By reducing the feature map in this way, the impact of the lightweight module on the overall model's inference speed can be minimized.

[0039] Then, BAM (337) is the feature map X of the first feature in which the channel dimension is compressed. s and feature map X of the second feature with reduced resolution d Merge to obtain the merged feature map Y.

[0040] Subsequently, BAM (337) learns the dependency between time and frequency by applying an attention mechanism to the merged feature map Y as shown in Equation 2. This operation is divided into a frequency-attention (f-attention) part that summarizes information for each frequency interval in the entire time domain of the signal and a time-attention (t-attention) part that summarizes information for each time interval in the entire frequency domain, thereby independently learning global information for the time-frequency interval.

[0041] Thus, BAM (337) applies an attention mechanism to the merged feature map Y to summarize Y in terms of time and frequency, combines the summarized feature maps using matrix multiplication, and then highlights regions useful for defect detection. Here, the output of the frequency-attention part is Z f ∈ and the output of the time-attention part is Z t ∈ Since, Z ∈ It has dimensions.

[0042] Therefore, BAM (337) adjusts the dimensions and resolution of the feature map Z as shown in Equation 3 and combines it with the input of each branch to create a feature map that highlights areas useful for defect detection. , Finally, output .

[0043] Additionally, the defect detection unit (330) further includes an auxiliary module (335), and the auxiliary module (335) improves the prediction performance of the splitting module (334) by providing frequency boundary information, which is dilated by extracting frequency boundaries from the spectrogram of the ground truth data as shown in FIG. 5 through a pre-prepared filter, for example, a Canny filter. The Canny filter can detect clear and accurate frequencies using a Gaussian filter and can extract thin boundaries using NMS (Non-Maximum Suppression).

[0044] The above auxiliary module (335) is activated only during the learning of the above splitting module (334) to help learn the frequency boundary information better, and thus has the advantage that no additional computation occurs during the inference step of the above splitting module (334).

[0045] Meanwhile, the model applied to the defect detection unit (330) is a classification loss function L that calculates the performance of the classification module (333) as shown in Equation 4 below. cls , a partitioning loss function L that calculates the performance of the partitioning module (334) seg and auxiliary loss function L that calculates the performance of the auxiliary module (335). aux It is trained using a loss function calculated as the weighted sum of.

[0046]

[0047] Here, λ1, λ2, and λ3 represent weights for each loss function, which are preset values ​​that can be adjusted according to the data.

[0048] In addition, the above loss function uses a focal loss function to resolve the data imbalance problem.

[0049] FIG. 6 is a flowchart showing the operation of an electric vehicle defect detection device according to an embodiment of the present invention.

[0050] The preprocessing unit of the electric vehicle defect detection device converts at least one signal detected from the motor and the reduction gear into a spectrogram (S601), and the feature fusion unit fuses the features extracted from at least one spectrogram converted in S601 and outputs the fused features (S603).

[0051] The defect detection unit extracts a first feature related to the type of defect from the fused feature output from S603 through a first branch of a pre-prepared model, inputs the first feature into the classification module of the model to output the type of defect, extracts a second feature related to the location of the defect from the fused feature through a second branch of the model, and inputs the feature formed by merging the first feature and the second feature into the segmentation module of the model to output a spectrogram that visualizes the type and location of the defect in the time-frequency domain. (S605)

[0052] The electric vehicle defect detection method of the present invention, as described above, can be implemented in the form of program instructions that can be executed through various computer components and recorded on a computer-readable recording medium. The computer-readable recording medium may include program instructions, data files, data structures, etc., either individually or in combination.

[0053] The program instructions recorded on the above-mentioned computer-readable recording medium may be those specifically designed and configured for the present invention, or they may be those known and available to those skilled in the art of computer software.

[0054] Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tapes; optical recording media such as CD-ROMs and DVDs; magneto-optical media such as floptical disks; and hardware devices specifically configured to store and execute program instructions such as ROM, RAM, and flash memory.

[0055] Examples of program instructions include machine code, such as that generated by a compiler, as well as high-level language code that can be executed by a computer using an interpreter, etc. The hardware device may be configured to operate as one or more software modules to perform processing according to the present invention, and vice versa.

[0056] Although various embodiments of the present invention have been illustrated and described above, the present invention is not limited to the specific embodiments described above. It is understood that various modifications can be made by those skilled in the art without departing from the essence of the invention as claimed in the claims, and such modifications should not be understood individually from the technical spirit or perspective of the present invention. Explanation of the symbols

[0058] 100: Powertrain 110: Motor 120: Reducer 200: Sensor section 210: Current signal 220, 230: Vibration signal 300: Processing unit 310: Preprocessing section 320: Feature Fusion Part 330: Defect detection unit 331: 1st Quarter 332: Second Quarter 333: Classification Module 334: Split Module 335: Auxiliary Module 336: ASPP Module 337: Interactive Focus Module

Claims

Claim 1 An electric vehicle defect detection device comprising: a device for detecting defects occurring in a drive motor and a reduction gear of an electric vehicle, the device comprising: a preprocessing unit that converts at least one signal detected from the motor and reduction gear into a spectrogram; a feature fusion unit that fuses features extracted from at least one converted spectrogram and outputs fused features; and a defect detection unit that extracts a first feature related to the type of defect from the fused features through a first branch of a pre-prepared model, inputs the first feature into a classification module of the model to output the type of defect, extracts a second feature related to the location of the defect from the fused feature through a second branch of the model, inputs the feature formed by merging the first feature and the second feature into a segmentation module of the model to output a spectrogram that visualizes the type and location of the defect in the time-frequency domain. Claim 2 An electric vehicle defect detection device according to claim 1, wherein the defect detection unit further includes a bilateral attention module (BAM), and the BAM applies an attention mechanism to a feature map in which the first feature and the second feature are merged to learn the dependency between time and frequency, summarizes the feature map in terms of time and frequency, and highlights a region useful for detecting the defect in the summarized feature map. Claim 3 An electric vehicle defect detection device according to claim 1, wherein the defect detection unit further includes an ASPP (Atrous Spatial Pyramid Pooling) module, and before merging the first feature and the second feature, enhances the contextual representation of the first feature through the ASPP module, upscales the first feature with enhanced contextual representation to restore the feature map size to its original resolution, and then merges it with the second feature. Claim 4 An electric vehicle defect detection device according to claim 1, wherein the defect detection unit further includes an auxiliary module, the auxiliary module provides frequency boundary information, which is dilated by extracting frequency boundaries from the spectrogram of ground truth data through a pre-prepared filter, to the second branch to improve the prediction performance of the segmentation module, and the auxiliary module is activated only during the learning of the segmentation module. Claim 5 An electric vehicle defect detection device according to claim 4, characterized in that the model is trained using a loss function calculated as a weighted sum of a classification loss function that calculates the performance of the classification module, a division loss function that calculates the performance of the division module, and an auxiliary loss function that calculates the performance of the auxiliary module. Claim 6 A method for detecting defects occurring in a drive motor and a reduction gear of an electric vehicle, comprising: a step in which a preprocessing unit converts at least one signal detected from the motor and the reduction gear into a spectrogram; a step in which a feature fusion unit fuses features extracted from the converted at least one spectrogram and outputs fused features; and a step in which a defect detection unit extracts a first feature related to the type of defect from the fused features through a first branch of a pre-prepared model, inputs the first feature into a classification module of the model to output the type of defect, extracts a second feature related to the location of the defect from the fused feature through a second branch of the model, inputs the feature formed by merging the first feature and the second feature into a segmentation module of the model to output a spectrogram that visualizes the type and location of the defect in the time-frequency domain. Claim 7 In claim 6, the defect detection unit further includes a bilateral attention module (BAM), and the step of outputting the spectrogram further includes the step of the BAM applying an attention mechanism to a feature map in which the first feature and the second feature are merged to learn the dependency between time and frequency, summarizing the feature map in terms of time and frequency, and highlighting a region useful for detecting the defect in the summarized feature map. Claim 8 An electric vehicle defect detection method according to claim 6, wherein the defect detection unit further includes an ASPP (Atrous Spatial Pyramid Pooling) module, and before merging the first feature and the second feature, enhances the contextual representation of the first feature through the ASPP module, upscales the first feature with enhanced contextual representation to restore the feature map size to its original resolution, and then merges it with the second feature. Claim 9 In claim 6, the defect detection unit further includes an auxiliary module, wherein the auxiliary module provides frequency boundary information, which is dilated by extracting frequency boundaries from the spectrogram of ground truth data through a pre-prepared filter, to the second branch to improve the prediction performance of the segmentation module, and the auxiliary module is activated only during the training of the segmentation module, an electric vehicle defect detection method. Claim 10 An electric vehicle defect detection method according to claim 9, characterized in that the model is trained using a loss function calculated as a weighted sum of a classification loss function that calculates the performance of the classification module, a division loss function that calculates the performance of the division module, and an auxiliary loss function that calculates the performance of the auxiliary module. Claim 11 A diagnostic system for a fault in an electric vehicle drive motor and a reduction gear, comprising: a sensor unit for detecting at least one signal generated from the motor and the reduction gear; and a processing unit for converting the detected at least one signal into a spectrogram, fusing features extracted from the converted at least one spectrogram to output a fused feature, extracting a first feature related to the type of the fault from the fused feature through a first branch of a pre-prepared model, inputting the first feature into a classification module of the model to output the type of the fault, extracting a second feature related to the location of the fault from the fused feature through a second branch of the model, and inputting a feature formed by merging the first feature and the second feature into a segmentation module of the model to output a spectrogram that visualizes the type and location of the fault in the time-frequency domain. Claim 12 A diagnostic system according to claim 11, wherein the processing unit further includes a bilateral attention module (BAM), and the BAM applies an attention mechanism to a feature map in which the first feature and the second feature are merged to learn the dependency between time and frequency, summarizes the feature map in terms of time and frequency, and highlights a region useful for detecting the defect in the summarized feature map. Claim 13 A diagnostic system according to claim 11, wherein the processing unit further includes an auxiliary module, the auxiliary module provides frequency boundary information, which is dilated by extracting frequency boundaries from the spectrogram of ground truth data through a pre-prepared filter, to the second branch to improve the prediction performance of the segmentation module, and the auxiliary module is activated only during the training of the segmentation module.

Images