DEVICE FOR LOCALISING NOISE IN A STEERING SYSTEM

Abstract

Device for locating a noise occurring in a steering system, the device comprising: a sound recording unit designed to detect a noise occurring in a steering system; a processing unit (200) configured to input data relating to the noise in the steering system into a neural network model configured to perform pre-learning and to localize a position or component where the noise occurs in the steering system, the noise being recorded by the sound recording unit; and a storage unit (300) in which the neural network model, which is designed to perform pre-learning, is stored, wherein the processing unit (200) is configured to perform preprocessing of generating an image resulting from converting the time domain data relating to the noise in the steering system into frequency domain data, wherein the noise is detected by the sound recording unit, and inputs the image generated by the preprocessing into the neural network model, wherein the neural network model is a convolutional neural network model comprising a convolution layer (61, 63), a pooling layer (62, 64) and a fully connected layer (65), characterized in that the convolution layer (61, 63) is configured to perform filtering of the image generated by preprocessing using a filter, to perform a convolution computation and to create a feature map for extracting a feature from the image generated by preprocessing, the pooling layer (62, 64) is designed such that it extracts a characteristic value for each region from the feature map created by the folding layer (61, 63) and reduces the size of the feature map. the fully connected layer (65) is designed in such a way that it further highlights the features and categorizes the highlighted features by applying an activation function, and With regard to the noise in the steering system, an output layer (66) is configured such that it locates and outputs the position or component where the noise occurs, based on a result of the categorization by the fully connected layer.

Description

BACKGROUND

[0001] Exemplary embodiments relate to a device for locating a noise in a steering system of a vehicle and, in particular, a device for locating a noise in a steering system with which a position or component where a noise occurs in a steering system can be localized using a neural network model. BACKGROUND DISCUSSION

[0002] A vehicle's steering system is a system that adjusts the vehicle's direction of travel according to the driver's steering input. In recent years, power steering systems, which increase the driver's effort when turning the steering wheel by means of motor power, have become widely used. Examples of these power steering systems include motor-driven power steering (MDPS) and electric power steering (EPS).

[0003] Such power steering systems, which utilize the engine's power, generally comprise a motor, a steering housing, a torque sensor, and an electronic control unit (ECU). The electronic control unit detects the steering wheel's rotational force via a torque sensor and controls the steering wheel torque by applying an electrical current to the motor according to the vehicle's speed. The steering housing receives the engine's power via a steering shaft, converts it into rotation, and drives the front wheels through an arm that includes a relay rod, a tie rod, a tie rod lever, and similar components.

[0004] The power steering system uses a deceleration mechanism, a column, a joint, and similar components to properly transmit the engine's power. Various abnormal sounds or noises occur in the connection points of these components or within the components themselves.

[0005] In the prior art, noise experts employ a method in which an anomalous tone or noise is used, with the aid of expensive analytical equipment, to locate the position or component where the anomalous tone or noise originates in the steering system. Specifically, in a prior art method, the anomalous tone or noise is reproduced in the steering system, the experts estimate the position of the anomalous tone or noise, a sensor is mounted at that position, and a sensing value from the sensor is analyzed.

[0006] In the state-of-the-art method, the noise expert performs the assessment based on their own determination. The ability to analyze the anomalous tone or noise varies among noise experts. Consequently, there is a high probability of an incorrect repair being carried out. Furthermore, it is very time-consuming to mount the sensor directly on a component where the anomalous tone or noise is estimated to occur and to reproduce the anomalous tone or noise for analysis. Moreover, the costs of noise analysis become excessively high with the state-of-the-art method, as an analysis is performed on every component where the anomalous tone or noise is estimated to occur.

[0007] The foregoing serves only to provide a better understanding of the background of the present invention and does not imply that the present invention falls within the scope of the prior art, which is already known to those skilled in the art.

[0008] The foregoing information set forth in this background section serves only to provide a better understanding of the background of the invention and may therefore contain information that does not represent the prior art. US 7,103,460 B1 discloses a device for locating a noise occurring in a steering system, having the features of the preamble of claim 1. Further devices for locating a noise occurring in a steering system are known from KR 10 2019 067 441 A, US 2019 / 0 295 567 A1, and US 2018 / 0 120 264 A1. OVERVIEW

[0009] Exemplary embodiments provide a device for locating a noise in a steering system, with which an anomalous tone or noise in the steering system is simply recorded by a sound recording unit (e.g. a microphone) using a neural network model, the recorded information is entered into the neural network model, which performs a learning process, and thereby a position or component where the anomalous tone or noise occurs in the steering system is localized.

[0010] Further embodiments of the invention are set forth in the following description and will be partly evident from the description or can be experienced during practical application of the invention. The object of the present invention is to provide an improved device for locating noise in the steering system. The invention is defined by the features of claims 1 and 2.

[0011] In at least one embodiment, a device for locating a noise occurring in a steering system is provided, the device comprising: a sound recording unit that detects a noise occurring in a steering system; a processing unit that inputs data relating to the noise in the steering system into a neural network model and locates a position or component where the noise occurs in the steering system, wherein the noise is recorded by the sound recording unit; and a storage unit in which the neural network model is stored. The neural network model can perform learning based on previously received input data.

[0012] In the device, the processing unit can perform preprocessing by converting the time domain data relating to the noise in the steering system into frequency domain data, where the noise is detected by the sound recording unit, and can input the preprocessed data into the neural network model.

[0013] In the device, the processing unit can perform preprocessing by applying a Mel-Frequency-Cepstrum-Coefficient (MFCC) technique to the data relating to the noise in the steering system and extracting a frequency feature of the noise in the steering system, where the noise is detected by the sound recording unit, and it can input the preprocessed data into the neural network model.

[0014] In the device, according to a first embodiment of the invention, the processing unit can perform preprocessing of generating an image resulting from converting the time domain data relating to the noise in the steering system into frequency domain data, wherein the noise is detected by the sound recording unit, and it can input the image generated by the preprocessing into the neural network model.

[0015] In the device, according to a second embodiment of the invention, the processing unit can perform preprocessing by applying an MFCC technique to the data relating to the noise in the steering system, extracting a frequency feature of the noise in the steering system, and generating the extracted frequency feature as an image, wherein the noise is detected by the sound recording unit, and it can input the image generated by the preprocessing into the neural network model.

[0016] In this device, the processing unit can input a region of the image generated by preprocessing into the neural network model.

[0017] According to the invention, the neural network model in the device is a convolutional neural network model comprising a convolution layer, a pooling layer, and a fully connected layer.

[0018] In the device, according to the invention, the convolution layer performs filtering of the image generated by the preprocessing using a filter, it performs a convolution calculation and it creates a feature map to extract a feature from the image generated by the preprocessing, the pooling layer extracts a characteristic value for each region from the feature map created by the convolution layer and it reduces the size of the feature map, the fully connected layer further highlights the features and it categorizes the highlighted features using an activation function, and with regard to the noise in the steering system, the output layer localizes and outputs the position or component where the noise occurs based on a result of the categorization by the fully connected layer.

[0019] The processing unit and storage unit of the device can be implemented in the form of a tablet personal computer (PC).

[0020] With the device for locating noise in the steering system, if the noise occurring in the steering system is detected by the sound recording unit and audio data relating to the detected noise is fed into the neural network model, a result can be obtained indicating the location or component where the anomalous sound or noise originates in the steering system. This significantly reduces the analysis time for noise diagnosis.

[0021] Furthermore, the device for locating noise in the steering system eliminates the need for a noise expert or a noise diagnostic device to pinpoint the noise within the vehicle. It also avoids the need to replace components estimated to be the source of the noise, replacing them one by one to pinpoint the location or component where the noise originates. This reduces the costs associated with noise localization and minimizes the frequency of unnecessary or incorrect repairs.

[0022] In a method not covered by the claims for locating a noise occurring in a steering system, the following is provided: receiving, from a sound recording device, a noise occurring in the steering system and converting the noise into a data stream representing the noise; inputting, by a processing unit, the data stream representing the noise into a data analysis model capable of performing pre-learning; and performing, by the data analysis model, an analysis of the data stream representing the noise and locating a position or component where the noise occurs in the steering system.

[0023] Effects achieved with the present invention are not limited to those described above, as long as they are within the scope of the claims.

[0024] It is understood that both the preceding general description and the following detailed description are exemplary and illustrative and provide a further explanation of the claimed invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] The accompanying drawings, which serve to provide a better understanding of the invention and are incorporated into and form part of this patent application, represent embodiments of the invention and, together with the description, serve to explain the principles of the invention. Fig. Figure 1 is a schematic view illustrating a steering system to be used for diagnosing the localization of an anomalous tone or noise by a noise localization device in a steering system. Fig. Figure 2 is a block diagram illustrating the device for locating the noise in the steering system according to one embodiment. Fig. Figure 3 is a flowchart illustrating an operation of a data preprocessing unit of the device for localizing the noise in the steering system according to one embodiment. Fig. Figure 4 is a diagram illustrating an example of a spectrogram showing the result of a fast Fourier transform by the data preprocessing unit of the device for localizing the noise in the steering system according to one embodiment. Fig. Figure 5 is a diagram illustrating an example of a spectrogram that is ultimately generated by the data preprocessing unit of the device for localizing the noise in the steering system according to one embodiment. Fig. Figure 6 is a view illustrating an example of a neural network model used in the device for localizing noise in the steering system according to an embodiment of the invention in which a neural network model is used. Fig. Figure 7 is a view illustrating an implementation example of the device for locating the noise in the steering system according to one embodiment. DETAILED DESCRIPTION OF THE EXECUTION FORMS SHOWN

[0026] A device for locating a noise in a steering system according to various embodiments is described in detail below with reference to the accompanying drawings.

[0027] In at least one embodiment of a device for localizing noise in the steering system, a neural network model is first created to locate a position or component where an anomalous tone / noise occurs in the component-based steering system. The device then enables the neural network model to perform training using training data that includes the anomalous tone or noise in the steering system and its location. The anomalous tone or noise in the steering system is recorded by a microphone or other type of sound recording unit or component. The device then inputs data regarding the anomalous tone or noise in the steering system to be diagnosed into the learning neural network model for localizing the anomalous tone or noise.Finally, the device enables the neural network model to locate the position of the anomalous tone / noise based on the input data of the anomalous tone / noise.

[0028] ] Fig. Figure 1 is a schematic view illustrating a steering system to be diagnosed for localizing the anomalous tone or noise by a noise localization device in a steering system.

[0029] The steering system in Fig. 1 is a steering system in which a so-called column-mounted motor-driven power steering system (C-MDPS) is used. In such a steering system, the motor unit 12 for the power steering is mounted on a steering column 11. Fig. Figure 1 represents the steering system in which the C-MDPS is used, which is only one of several steering systems in which the present invention is applied. The present invention is not limited to the steering system of Fig. 1 limited.

[0030] With reference to Fig. When a driver turns the steering wheel, a torque sensor measures the torque generated by the driver and transmits a torque measurement to a steering system control unit. The steering system control unit then controls the motor unit 12 to generate a steering force corresponding to the measured torque. The motor unit 12 generates the steering force applied to the steering column 11. The steering force generated by the motor unit 12 is transmitted via a connecting rod, which has a universal joint 14, to a gearbox 16. A rack and pinion mechanism within the gearbox 16 converts a rotary motion into a linear motion, thereby steering the vehicle.

[0031] In this steering system structure, the abnormal tone or noise occurs primarily in a deceleration unit 13 provided on the motor unit 12, the universal joint 14, a bracket 15 holding a rack inside the gearbox 16, and the like.

[0032] In one embodiment, data relating to the anomalous tone or noise occurring in the steering system and information about the position or component where the anomalous tone or noise occurs are pre-acquired and used as learning data that enables the neural network model to perform learning.

[0033] Fig. Figure 2 is a block diagram illustrating the device for locating the noise in the steering system according to one embodiment.

[0034] According to Fig. In one embodiment, the device for localizing noise in the steering system comprises a microphone 100, a processing unit 200, and a storage unit 300. The microphone 100 detects the anomalous tone or noise occurring in the steering system. The processing unit 200 uses the anomalous tone or noise detected by the microphone 100 in the neural network model that performs pre-learning and localizes the position or component where the detected anomalous tone or noise occurs in the steering system. The neural network model that performs pre-learning is stored in the storage unit 300.

[0035] The Microphone 100 is a device for detecting anomalous sounds or noises occurring in the steering system. A cost-effective, commercially available microphone or an expensive, high-performance microphone can be used, as the Microphone 100 only needs to be capable of detecting the anomalous sound or noise that occurs during operation of the steering system in a test environment, such as a vehicle service center.

[0036] The processing unit 200 can include a data preprocessing unit 210. The data preprocessing unit 210 converts the data relating to the anomalous tone or noise recorded by the microphone 100 into a format suitable for training by the neural network model or for use with the neural network model.

[0037] The Processing Unit 200 primarily performs two types of functions. First, the Processing Unit 200 enables the neural network model for localizing the position where the anomalous tone or noise occurs in the steering system to perform the learning process. Second, the Processing Unit 200 uses the data regarding the noise in the steering system being diagnosed to localize the anomalous tone or noise in the neural network model performing the learning process and localizes the position or component where the anomalous tone or noise occurs in the steering system being diagnosed.

[0038] For this purpose, the processing unit 200 can include a machine learning unit 220 and a noise localization unit 230. The machine learning unit 220 performs various types of calculations and processing required for learning by the neural network model. The noise localization unit 230 uses the anomalous tone or noise in the steering system in the neural network model that performs the learning process and performs various types of calculations and processing required to localize the position where the anomalous tone or noise occurs.

[0039] The neural network model used to locate the position or component where the anomalous tone or noise occurs in the steering system is stored in memory unit 300. Furthermore, the training data used for machine learning by the neural network model, specifically the training data resulting from the conversion by data preprocessing unit 210 of processing unit 200, is stored in memory unit 300. Additionally, various data required for training and use by the neural network model are stored in memory unit 300. Data preprocessing

[0040] The data preprocessing unit 210 of the processing unit 200 performs preprocessing in which training data and actual measured noise data relating to a component are converted into suitable formats. The training data is input for the neural network model to learn how to localize the noise in the steering system. The actual measured noise data relating to the component are input into the neural network model.

[0041] The data input into the data preprocessing unit 210 is data relating to the anomalous tone or noise picked up by microphone 100 in the steering system. This data is represented in a time domain. First, the data preprocessing unit 210 converts this time-domain data into frequency-domain data. In this process, the data preprocessing unit 210 applies a Mel-Frequency-Cepstrum-Coefficient (MFCC) technique to extract a characteristic from the audio data.

[0042] Fig. Figure 3 is a flowchart illustrating an operation of the data preprocessing unit 210 of the device for localizing the noise in the steering system according to one embodiment. The flowchart in Fig. Section 3 describes the steps of applying the MFCC technique to data relating to the noise detected by microphone 100 in the steering system and extracting and mapping a frequency domain feature.

[0043] The audio data is not non-stationary (i.e., it can have different frequency domain characteristics over time). Therefore, the data preprocessing unit 210 performs frame forming (step 211) and window forming (step 212) for conversion to the frequency domain for a short period during which the audio data is considered stationary. The parameters of the frame forming and window forming (steps 211 and 212) are approximately matched using an experimental procedure.

[0044] The data preprocessing unit 210 then performs a fast Fourier transform (FFT) on each frame (step 213).

[0045] Fig. Figure 4 is a diagram illustrating an example of a spectrogram showing the result of an FFT by the data preprocessing unit 210 of the device for localizing the noise in the steering system according to an embodiment.

[0046] A horizontal axis of the spectrogram in Fig. The number 4 represents a time frame, and its vertical axis represents a frequency. The intensity of a white color indicates a frequency magnitude at any given time.

[0047] Subsequently, the data preprocessing unit 210 applies a Mel filter bank to the result of the FFT (step 214), performs a discrete cosine transform (DCT) on a result of filtering through the Mel filter bank, and finally creates a spectrogram showing the frequency feature (step 215).

[0048] Fig. Figure 5 is a diagram illustrating an example of a spectrogram that is ultimately generated by the data preprocessing unit 210 of the device for localizing the noise in the steering system according to one embodiment.

[0049] A horizontal axis of the spectrogram in Fig. The number 5 represents a time frame, and its horizontal axis represents a frequency. The intensity of a white color indicates the frequency magnitude at each point in time.

[0050] The above description shows that the MFCC technique is used in all steps from the frame formation and windowing of the data relating to the noise input from microphone 100 to the final execution of the DCT and output of the DCT result.

[0051] The data preprocessing unit 210 identifies a region A of the spectrogram, which exhibits the frequency feature derived by applying the MFCC technique, as a region to be input into the neural network model. An image of region A of the spectrogram, generated by the preprocessing, is created in the storage unit 300. The size of region A is then appropriately calibrated using an experimental procedure. Machine learning through the neural network model

[0052] The machine learning unit 220 is an element that enables the neural network model stored in the memory unit 300 to perform the learning.

[0053] The data used for learning by the neural network model include data regarding the noise picked up by microphone 100 in the steering system and data regarding the position or component where the noise occurs in the steering system.

[0054] This means that the data to be input for learning is data relating to the noise recorded by microphone 100, and the data to be output for learning is data relating to the position or component where the noise occurs in the steering system. Specifically, the data to be input for learning is an image of region A of the spectrogram generated by the data preprocessing unit 210.

[0055] The learning data, which includes the data to be inputted for learning and the data to be output for learning, is acquired by testing various steering systems and the like.

[0056] Fig. Figure 6 is a view illustrating an example of a neural network model used in the device for localizing noise in the steering system according to an embodiment of the invention in which a neural network model is used.

[0057] According to Fig. 6 The neural network model used in the device for localizing the noise in the steering system according to the embodiment is implemented as a convolutional neural network (CNN) comprising convolution layers 61 and 63, pooling layers 62 and 64, a fully connected layer 65 and an output layer 66.

[0058] Convolution layers 61 and 63 perform filtering on an input image A using a filter to extract a feature from the input image generated by the data preprocessing unit 210, perform a convolution calculation, and create a feature map. For example, k feature maps are created with k filters. The filter initially takes a suitable starting value as a numerical value (weighting factor) and then adapts it to a suitable value through learning.

[0059] Pooling layers 62 and 64 extract a characteristic value for each region from the feature maps created by convolutional layers 61 and 63. Pooling layers 62 and 64 perform maximum pooling, average pooling, or similar operations, thereby reducing the size of the feature map and the computational effort required. Maximum pooling involves filtering the feature map using a filter, and the maximum value for a region within that region is determined as the characteristic value. Average pooling involves filtering the feature map using a filter, and the average value for a region within that region is determined as the characteristic value.

[0060] For example, in Fig. Figure 6 shows two folding layers, folding layers 61 and 63, and two pooling layers, pooling layers 62 and 64, but the number of folding layers and the number of pooling layers can be increased or decreased as necessary.

[0061] The feature map output to pooling layer 64 undergoes a flattening process and is then fed into the fully connected layer 65. The flattening process converts the two-dimensional feature map output by pooling layer 64 into a one-dimensional feature map. Features of nodes contained in the fully connected layer 65 are further highlighted and categorized using an activation function. These categorized features are then transferred to output layer 66. Regarding the anomalous tone or noise in the steering system, output layer 66 locates the position or component where the noise originates and outputs a localization result.

[0062] The neural network model is enabled by the machine learning unit 220 to perform learning. For example, the machine learning unit 220 enables the neural network model to learn a weighting factor for a connection between the layers of the neural network model or between the nodes, a parameter for a node state, and the like, using a delta rule and backpropagation learning, which are typical supervised learning techniques.

[0063] After the learning process by the neural network model has finished, the machine learning unit stores 220 different parameters, weighting factors, and the like, which are recorded in the neural network model, in the storage unit 300. Localizing sound using the neural network model

[0064] The noise localization unit 230 described above locates the position or component where the noise occurs in the steering system using the neural network model that performs the learning.

[0065] When the microphone 100 detects the noise occurring in the steering system to be diagnosed for localizing the anomalous tone or noise, the data preprocessing unit 210 performs preprocessing on the data relating to the detected noise, it generates the spectrogram, which is an image showing the frequency feature, and it then stores region A of the spectrogram in the storage unit 300 or makes region A of the same available to the noise localization unit 230.

[0066] Subsequently, the noise localization unit 230 inputs pre-processed image data into the neural network model stored in the storage unit 300, performs a computational processing on each layer of the neural network model and outputs a result of the neural network model that localizes the position or component where the noise occurs in the steering system.

[0067] Fig. Figure 7 is a view illustrating an implementation example of the device for locating the noise in the steering system according to one embodiment.

[0068] As in Fig. As shown in Figure 7, the device for localizing the noise in the steering system according to an embodiment described above is implemented as a combination of a portable tablet PC 700 and the microphone 100 in the test area for actually localizing the anomalous tone or noise.

[0069] A central processing unit (CPU) included in the Tablet PC 700 serves as processing unit 200. A storage device included in the Tablet PC 700, such as flash memory, serves as storage unit 300.

[0070] An application for executing a data preprocessing algorithm and an algorithm for locating the anomalous tone or noise, using the training data described above, is stored in memory unit 300 in the tablet PC 700. A user runs the application and records the anomalous tone or noise from a vehicle undergoing testing using microphone 100 connected to the tablet PC 700. The application then executes the data preprocessing algorithm and the algorithm for locating the anomalous tone or noise. This allows the position, component, and other details where the anomalous tone or noise originates in the steering system of a real vehicle to be determined.

[0071] Specifically, the application records the anomalous sound or noise picked up by microphone 100, analyzes the anomalous sound or noise, and displays a pattern of the anomalous sound or noise. Furthermore, the application executes the machine learning algorithm for localizing the anomalous sound or noise, thus locating the position, component, and other factors where the anomalous sound or noise occurs.

[0072] The device for localizing noise in the steering system, according to the embodiments described above, can be implemented in the form of various computing devices. In particular, the device for localizing noise in the steering system, according to one or more embodiments, can also be implemented as a personal device (smartphone) equipped with a microphone, memory, and a processor. Learning by the neural network model requires significant computational power. Therefore, a high-performance computing device enables the neural network model to perform the learning process. The neural network model that performs the learning and the algorithm for performing the preprocessing are stored in a personal device.This personal device allows for the immediate localization of a position or component where an anomalous tone or noise occurs in a steering system, on a vehicle production line, in a service center, or the like.

[0073] As described above, if a noise occurring in a vehicle's steering system is detected by a microphone and audio data relating to the detected noise is then input into a neural network model, the noise localization device in one embodiment can capture a result of localizing a position or component where the anomalous sound or noise originates in the steering system, thus considerably reducing the time required for noise diagnosis. Furthermore, no noise expert or noise diagnostic device is required to locate the noise in the steering system. It is also unnecessary to replace components where the noise is estimated to originate, a laborious process that is performed one by one to pinpoint the location or component where the noise occurs.This can reduce the costs of locating the noise. Furthermore, it can reduce the frequency of unnecessary or incorrect repairs.

Claims

[1] Device for locating a noise occurring in a steering system, the device comprising: a sound recording unit designed to detect a noise occurring in a steering system; a processing unit (200) configured to input data relating to the noise in the steering system into a neural network model configured to perform pre-learning and to localize a position or component where the noise occurs in the steering system, the noise being recorded by the sound recording unit; and a storage unit (300) in which the neural network model, which is designed to perform pre-learning, is stored, wherein the processing unit (200) is configured to perform preprocessing of generating an image resulting from converting the time domain data relating to the noise in the steering system into frequency domain data, wherein the noise is detected by the sound recording unit, and inputs the image generated by the preprocessing into the neural network model, wherein the neural network model is a convolutional neural network model which has a convolutional layer (61, 63), a pooling layer (62, 64) and a fully connected layer (65), characterized by , that the convolution layer (61, 63) is designed such that it performs filtering of the image generated by the preprocessing using a filter, performs a convolution calculation and creates a feature map to extract a feature from the image generated by the preprocessing, the pooling layer (62, 64) is designed such that it extracts a characteristic value for each region from the feature map created by the folding layer (61, 63) and reduces the size of the feature map. the fully connected layer (65) is designed in such a way that it further highlights the features and categorizes the highlighted features by applying an activation function, and With regard to the noise in the steering system, an output layer (66) is configured such that it locates and outputs the position or component where the noise occurs, based on a result of the categorization by the fully connected layer. [2] Device for locating a noise occurring in a steering system, the device comprising: a sound recording unit designed to detect a noise occurring in a steering system; a processing unit (200) configured to input data relating to the noise in the steering system into a neural network model configured to perform pre-learning and to localize a position or component where the noise occurs in the steering system, the noise being recorded by the sound recording unit; and a storage unit (300) in which the neural network model, which is designed to perform pre-learning, is stored, wherein the processing unit (200) is configured to perform preprocessing by applying a Mel-Frequency-Cepstrum-Coefficient technique to the data relating to the noise in the steering system, extracts a frequency feature of the noise in the steering system and generates the extracted frequency feature as an image, wherein the noise is detected by the sound recording unit, and inputs the image generated by the preprocessing into the neural network model, wherein the neural network model is a convolutional neural network model which has a convolutional layer (61, 63), a pooling layer (62, 64) and a fully connected layer (65), characterized by, that the convolution layer (61, 63) is designed such that it performs filtering of the image generated by the preprocessing using a filter, performs a convolution calculation and creates a feature map to extract a feature from the image generated by the preprocessing, the pooling layer (62, 64) is designed such that it extracts a characteristic value for each region from the feature map created by the folding layer (61, 63) and reduces the size of the feature map. the fully connected layer (65) is configured to further highlight the features and categorize the highlighted features using an activation function, and with regard to the noise in the steering system, an output layer (66) is configured to localize and output the position or component where the noise occurs, based on a result of the categorization by the fully connected layer. [3] Device according to claim 1, wherein the processing unit (200) is configured to input a region of the image generated by preprocessing into the neural network model. [4] Device according to claim 2, wherein the processing unit (200) is configured to input a region of the image generated by preprocessing into the neural network model. [5] Device according to claim 1, wherein the processing unit (200) and the storage unit are implemented in the form of a tablet PC. [6] Device according to claim 1, wherein the sound recording unit comprises a microphone (100).

Images