DETERIORATION DEGREE PREDICTION METHOD AND DETERIORATION DEGREE PREDICTION SYSTEM

Abstract

A deterioration degree of a noise source is predicted by performing deep learning using frequency data indicating vibration of the noise source during a period and deterioration data indicating deterioration of the noise source during the predetermined period to obtain a condition determination model. Initial vibration signals are measured at the noise point during a monitoring period. The initial vibration signals are converted to a frequency domain, and the frequency-domain-converted initial vibration signals are input to the condition determination model to obtain time-series deterioration data of deterioration of the noise source during the monitoring period. A deterioration prediction model is obtained by deep learning using the time-series deterioration data.The deterioration prediction model predicts deterioration data that indicate the expected deterioration of the noise source during a forecast period. The remaining lifetime of the noise source can be calculated.

Description

Technical FieldThe present invention relates to a deterioration degree prediction method and a deterioration degree prediction system.BackgroundIn recent years, the comfort of vehicle drivers has been considered important, and the noise and vibration (vibration) perceived by the drivers may play an important role in the marketability of vehicles. A variety of sounds and / or vibrations may be generated in vehicles.In particular, in the event of a deterioration of the drive system of a vehicle, this deterioration can lead to noises and / or vibrations and / or to changes in the noises and / or vibrations. It is useful to determine (e.g., detect) this to deal with / counteract the degradation (e.g., replacement, update, and / or maintenance of the component).Brief Discussion of the InventionThe following brief discussion provides a simplified overview of certain features and characteristics. The brief discussion is not a comprehensive survey and is not intended to identify important or critical elements.Systems, apparatus, and methods are described for predicting a degree of degradation. A method may include: determining a state determination model by performing deep learning (also referred to as deep learning) using: a plurality of frequency data elements indicating a plurality of vibrations of a noise source during a predetermined period and a plurality of degradation data elements indicating respective degrees of degradation of the noise source during the predetermined period; measuring a plurality of first vibration signals from the noise source at a noise point during a monitoring period; determining (e.g., estimating) time-series degradation data indicating degrees of degradation of the noise source during the monitoring period by inputting a plurality of first frequency data elements, which are / have been obtained by converting the plurality of first vibration signals into a frequency range, into the state determination model, determining a deterioration prediction model by performing deep learning using the time series deterioration data, predicting, based on the deterioration prediction model, deterioration data indicating degrees of deterioration of the noise source at at least one target time point within a prediction period after the monitoring period, and calculating, based on the time series deterioration data and the deterioration data at the target time point, a remaining life (e.g., remaining life also) of the noise source.Additionally or alternatively, a deterioration degree prediction system may include an input unit configured to acquire vibration signals from a noise point configured to measure noise from a noise source, a determination model learning unit configured to determine a state determination model (e.g., state estimation model) by performing deep learning (deep learning) using: a plurality of frequency data items indicating a plurality of vibrations of the noise point during a predetermined period and a plurality of deterioration data items indicating deterioration degrees of the noise source during the predetermined period, and a controller configured to measure a plurality of first vibration signals at the noise point during a monitoring period, A time series degradation data indicating degrees of degradation of the noise source during the monitoring period to be determined (e.g., estimated) by inputting a plurality of first frequency data items into the state determination model, wherein the plurality of first frequency data items is / have been obtained by converting the plurality of first vibration signals into a frequency range, predicting degradation data indicating a degree of degradation of the noise source at at least one target time point within a prediction period after the monitoring period by a degradation prediction model performing deep learning using the time series degradation data, and calculating the remaining life of the noise source based on the time series degradation data and the predicted degradation data at the at least one target time point.These and other features and advantages are described in more detail below.Brief Description of the DrawingsFIG. 1 is a block diagram schematically illustrating a deterioration degree prediction system according to an example. FIG. 2 is a drawing illustrating an example of a state determination model in the deterioration degree prediction system according to the example. FIG. 3 is a drawing illustrating an example of the output of the state determination model shown in FIG. 2. FIG. 4 is a drawing illustrating an example of a deterioration prediction model in the deterioration degree prediction system according to the example. FIG. 5 is a drawing illustrating an example of a plurality of LSTM cells in FIG. 4. FIG. 6 is a drawing illustrating a detailed configuration of a controller illustrated in FIG. 1. FIG. 7 is a drawing illustrating an example of a graph explaining the data generated by the deterioration degree prediction system according to the example. FIG. 8 is a flowchart of a deterioration degree prediction method according to an example.Detailed DescriptionHereinafter, examples in this specification will be described in detail with reference to the accompanying drawings. The same or similar constituent elements are denoted by the same or similar reference numerals, and repeated description will not be made. As used herein, the suffix "module" and / or the suffix "unit" is given to components and used interchangeably only to facilitate the creation of this description, and the suffixes themselves do not have mutually distinguishable meanings or roles. Further, in describing examples disclosed in this specification, detailed descriptions of known technologies will be omitted when it is determined that the specific description thereof may obscure the gist of the examples disclosed in this specification. Moreover, the accompanying drawings are for easy understanding of the examples disclosed in the present specification, and the technical spirit disclosed in the present specification is not limited by the accompanying drawings, and it is to be understood that the present disclosure includes all modifications, equivalents, and substitutes included within the spirit and technical scope of the present disclosure and invention.Terms having an ordinal number such as first and second are used to describe various constituent elements, but the constituent elements are not limited by the terms. The terms are used only to distinguish one component from another component.In the present application, the terms "including", "comprising", and "with" are intended to indicate the presence of features, numbers, steps, operations, components, and components described in the specification, or a combination thereof, and do not exclude the possibility of the presence or addition of one or more other features, numbers, steps, operations, components, and components, or a combination thereof in advance.In a component for controlling other components under a certain control condition, according to an example, a program may be installed which is implemented as a set of instructions / commands specifying a control algorithm required for controlling the other components. The control component may generate output data by processing input data and / or stored data according to the program (e.g., a stored and / or installed program). The control component may include a non-volatile memory for storing programs and a (e.g., temporary) memory for storing data. Instructions for implementing the program may be stored in a non-transitory computer readable storage medium (e.g., the non-transitory memory).Disclosed herein are methods, apparatus, and / or systems for determining and / or predicting future degradations in one or more components of a vehicle and / or for determining and / or predicting the life remaining of the vehicle propulsion system.FIG. 1 is a block diagram schematically illustrating a deterioration degree prediction system according to an example.A deterioration degree prediction system 10 according to an example of the present disclosure may include a computer system for implementing various examples to be described in this specification. In particular, the degradation level prediction system 10 may include at least one processor and a memory, and the memory may include computer readable instructions, and at least one processor may be configured to perform the instructions stored in the memory, whereby various examples to be described in this specification may be implemented.A "processor" may refer to an apparatus for processing arithmetic operations, logical operations, determination operations, etc. to provide at least one function. The processor may be implemented in the form of hardware, software, or a combination of hardware and software. A processor may be implemented in the form of, for example, software executable in a predetermined area of a memory, such as a task (task), a class, a subroutine, an operation, an object, an execution thread, and a program, or in the form of hardware, such as a field programmable gate array (FPGA) and an application specific integrated circuit (ASIC), or in the form of a combination of software and hardware. The processor may be embodied on a computer readable storage medium and / or distributed among a plurality of computers.For example, the degradation level prediction system 10 may be implemented in the form of a supercomputer, a server, a mainframe computer, a workstation, a personal computer, a laptop computer, a tablet computer, an embedded computer (also "embedded computer"), a portable computer, etc. The degradation level prediction system 10 may include an output device and / or component configured to output one or more results (e.g., an indicator and / or a notification of the degradation level and / or life of a vehicle component) as described herein. However, the scope of the present disclosure is not limited thereto, and the degradation level prediction system 10 may be any data processing device / computing device capable of implementing various functions / methods described in this specification.The deterioration degree prediction system 10 may be implemented in the form of a single data processing apparatus or in the form of a plurality of data processing apparatuses.When a vehicle is driven, noise may be generated in various places. At least one sound generation point in a vehicle (hereinafter referred to as a "sound source") may indicate various points in the vehicle at which sound may be generated, such as parts of the vehicle drive, and for example, a sound source may be, for example, the motor reducer (e.g., a geared motor and / or reducer for a motor, e.g., an electric motor), the brake, a tire, or the like of the vehicle. When referred to herein as "noise", this may be considered synonymous with "noise", and the term "noise" may be interchangeable with the term "noise".Hereinafter, for convenience of description, a point that is within a predetermined range (e.g., in, on, within a predetermined distance) from the position of a part corresponding to the noise source (e.g., the engine reducer of the vehicle, etc.) will be referred to as a "noise point".In this specification, the deterioration degree prediction system 10 is described as predicting (hereinafter, briefly, only predicting) the deterioration degree of a vehicle, but the disclosure is not limited thereto. The degradation level prediction system 10 may be used to predict the degradation level of any other motorized device.The degradation level prediction system 10 may acquire vibration signals of a noise point at a number n of times (where n is a natural number greater than or equal to 2), determine (e.g., estimate) degradation levels of the noise point (and / or the noise source) from data indicated by the vibration signals of the noise point based on deep learning (also referred to as "deep learning" from English), and / or predict the degradation level of the noise source at a target time point in the future based on time series data indicating the degradation levels of the noise source from the past to the current time point. Hereinafter, the current time point may refer to the latest time point among the n time points indicated by the vibration signals of the sound point acquired by the deterioration degree prediction system 10, and a target time point may include at least one time point falling within a predetermined prediction period after the current time point for which the deterioration degree of the sound point (and / or the sound source) is / is to be predicted.Deep learning (deep learning) may be implemented via numerous models, such as convolutional neural network (CNN) models, convolutional neural network (MLP) models, multilayer perceptron (MLP) models, deep neural network regression models, long short term memory (LSTM) models, Convolutional neural network models, etc., may extract a feature map from data and / or simplify (reduce) the feature map and / or convert the dimension of the simplified feature map to a one-dimensional vector through a flattening layer (also called "flattening layer" or "flattening layer" or "vectoring layer", for example, in technical terms) to perform a learning operation.Further, the degradation level prediction system 10 may determine (e.g., estimate) whether the degradation level of a noise source (e.g., the engine reducer) is normal or abnormal from vibration signals associated with the noise point, and / or determine what (e.g., how high) the degradation level is when the degradation level is abnormal.For example, the degradation level prediction system 10 may determine what (e.g., how high) the degradation level of the engine reducer of the vehicle will be at a target time in the future from data indicating the vibration of the noise generated by the engine reducer of the vehicle.Referring to FIG. 1, the deterioration degree prediction system 10 may include an input unit 100, a determination model learning unit (e.g., also determination model learning unit) 200, a prediction model learning unit (e.g., also prediction model learning unit) 300, a controller 400, and / or a storage unit 500. The input unit 100, the determination model learning unit 200, the prediction model learning unit 300, the controller 400, and / or the storage unit 500 may be implemented as one or more processors, and a memory storing instructions that, when executed by the one or more processors, cause the degradation level prediction system 10 to perform one or more operations / operations described herein.At least one sensor mounted at a noise point in the vehicle may measure (e.g., at that noise point) a vibration signal (e.g., also vibration signal). For example, an accelerometer located at a noise point in the vehicle may generate three-axis data indicative of vibration signals related to the x-axis, y-axis, and z-axis.The input unit 100 may acquire vibration signals measured at a sound point corresponding to a sound point generating the vibration signals (e.g., configured to detect vibrations of the sound point generating the vibration signals) from at least one sensor. Specifically, the input unit 100 may acquire vibration signals from at least one sensor within a predetermined range from a noise source corresponding to (e.g., belonging to) the noise point. The vibration signals obtained by the input unit 100 may be measured and / or obtained in real time.The vibration signals measured at the noise point may be time series data. For example, a vibration signal measured at the noise point may include data indicating the vibrations measured for each unit time (also, unit time) during a predetermined monitoring period from the time t_ 0 to the time t_n after the time t_ 0. The time t_ 0 may be the initial time at which at least one sensor has started collecting vibration signals, and the time t_n may be the current time. Hereinafter, for convenience of explanation, the time point t_m (where m is a natural number greater than 1) may be a time point that is (m-1) times the unit time after the time point t_ 1. For example, when the unit time is 1 hour, the time point t_ 5 may be a time point 4 hours later than the time point t_ 1.The input unit 100 may receive (e.g., from an external device) data indicating information on the state of the sound spot associated with a time at which one or more sensors measure vibration signals. Here, failure information may include data indicating the type of the failure (degradation) in the vehicle, the presence or absence of a failure, the degree of degradation, and so forth.For example, the input unit 100 (e.g., from an external device) may obtain data indicating failure information (e.g., on and / or on the motor gear reducer in the vehicle) at a first time when a first vibration signal is measured.However, the disclosure is not limited thereto. The storage unit 500 may include (e.g., store) vibration signals collected from the sound point and / or data indicating information on the state of the sound point at the time of measuring each vibration signal. For example, the storage unit 500 may include a vibration signal indicating noise at the engine reducer, which is a noise source in model A, of a vehicle of model A and / or data indicating failure information about the engine reducer of the vehicle at the time of measurement of the vibration signal.Hereinafter, for convenience of description, it is assumed that the controller 400 acquires vibration signals and data indicating corresponding failure information from the input unit 100.The controller 400 may receive vibration signals obtained / obtained from the input unit 100.The controller 400 may convert one or more (e.g., each) of the vibration signals collected from a sound point into data in a frequency range. The manner in which the controller 400 converts vibration signals into data in a frequency range may be realized by a fast Fourier transform (FFT for short).For convenience of description, data in a frequency range to which the controller 400 converts a vibration signal at a noise point will be referred to as noise point data hereinafter. The controller 400 may store, in the storage unit 500, noise point data corresponding to vibration signals at a noise point acquired in real time.In the following, noise point data can be / are linked to data which have information about the state of the noise point at the time of measuring the noise point data. For example, data indicating fault information about the in-vehicle engine reducer at a first time may be associated with noise point data indicating vibration at the first time.The determination model learning unit (e.g., also estimation model learning unit) 200 may acquire noise point data from the storage unit 500. Noise point data acquired by the determination model learning unit 200 may be data marked (e.g., marked / marked) with the state information of the noise point according to the measurement timing. Some of the data labeled with the state information of the sound point according to the timings at which the vibration signals at the sound point were measured may be a learning data set, and others may be a test data set.The determination model learning unit 200 may generate a state determination model 2000 based on the noise point data labeled with the state information of the noise point, and / or train the state determination model by deep learning. The state determination model 2000 may be implemented in the form of a CNN model, an MLP model, a deep neural network regression model, etc.The determination model learning unit 200 may train the state determination model 2000 using the learning data set. The determination model learning unit may test the state determination model 2000 using the test data set. The state determination model 2000 may be implemented by a regression model for predicting the life of the noise source belonging to the noise point. The state determination model 2000 may use the learning data set to learn, by deep learning, a method of predicting the state information of the sound point from data indicating the vibration (e.g., vibration) of the sound point.The determination model learning unit 200 may provide frequency data indicating the vibration measured by the noise source at a first time to the state determination model 2000, and / or derive an error between the prediction data output by the state determination model 2000 and the degree of deterioration of the noise source at the first time, and / or modify the state determination model 2000 based on the derived error.The state determination model 2000 may be a transfer function configured (e.g., configured) to output the state information of the sound point as an input based on frequency data indicating vibration of the sound point. The state determination model 2000 may be a learning function configured to output the state information of the sound point based on the frequency data indicating the vibration of the sound point as an input.Hereinafter, the learned (e.g., learned / trained) and / or tested state determination model 2000 will be referred to as the learned state determination model 2000 for simplicity.The determination model learning unit 200 may pass the learned state determination model 2000 to the controller 400.The controller 400 may control operations of the input unit 100, the determination model learning unit 200, the prediction model learning unit 300, and / or the storage unit 500. The controller 400 may be implemented in the form of hardware or software or a combination of hardware and software. The control device 400 can be implemented in the form of a processor (e.g. a microprocessor).The controller 400 may take control of the entire process of learning and testing the determination model learning unit 200. The controller 400 may use the learned state determination model 2000 to control the operation of detecting data indicating the degree of deterioration of the noise source at the time of measurement of the vibration signal, based on the data indicating the vibration of the noise point.The controller 400 may determine (e.g., estimate) data indicating the degree of degradation of the noise source using the learned state determination model 2000 and based on the data indicating the vibration measured at the noise point at any time. Therefore, if the learned state determination model 2000 receives frequency data indicating vibrations measured for each of a number n of time units within a predetermined monitoring period (e.g., from the time t_ 0 to the time t_n, hereinafter referred to as "time series frequency data"), it may determine (e.g., estimate) data indicating degrees of degradation of the noise source corresponding to the number n of time units falling within the monitoring period from the time t_ 0 to the time t_n (hereinafter referred to as "time series degradation data", or also "time series degradation data"). Here, n may be the number of time units. For example, when the predetermined time period is 10 hours and each unit time is 1 hour, the time series frequency data and the time series deterioration data may each include data items belonging to the 10 times.The controller 400 may input the time series frequency data to the learned state determination model 2000 and store the time series degradation data estimated by the state determination model 2000 in the storage unit 500. The storage unit 500 may store the time series degradation data in real time.For example, the storage unit 500 may store the degree of degradation of the noise source determined (e.g., estimated) by the state determination model 2000 at time t_ 1, and store the degree of degradation of the noise source determined (e.g., estimated) by the state determination model 2000 at time t_ 2 after time t_ 1.The storage unit 500 may store storage media of at least one kind of storage media including storing a flash type, a hard disk type, a micro type, and a card type (e.g., an SD card (SD=secure digital) or an XD card (XD=eXtreme digital)), and the like, and storing a RAM type (random access memory), an SRAM type (static RAM), a ROM type (read only memory), a PROM type (programmable read only memory), an EEPROM type (electronically erasable PROM), an MRAM type (magnetic RAM), a magnetic disk type and an optical disk type.The controller 400 may predict, based on the data indicating the determined (e.g., estimated) degree of degradation of the noise source, the degree of degradation of the noise source at a target time point that is after the time point at which a vibration signal was measured. The controller 400 may use a deterioration prediction model 3000 generated by the prediction model learning unit 300 to predict the degree of deterioration of the noise source at a target timing.The degradation prediction model 3000 may be implemented in the form of an LSTM model, for example.The prediction model learning unit 300 may retrieve the time-series degradation data from the storage unit 500. The prediction model learning unit 300 may use the time-series degradation data to train the degradation prediction model 3000. The degradation prediction model 3000 may include a plurality of LSTM blocks, and each of the plurality of LSTM blocks may include a plurality of LSTM cells. The degradation prediction model 3000 may learn a data characteristic (history) over time indicated by the time-series degradation data.The deterioration prediction model 3000 may input the time-series deterioration data corresponding to the number of n times falling within the monitoring period from the time t_ 0 to the time t_n to the cells corresponding to the times among a plurality of LSTM cells belonging to one of the plurality of LSTM blocks by deep learning.The deterioration prediction model 3000 may learn the deterioration history of the noise source from the time-series deterioration data corresponding to the plurality of times falling within the monitoring period from the time t_ 0 to the time t_n by the plurality of LSTM cells belonging to each of the plurality of LSTM blocks, and predict the degree of deterioration of the noise source corresponding to a target time point later than the time t_n based on the deterioration history.From time-series data belonging to a plurality of successive times (e.g., a plurality of times falling within the monitoring period from time t_ 0 to time t_n), the deterioration prediction model 3000 may predict data indicating the degree of deterioration of the noise source at a time point that is later / later than the plurality of times (e.g., a target time point after time t_n, which is the last time point in the monitoring period). The degradation prediction model 3000 may be a transfer function that outputs the degradation degree of the noise source according to a target timing that is after the time t_n at which the time series degradation data has been acquired. The degradation prediction model 3000 may use the time-series degradation data to learn how to predict, by deep learning, the degradation degree of the noise source that belongs to each target time point included in a prediction period after the time point t_n.The prediction model learning unit 300 may provide the learned deterioration prediction model 3000 to the controller 400.The controller 400 may take over the control of the entire learning of the prediction model learning unit 300. The controller 400 may use the learned deterioration prediction model 3000 to control the operation of detecting data indicating the degree of deterioration of the noise source at a later time than the times indicated by the time-series deterioration data from the time-series deterioration data. The controller 400 may predict data indicating the degree of degradation of the noise source at a target time point at the noise point using the learned degradation prediction model 3000.The controller 400 may input the data indicating the degrees of deterioration of the noise source according to the units of time corresponding to the monitoring period from the time t_ 0 to the time t_n to the learned deterioration prediction model 3000, and derive the degree of deterioration of the noise source according to a target time point that is after the time t_n from the deterioration prediction model 3000, and store the derived data in the storage unit 500.The storage unit 500 may store the vibration signals of the noise source acquired by the input unit 100, the learned state determination model 2000, and the learned deterioration prediction model 3000. The storage unit 500 may store the time-series deterioration data and data indicating the degree of deterioration of the noise source at the target time point as predicted by the state determination model 2000.FIG. 2 is a drawing illustrating an example of a state determination model in the deterioration degree prediction system according to the example.Referring to FIG. 2, the state determination model 2000 may include a plurality of FC layers 211 to 214 (FC here standing for "fully connected", for example) and a plurality of layers 2113 to 2115.The determination model learning unit 200 may train the state determination model 2000 by inputting time-series frequency data 201 to the state determination model 2000.The time-series frequency data 201 may include a plurality of sound point frequency data items corresponding to (e.g., corresponding to) the vibration signals actually measured at the sound point and label data items (e.g., label data items) indicating the state information of the sound source at the times of measurement of the vibration signals, and the label data items may be respectively associated with the corresponding sound point frequency data items. Each of the plurality of noise point frequency data items may include data in a two-dimensional 3-by-m matrix corresponding to the three axes (x-axis, y-axis, and z-axis) (where m is a natural number greater than 1). The time-series frequency data 201 may be, for example, a 3-by-2048 two-dimensional matrix.The first FC layer 211 may include a plurality of layers 2101 to 2103. The plurality of layers 2101 to 2103 may have a weight and a so-called bias (e.g. as a so-called bias neuron, for example in the form of a constant input, a threshold value or the like). To each of the plurality of layers 2101 to 2103, data belonging to one axis may be input from the time-series frequency data 201.In the following description, it is assumed that the time-series frequency data 201 having a two-dimensional 3-by-248 matrix is input to the first FC layer 211. The layer 2101 may be input with a 2048 bit one-dimensional vector indicating a first line corresponding to the x-axis of the time-series frequency data 201. The layer 2102 may be input with a 2048 bit one-dimensional vector indicating a second line corresponding to the y-axis of the time-series frequency data 201. The layer 2103 may be input with a 2048 bit one-dimensional vector indicating a third line corresponding to the z-axis of the time-series frequency data 201.Each of the plurality of layers 2101 to 2103 may include a predetermined number of nodes and perform matrix calculation on input data and output data corresponding to the number of nodes. For example, each of the plurality of layers 2101 to 2103 may have 6400 nodes, and the layer 2101 may output a one-dimensional 6400 bit vector with respect to the x-axis, and the layer 2102 may output a one-dimensional 6400 bit vector with respect to the y-axis, and the layer 2103 may output a one-dimensional 6400 bit vector with respect to the z-axis.The second FC layer 212 may include a plurality of layers 2104- 2106. The plurality of layers 2104- 2106 may have a weight and bias. To each of the plurality of layers 2104 to 2106, the output of an associated one of the plurality of layers 2101 to 2103 may be input. For example, the layer 2104 may be input with the output of the layer 2101. The layer 2105 may be input with the output of the layer 2102. Layer 2106 may be input with the output of layer 2103Each of the plurality of layers 2104 to 2106 may include a predetermined number of nodes and perform matrix calculation on input data and output data corresponding to the number of nodes. For example, each of the plurality of layers 2104- 2106may have 3200 nodes, and the layer 2104may output a 3200-bit one-dimensional vector with respect to the x-axis, and the layer 2105may output a 3200-bit one-dimensional vector with respect to the y-axis, and the layer 2106may output a 3200-bit one-dimensional vector with respect to the z-axis.The third FC layer 213 may include a plurality of layers 2107 to 2109. The plurality of layers 2107 to 2109 may have a weight and a bias. To each of the plurality of layers 2107 to 2109, output of an associated one of the plurality of layers 2104 to 2106 may be input. For example, the layer 2107 may be input with the output of the layer 2104. The layer 2108 may be input with the output of the layer 2105. The layer 2109 may be input with the output of the layer 2106.Each of the plurality of layers 2107 to 2109 may include a predetermined number of nodes and perform matrix calculation on input data and output data corresponding to the number of nodes. For example, each of the plurality of layers 2107 to 2109 may have 1600 nodes, and the layer 2107 may output a 1600-bit one-dimensional vector with respect to the x-axis, and the layer 2108 may output a 1600-bit one-dimensional vector with respect to the y-axis, and the layer 2109 may output a 1600-bit one-dimensional vector with respect to the z-axis.The fourth FC layer 214 may include a plurality of layers 2110 to 2112. The plurality of layers 2110 to 2112 may have a weight and a bias. To each of the plurality of layers 2110 to 2112, the output of an associated one of the plurality of layers 2107 to 2109 may be input. For example, the layer 2110 may be input with the output of the layer 2107. The layer 2111 may be input with the output of the layer 2108. The layer 2112 may be input with the output of the layer 2109.Each of the plurality of layers 2110 to 2112 may include a predetermined number of nodes and perform matrix calculation on input data and output data corresponding to the number of nodes. For example, each of the plurality of layers 2110 to 2112 may have 400 nodes, and the layer 2110 may output a 400-bit one-dimensional vector with respect to the x-axis, and the layer 2111 may output a 400-bit one-dimensional vector with respect to the y-axis, and the layer 2112 may output a 400-bit one-dimensional vector with respect to the z-axis.The determination model learning unit 200 may normalize the outputs of the plurality of layers 2110 to 2112 and convert the normalized outputs into one-dimensional data and input the one-dimensional data into the layer 2113. For example, the layer 2113 may be input with a 1200-bit one-dimensional vector obtained by flattening (e.g., also "flattening" or "vectoring", for example, reducing the dimensionality of the input vectors / matrices in different ways) the 400-bit one-dimensional vector output by the layer 2110, the 400-bit one-dimensional vector output by the layer 2111, and the 400-bit one-dimensional vector output by the layer 2112.Each of the plurality of layers 2113 to 2115 may include a predetermined number of nodes and perform matrix calculation on input data and output data corresponding to the number of nodes. For example, layer 2113 may have 256 nodes and output a 256-bit one-dimensional vector, and layer 2114 may have 256 nodes and output a 256-bit one-dimensional vector, and layer 2115 may have 1 node and output a 1-bit one-dimensional vector.Data output from the layer 2115 may indicate the degree of malfunction of the noise source. For example, if data output from layer 2115 is 0, this may indicate that the engine reducer is normal (e.g., has a normal operating state), and if data output from layer 2115 falls within a range greater than 0 and equal to or less than 0.04, this may indicate that the engine reducer is in a first degree of degradation, and if data output from layer 2115 falls within a range greater than 0.04 and equal to or less than 0.28, this may indicate that the engine reducer is in a second degree of degradation, and if data output from layer 2115 falls within a range greater than 0.28 and equal to or less than 1, this may indicate that the engine reducer is in a third degree of degradation. In this case, the first degree of deterioration indicates a lower degree of deterioration compared to the second degree of deterioration, and the second degree of deterioration indicates a lower degree of deterioration compared to the third degree of deterioration.The determination model learning unit 200 may compare prediction data output from the layer 2115 when a first data item among the plurality of noise point frequency data items has been input to the state determination model 2000 with a label data item corresponding to the first data item in the learning data set. The output of the layer 2115 may be the output of the state determination model 2000.The determination model learning unit 200 may derive an error between the prediction data output from the state determination model 2000 and the label data (also, for example, "labeling data"), and output the state determination model 2000 based on the derived error.For example, the determination model learning unit 200 may train the plurality of FC layers 211 to 214 and the plurality of layers 2113 to 2115 so as to minimize a loss function (Loss Function) value indicating the difference between the prediction data output from the layer 2115 and the label data.Further, the determination model learning unit 200 may test the state determination model 2000 by inputting the test data set including the time series frequency data 201 to the state determination model 2000.The determination model learning unit 200 may predict the degree of malfunction of the noise source indicated by each of a plurality of noise point frequency data items by the state determination model 2000.The determination model learning unit 200 may provide the state determination model 2000 generated by deep learning to the controller 400. The storage unit 500 may store the learned state determination model 2000.In this way, the controller 400 may provide the learning data set and the test data set to the determination model learning unit 200, and receive the state determination model 2000 from the determination model learning unit 200.The controller 400 may transmit the time series frequency data to the learned state determination model 2000 and acquire the time series degradation data corresponding to the time series frequency data.When the learned state determination model 2000 receives noise point data, it may determine (e.g., estimate) the degree of degradation of the noise source according to the noise point data. The storage unit 500 may store the degree of deterioration of the noise source output from the state determination model 2000.When the learned state determination model 2000 receives the time-series frequency data, it may further determine (e.g., estimate) time-series degradation data according to the time-series frequency data. The storage unit 500 may store the time series degradation data output from the state determination model 2000.FIG. 3 is a drawing illustrating an example of the output of the state determination model shown in FIG. 2.Referring to FIG. 3, the state determination model 2000 may output the degree of degradation of the noise source, and the degree of degradation of the noise source may be one of the first degree of degradation, the second degree of degradation, and the third degree of degradation in this case.In the example of FIG. 2, the state determination model 2000 may output a value indicating the degree of degradation of the engine reducer. When the data output from the state determination model 2000 is 0, it may indicate that the engine reducer is normal (e.g., has a normal operating state), and when the data output from the state determination model 2000 falls within a range greater than 0 and equal to or less than 0.1, it may indicate that the engine reducer is in a first degree of degradation, and when the data output from the state determination model 2000 falls within a range greater than 0.1 and equal to or less than 0.6, it may indicate that the engine reducer is in a second degree of degradation, and when the data output from the state determination model 2000 falls within a range greater than 0.6 and equal to or less than 1, it may indicate that the engine reducer is in a third degree of degradation. In this case, the first degree of deterioration indicates a lower degree of deterioration compared to the second degree of deterioration, and the second degree of deterioration indicates a lower degree of deterioration compared to the third degree of deterioration. In other words, the threshold for the first degree of degradation may be 0.1, the threshold for the second degree of degradation may be 0.6, and the threshold for the third degree of degradation may be 0.9.Referring back to FIG. 3, most values output from the state determination model 2000 that received noise point data indicating that the motor reducer was normal were values close to 0.Most values output from the state determination model 2000 that received noise point data (deterioration degree 1) indicating that the engine reducer is in the first deterioration degree were values close to 0.1.Most values output from the state determination model 2000 that received noise point data (deterioration degree 2) indicating that the engine reducer is in the second deterioration degree were values close to 0.6.Most values output from the state determination model 2000 that received noise point data (deterioration degree 3) indicating that the engine reducer is in the third deterioration degree were values close to 0.9.Referring to FIG. 3, the values output from the state determination model 2000 have a tendency similar to that of the degrees of deterioration of the noise source belonging to the noise point data input to the state determination model 2000. Therefore, the state determination model 2000 may be considered capable of determining (e.g., estimating) the degree of degradation of the engine reducer (noise source) with a significant degree of accuracy.FIG. 4 is a drawing illustrating an example of a deterioration prediction model in the deterioration degree prediction system according to the example.Referring to FIG. 4, the degradation prediction model 3000 may include a plurality of LSTM blocks 311 and 312, a flattening layer (also called "flattening layer" or "vectoring layer", for example, in technical terms) 313, and an FC layer 314.The prediction model learning unit 300 may provide the time-series degradation data to the degradation prediction model 3000, and may predict degradation data indicating the degree of degradation of the noise source at a second time point (e.g., the time point t_ 5) that is after a first time period (e.g., from the time point t_ 0 to the time point t_ 4) corresponding to a plurality of consecutive first time points, based on data belonging to the first time period (e.g., from the time point t_ 0 to the time point t_ 4) among the time-series degradation data, with reference to the LSTM-type degradation prediction model 3000. The prediction model learning unit 300 may modify the deterioration prediction model 3000 based on an error between the data belonging to the time-series deterioration data at the second time point (e.g., the time point t_ 5) and the deterioration data predicted as the degree of deterioration of the noise source at the second time point (e.g., the time point t_ 5).In FIG. 4, the number of the LSTM blocks 311 and 312 is shown to be two, but this is for convenience of explanation only, and the disclosure is not limited thereto.In the following description, it is assumed that in the deterioration prediction model 3000 shown in FIG. 4, the epoch (also known in the art as "epoch") is 1000 and the batch size is 64, the optimizer is an adam optimizer (Ir=0.001), and the loss is an average absolute error.Time-series deterioration data x 1 to x n, which are associated with a number n of times belonging to the monitoring period from the time t_ 0 to the time t_n, may be used as input data 301 for the deterioration prediction model 3000. The time series deterioration data x 1 to x n may indicate the degrees of deterioration of the noise source at the respective timings. For example, data x 1 may indicate the degree of degradation of the noise source at time t_ 1, and data x 2 may indicate the degree of degradation of the noise source at time t_ 2.The plurality of LSTM blocks 311 and 312 may include a plurality of LSTM cells 3101 to 3110. In particular, the LSTM block 311 may include the plurality of LSTM cells 3101 to 3105, and the LSTM block 312 may include the plurality of LSTM cells 3106-3110. In the plurality of LSTM cells 3101 to 3110, the data dimension size value (h) of an LSTM cell state may be 64.Hereinafter, in the description of FIG. 4, for convenience of description, n=5 is assumed, and in FIG. 4, the number of LSTM cells 3101 to 3110 included in each of the plurality of LSTM blocks 311 and 312 is shown to be 5, which corresponds to n=5, but the disclosure is not limited thereto.The LSTM block 311 may receive time-series degradation data (e.g., data xt where t is a natural number equal to or greater than 1 and equal to or less than n) belonging to time point data indicating a time point (e.g., a time point t) among the time-series degradation data x 1 to x n. The plurality of LSTM cells 3101 to 3105 may refer to identical cells having a parameter, and their outputs differ depending on an input data (the input data).Hereinafter, an example of the configuration of each of the plurality of LSTM cells 3101 to 3110 will be described with reference to FIG. 5.FIG. 5 is a drawing illustrating an example of a plurality of LSTM cells in FIG. 4.Referring to FIG. 5, each of the plurality of LSTM cells 3101 to 3110 may receive input data xt corresponding to the time t, as well as the previous cell long-term state c t-1 and the previous cell short-term state h t-1 and output the current cell long-term state c t the current cell short-term state ht, and output data y t respectively. In this case, the long-term states and the short-term states can specify features / properties in the trend of the input data xtover time. Each of the plurality of LSTM cells 3101 to 3110 may determine a weight and a bias corresponding to each of a plurality of parameters so as to be able to output output data, a long-term state, and a short-term state belonging to a later time than the time indicated by the input data xt, on the basis of the long-term state at the previous time and the short-term state at the previous time.Next, the operations of the plurality of LSTM cells 3101 to 3110 will be described with reference to FIGS. 4 and 5.Hereinafter, for convenience of description, it is assumed that the input data associated with the plurality of LSTM cells 3101 to 3105 is the time-series deterioration data x 1 to x n and that the long-term state c t associated with the plurality of LSTM cells 3101 to 3105 is a long-term state c1 t and that the short-term state ht associated with the plurality of LSTM cells 3101 to 3105 is a short-term state h1t, and that the output data y t associated with the plurality of LSTM cells 3101 to 3105 is output data y1 t. Further, it is assumed that the input data associated with the plurality of LSTM cells 3106-3110 is the output data y1t-y1 n of the plurality of LSTM cells 3101-3105, and that the long-term state c t associated with the plurality of LSTM cells 3106-3110 is a long-term state c2t, and that the short-term state ht associated with the plurality of LSTM cells 3106-3110 is a short-term state h2t, and that the output data y associated with the plurality of LSTM cells 3106-3110 is t output data y2t.The LSTM cell 3101 may receive the time-series degradation data x 1, the previous cell long-term state c 1 0 and the previous cell short-term state h 1 0 according to (e.g., at) the time t_ 1, and output the current cell long-term state c 1 1 the current cell short-term state h 1 1 and the output data y 1 1. The previous cell long term state c 1 0 and the previous cell short term state h 10 may be determined / set as initial information in advance.The LSTM cell 3102 may receive the time-series degradation data x 2, the previous cell long-term state c 1 1 and the previous cell short-term state h 1 1 according to (e.g., at) time t_ 2, and output the current cell long-term state c 1 2 the current cell short-term state h 1 2 and the output data y 1 2.The LSTM cell 3103 may receive the time-series degradation data x 3, the previous cell long-term state c 1 2 and the previous cell short-term state h 1 2 according to (e.g., at) time t_ 3, and output the current cell long-term state c 1 3 the current cell short-term state h 1 3 and the output data y 1 3.The LSTM cell 3104 may receive the time-series degradation data x 4, the previous cell long-term state c 1 3 and the previous cell short-term state h 1 3 according to (e.g., at) time t_ 4, and output the current cell long-term state c 1 4 the current cell short-term state h 1 4 and the output data y 1 4.The LSTM cell 3105 may receive the time-series degradation data x 5, the previous cell long-term state c 1 4 and the previous cell short-term state h 1 4 according to (e.g., at) time t_ 5, and output the current cell long-term state c 1 5 the current cell short-term state h 1 5 and the output data y 1 5.The plurality of LSTM cells 3101 to 3105 may learn how to determine the short-term states h 1 1 to h 1 5, the long-term states c 1 1 to c 1 5 and the output data y 1 1 to y 1 5 of the cells at each time based on the time series degradation data x 1 to x n respectively. The output data y 1 1 to y 1 5 may be determined based on the weights and multiple biases of the input data xt, the long-term states c 1 1 to c 1 5 and the short-term states h 1 1 to h 1 5.The output data of the LSTM block 311 may be input to the LSTM block 312. The plurality of LSTM cells 3106-3110 may refer to identical cells having a parameter, and their outputs differ depending on an input data (input data).The LSTM cell 3106 may receive the output data y 1 1 of the LSTM cell 3101, the long-term state c 2 0 of the previous cell, and the short-term state h 2 0 of the previous cell according to (e.g., at) time t_ 1, and output the long-term state c 2 1 of the current cell, the short-term state h 2 1 of the current cell, and the output data y 2 1. The previous cell long term state c 2 0 and the previous cell short term state h 2 0 may be determined / set as initial information in advance.The LSTM cell 3107 may receive the output data y1 2 of the LSTM cell 3102, the long-term state c2 1 of the previous cell, and the short-term state h2 1 of the previous cell according to (e.g., at) time t_2, and output the long-term state c2 2 of the current cell, the short-term state h2 2 of the current cell, and the output data y2 2.The LSTM cell 3108 may receive the output data y1 3 of the LSTM cell 3103, the long-term state c2 2 of the previous cell, and the short-term state h2 2 of the previous cell according to (e.g., at) time t_3, and output the long-term state c2 3 of the current cell, the short-term state h2 3 of the current cell, and the output data y2 3.The LSTM cell 3109 may receive the output data y1 4 of the LSTM cell 3104, the long-term state c2 3 of the previous cell, and the short-term state h2 3 of the previous cell according to (e.g., at) time t_4, and output the long-term state c2 4 of the current cell, the short-term state h2 4 of the current cell, and the output data y2 4.The LSTM cell 3110 may receive the output data y 1 5 of the LSTM cell 3105, the long-term state c 2 4 of the previous cell, and the short-term state h 2 4 of the previous cell according to (e.g., at) time t_ 5, and output the long-term state c 2 5 of the current cell, the short-term state h 2 5 of the current cell, and the output data y 2 5.The plurality of LSTM cells 3106-3110 may learn how to determine the short-term states h2 1 to h2 5, the long-term states c2 1 to c2 5 and the output data y2 1 to y2 5 of the cells at each time based on the output data y1 1 to y1 5 of the plurality of LSTM cells 3101 to 3105. The output data y 2 1 to y 2 5 may be determined based on the weights and multiple biases of the input data y 1 1 to y 1 5, the long-term states c 2 1 to c 2 5 and the short-term states h 2 1 to h 2 5.Referring to FIG. 5, each of the plurality of LSTM cells 3101 to 3110 may include an input gate (input gate), a forget gate (output gate) and an output gate (ft, an input gate it, a hyperbolic tangent input layer gt, and an output gate layer o t.The visitor layer ft may indicate data obtained based on values obtained by applying the weights and biases for the input data x t and the short-term state h t-1 of the previous cell input to a sigmoid layer σ. The visitor layer ft may determine the weight and bias values by learning (e.g., by learning).The input port layer it may indicate data obtained based on values obtained by applying the weights and biases for the input data x t and the short-term state h t-1 of the previous cell input to a sigmoid layer σ. The input port layer it may determine the weight and / or bias values via learning (e.g., through learning).The hyperbolic tangent input layer gt may indicate data obtained based on values obtained by applying the weights and biases for the input data xt and the short-term state h t-1 of the previous cell input to a hyperbolic tangent layer tanh. The hyperbolic tangent input layer gt may determine the weight and / or bias values via learning (e.g., by learning).The output port layer o t may indicate data obtained based on values obtained by applying the weights and biases for the input data x t and the short-term state h t-1 of the previous cell input to a sigmoid layer σ. The output port layer o t may determine the weight and bias values via learning (e.g., by learning).The input port may output a value that is the product of the long term state c t-1 of the previous cell and a value of the visitor layer (f t).The visitor may multiply a value of the input port layer (it) and a value of the hyperbolic tangent input layer (gt), and / or add the output of the input port (the product of the long-term state c t-1 of the previous cell and the value of the visitor layer (ft)) to the product, thereby updating the long-term state c t of the current cell.The output port may update the short-term state h t-1 of the current cell by multiplying the data y t, which is output based on the long-term state c t of the current cell input to the hyperbolic tangent layer tanh, by a value of the output port layer (o t).Referring back to FIG. 4, the flattening layer 313 may flatten the output data y 2 1 to y 2 n, which are output from the plurality of LSTM cells 3106-3110. Flattening layer 313 may receive output data y 2 1 to y 2 n and generate a one-dimensional vector. For example, flattening layer 313 may receive output data y 2 1 to y 2 5 and generate a 5-bit one-dimensional vector. The flattening layer 313 may generate output data by flattening a number n of data into a one-dimensional vector by dimensional transformation while keeping the size of the input data equal.The FC layer 314 may be completely connected to the flattening layer 313, and may output data indicating the degree of deterioration of the noise source associated with each of the plurality of times t_ 0 to t_n and the target times that are after the time t_n. The FC layer 314 may receive the output of the flattening layer 313 as input data and output a one-dimensional vector by performing an operation on the input data. The output of the FC layer 314 may be the output of the degradation prediction model 3000.For example, an FC layer 314 may receive a one-dimensional n-bit vector as input and generate a one-dimensional n-bit vector by performing an operation Wx+b. Here, x, W, and b may respectively indicate an input, a weight, and a bias associated therewith. The FC layer 314 may determine the weight and bias values by learning.The degradation prediction model 3000 may generate output data 302 including the time-series degradation data belonging to the predetermined monitoring period indicated by the input data 301 and time-series degradation data belonging to a predetermined prediction period that is after the predetermined monitoring period and predicted from the input data 301.The controller 400 may input first input data 301 including deterioration data belonging to at least one time point among the time-series deterioration data x 1 to x n and output data (e.g., the data x n+1), which has been output from the deterioration prediction model 3000 at the previous time point, to the deterioration prediction model 3000, thereby generating output data 302 indicating the degree of deterioration of the noise source at a time point after the time point indicated by the first input data 301. The controller 400 may repeatedly collect output data from the degradation prediction model 3000 by delaying the timing of the input data 301 by one unit time until the timing indicated by the output data 302 reaches the final timing belonging to the predetermined prediction period.For example, if the input data 301 is the time-series degradation data x 1 to x 5, which belong to the period from the time t_ 0 to the time t_ 5, the prediction data 302 output from the degradation prediction model 3000 may include the degree of degradation of the noise source at the time t_ 6. If the input data 301 is the time-series deterioration data x 2 to x 6, which are the period from the time t_ 2 to the time t_ 6, the prediction data 302 output from the deterioration prediction model 3000 may include the degree of deterioration of the noise source at the time t_ 7. When the predetermined prediction period is a period from time t_ 6 to time t_ 10, the controller 400 may input the time-series deterioration data x 4 to x 9, which are related to the period from time t_ 4 to time t_ 9, to the deterioration prediction model 3000, and repeatedly collect output data from the deterioration prediction model 3000 until the degree of deterioration of the noise source is predicted at time t_ 10.The controller 400 may output time-series data indicating the degrees of deterioration of the noise source according to the predetermined monitoring period (the period from the time t_ 0 to the time t_n) and the predetermined prediction period (the period from the time t_(n+1) to the time t_(n+k), where k is a natural number greater than 2) (hereinafter referred to as "time-series prediction data") from the deterioration prediction model 3000.The time series prediction data may include data indicating the degrees of deterioration of the noise source for the period from time t_ 0 to time t_(n+k). The flattening layer 313 and the FC layer 314 flatten ("flatten"), for example, an n-bit one-dimensional vector belonging to the predetermined monitoring period (the period from the time t_0 to the time t_n) and a k-bit one-dimensional vector belonging to the predetermined prediction period (the period from the time t_(n+1) to the time t_(n+k)), thereby generating time-series prediction data in the form of an (n+k) one-dimensional vector (in other words, for example, the flattening layer 313 and the FC layer 314 convert the n-bit one-dimensional vector belonging to the predetermined monitoring period and the k-bit one-dimensional vector belonging to the predetermined prediction period, a one-dimensional (n+k)-bit vector um).The storage unit 500 may store the time series prediction data predicted from the state determination model 2000.As described above, in one example, the time-series degradation data output from the state determination model 2000 may be used as learning data for learning the degradation prediction model 3000. However, the example is not limited thereto, and in some examples, the storage unit 500 may store data indicating the degrees of deterioration of the noise source in association with the plurality of times belonging to the predetermined period, and the deterioration prediction model 3000 may learn how to predict data at a later time from the characteristics / characteristics of the data stored in the storage unit 500 over time, the data indicating the degrees of deterioration of the noise source in association with the plurality of times.FIG. 6 is a drawing illustrating a detailed configuration of a controller illustrated in FIG. 1.Referring to FIG. 6, the controller 400 may include a state determination unit (e.g., state estimation unit) 410, a deterioration degree prediction unit 420, and / or a remaining lifetime calculation unit 430.The state determination unit 410 may generate and / or train the state determination model 2000 as described above and / or estimate the state information of the noise source from the learned state determination model 2000.The deterioration degree prediction unit 420 may generate and / or train the deterioration prediction model 3000 as described above and predict the deterioration degree of the noise source according to a target timing from the learned deterioration prediction model 3000.Next, the state information of the noise source acquired (e.g., estimated) by the state acquisition unit 410 and the deterioration degrees of the noise source predicted by the deterioration degree prediction unit 420 will be described with reference to FIG. 7.FIG. 7 is a drawing illustrating an example of a graph explaining the data generated by the deterioration degree prediction system according to the example.Referring to FIG. 7, the remaining lives (RUL, for short) corresponding to the time-series degradation data (degradation data) obtained (e.g., estimated) by the state obtaining unit 410 according to the number of cycles of the vehicle corresponding to a time (time) and / or the degradation degree of the noise source (prediction data) predicted by the degradation degree predicting unit 420 are shown.In the following description, it is assumed that the input data is data corresponding to a period from the 0thcycle to the 1000thcycle, and refer to FIG. 7.Referring to FIG. 7, with time, the remaining lives (RUL) corresponding to the time-series deterioration data (degradation data) acquired (e.g., estimated) by the state acquisition unit 410 in association with each time point from the 0th cycle to the 1000th cycle and / or the remaining lives (RUL) corresponding to the degrees of deterioration of the noise source (prediction data) predicted by the degree-of-deterioration prediction unit 420 in association with the 1000th cycle to the 1500th cycle tend to decrease.In addition, the time-series degradation data (degradation data) and / or the (e.g., predicted) degradation levels of the noise source (prediction data) may have similar tendencies to the actual data (true value) indicating the actual malfunction (e.g., an actual failure) of the noise source. Therefore, in the example, the determination (estimation) and prediction of the degree of degradation of the noise source using the state determination model 2000 and the degradation prediction model 3000 may be considered to be possible with a considerable degree of accuracy.Referring back to FIG. 6, the remaining lifetime calculation unit 430 may calculate the remaining lifetime at a reference time point corresponding to target times, based on the state information of the noise source acquired (e.g., estimated) by the state acquisition unit 410 and the deterioration degree of the noise source predicted by the deterioration degree prediction unit 420.The remaining life calculation unit 430 may perform curve adjustment on the time-series deterioration data belonging to the monitoring period from the time t_ 0 to the time t_n and the data indicating the degrees of deterioration of the noise source after the time t_n, thereby generating a trend curve. The remaining lifetime calculation unit 430 may generate a trend curve corresponding to the time-series prediction data by acquiring (e.g., estimating) a series of linear relationships of data on the basis of the time-series prediction data. The manner in which the remaining lifetime calculation unit 430 performs curve fitting may be implemented by various algorithms.The remaining life calculation unit 430 may calculate the remaining life at the current time based on the trend curve generated by the curve fitting and the deterioration degree of the noise source at the current time. The remaining life calculation unit 430 may derive the deterioration degree of the noise source at the reference time from the trend curve. Here, the reference time point may be one of at least one time point belonging to the target time points. The degree of deterioration of the noise source derived from the trend curve may be different from the degree of deterioration of the noise source at the reference time point among the time series prediction data. This is because the trend curve is a curve generated from the time-series prediction data predicted by the deterioration prediction model 3000 by the curve fitting.The remaining life calculation unit 430 may derive the remaining life of the noise source at the current time based on the degree of deterioration of the noise source at the current time and the degree of deterioration of the noise source at the reference time derived from the trend curve. The manner in which the remaining lifetime calculation unit 430 calculates the remaining lifetime may be implemented by various algorithms.For example, when it is assumed that the time point t=200 is the current time point, the remaining life calculation unit 430 may derive the degree of deterioration of the motor reducer at the time point t=400 from the trend curve and calculate the remaining life of the motor reducer at the current time point based on the degree of deterioration of the motor reducer at the current time point and the degree of deterioration of the motor reducer at the time point t=400.FIG. 8 is a flowchart of a deterioration degree prediction method according to an example.The deterioration degree prediction system 10 has been described, so redundant description will be omitted below. A deterioration degree prediction method to be described below may be a method performed by at least one processor included in the deterioration degree prediction system 10.Referring to FIG. 8, the sensors may collect vibration signals from the noise point (S 100).The sensors may collect vibration signals at the noise point. The input unit 100 may acquire the collected vibration signals from the sensors.The controller 400 may convert the vibration signals into data in the frequency domain (S 200).The determination model learning unit 200 may generate and train the state determination model 2000 to determine (e.g., estimate) the degree of degradation by deep learning under the control of the controller 400 (S 300).The state determination model learning unit 200 may determine the state determination model 2000 on the basis of learning data in which noise point data indicating vibration signals at a first time point is marked ("labeled") with the degree of deterioration of the noise source at the first time point.The state determination unit 410 may determine (e.g., estimate) the time-series degradation data indicating the degrees of degradation of the noise source according to the time units, based on the learned state determination model 2000 during the monitoring period from the time point t_ 0 to the time point t_n, and store the determined (e.g., estimated) data in the storage unit 500 (S 400). The state determination unit 410 may provide the time series frequency data belonging to the monitoring period from the time t_ 0 to the time t_n to the state determination model 2000, and acquire the time series degradation data output from the state determination model 2000.The prediction model learning unit 300 may generate and train the degradation prediction model 3000 to predict the degradation state of the noise source at a target timing that is after the current timing by deep learning under control of the controller 400 (S 500).The prediction model learning unit 300 may determine the deterioration prediction model 3000 based on the time-series deterioration data output from the state determination model 2000.The deterioration degree prediction unit 420 may predict the deterioration degrees of the noise source belonging to target timings after the time t_n and store the predicted data in the storage unit 500 based on the learned deterioration prediction model 3000 (S 600).The deterioration degree prediction unit 420 may provide the time-series deterioration data to the deterioration prediction model 3000 and derive the time-series prediction data output from the deterioration prediction model 3000. The degradation prediction model 3000 may flatten the time-series degradation data and the degradation degrees of the noise source at the target timings belonging to the prediction period that is after the monitoring period indicated by the time-series degradation data, thereby generating the time-series prediction data.The remaining life calculation unit 430 may calculate the remaining life of the noise source at the current time based on the state information of the noise source acquired (e.g., estimated) by the state acquisition unit 410 and the deterioration degree of the noise source predicted by the deterioration degree prediction unit 420 (S 700).The remaining lifetime calculation unit 430 may generate the trend curve according to the time series prediction data from the time series prediction data. The remaining life calculation unit 430 may calculate the remaining life at the current time based on the trend curve generated by the curve fitting and the deterioration degree of the noise source at the current time.The remaining lifetime calculation unit 430 may derive the degree of deterioration associated with the reference time from the trend curve. The remaining life calculation unit 430 may derive the remaining life of the noise source at the current time based on the deterioration degree of the noise source at the current time and the deterioration degree of the noise source at the reference time derived from the trend curve.The present disclosure seeks to provide a deterioration degree prediction method and a deterioration degree prediction system capable of measuring the noise detected from a noise point in a vehicle and determining (e.g., estimating) the deterioration degree of the noise source based on the measured noise and predicting a deterioration degree at a time in the future.A method according to a feature of the present disclosure is a deterioration degree prediction method that is performed (e.g., applied) to a noise point generating noise by at least one processor, and includes a step of determining a state determination model by performing deep learning using a plurality of frequency data items indicating a plurality of vibrations of the noise point during a predetermined period and a plurality of deterioration data items indicating the degrees of deterioration of the noise point (and / or the noise source) during the predetermined period, a step of measuring a plurality of first vibration signals at the noise point during a monitoring period, a step of determining (e.g., estimating) time-series deterioration data, which indicate the degrees of deterioration of the noise point (and / or the noise source) during the monitoring period by inputting a plurality of first frequency data items in the frequency range obtained by converting the plurality of first vibration signals into the state determination model, a step of determining a deterioration prediction model by performing deep learning using the time series deterioration data, a step of predicting deterioration data indicating the degrees of deterioration of the noise point (and / or the noise source) of at least one target time point belonging to a prediction period after the monitoring period from the deterioration prediction model, and a step of calculating the remaining life of the noise source based on the time series degradation data and the degradation data at the target timing.The step of acquiring the state acquisition model may include a step of providing a frequency data item indicating vibration measured from the noise point at a first time as an input to the state acquisition model, a step of predicting a first deterioration data item indicating the degree of deterioration of the noise point (and / or the noise source) according to the frequency data item at the first time, and a step of modifying the state acquisition model depending on an error between the data item indicating the degree of deterioration of the noise point (and / or the noise source) at the first time and the first deterioration data item.The step of obtaining the deterioration prediction model may include a step of providing the time-series deterioration data as an input to the deterioration prediction model, a step of predicting a first deterioration data item indicating the degree of deterioration of the noise point (and / or the noise source) at a second time point that is after a first time period corresponding to a plurality of consecutive first time points, according to data belonging to the first time period among the time-series deterioration data in a long-short-term memory (LSTM) manner, and a step of modifying the deterioration prediction model depending on an error between the data, which belong to the time-series deterioration data at the second time and the first deterioration data item.The step of predicting the degradation data indicative of the degradation levels of the noise point (and / or noise source) may include a step of generating time-series prediction data by flattening (e.g., also "flattening" or "vectoring", for example, reducing the dimensionality of the input vectors / matrices in different ways) the time-series degradation data and the data indicative of the degradation levels of the noise point (and / or noise source) according to the prediction period.The step of calculating the remaining life of the noise source may include: a step of generating a trend curve according to the time series prediction data by performing curve fitting on the time series prediction data; a step of deriving a first degree of deterioration of the noise source according to a reference time among the target times from the trend curve; and a step of calculating the remaining life of the noise source at the current time based on the degree of deterioration of the noise source at the current time and the first degree of deterioration of the noise source.A deterioration degree prediction system according to another feature of the present disclosure includes an input unit that acquires vibration signals from a sound point at which noise is generated, a determination model learning unit that determines a state determination model (e.g., state estimation model) by performing deep learning (deep learning) using a plurality of frequency data items indicating a plurality of vibrations of the sound point during a predetermined period and a plurality of deterioration data items indicating the deterioration degrees of the sound point (and / or the sound source) during the predetermined period, and a controller that measures a plurality of first vibration signals at the sound point during a monitoring period and time-series deterioration data, which indicate the degrees of deterioration of the noise point (and / or the noise source) during the monitoring period, acquires (e.g., estimates) by inputting a plurality of first frequency data items in a frequency range obtained by converting the plurality of first vibration signals into the state acquisition model, and deterioration data indicating the degrees of deterioration of the noise point (and / or the noise source) at at least one target time point belonging (r) to a prediction period later / after the monitoring period from a deterioration prediction model that performs deep learning using the time-series deterioration data, predicts, and calculates the remaining life of the noise source based on the time-series deterioration data and the deterioration data at the target time point.The determination model learning unit may provide a frequency data item indicating vibration measured from the noise point at a first time as an input to the state determination model and predict a first deterioration data item indicating the degree of deterioration of the noise point (and / or the noise source) according to the frequency data item at the first time, and modify the state determination model according to an error between the data item indicating the degree of deterioration of the noise point (and / or the noise source) at the first time and the first deterioration data item.The deterioration degree prediction system may further include a prediction model learning unit that supplies the time-series deterioration data as an input to the long-short-term memory (LSTM) type deterioration prediction model, and predicts a first deterioration data item indicating the degree of deterioration of the noise point (and / or the noise source) at a second time point that is after a first time period corresponding to a plurality of consecutive first times, according to data belonging to the first time period among the time-series deterioration data, and modifies the deterioration prediction model according to an error between data belonging to among the time-series deterioration data at the second time point and the first deterioration data item.The degradation prediction model may generate time-series prediction data by flattening (e.g., also "flattening" or "vectoring", for example, reducing dimensionality of the input vectors / matrices in different ways) the time-series degradation data and data indicative of the degradation levels of the noise point (and / or noise source) according to (e.g., in) the prediction period.The controller may generate a trend curve corresponding to the time series prediction data by performing curve adjustment on the time series prediction data, and derive, from the trend curve, a first sound source degradation degree belonging to a target time point among the target times, and calculate the remaining life of the sound source at the current time point based on the sound source degradation degree at the current time point and the first sound source degradation degree.According to the present disclosure, it is possible to determine an abnormality (abnormality) in a vehicle drive system from a vibration signal of a noise point.According to the present disclosure, it is possible to more accurately estimate an abnormality in a vehicle drive system from vibration of a noise point using a deep learning model.According to the present disclosure, real-time data is used in a vehicle drive system in actual operation. Therefore, it is possible to estimate the degradation state of a component of a vehicle in consideration of the current state of the vehicle and predict a degradation timing in the future or estimate the remaining life at a timing in the future.According to the present disclosure, it is possible to accurately determine degradation in a vehicle drive system, so that it is possible to appropriately deal with / counteract degradation. This can reduce the time and cost for vehicle maintenance.According to the present disclosure, thereby, the riding comfort of the vehicle operator can be improved, and the satisfaction of the vehicle users can be improved.Although the present disclosure has been described in connection with practical examples, the disclosure is not limited to these examples. On the contrary, it is intended to cover various modifications and equivalent arrangements included within the spirit and scope of the appended claims.

Claims

A method comprising: determining a state determination model (2000) by performing deep learning, also called deep learning, using: a plurality of frequency data items indicating a plurality of vibrations of a noise source during a predetermined period and a plurality of degradation data items indicating respective degrees of degradation of the noise source during the predetermined period; measuring a plurality of first vibration signals from the noise source at a noise point during a monitoring period; determining, in particular estimating, time-series degradation data indicating degrees of degradation of the noise source during the monitoring period by inputting a plurality of first frequency data items obtained by converting the plurality of first vibration signals into a frequency range into the state determination model (2000), determining a degradation prediction model ( 3000) by performing deep learning using the time series degradation data, predicting, based on the degradation prediction model ( 3000), degradation data indicating degradation degrees of the noise source at at least one target time point within a prediction period after the monitoring period, and calculating, based on the time series degradation data and the degradation data at the target time point, a remaining life of the noise source.The method of claim 1, wherein the determining the state determination model (2000) comprises: providing a frequency data element indicating vibration measured from the noise point at a first time as an input to the state determination model (2000); predicting a first degradation data element indicating a first degree of degradation of the noise source according to the frequency data element at the first time; and modifying the state determination model (2000) based on an error between a data element indicating the degree of degradation of the noise source at the first time and the first degradation data element.The method according to claim 1 or 2, wherein the determining the degradation prediction model (3000) comprises: providing the time-series degradation data as an input to the degradation prediction model (3000); predicting, by the degradation prediction model (3000), a first degradation data item indicating a second degradation degree of the noise source at a second time point that is after a first time period corresponding to a plurality of consecutive first time points according to data belonging to the first time period among the time-series degradation data; and modifying the degradation prediction model (3000) depending on an error between the data belonging among the time-series degradation data at the second time point and the first degradation data item.The method of claim 3, wherein the degradation prediction model (3000) comprises an LSTM model, wherein LSTM stands for short term long memory.The method of any one of claims 1 to 4, wherein predicting the degradation data indicating the degradation degrees of the noise source comprises generating time-series prediction data by flattening, for example flattening or vectoring, the time-series degradation data and data indicating the degradation degrees of the noise source according to the prediction period.The method of claim 5, wherein the calculating the remaining life of the noise source comprises: generating a trend curve according to the time-series prediction data by performing curve fitting on the time-series prediction data, deriving a first degree of degradation of the noise source according to a reference time point among the at least one target time point from the trend curve, and determining, based on the degree of degradation of the noise source at the current time point and the first degree of degradation of the noise source, the remaining life of the noise source at the current time point.The method of any of claims 1 to 6, wherein the noise source comprises one or more of an engine reducer, a brake, or a tire.A deterioration degree prediction system (10) comprising: an input unit (100) configured to acquire vibration signals from a noise point configured to measure noise from a noise source; a determination model learning unit (200) configured to determine a state determination model (2000) by performing deep learning, also called deep learning, using: a plurality of frequency data items indicating a plurality of vibrations of the noise point during a predetermined period and a plurality of deterioration data items indicating deterioration degrees of the noise source during the predetermined period; and a controller (400) configured to: measure a plurality of first vibration signals at the noise point during a monitoring period, time-series deterioration data, which degrees of deterioration of the noise source during the monitoring period are to be determined by inputting a plurality of first frequency data items into the state determination model (2000), the plurality of first frequency data items being obtained by converting the plurality of first vibration signals into a frequency range, predicting deterioration data indicating a degree of deterioration of the noise source at at least one target time point within a prediction period after the monitoring period by a deterioration prediction model (3000) performing deep learning using the time series deterioration data, and calculating the remaining life of the noise source based on the time series deterioration data and the predicted deterioration data at the at least one target time point.The deterioration degree prediction system (10) according to claim 8, wherein the determination model learning unit (200) is configured to: provide a frequency data item indicating vibration measured from the noise point at a first time as an input to the state determination model (2000); predict a first deterioration data item indicating a first deterioration degree of the noise source according to the frequency data item at the first time; and modify the state determination model (2000) according to an error between a data item indicating the deterioration degree of the noise source at the first time and the first deterioration data item.The deterioration degree prediction system (10) according to claim 8 or 9, further comprising: a prediction model learning unit (300) configured to: provide the time-series deterioration data as an input to the deterioration prediction model (3000); predict, based on data belonging to a first time period among the time-series deterioration data, a second deterioration data item indicating a deterioration degree of the noise source at a second time point that is after the first time period, the first time period corresponding to a plurality of consecutive first time points; and modify the deterioration prediction model (3000) depending on an error between a portion of the time-series deterioration data according to the second time point and the second deterioration data item.The degradation level prediction system (10) of claim 10, wherein the degradation prediction model (3000) comprises an LSTM model, wherein LSTM stands for short term long memory.The degradation level prediction system (10) according to any one of claims 8 to 11, wherein the degradation prediction model (3000) is configured to generate time-series prediction data by flattening, for example flattening or vectoring, the time-series degradation data and data indicating the degradation levels of the noise source according to the prediction period.The deterioration degree prediction system (10) according to claim 12, wherein the controller (400) is configured to: generate a trend curve according to the time series prediction data by performing curve fitting on the time series prediction data, derive, based on the trend curve, a first deterioration degree of the noise source according to a reference time point among the at least one target time point, and calculate, based on the deterioration degree of the noise source at a current time point and the first deterioration degree of the noise source, the remaining life of the noise source at the current time point.The degradation level prediction system (10) of any of claims 8 to 13, wherein the noise source comprises one or more of an engine reducer, a brake, or a tire.

Images