Noise data artificial intelligence device and preprocessing method for identifying problematic noise sources

Abstract

A noise data artificial intelligence learning method for identifying the source of problem noise may include a noise data preprocessing method for identifying the source of problem noise, the method comprising: selecting a unit frame of problem noise from noise sampled over time; dividing the unit frame into N segments; analyzing the frequency characteristics of each of the N segments, and extracting the frequency components of each segment through a logarithmic Mel filter; and outputting feature parameters as a representative frame by averaging information of the N segments, wherein artificial intelligence learning is performed by applying a bidirectional RNN based on the feature parameters extracted according to time changes according to the noise data preprocessing method.

Description

Technical Field

[0001] The present invention relates to a noise data artificial intelligence device and preprocessing method for identifying problem noise sources, for example, a deep learning-based artificial intelligence problem noise source diagnosis technology for irregular noise such as vehicle noise, especially powertrain noise Background Art

[0002] A vehicle is a finished body of a vehicle into which many parts are assembled.

[0003] A vehicle's powertrain has inherent noise and vibration characteristics.

[0004] However, the noise information generated by the combination of many parts is very large.

[0005] This noise is different from the working noise under normal circumstances and the problem noise under abnormal circumstances.

[0006] When a problem noise occurs, the vehicle is often in an abnormal state.

[0007] However, it is very difficult to determine which part of the vehicle has a fault by only using problematic noise information generated by a vehicle having a very complicated structure.

[0008] For example, in the case of a vehicle's powertrain, it can be difficult to locate the area where problematic noise occurs.

[0009] This is because the problematic noise of the vehicle occurs irregularly.

[0010] Therefore, usually, noise experts diagnose the problem noise through sound evaluation and find the source of the problem noise based on their past experience.

[0011] Currently, even with many noise experts using sensors or other test conditions, it can take a long time.

[0012] However, it is difficult to accurately diagnose the source of problematic noise.

[0013] Recently, efforts have been made to find the source of the problematic noise by using noisy data, but no adequate solution has been found.

[0014] The information included in this Background section of the invention is only for enhancement of understanding of the general background of the invention and should not be taken as an acknowledgement or any form of suggestion that this information forms the prior art already known to a person skilled in the art. Summary of the invention

[0015] Various aspects of the present invention are directed to providing a noise data artificial intelligence apparatus and preprocessing method for identifying sources of problematic noise, so as to maximize the asset value of accumulated noise big data for each type of noise data by using artificial intelligence techniques.

[0016] The present invention includes a noise data artificial intelligence learning method for identifying the source of problem noise, the method comprising: a noise data preprocessing method for identifying the source of problem noise, comprising: selecting a unit frame of problem noise in noise sampled over time; dividing the unit frame into N segments; analyzing the frequency characteristics of each of the N segments, and extracting the frequency components of each segment by applying a logarithmic Mel filter; and outputting feature parameters as a representative frame by averaging information on the N segments, wherein artificial intelligence learning based on feature parameters extracted according to time changes according to the noise data preprocessing method applies a bidirectional RNN.

[0017] Furthermore, the samples are taken over a range twice that of the problem frequency band.

[0018] Furthermore, an overlap is set between a unit frame and a next unit frame over time.

[0019] Furthermore, AI learning is also configured to apply deep neural networks (DNN).

[0020] In addition, artificial intelligence learning also applies attention mechanisms.

[0021] Furthermore, the artificial intelligence learning is configured to additionally apply an early stage ensemble algorithm.

[0022] Furthermore, while the time axis of problematic noise learning data is continuously collected, an integrated model of the jointly trained RNNs algorithm is additionally applied, which uses the time frequency map and the engine RPM frequency map for improving accuracy.

[0023] In addition, in a noise data artificial intelligence diagnostic device implemented using a noise data artificial intelligence learning method for identifying the source of problematic noise, the noise of the vehicle or powertrain is directly measured through the input device of the device, or stored noise data is provided through a storage medium.

[0024] First, it is possible to apply the concept of a multi-diagnosis system configured to output diagnosis results for noise source inputs from 1st to nth in a probabilistic technique considering complex noise sources in a powertrain, thereby diagnosing the source of problematic noise with low characteristics.

[0025] Secondly, it is possible to generate a learned model based on big data of powertrain problem noise, thereby probabilistically and accurately diagnosing complex and irregular powertrain noise or vehicle noise.

[0026] Third, the source of the problematic noise can be accurately diagnosed in a very short time.

[0027] Fourth, GRU, DNN, attention mechanism, and early integration algorithms can be used to prevent data loss in the past.

[0028] Fifth, the learning model for irregular noise can be accurately established by applying an attention mechanism to additionally apply weights of important parts of the learning data.

[0029] Sixth, it can effectively improve the ability of people (professional engineers) to analyze data and determine the required work.

[0030] The method and apparatus of the present invention have other features and advantages which will be more particularly described or apparent from the accompanying drawings or in the accompanying drawings which together with the following detailed description herein serve to explain certain features. Principles of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 is an entire flow chart according to an exemplary embodiment of the present invention.

[0032] Figure 2 Noise data preprocessing for each type according to an exemplary embodiment of the present invention is shown.

[0033] Figure 3 A process of performing diagnosis by inputting any noise data into an artificial intelligence learning model according to an exemplary embodiment of the present invention is shown.

[0034] Figure 4 A time-frequency diagram used in artificial intelligence learning according to an exemplary embodiment of the present invention is shown.

[0035] Figure 5 An engine RPM-frequency graph according to an exemplary embodiment of the present invention is shown.

[0036] It will be appreciated that the drawings are not necessarily drawn to scale, and that they present somewhat simplified representations of various features illustrating the basic principles of the invention. The specific design features of the present invention as incorporated herein, including, for example, specific dimensions, orientations, locations, and shapes will be determined in part by the specific intended application and use environment.

[0037] In the drawings, like reference numerals refer to the same or equivalent parts of the present invention throughout the several figures of the drawing. DETAILED DESCRIPTION

[0038] Reference will now be made in detail to various embodiments of the present invention, examples of which are shown in the accompanying drawings and described below. Although the present invention will be described in conjunction with exemplary embodiments of the present invention, it should be understood that this specification is not intended to limit the present invention to those exemplary embodiments. On the other hand, the present invention is intended to cover not only exemplary embodiments of the present invention, but also various alternatives, modifications, equivalents and other embodiments that may be included within the spirit and scope of the present invention as defined in the appended claims.

[0039] Various modifications can be made and various embodiments can be produced according to various aspects of the present invention, so that specific embodiments are shown in the drawings and described in detail in the specification. However, it is understood that it is not intended to limit the present invention to a specific including form, but to include all modifications, equivalent forms and alternative forms that fall within the spirit and technical scope of the present invention.

[0040] In describing each of the figures, the same reference numerals are used for the same elements.

[0041] The terms "first", "second", etc. may be used to illustrate different components, but these components may not be limited by the terms. These terms are used to distinguish one element from another.

[0042] For example, without departing from the scope of the present invention, a first component may be referred to as a second component, and similarly, a second component may be referred to as a first component. The term "and / or" includes a combination of a plurality of associated listed items or any one of the plurality of associated listed items.

[0043] Unless otherwise defined, all terms used herein, including technical and scientific terms, have the same meaning as commonly understood by one of ordinary skill in the art to which exemplary embodiments of the present invention belong.

[0044] It will be further understood that terms, such as those defined in commonly used dictionaries, may be additionally interpreted as having meanings consistent with their meanings in the context of the relevant art, and will not be interpreted in an idealized or overly formal sense unless expressly defined in the application.

[0045] A noise data preprocessing method of the present invention for identifying sources of problematic noise will be described.

[0046] Any noise is noise for which no information is known regarding the type and source of the problem noise required for diagnosis.

[0047] As will be described later, the present invention may use measured engine RPM information concurrently with any noise.

[0048] Also, if desired, diagnosis can be performed without engine RPM information by using any noise as input.

[0049] Figure 1 is a flow chart of the entirety of an exemplary embodiment of the present invention. In the present case, reference is made to Figure 2 The entire flow chart of the noise data preprocessing device 100-1 is described.

[0050] First, any noise S1 is input through the noise data input unit 110 of the noise data preprocessing apparatus 100 - 1 .

[0051] Among the noises sampled over time by the noise data learning unit 100 , the unit frame S2 is selected for the problem noise.

[0052] The unit frame is divided into N segments S3 by the segment divider 200 .

[0053] Currently, sampling over time is performed over a range that is twice the frequency band in question.

[0054] In an exemplary embodiment of the present invention, the sampling operation is referred to as pre-processing of noisy big data.

[0055] Considering the characteristics of power transmission, diagnostic learning requires preprocessing of big data, and it is important to analyze the noise characteristics of power transmission.

[0056] Preprocessing of noisy big data is used for learning to extract feature parameters for learning with high accuracy.

[0057] Preferably, as a method used in preprocessing of noise big data, resampling is performed by frequency sampling which is twice the frequency of noise to be diagnosed.

[0058] Furthermore, an overlap is set between a unit frame having a time and a next unit frame.

[0059] There may also be overlap between unit frames.

[0060] Even if one unit frame is divided into N segments, there may be overlaps.

[0061] The overlap is to prevent data loss.

[0062] Through the segment analyzer 300 and the average value generator 400, the frequency characteristic of each segment is analyzed by applying a logarithmic Mel filter and the frequency component S4 is extracted.

[0063] The log-Mel filter is used to use both low-frequency components and high-frequency components.

[0064] Through the learning model integrator 500, the learning model generator 600 and the feature parameter extractor 610, a feature vector (feature parameter) is output as a representative frame S5 by averaging information related to the segment.

[0065] The characteristic parameter is a dimensionless concept, the characteristic vector is a concept having a dimension, and the characteristic parameter may be a higher concept than the characteristic vector.

[0066] However, in this article, eigenvectors and eigenparameters can be understood as similar concepts.

[0067] In an exemplary embodiment of the present invention, the sampling operation is referred to as pre-processing of noisy big data.

[0068] At the same time, the output feature vector can be used as an input value for artificial intelligence learning of the artificial intelligence learning machine 700.

[0069] The artificial intelligence logic 710 of the artificial intelligence learning machine 700 ultimately identifies the source of the problematic noise by using the feature vector S6.

[0070] Artificial intelligence logic 710 may be understood in a similar manner to deep learning or machine learning, machine learning, and the like.

[0071] Artificial intelligence learning utilizes characteristic parameters according to changes over time, so all irregular characteristics that are the noise characteristics of the powertrain can be confirmed.

[0072] Currently, artificial intelligence learning can apply bidirectional RNN (GRU: gated recurrent unit).

[0073] The Gated Recurrent Unit (GRU) is one of the recurrent neural network (RNN) technologies, so it is possible to build artificial intelligence learning models based on changes in time.

[0074] At the same time, it can be considered that artificial intelligence learning can additionally apply deep neural networks (DNNs).

[0075] A deep neural network (DNN) may be used as an algorithm for improving the accuracy of the many-to-one probability.

[0076] At present, it is probably best to apply a soft max technique.

[0077] The many-to-one diagnostic technique is a technique for diagnosing the source of problematic noise in a time stream.

[0078] Furthermore, it may be preferred that the AI ​​learning additionally applies an attention mechanism.

[0079] The attention mechanism can be used to improve the information loss of past time data.

[0080] Attention mechanism is a learning technique for weighting important feature vectors over time.

[0081] That is, the attention mechanism can build an accurate learning model for irregular noise by additionally applying weights for important parts of the learning data.

[0082] Currently, it is preferably applied by combining bidirectional RNN (GRU).

[0083] Therefore, AI learning can additionally apply early integration algorithms.

[0084] The early integration algorithm is a weighted learning technique that is used to emphasize the importance of past time data in the problem of recurrent neural network (RNN), which is a higher-level concept of GRU.

[0085] That is, the information can be maintained using early integration techniques by equally dividing the weights of the initial and final times.

[0086] Figure 2 Noise data preprocessing for each type according to an exemplary embodiment of the present invention is shown.

[0087] Figure 3 A process of performing diagnosis by inputting any noise data into an artificial intelligence learning model according to an exemplary embodiment of the present invention is shown.

[0088] Reference Figure 2 and Figure 3 The noise data artificial intelligence diagnosis device 1 can provide a device implemented using a noise data artificial intelligence learning method for identifying the source of the problem noise. The noise data artificial intelligence diagnosis device 1 includes an input device 3, a noise data preprocessing device 100-1, an artificial intelligence learning machine 700, a problem noise source identifier 800 and a problem noise determiner 900.

[0089] The input device 3 can directly measure the noise of the vehicle or powertrain through the noise measurement sensor 3-1, or provide stored noise data through the storage medium 3-2. That is, the storage medium 3-2 stores the noise data, and therefore, the noise data can also be applied to the artificial intelligence learning model.

[0090] The noise data preprocessing device 100 - 1 includes a noise data learning unit 100 , a noise data input unit 110 , a segment divider 200 , a segment analyzer 300 , an average value generator 400 , a learning model integrator 500 , a learning model generator 600 , and a feature parameter extractor 610 .

[0091] The operation of the noise data artificial intelligence diagnostic device 1 is as follows.

[0092] First, noise data for each question type is learned for each unit time by the noise data learning unit 100 .

[0093] The noise data to be input to the noise data input unit 110 is any noise.

[0094] Currently, select the unit frame and select the overlapping range.

[0095] There are various types of vehicle noise, and the types of vehicle noise that occurs irregularly are as follows.

[0096] That is, as the problem noise generated according to the time length, there are problem noise generated in a short time and problem noise generated in a long time.

[0097] Furthermore, as problematic noise generated according to frequency components, there is problematic noise of a specific frequency component.

[0098] Each type of noise data preprocessing of the present invention takes into account both the time characteristics and the frequency component characteristics.

[0099] That is, the time axis is divided into one frame unit to include noise characteristics that change over time and from low-frequency components to high-frequency components.

[0100] The reason for dividing into one-frame units is to accurately determine the noise characteristics that vary with time.

[0101] Therefore, an overlap between one divided frame and the next frame is set.

[0102] Currently, the overlap ratio is applied evenly.

[0103] One frame is divided into N segments (ie, nine segments in total) by the segment divider 200, and the frequency characteristics of each segment are analyzed by the segment analyzer 300 to apply a log-Mel filter.

[0104] The result of applying the log-Mel filter is output as the average value of the average value generator 400 .

[0105] This corresponds to a post-processing operation for increasing the efficiency of frequency characteristics from a low frequency band to a high frequency band.

[0106] Thereafter, features of each data over time are extracted and then integrated into a learning model of the learning model integrator 500.

[0107] Next, a learning feature learning model based on a feature vector (parameter) using vibration big data collected when the same fault phenomenon occurs is generated by the learning model generator 600 .

[0108] The artificial intelligence learning machine 700 may identify the source of the problematic noise by using the extracted feature vector (parameter) of the feature parameter extractor 610 , thereby obtaining the classification result of the problematic noise source identifier 800 .

[0109] In addition, the artificial intelligence learning machine 700 can output highly meaningful data among all noises as the focus result of the problem noise determiner 900.

[0110] The focus results of the problem noise determiner 900 can be used to apply a technique for use when a noise, vibration, harshness (NVH) expert group analyzes noise and compensates for the disadvantage that the weight of the latest data is increased and the weight of the initial data is reduced when analyzing irregular time data.

[0111] Figure 4 A time-frequency graph 711 used for artificial intelligence learning according to an exemplary embodiment of the present invention is shown. Figure 5 An engine RPM-frequency graph 721 is shown according to an exemplary embodiment of the present invention.

[0112] When continuously collecting a timeline of problematic noisy learning data, an Ensemble model of the jointly trained RNNs algorithm can be used to improve accuracy.

[0113] The ensemble model of the jointly trained RNNs algorithm is an algorithm that utilizes the time-frequency map 711 and the engine RPM-frequency map 721.

[0114] The integrated model of jointly trained RNNs algorithms is also called the joint temporal and RPM graph analysis technique.

[0115] Currently, the case where the time axis of the problem noise learning data is continuously collected represents the case where synchronization is performed based on the same time.

[0116] For example, when the time axis of all learning data and input data for diagnosis is 10 seconds, in the case where data is measured at the same time, noise data is collected regularly on the time axis.

[0117] Meanwhile, as yet another exemplary embodiment of an apparatus implemented using a noise data artificial intelligence learning method for identifying problematic noise sources, any noise and engine RPM signal may be applied to a measurable portable device.

[0118] As a result, any measured noise can be analyzed by applying a PC-based or analyzer-based system.

[0119] The present invention is a technology that accurately diagnoses the source of problematic noise in a very short time by using noise data, and can output the diagnosis result of the problematic noise source identifier 800 in about 5 seconds.

[0120] For convenience of explanation and accurate definition of the appended claims, the terms "upper", "lower", "inwardly", "outwardly", "upwardly", "downwardly", "up", "lower", "front", "rear", "back", "inner", "external", "inwardly", "outwardly", "inner", "external", "inner", "outer", "forward" and "rearward" are used to describe features of the exemplary embodiments with reference to the positions of the features as shown in the accompanying drawings. It will also be understood that the term "connect" or its derivatives refer to both direct and indirect connections.

[0121] The foregoing descriptions of specific exemplary embodiments of the present invention have been given for the purpose of illustration and description. They are not intended to be exhaustive or to limit the present invention to the precise form disclosed, and it is apparent that many modifications and variations are possible in light of the above teachings. Exemplary embodiments are selected and described to explain certain principles of the present invention and their practical applications, so that other persons skilled in the art can make and utilize various exemplary embodiments of the present invention and various alternatives and modifications thereof. The scope of the present invention is intended to be limited by the appended claims and their equivalents.

Claims

1. A noise data artificial intelligence learning method for identifying and diagnosing the source of vehicle powertrain problem noise, the method comprising: A noise data preprocessing method for identifying the source of problematic noise, the preprocessing method comprising: Selecting a unit frame of the problem noise among the noise sampled over time; Dividing the unit frame into N segments; analyzing frequency characteristics of each of the N segments and extracting frequency components of each segment by applying a log-Mel filter; and By averaging the information of N fragments, as a representative frame, the feature parameters are output. Among them, the artificial intelligence learning applications based on the feature parameters extracted over time through the noise data preprocessing method include the bidirectional recurrent neural network RNN ​​of the gated recurrent unit GRU; Wherein, the artificial intelligence learning is further configured to apply a deep neural network DNN to improve the probability accuracy of the many-to-one diagnostic technology; wherein the artificial intelligence learning is configured to additionally apply an attention mechanism to improve information loss of past time data; Wherein, the artificial intelligence learning is configured to additionally apply an early integration algorithm to emphasize the importance of past time data in a recurrent neural network (RNN) problem; Therein, when the time axis of the problem noise learning data is continuously collected, an integrated model of the RNN algorithm is additionally applied, and the integrated model of the RNN algorithm jointly uses the time frequency map and the engine RPM-frequency map in order to improve accuracy.

2. The noise data artificial intelligence learning method for identifying and diagnosing the source of vehicle powertrain problem noise according to claim 1, in, The sampling is configured to sample within a range twice that of the problematic frequency band.

3. The noise data artificial intelligence learning method for identifying and diagnosing the source of vehicle powertrain problem noise according to claim 1, in, Sets an overlap between a unit frame and the next unit frame according to time.

4. The noise data artificial intelligence learning method for identifying and diagnosing the source of vehicle powertrain problem noise according to claim 1, in, The AI ​​logic of the AI ​​learning machine ultimately identifies the source of the problematic noise by using the feature vector.

5. The noise data artificial intelligence learning method for identifying and diagnosing the source of vehicle powertrain problem noise according to claim 1, in, Many-to-one diagnostic techniques are used to diagnose the source of problematic noise in the time stream.

6. The noise data artificial intelligence learning method for identifying and diagnosing the source of vehicle powertrain problem noise according to claim 1, in, The attention mechanism builds an accurate learning model for the problem noise by additionally applying weights to important parts of the learning data.

7. The noise data artificial intelligence learning method for identifying and diagnosing the source of vehicle powertrain problem noise according to claim 1, in, Early ensemble algorithms were used to maintain information by equally dividing the weights of the initial and final times in the learning data.

8. A noise data artificial intelligence diagnostic device, used to implement the noise data artificial intelligence learning method for identifying and diagnosing the source of vehicle powertrain problem noise according to claim 1, in, The noise of the vehicle or powertrain is measured directly via an input device of the apparatus, or stored noise data is provided via a storage medium.

Images