LEARNING DEVICE, SQUEAK NOISE PREDICTION DEVICE, LEARNING METHOD, AND SQUEAK NOISE PREDICTION METHOD

Abstract

A learning device may include one or more processors and a storage medium storing computer-readable instructions that enable the one or more processors to preprocess one or any combination of first raw data extracted from a braking device of a vehicle, second raw data comprising wheel data associated with a wheel of the vehicle, and third raw data measured by an external sensor of the vehicle to obtain training input data at a target time at which the braking device is operating, apply the training input data and target data corresponding to the squeal noise generated by the training input data to a squeal noise prediction model to obtain temporary output data, and train the squeal noise prediction model based on a first loss value,which is generated by applying the temporary output data and the target data to a first loss function associated with the regression.

Description

TECHNICAL FIELDThe present disclosure relates to a learning device for predicting quiet sounds in a vehicle.BACKGROUNDA brake device of a vehicle is a device that converts the rotational kinetic energy of the vehicle wheel into thermal energy by friction between a brake disk and a brake pad included in the brake device. Therefore, when the resonance occurs in the brake device due to the excitation energy between the brake disk and the brake pad, and when the resonance is larger than the damping limit of the system, noise may be generated.If the noises correspond to the frequency range of 1 KHz to 16 KHz, they are quiet noises. This results in great unsatisfaction among users because the sound image is unpleasant. Although the vehicle is constructed such that quieting noises of the brake device are avoided in the developing phase of the vehicle, noises which are not generated in the developing phase of the vehicle can be generated because the brake disk and the brake pad included in the device wear out with the lapse of time and the characteristics of the brake device also change.In order to solve such a problem, a technology for predicting quiet noises and a technology for reducing the predicted quiet noises must be developed.SUMMARYThe present disclosure relates to a learning device, a quiet sound prediction device, a learning method, and a quiet sound prediction method, and more particularly relates to technologies for predicting quiet sounds.Some embodiments of the present disclosure may solve the above-mentioned problems occurring in the related art while maintaining advantages achieved by the related art.An embodiment of the present disclosure may provide a learning device for training a quiet sound prediction model based on training input data generated by preprocessing first raw data, second raw data, and third raw data to reflect a relationship between a rapid change in the friction coefficient between a brake disk and a brake pad and a torque estimate in the quiet sound prediction device, a learning method, and a quiet sound prediction method.An embodiment of the present disclosure may provide a learning device for training a quiet sound prediction model based on a first loss function associated with regression or a second loss function associated with classification to increase user satisfaction using a quiet sound prediction model with better performance than training the quiet sound prediction model using a loss function, a quiet sound prediction device, a learning method, and a quiet sound prediction method.An embodiment of the present disclosure may provide a learning device for performing online learning of a quiet sound prediction model using the obtained data after training the quiet sound prediction model to reduce quiet sounds that can be generated by a brake device that ages according to a use condition of a user, a model generation device, a learning method, and a quiet sound prediction method.Technical problems to be solved by some embodiments of the present disclosure are not necessarily limited to the above-mentioned problems, and solutions provided by one embodiment for other technical problems not mentioned herein can be clearly understood by a person skilled in the art to which the present disclosure pertains from the following description.According to an embodiment of the present disclosure, a learning device may include a memory storing computer-executable instructions and at least one processor that accesses the memory and executes the instructions. The at least one processor may pre-process at least one of first raw data extracted by a brake device of a vehicle, second raw data comprising data elements associated with a wheel of the vehicle, third raw data measured by an external sensor of the vehicle, or any combination thereof to obtain training input data at a target time point at which the brake device is operated, apply the training input data and target data corresponding to the quiet sound generated by the training input data to a quiet sound prediction model to obtain temporary output data, and may train the quiet sound prediction model based on a first loss value obtained by applying the temporary output data and the target data to a first loss function associated with the regression.In one embodiment, the at least one processor may generate the first raw data comprising at least one of oil pressure data associated with oil pressure applied to the brake device, brake disk temperature data, torque data associated with torque applied to a brake disk included in the brake device, or any combination thereof, the second raw data comprising at least one of wheel speed data associated with a speed of the wheel, or roll circumference data of a tire connected to the wheel, or any combination thereof, at the target time, and may generate the third raw data comprising at least one of outside air temperature data from the external sensor, or humidity data from the external sensor, or any combination thereof, at the target time.In one embodiment, the at least one processor may generate first sub-input data comprising the square of the brake disc temperature data and the square of the wheel speed data, may generate second sub-input data comprising a change between the wheel speed data and the wheel speed data at a later time after the target time, a change between the brake disc temperature data and the brake disc temperature data at the later time, a change between the oil pressure data and the oil pressure data at the later time, and a change between the torque data and the torque data at the later time, may generate third sub-input data, the first torque estimation data generated based on the change between the wheel speed data and the wheel speed data at the later time, and second torque estimation data, which are generated based on the first torque estimation data and the brake disc temperature data, and may connect the first sub input data, the second sub input data, and the third sub input data to obtain the training input data.In one embodiment, the at least one processor may train the quiet sound prediction model based on a second loss value obtained by applying the temporary output data and the target data to a second loss function associated with the classification.In an embodiment, the at least one processor may determine a classification class of the training input data based on a comparison between the temporary output data and a predetermined threshold, and may train the quiet sound prediction model based on a second loss value obtained by applying the classification class and the output data to the second loss function.According to an embodiment of the present disclosure, a quiet sound prediction apparatus may include a memory storing computer-executable instructions and at least one processor that accesses the memory and executes the instructions. The at least one processor may apply at least one of data extracted from a brake device of a vehicle, data corresponding to a wheel of the vehicle, or data measured by an external sensor of the vehicle, or any combination thereof, to a trained quiet sound prediction model to obtain an expected probability indicating a probability that quiet sounds of the brake device are generated, determine that the quiet sound is generated in the brake device based on the expected probability being greater than a predetermined threshold, and determine an oil pressure control mode of the vehicle based on at least one of a stability control mode of the vehicle, an outside air temperature measured by the external sensor, or a travel time of the vehicle, or any combination thereof based on the vehicle, determining the quiet sound generated in the brake device.In one embodiment, the at least one processor may determine the oil pressure control mode based on a first sub-condition for comparing the outside air temperature to a first threshold and a second sub-condition for comparing the travel time to a second threshold based on the stability control mode being disabled.In an embodiment, the at least one processor may determine the oil pressure control mode as a first oil pressure control mode based on the first sub-condition being satisfied and the second sub-condition being satisfied, may determine the oil pressure control mode as a second oil pressure control mode based on the first sub-condition being satisfied and the second sub-condition being unsatisfered, and may determine the oil pressure control mode as a third oil pressure control mode based on the first sub-condition being unsatisfered and the second sub-condition being unsatisfered.In one embodiment, the at least one processor may determine a torque compensation amount corresponding to the determined oil pressure control mode based on an oil pressure reduction amount corresponding to the determined oil pressure control mode, a friction coefficient between a brake disk and a brake pad included in the brake device, and a piston surface of the brake disk included in the brake device.In one embodiment, the at least one processor may apply the torque compensation amount to at least one of a motor for generating a braking force by regenerative braking in the vehicle, a motor for generating a braking force using an electronic parking brake of the vehicle, or a transmission for generating a braking force using an engine brake of the vehicle in the vehicle, or any combination thereof, to generate a braking force lost by the oil pressure reduction amount for braking the vehicle.According to an embodiment of the present disclosure, a learning method may include preprocessing at least one of first raw data extracted from a brake device of a vehicle, second raw data including data items associated with a wheel of the vehicle, third raw data measured from an external sensor of the vehicle, or any combination thereof to obtain training input data at a target time point at which the brake device is operating, applying the training input data and the target data corresponding to the quiet sound generated by the training input data to a quiet sound prediction model to obtain temporary output data, and training the quiet sound prediction model based on a first loss value, which is obtained by applying the temporary output data and the target data to a first loss function associated with the regression.In an embodiment, obtaining the training input data may include generating the first raw data including at least one of oil pressure data associated with the oil pressure applied to the brake device, brake disk temperature data, torque data associated with a torque applied to a brake disk included in the brake device, or any combination thereof, generating the second raw data including at least one of wheel speed data associated with a speed of the wheel or roll circumference data of a tire connected to the wheel, or any combination thereof, at the target time, and generating the third raw data including at least one of outside air temperature data from the external sensor or humidity data from the external sensor, or any combination thereof, at the target time.In an embodiment, the learning method may further include generating first sub-input data including the square of the brake disk temperature data and the square of the wheel speed data, generating second sub-input data including a change between the wheel speed data and the wheel speed data at a subsequent time after the target time, a change between the brake disk temperature data and the brake disk temperature data at the subsequent time, a change between the oil pressure data and the oil pressure data at the subsequent time, and a change between the torque data and the torque data at the subsequent time, generating third sub-input data including first torque estimation data generated based on the change between the wheel speed data and the wheel speed data at the subsequent time, and second torque estimation data, generating based on the first torque estimation data and the brake disc temperature data, and connecting the first sub-input data, the second sub-input data, and the third sub-input data to obtain the training input data.In an embodiment, training the quiet sound prediction model may include training the quiet sound prediction model based on a second loss value obtained by applying the temporary output data and the target data to a second loss function associated with the classification.In an embodiment, training the quiet sound prediction model may include determining a classification class of the training input data based on a comparison between the temporary output data and a predetermined threshold value, and training the quiet sound prediction model based on a second loss value obtained by applying the classification class and the output data to the second loss function.According to an embodiment of the present disclosure, a quiet sound prediction method may include applying at least one of data extracted from a brake device of a vehicle, data corresponding to a wheel of the vehicle, or data measured by an external sensor of the vehicle, or any combination thereof, to a trained quiet sound prediction model to obtain an expected probability indicating a probability that quiet sounds of the brake device are generated, determining that the quiet sound is generated in the brake device based on the expected probability being greater than a predetermined threshold, and determining an oil pressure control mode of the vehicle based on at least one of a stability control mode of the vehicle, an outside air temperature measured by the external sensor, a travel time of the vehicle or any combination thereof based on the quiet sound being generated in the brake device.In one embodiment, determining the oil pressure control mode of the vehicle may include determining the oil pressure control mode based on a first sub-condition for comparing the outside air temperature to a first threshold and a second sub-condition for comparing the travel time to a second threshold based on the stability control mode being disabled.In one embodiment, the determining of the oil pressure control mode of the vehicle may include determining the oil pressure control mode as a first oil pressure control mode based on the first sub-condition being satisfied and the second sub-condition being satisfied, determining the oil pressure control mode as a second oil pressure control mode based on the first sub-condition being satisfied and the second sub-condition being unsatisfered, and determining the oil pressure control mode as a third oil pressure control mode based on the first sub-condition being unsatisfered and the second sub-condition being unsatisfered.In an embodiment, the quietness prediction method may further include determining a torque compensation amount corresponding to the determined oil pressure control mode based on an oil pressure reduction amount corresponding to the determined oil pressure control mode, a friction coefficient between a brake disk and a brake pad included in the brake device, and a piston surface of the brake disk included in the brake device.In one embodiment, the quiet sound prediction method may further include applying the torque compensation amount to at least one of a motor for generating a braking force by regenerative braking in the vehicle, a motor for generating a braking force using an electronic parking brake of the vehicle, a transmission for generating a braking force using an engine brake of the vehicle in the vehicle, or any combination thereof, to generate a braking force lost by the oil pressure reduction amount for braking the vehicle.BRIEF DESCRIPTION OF THE DRAWINGSThe above and other features and advantages of the present disclosure will become more apparent from the following detailed description taken in conjunction with the accompanying drawings. FIG. 1 illustrates a learning device for training a quiet sound prediction model according to an embodiment of the present disclosure; FIG. 2 is a flowchart for describing a method for training a quiet sound prediction model according to an embodiment of the present disclosure; FIG. 3 is a drawing illustrating a method of obtaining data items for training a quiet sound prediction model in a learning device according to an embodiment of the present disclosure; FIG. 4 is a drawing for describing training input data in a learning device according to an embodiment of the present disclosure; FIG. 5 is a flowchart for describing a method for predicting quiet sounds using a trained quiet sound prediction model in a quiet sound prediction apparatus according to an embodiment of the present disclosure; FIG. 6 is a flowchart for describing a method for determining a torque compensation amount in a quiet sound prediction apparatus according to an embodiment of the present disclosure; FIG. 7 is a flowchart for describing a method for determining an oil pressure control mode in a quiet sound prediction device according to an embodiment of the present disclosure; FIG. 8 is a flowchart for describing a method of braking a vehicle in a quiet sound prediction apparatus according to an embodiment of the present disclosure; and FIG. 9 illustrates a computer system that may be associated with a learning device, a quiet sound prediction device, a learning method, and a quiet sound prediction method, in accordance with an embodiment of the present disclosure.With respect to the description of the drawings, the same or similar terms may be used for the same or similar components.DETAILED DESCRIPTION OF EMBODIMENTSHereinafter, some embodiments of the present disclosure will be described in detail with reference to the drawings. In adding reference numerals to the components of each drawing, it should be noted that an identical component may be denoted by like reference numerals even though it is illustrated in different drawings. Moreover, detailed description of known features or functions may be omitted so as not to unnecessarily obscure the gist of the present disclosure. Hereinafter, various embodiments of the present disclosure will be described with reference to the accompanying drawings. It should be understood, however, that the embodiments are not necessarily intended to limit the present disclosure to particular embodiments, and that one embodiment may include various modifications, equivalents, and / or alternatives to other embodiments of the present disclosure. With reference to the description of the drawings, similar components may be identified by similar reference numerals.In the present disclosure, the terms such as "first / first / first", "second / second / second", "A", "B", "(a)", "(b)", and the like may be used herein. Such terms may be used only to distinguish one component from another component, but do not necessarily limit the respective components regardless of the order or priority of the respective components. Moreover, unless otherwise defined, terms used herein, including technical and scientific terms, may have the same meaning as commonly understood by one of ordinary skill in the art to which the present disclosure pertains. Such terms as defined in a generally used dictionary may be interpreted as having the same meaning as the contextual meanings in the relevant art. For example, the terms such as "first / first / first", "second / second / second", "1.", "2." or the like used in the present disclosure may be used to refer to various components regardless of the order and / or priority, and to distinguish one component from another component, but do not necessarily limit the components. For example, a first user device and a second user device indicate different user devices, regardless of order and / or priority. For example, without departing from the scope of the present disclosure, a first component may be referred to as a second component, and similarly, a second component may be referred to as a first component.In the present disclosure, the terms "has / have / have", "may have / have", "comprises / comprises" and "comprises / has", or "may comprise" and "may comprise" indicate the presence of corresponding features (e.g., components such as numerical values, functions, operations, or parts), but do not exclude the presence of additional features.It can be understood that when a component (e.g., a component) is referred to as being "(operatively or communicatively) coupled to / on" or "connected to" another component (e.g., a second component), it may be directly coupled to or connected to / on the other component or an intervening component (e.g., a third component) may be present. On the other hand, when a component (e.g., a first component) is referred to as being "directly coupled" or "directly connected" to / at another component (e.g., a second component), it can be understood that there is no intermediate component (e.g., a third component).According to the situation, the term "configured to" used in the present disclosure may be used interchangeably, for example, with the terms "suitable for", "having the capability to", "configured to", "adapted to", "made / done to", or "capable".The term "configured to" does not necessarily mean "specifically designed to" in the hardware. Rather, the term "a device configured to" may mean that the device is "capable of cooperating with another device or other parts. For example, a "processor configured to perform A, B, and C" may mean a general-purpose processor (e.g., a central processing unit (CPU) or an application processor) that can perform corresponding operations by executing one or more software programs stored by a specific processor (e.g., an embedded processor) for performing a corresponding operation or a storage device. The terms in the singular also include the plural, unless the context clearly indicates otherwise. All terms used herein, including technical or scientific terms, may have the same meaning as commonly understood by one of ordinary skill in the art described in the present disclosure. It will be further understood that terms defined in a dictionary and commonly used should also be interpreted as being common in the relevant art, rather than in an idealized or overly formal sense unless expressly so defined herein in various embodiments of the present disclosure. In some instances, although terms are defined in the specification, they should not be construed to exclude embodiments of the present disclosure.In the present disclosure, the terms "A or B", "at least one of A or / and B", or "one or more of A or / and B", and the like, may include any and all combinations of the associated listed elements. For example, the term "A or B", "at least one of A and B", or "at least one of A or B" may refer to all of the case (1) in which at least one A is included, the case (2) in which at least one B is included, or the case (3) in which both at least one A and at least one B are included. Moreover, in describing an embodiment of the present disclosure, each of the terms such as "A or B", "at least one of A and B", "at least one of A or B", "A, B or C", "at least one of A, B and C", "at least one of A, B or C", and "at least one of A, B, C or any combination thereof" may include any or all possible combinations of the elements enumerated together in a corresponding one of the terms. Specifically, the term such as "at least one of A, B, or C, or a combination thereof" may include, for example, more than one "A", "A", more than one "B", "B", more than one "C", or "C", or "AB", or "ABC", which is a combination thereof.Hereinafter, some embodiments of the present disclosure will be described in detail with reference to FIGS. 1 to 9.FIG. 1 illustrates a learning apparatus for training a quiet sound prediction model according to an embodiment of the present disclosure.A learning device 100 according to an embodiment may include a processor 110 and a memory 120 including instructions 122, one or both of which may be present in a plurality or include multiple components.The learning device 100 may indicate a device that trains the quiet sound prediction model. For example, the learning device 100 may perform the following operations to train the quiet sound prediction model. The learning device 100 may pre-process data items from raw data received from a braking device of a vehicle, a wheel of the vehicle, and an external sensor of the vehicle to obtain training input data. The learning device 100 may apply the training input data and target data to the quiet sound prediction model to obtain temporary output data. The learning device 100 may train the quiet sound prediction model based on a loss function obtained by applying the temporary output data and the target data to a loss function.The learning device 100 may pre-process at least one of first raw data extracted by the brake device of the vehicle, second raw data including data items associated with the wheel of the vehicle, third raw data measured by the external sensor of the vehicle, or any combination thereof to obtain the training input data at a target time when the brake device is in operation. The target time point may indicate, for example, a certain time point at which a brake signal is input to the brake device. The first raw data may include oil pressure data, brake disk temperature data, and torque data. The second raw data may include wheel speed data and roll circumference data. The third raw data may include outdoor air temperature data and humidity data. A detailed method for generating the first to third raw data will be described below with reference to FIG. 3.The learning device 100 may apply the training input data and the target data corresponding to quiet sounds generated by the training input data to the quiet sound prediction model to obtain the temporary output data. For example, the temporary output data may be an output generated in response to the training input data by the quiet sound prediction model during the training.The learning device 100 may train the quiet sound prediction model based on a loss value obtained by applying the temporary output data and the target data to a first loss function associated with the regression or a second loss function associated with the classification. For example, the learning device 100 may apply the temporary output data and the target data to the first loss function to obtain a first loss value. Otherwise, the learning device 100 may compare the temporary output data with the target data to determine a classification class of the training input data. The learning device 100 may apply the classification class, the temporary output data, and the target data to the second loss function to obtain a second loss value. The learning device 100 may train the quietness prediction model based on the first loss value or the second loss value. A detailed description thereof will be given below with reference to FIG. 2.The processor 110 may execute software and may control at least one other component (e.g., a hardware or software component) connected to the processor 110. In addition, processor 110 may perform a variety of data processings or calculations. For example, the processor 110 may store the first raw data, the second raw data, the third raw data, and the like in the memory 120. As a reference, the processor 110 may perform all operations performed by the learning device 100. Therefore, for convenience of description in the specification, the operation performed by the learning device 100 will be mainly described as an operation performed by the processor 110.For convenience of description in the specification, the processor 110 is mainly described as a processor in a non-limiting manner. For example, the learning device 100 may include at least one processor. Each of the at least one processor may perform all operations associated with training the quiet sound prediction model.The memory 120 may temporarily and / or permanently store various items of data and / or sub-information required to train the quiet sound prediction model. For example, the memory 120 may store the first raw data, the second raw data, the third raw data, and the like.FIG. 2 is a flowchart for describing a method for training a quiet sound prediction model according to an embodiment of the present disclosure.In operation 210, a learning device (e.g., a learning device 100 of FIG. 1 ) according to an embodiment may pre-process at least one of first raw data extracted from a braking device of a vehicle, second raw data comprising data elements associated with a wheel of the vehicle, third raw data measured from an external sensor of the vehicle, or any combination thereof to obtain training input data at a target time when the braking device is operating.In operation 220, the learning device may apply the training input data and target data corresponding to the quiet sound generated by the training input data to a quiet sound prediction model to obtain temporary output data. For example, the quiet sound prediction model may indicate a model that has been and / or is trained using machine learning and may be a trained machine learning model that outputs a training output (e.g., a quiet sound prediction probability) from a training input (e.g., the training input data).In operation 230, the learning device may train the quietness prediction model based on a first loss value obtained by applying the temporary output data and the target data to a first loss function associated with the regression. However, the method for training the quiet sound prediction model is not limited thereto. For example, the learning device may train the quiet sound prediction model based on a second loss value obtained by applying the temporary output data and the target data to a second loss function related to the classification.For example, the quiet sound prediction model may include a neural network. The neural network may include a plurality of layers. Each layer may include a plurality of nodes. The node may have a node value determined based on an activation function. A node of any layer may be connected to a node (e.g., another node) of another layer via a connection (e.g., a connection edge (connection edge)) with a connection weight. The node value of the node can be relayed or propagated to other nodes via the connection. In an inference operation of the neural network, node values may be passed from a previous layer toward a next layer.The calculation of the forward propagation in the quiet sound prediction model may indicate, for example, a calculation of relaying a node value based on input data in the direction facing from the input layer of the quiet sound prediction model to the output layer. In other words, a node value of the node may be passed (e.g., propagate forward) to a node (e.g., a next node) of a next layer that is connected to the node via the link edge. For example, the node may receive a link weight weighted value from a previous node (e.g., a plurality of nodes) connected via the link edge.The node value of the node may be determined based on applying an activation function to the sum (e.g., weighted sum) of the weighted values received from previous nodes. The neural network parameter may include the above-mentioned link weighting. The neural network parameter may be updated to change in a direction in which the value of a loss function (e.g., the first loss function or the second loss function) is sought. A detailed description relating to the loss function will be given below with reference to FIG. 3.The machine learning model (e.g., the trained quiet noise prediction model) may be generated using machine learning. A learning algorithm may include, for example, but is not limited to supervised learning, unsupervised learning, semi-supervised learning, or reinforcement learning.The learning algorithm may include a machine learning system including an algorithm for performing training on data. Such an algorithm may include, for example, a function of operating an artificial intelligence (AI) computer or an external program automatically learned by the computer without automatic thinking, automatic adaptation, automatic determination, automatic learning, or an external program or the artificial intelligence (AI), or any combination thereof. Machine learning may include, for example, classification, regression analysis, feature learning, online learning, autonomous learning, supervised learning, cluster analysis, dimension reduction, structure prediction, abnormal behavior detection, or a neural network, or any combination thereof.The machine learning model may include a plurality of artificial neural network layers. Specifically, the trained quiet noise prediction model may include a common layer having at least one convolution operation and a plurality of classification layers connected to the common layer. For example, an artificial neural network may include, but is not limited to, a combination of at least one of a deep neural network (DNN), a convolutional neural network (CNN), a U-network for image segmentation (U-network), a recurrent neural network (RNN), a constrained Boltzmann machine (RBM), a deep belief network (DBN), a bidirectional recurrent deep neural network (BRDNN) or deep Q-networks, or any combination thereof. For example, the trained quiet noise prediction model may be a mixed effect, reflection machine learning model. Herein, the mixed effect and reflection machine learning model may include a regression methodology for outputting a prediction probability for quiet sounds from the input data and a methodology that takes into account a distribution-based random effect. To illustrate, the mixed effect and reflection machine learning model may include linear regression, logistic regression, classification and regression tree algorithm, support vector machine (SVM), naive bayes, K-nearest neighbor, random forest algorithm, XGBoost, LightGBM, mixed effects random forests (MERF), mixed effects light gradient boosting (MELGB), or the like.FIG. 3 is a drawing illustrating a method of obtaining data items for training a quiet sound prediction model in a learning device according to an embodiment of the present disclosure.A learning device (e.g., a learning device 100 of FIG. 1 ) according to an embodiment may pre-process a raw dataset 330 received from a brake device 310 of a vehicle 300, a wheel 320 of the vehicle 300 and an external sensor of the vehicle 300 to obtain training input data 340 at a target time at which the brake device 310 is operating.The learning device may generate first raw data including at least one of oil pressure data (e.g., shown as "X3 pressure" in FIG. 3 ) associated with the oil pressure applied to the brake device 310, brake disc temperature data (e.g., shown as "X2 Temp" in FIG. 3 ), torque data (e.g., shown as "X4 torque" in FIG. 3 ), or any combination thereof.The learning device may generate, at the target time, second raw data including at least one of wheel speed data (e.g., shown as "X1 speed" in FIG. 3 ) associated with a speed of the wheel 320, or rolling circumference data (e.g., shown as "X6 rolling circumference" in FIG. 3 ) of a tire combined with the wheel 320, or any combination thereof.The learning device may generate, at the target time, third raw data including at least one of outside air temperature data (e.g., shown as "X 5 outside air temperature" in FIG. 3 ) from an external sensor of the vehicle 300 or humidity data (e.g., shown as "X 7 humidity" in FIG. 3 ) from the external sensor of the vehicle 300, or any combination thereof.The learning device may pre-process the raw data set 330 including the first raw data, the second raw data, and the third raw data to obtain the training input data 340. For example, the learning device may connect a plurality of pieces of partial input data generated by preprocessing the raw dataset 330 to obtain the training input data 340.The learning device may generate first partial input data including the square of the temperature data of the brake disc (i.e., the process of preprocessing the temperature data of the brake disc as described below) and the square of the wheel speed data.The learning device may generate second partial input data including a change between the wheel speed data and the wheel speed data at a subsequent time after the target time, a change between the temperature data of the brake disc and the temperature data of the brake disc at the subsequent time, a change between the oil pressure data and the oil pressure data at the subsequent time, and a change between the torque data and the torque data at the subsequent time.The learning device may generate third partial input data including first torque estimation data generated based on the change between the wheel speed data and the wheel speed data at the subsequent time after the target time and second torque estimation data generated based on the first torque estimation data and the temperature data of the brake disc. A detailed description of the first sub-input data, the second sub-input data, and the third sub-input data will be described below with reference to FIG. 4.The learning device may connect the first sub input data, the second sub input data, and the third sub input data to obtain the training input data 340. However, the method of obtaining the training input data 340 is not limited thereto. For example, the learning device may obtain the training input data 340 by preprocessing raw torque data (not shown) obtained based on a torque applied to the brake device 310, a torque applied to the wheel 320, a torque applied to a regenerative braking motor 370, or any combination thereof.Specifically, the learning device may connect the first sub input data, the second sub input data, and the third sub input data to obtain first training input data (not shown). The learning device may obtain second training input data (not shown) by preprocessing the raw torque wheel data. The learning device may apply the first training input data to the quiet sound prediction model 350 to perform training of the quiet sound prediction model 350. The learning device may apply the second training input data to the quiet sound prediction model 350 to perform training of the quiet sound prediction model 350 based on a time for the above-mentioned training being greater than a predetermined training time.The learning device may train the quiet sound prediction model 350 using the first training input data to reflect a relationship between data items obtained from the vehicle 300 and quiet sound in the quiet sound prediction model 350. Thereafter, the learning device may train the quiet sound prediction model 350 using the first training input data to reflect a relationship between torque data items obtained from the vehicle 300 and quiet sound in the quiet sound prediction model 350.The learning device may apply the obtained training input data 340 and the target data corresponding to the quiet sound generated by the training input data 340 to the quiet sound prediction model 350. The learning device may apply the temporary output data obtained from the quiet sound prediction model 350 and the target data to a loss function to train the quiet sound prediction model 350.The learning device may use a first loss function or a second loss function to train the quiet sound prediction model 350. The first loss function may be, for example, a regression-associated loss function that includes a mean absolute error (MAE), a mean squared error (MSE), a root mean square error (RMS), or the like. The second loss function may be a loss function associated with a classification, including, for example, binary cross entropy, categorial cross entropy, or the like. The learning device may use at least one of the first loss function or the second loss function to train the quietness prediction model 350 based on a condition specified by a user.In particular, when the second loss function is used to train the quiet sound prediction model 350, the learning device may determine a classification class of the training input data 340 based on a comparison between the temporary output data and a predetermined threshold. The learning device may train the quiet sound prediction model 350 based on a loss value obtained by applying the classification class and the target data to the second loss function.The classification class of the training input data 340 may be determined by Table 1 below. [Table 1] Table 1] [Table 1] Table 1]0Determining that noise is not generated predict noise of 25% or less of the noise generation determination in dB1Determining that noise is not generated predict noise of 50% or less of the noise generation determination in dB2Determining that noise is not generated predict noise of 75% or less of the noise generation determination in dB3Determining that noise is generated predict noise of 125% or less of the noise generation determination in dB4Determining that noise is generated predicts noise of 150% or less of the noise generation determination in dBHerein, the classification class being "0" may mean that the training input data 340 includes a class in which a quiet sound is not generated, which may mean that a noise of 25% or less of the quiet sound generation determination is predicted in dB. The classification class being "1" may mean that the training input data 340 includes the class in which the quiet sound is not generated, which may mean that a sound of 50% or less of the quiet sound generation determination is predicted. The classification class being "2" may mean that the training input data 340 includes the class in which the quiet sound is not generated, which may mean that a noise of 75% or less of the quiet sound generation determination is predicted in dB. The classification class being "3" may mean that the training input data 340 includes a class in which the quiet sound is generated, which may mean that a noise of 125% or less of the quiet sound generation determination is predicted in dB. The classification class being "4" may mean that the training input data 340 includes the class in which the quiet sound is generated, which may mean that a noise of 150% or less quiet sound generation determination is predicted in dB.The learning device may apply the result output from the quiet sound prediction model 350, trained by at least one of the first loss function or the second loss function (e.g., an expected probability of quiet sounds or whether quiet sounds are generated), to control logic 360. The learning device may apply the output result to the control logic 360 and thus adjust and / or correct oil pressure braking in the braking device 310 or regenerative braking in a regenerative braking motor. A detailed description will be given hereinbelow with reference to FIGS. 7 to 8.FIG. 4 is a drawing for describing training input data in a learning device according to an embodiment of the present disclosure.Referring to FIG. 4, the training input data 400 may include data items obtained by preprocessing raw data received from a brake device of a vehicle, a wheel of the vehicle, and an external sensor of the vehicle.The training input data 400 may include the following variables. For example, "Bremsscheibe_1c_2d" may refer to the square of the brake disc temperature data. Herein, a value calculated by squaring the brake disk temperature data may include a relationship between a width in which the temperature of a brake disk changes and a fluctuation width of a friction coefficient between the brake disk and a brake pad.In the training input data 400, "deceleration" may refer to a change (which may indicate a difference, for example, hereinafter referred to as "change") between wheel speed data at a time "t" and wheel speed data at a time "t-n". The change between the pieces of wheel speed data may include, for example, a deceleration having a linear relationship to the coefficient of friction between the brake disc and the brake pad based on a linear relationship between torque and deceleration.In the training input data 400, "Temp_Rate" may be a change in the brake disk temperature data, which may refer to a change between brake disk temperature data at time "t" and brake disk temperature data at time "t-n". The change between pieces of brake disk temperature data may include, for example, a relationship between a width in which the temperature of the brake disk changes and a fluctuation width of a friction coefficient between the brake disk and the brake pad.In the training input data 400, "Temp_Ruck" may be a change between a change in the brake disk temperature data at time "t" and a change in the brake disk temperature data at time "t-n". For example, the change between the changes in the brake disk temperature data may include a relationship between a change in the changed target and the change in the brake disk temperature data.In the training input data 400, "Pressure_Rate" may be a change between oil pressure data at time "t" and oil pressure data at time "t-n". The change between the oil pressure data may include, for example, a physical phenomenon in which the friction coefficient between the brake disc and the brake pad is influenced by a pressure.In the training input data 400, "Torque_Schätz" may be first torque estimation data that may refer to a relationship between a change in the wheel speed data and oil pressure data. For example, the first torque estimate data may indicate data generated based on a characteristic of the deceleration having a linear relationship to the torque.In the training input data 400, "Torque_Schätz_Temp" may be second torque estimation data that may relate to the result of multiplying the first torque estimation data by the brake disc temperature data. For example, the second torque estimation data may indicate data generated based on a feature reflecting the influence of the temperature in the deceleration feature having the linear relationship with the torque.In the training input data 400, "kE" may be the kinetic energy of a vehicle, which may refer to the square of the wheel speed data. In the training input data 400, "kE_CumSum" may be a value in which the kinetic energy of the vehicle is accumulated, which may refer to the sum of the kinetic energy of the vehicle during a predetermined duration.In the training input data 400, "Torque_Rate" may be a change between torque data at time "t" and torque data at time "t-n". Here, the torque data at time "t" and the torque data at time "t-n" may indicate first torque data at time "t" and first torque data at time "t-n", respectively, or second torque data at time "t" and second torque data at time "t-n", respectively. The torque data may include, for example, a feature in which the coefficient of friction between the brake disk and the brake pad is unstable when the torque applied to the brake device greatly changes.However, the variables included in the training input data 400 are not limited thereto. For example, a learning device may pre-process data items from raw data associated with the braking device of the vehicle to obtain the training input data 400. Thus, for convenience, the variables included in the training input data 400 are mainly described in the specification as the variables shown in FIG. 4.FIG. 5 is a flowchart for describing a method for predicting quiet sounds using a trained quiet sound prediction model in a quiet sound prediction apparatus according to an embodiment of the present disclosure.In operation 510, the quiet sound prediction device according to an embodiment may apply at least one of data extracted from a braking device of a vehicle, data corresponding to a wheel of the vehicle, data measured from an external sensor of the vehicle, or any combination thereof to a trained quiet sound prediction model (e.g., quiet sound prediction model 350 of FIG. 3 ) to obtain an expected probability indicating a probability that a quiet sound of the braking device will be generated.For example, the quiet sound prediction device may include a processor and memory having instructions. The processor may execute software and may control at least one other component (e.g., a hardware or software component connected to the processor). In addition, the processor may perform a variety of data processings or calculations. For example, the processor may store the data extracted by the brake device of the vehicle, the data corresponding to the wheel of the vehicle, the data measured by the external sensor of the vehicle, and the trained quiet sound prediction model in the memory. The memory may temporarily and / or permanently store various items of data and / or partial information required to draw conclusions about the trained quiet sound prediction model.In operation 520, the quietness predictor may determine that the quietness is generated in the brake device based on the expected probability being greater than a predetermined threshold. For example, the quiet sound prediction device may determine that the quiet sound is generated in the braking device when the expected probability obtained from the trained quiet sound prediction model is greater than the user-predetermined threshold.In operation 530, the quietness predictor may determine an oil pressure control mode of the vehicle based on at least one of a stability control mode of the vehicle, an outside air temperature measured by the external sensor, a travel time of the vehicle, or any combination thereof, that the quietness is generated in the brake device. The oil pressure control mode may be, for example, a mode in which the level of the oil pressure applicable to the brake device is determined. A detailed description thereof will be given below with reference to Figs. 8 to 9.FIG. 6 is a flowchart for describing a method for determining a torque compensation amount in a quiet sound prediction apparatus according to an embodiment of the present disclosure.In operation 610, the quiet sound predictor may output a driving state of the vehicle. For example, the vehicle driving state may include data extracted from a brake device of a vehicle, data corresponding to a wheel of the vehicle, and data measured from an external sensor of the vehicle. The quiet sound prediction device may apply the outputted vehicle running state to a trained quiet sound prediction model.In operation 620, the quiet sound predictor may determine whether a quiet sound is being generated. For example, the quiet sound prediction device may apply the driving state of the vehicle to the trained quiet sound prediction model and thus obtain an expected probability of quiet sounds in the brake device. The quiet sound prediction device may determine that the quiet sound is generated in the brake device based on the expected probability being greater than a predetermined threshold. On the basis that the quiet sound is generated in the brake device, the quiet sound prediction device can perform the following operations.In operation 630, the quiet sound predictor may determine whether a stability control mode of the vehicle is in an inactive state. Based on the stability control mode of the vehicle being in the inactive state, the quietness prediction device may perform the following operations.In operation 640, the quietness predictor may obtain an outside air temperature measured by an external sensor and a travel time of the vehicle. As a result, the quiet sound predictor may reduce oil pressure and compensate for torque using an actuator in operation 650. A detailed description thereof will be given with reference to FIG. 8.FIG. 7 is a flowchart for describing a method for determining an oil pressure control mode in a quiet sound prediction device according to an embodiment of the present disclosure.The quiet sound prediction device according to an embodiment may determine an oil pressure control mode based on a first sub-condition for comparing an outside air temperature with a first threshold and a second sub-condition for comparing a travel time with a second threshold based on a stability control mode being disabled.In operation 710, the quietness predictor may compare the outdoor air temperature to the first threshold (e.g., shown as 0 degrees Celsius in FIG. 7 ). For example, if the outside air temperature is greater than or equal to the first threshold, the quietness predictor may compare a travel time (e.g., shown as dwell time in FIG. 7 ) of a vehicle to a second threshold (e.g., shown as 2 hours in FIG. 7 ) in operation 720. When the outside air temperature is greater than or equal to the first threshold and the travel time of the vehicle is less than or equal to the second threshold, the quietness prediction device may determine the oil pressure control mode of a brake device as a third oil pressure control mode.In operation 730, the quietness predictor may determine the oil pressure control mode as a first oil pressure control mode based on the first precondition being met and the second precondition being met. To illustrate, the quietness prediction device may determine the oil pressure control mode as a second oil pressure control mode based on the first precondition being satisfied and the second precondition not being satisfied. For illustrative purposes, based on the first precondition not being satisfied and the second precondition not being satisfied, the quietness prediction device may determine the oil pressure control mode as the third oil pressure control mode.As a reference, for convenience of description in the specification, the first oil pressure control mode may indicate a mode for applying an oil pressure reduced by 5 bar or more from the oil pressure applied to the brake device to the brake device, the second oil pressure control mode may indicate a mode for applying an oil pressure reduced by 3 bar or more and 5 bar or less from the oil pressure applied to the brake device to the brake device, and the third oil pressure control mode may indicate a mode for applying an oil pressure reduced by 3 bar or less from the oil pressure applied to the brake device to the brake device.FIG. 8 is a flowchart for describing a method of braking a vehicle in a quiet sound prediction apparatus according to an embodiment of the present disclosure.In operation 810, the quiet sound prediction device according to an embodiment may determine an oil pressure control mode of a vehicle. For example, as described above with reference to FIG. 7, the quietness prediction device may determine the oil pressure control mode based on the result of comparing the outside air temperature with the first threshold and comparing the travel time with the second threshold based on the stability control mode being deactivated.In operation 820, the quietness prediction device may determine a torque compensation amount corresponding to the determined oil pressure control mode based on an oil pressure reduction amount corresponding to the determined oil pressure control mode, a coefficient of friction between a brake disc and a brake pad included in the brake device, and a piston surface of the brake disc included in the brake device. For example, when the determined oil pressure control mode is a first oil pressure control mode, the quietness prediction device may determine a torque compensation amount corresponding to an oil pressure reduced by 5 bar or more.In operation 830, the quietness prediction device may generate a braking force lost by the oil pressure decrease amount to decelerate the vehicle. For example, the quietness prediction device may apply the determined torque compensation amount to at least one of a motor for generating a braking force by regenerative braking in the vehicle, a motor for generating a braking force using an electronic parking brake of the vehicle, or a transmission for generating a braking force by engine braking of the vehicle in the vehicle, or any combination thereof, and thus generate a braking force lost by the oil pressure reduction amount for braking the vehicle.FIG. 9 is a drawing illustrating a computer system associated with a learning device, a quiet sound prediction device, a learning method, and a quiet sound prediction method according to an embodiment of the present disclosure.Referring to FIG. 8, a computer system 1000 may include, via the learning device, the quiet sound prediction device, the learning method, and the quiet sound prediction method, at least a processor 1100, a memory 1300, a user interface input device 1400, a user interface output device 1500, a storage 1600, and a network interface 1700, which are connected to each other via a bus 1200, wherein any combination or all of the components thereof may be present in a plurality or include multiple components thereof.The processor 1100 may be a central processing unit (CPU) or a semiconductor device that processes instructions stored in the memory 1300 and / or the storage 1600. The memory 1300 and the storage 1600 may include various types of volatile or nonvolatile storage media. For example, the memory 1300 may include a read only memory (ROM) 1310 and a random access memory (RAM) 1320.Accordingly, the operations of the method or the algorithm described in connection with the embodiments disclosed in the specification may be realized directly with a hardware module, a software module, or a combination of the hardware module and the software module executed by the processor 1100. The software module may reside on a storage medium (i.e., the random access memory 1300 and / or the storage 1600) such as a RAM, a flash memory, a ROM, an EPROM, an EEPROM, a register, a hard disk, a removable disk, and a CD-ROM.The example storage medium may be coupled to the processor 1100. The processor 1100 may read information from the storage medium and write information to the storage medium. Alternatively, the storage medium may also be integrated into the processor 1100. The processor and the storage medium may be incorporated in an application specific integrated circuit (ASIC). The ASIC may be housed within a user terminal. In another case, the processor and the storage medium may be housed as separate components in the user terminal.Although the present disclosure has been described above with reference to exemplary embodiments and the accompanying drawings, the present disclosure is not necessarily limited thereto, but can be variously modified and modified by a person skilled in the art to which the present disclosure pertains without departing from the spirit and scope of the present disclosure claimed in the following claims.The embodiments described above may be implemented with hardware components, software components, and / or a combination of hardware components and software components. For example, the apparatuses, methods, and components described in the embodiments may be realized using general-purpose or special-purpose computers, such as a processor, a controller, an arithmetic logic unit (ALU), a digital signal processor, a microcomputer, a field programmable array (FPGA), a programmable logic unit (PLU), a microprocessor, or any device capable of executing and responding instructions. A processing unit may perform an operating system (OS) or a software application running on the OS. In addition, in response to execution of software, the processing unit can access, store, manipulate, process, and generate data. It can be understood by a person skilled in the art that the processing unit can comprise a plurality of processing elements and / or a plurality of types of processing elements, even if a single processing unit can be represented for better understanding. For example, the processing unit may include a plurality of processors, or a processor and a controller. The processing unit may also have another configuration of processing, such as a parallel processor, together or distributed.Software may include computer programs, codes, instructions, or one or more combinations thereof, and may form a processing unit to operate in a desired manner, or may direct the processing unit independently or jointly. Software and / or data may be embodied permanently or temporarily in any type of machine, component, physical equipment, virtual equipment, computer storage media or devices, or transmitted signal waves to be interpreted by the processing unit or to provide instructions or data to the processing unit. Software may be distributed over computer systems interconnected via networks and may be stored or executed in distributed form. Software and data may be recorded on a computer readable storage medium.The methods according to embodiments may be realized in the form of program instructions executable over various computer systems and capable of being recorded on computer readable media. The computer readable media may include program instructions, data files, data structures, and the like, alone or in combination, and the program instructions recorded on the media may be specifically designed and embodied for example, or may be known and usable by one of ordinary skill in the art of computer software. Examples of computer readable media include magnetic media such as hard disks, floppy disks, and magnetic tapes; optical media such as compact disc read only memory (CD-ROM) disks and digital versatile disks (DVDs); magneto-optical media such as floppy disks; and hardware devices specifically configured to store and execute program instructions, such as read only memory (ROM), random access memory (RAM), flash memory, and the like. Program instructions include both machine code as generated by a compiler and higher level code executable by the computer using an interpreter.The above-described hardware devices may be configured to act as one or a plurality of software modules to perform the operations of the embodiments, or vice versa.Although the embodiments are described with reference to limited drawings, it will be understood by those skilled in the art that the embodiments may be variously changed or modified based on the above description. For example, suitable effects can be obtained even when the above processes and methods are performed in an order other than described above and / or the above components, such as systems, structures, devices, or circuits, are combined or coupled in forms and modes other than described above, or are replaced or replaced with other components or equivalents.Next, a description will be given of advantages of the learning device, the quiet sound prediction device, the learning method, and the quiet sound prediction method according to an embodiment of the present disclosure.According to at least one of embodiments of the present disclosure, the learning device may train a quiet sound prediction model based on training input data generated by preprocessing first raw data, second raw data, and third raw data, and thus reflect a relationship between a rapid change in the friction coefficient between the brake disk and the brake pad and a torque estimate in the quiet sound prediction model.According to at least one of embodiments of the present disclosure, the learning device may train the quiet sound prediction model based on a first loss function associated with regression or a second loss function associated with classification, and thus increase user satisfaction by using the quiet sound prediction model with better performance than if the quiet sound prediction model was trained using a loss function.According to at least one of embodiments of the present disclosure, the learning device may perform online learning of the quiet sound prediction model using the obtained data after training the quiet sound prediction model, and thus reduce quiet sounds that may be generated by the brake device that ages according to a use state of the user.Other implementations, other embodiments, and equivalents to the claims are within the scope of the following claims.Accordingly, the embodiments of the present disclosure are not intended to limit the technical spirit of the present disclosure, but are provided for illustration. The scope of the present disclosure may be construed based on the appended claims, and technical ideas within the scope corresponding to the claims may be included in the scope of the present disclosure.SIGNIFICANT FIGURE: FIG. 3300 Vehicle 310 Brake device 320 Wheel 330 Raw data set 340 Training input data 370 Regenerative braking motor 400 Training input data

Claims

A learning device comprising: one or more processors; and a storage medium storing computer readable instructions that, when executed by the one or more processors, enable the one or more processors to: preprocess one or a combination of first raw data extracted by a brake device of a vehicle, second raw data including wheel data associated with a wheel of the vehicle, and third raw data measured by an external sensor of the vehicle to obtain training input data at a target time when the brake device operates, apply the training input data and target data corresponding to the quiet sound generated by the training input data to a quiet sound prediction model to obtain temporary output data, and apply the quiet sound prediction model based on a first loss value, which is obtained by applying the temporary output data and the target data to a first loss function which is associated with a regression.The learning device of claim 1, wherein the instructions further enable the one or more processors to: generate the first raw data comprising at least one of or a combination of oil pressure data associated with oil pressure applied to the brake device, brake disk temperature data, and torque data associated with torque applied to a brake disk included in the brake device; generate, at the target time, the second raw data comprising at least one of or both of wheel speed data associated with a speed of the wheel and roll circumference data of a tire connected to the wheel; and generate, at the target time, the third raw data comprising at least one of or both of outside air temperature data from the external sensor and humidity data from the external sensor.The learning device of claim 2, wherein the instructions further enable the one or more processors to: generate first sub-input data comprising a first square of the brake disk temperature data and a second square of the wheel speed data; generate second sub-input data comprising a first change in the wheel speed data at the target time and at a subsequent time, the subsequent time following the target time, a second change in the brake disk temperature data at the target time and at the subsequent time, a third change in the oil pressure data at the target time and at the subsequent time, and a fourth change in the torque data at the target time and at the subsequent time; third sub-input data comprising first torque estimation data generated based on the first change in the wheel speed data and second torque estimation data generated based on the first torque estimation data and the brake disc temperature data; and connecting the first sub-input data, the second sub-input data, and the third sub-input data to obtain the training input data.The learning device of claim 1, wherein the instructions further enable the one or more processors to train the quiet sound prediction model based on a second loss value obtained by applying the temporary output data and the target data to a second loss function associated with the classification.The learning device of claim 4, wherein the instructions further enable the one or more processors to: determine a classification class of the training input data based on a comparison between the temporary output data and a predetermined threshold; and train the quietness prediction model based on a second loss value obtained by applying the classification class and the output data to the second loss function.A quiet sound prediction device, comprising: one or more processors; and a storage medium storing computer readable instructions that, when executed by the one or more processors, enable the one or more processors to: apply, to a trained quiet sound prediction model, one or a combination of brake data extracted from a brake device of a vehicle, wheel data corresponding to a wheel of the vehicle, and external sensor data measured by an external sensor of the vehicle to obtain an expected probability indicating a probability that a quiet sound of the brake device is generated based on the expected probability being greater than a predetermined threshold, determine that the quiet sound is generated in the brake device, and determining an oil pressure control mode of the vehicle based on one or a combination of a stability control mode of the vehicle, an outside air temperature measured by the external sensor, and a travel time of the vehicle based on the generation of the quiet sound in the brake device.The quiet noise predictor of claim 6, wherein the instructions further enable the one or more processors to determine the oil pressure control mode based on a first sub-condition for comparing the outside air temperature to a first threshold, based on a second sub-condition for comparing the travel time to a second threshold, and based on the stability control mode being disabled.The quiet sound prediction apparatus of claim 7, wherein the instructions further enable the one or more processors to: determine the oil pressure control mode as a first oil pressure control mode based on the first precondition being met and the second precondition not being met; determine the oil pressure control mode as a second oil pressure control mode based on the first precondition being met and the second precondition not being met; and determine the oil pressure control mode as a third oil pressure control mode based on the first precondition not being met and the second precondition not being met.The quietness prediction device of claim 7, wherein the instructions further enable the one or more processors to determine a torque compensation amount corresponding to the determined oil pressure control mode based on an oil pressure reduction amount corresponding to the determined oil pressure control mode, a coefficient of friction between a brake disc and a brake pad included in the brake device, and a piston surface of the brake disc included in the brake device.The quiet sound prediction apparatus of claim 9, wherein the instructions further enable the one or more processors to apply the torque compensation amount to one or a combination of a regenerative braking force motor for generating a regenerative braking force by regenerative braking in the vehicle, an electronic parking brake motor for generating a force of the electronic parking brake by an electronic parking brake of the vehicle, or a transmission for generating a transmission braking force by an engine brake of the vehicle to generate a compensation braking force lost by the oil pressure reduction amount for braking the vehicle.A learning method, comprising: preprocessing one or a combination of first raw data extracted from a brake device of a vehicle, second raw data including wheel data associated with a wheel of the vehicle, and third raw data measured by an external sensor of the vehicle to obtain training input data at a target time at which the brake device is in operation; applying the training input data and the target data corresponding to the quiet sound generated by the training input data to a quiet sound prediction model to obtain temporary output data; and training the quiet sound prediction model based on a first loss value obtained by applying the temporary output data and the target data to a first loss function associated with the regression.The learning method of claim 11, wherein obtaining the training input data comprises: generating the first raw data comprising one or a combination of oil pressure data associated with oil pressure applied to the brake device, brake disk temperature data, and torque data associated with torque applied to a brake disk included in the brake device; generating the second raw data comprising one or both of wheel speed data associated with a speed of the wheel and roll circumference data of a tire connected to the wheel at the target time; and generating the third raw data comprising one or both of outside air temperature data from the external sensor and humidity data from the external sensor at the target time.The learning method of claim 12, further comprising: generating first sub-input data comprising a first square of the brake disk temperature data and a second square of the wheel speed data; generating second sub-input data comprising a first change in the wheel speed data at the target time and at a subsequent time, the subsequent time following the target time, a second change in the brake disk temperature data at the target time and at the subsequent time, a third change in the oil pressure data at the target time and at the subsequent time, and a fourth change in the torque data at the target time and at the subsequent time; generating third sub-input data comprising first torque estimation data generated based on the first change in the wheel speed data and second torque estimation data generated based on the first torque estimation data and the brake disc temperature data; and connecting the first sub-input data, the second sub-input data, and the third sub-input data to obtain the training input data.The learning method of claim 11, wherein training the quiet sound prediction model comprises training the quiet sound prediction model based on a second loss value obtained by applying the temporary output data and the target data to a second loss function associated with the classification.The learning method of claim 14, wherein training the quiet sound prediction model further comprises: determining a classification class of the training input data based on a comparison between the temporary output data and a predetermined threshold; and training the quiet sound prediction model based on a second loss value obtained by applying the classification class and the output data to the second loss function.A quiet sound prediction method comprising: applying one or a combination of brake data extracted from a brake device of a vehicle, wheel data corresponding to a wheel of the vehicle, and external sensor data measured by an external sensor of the vehicle to a trained quiet sound prediction model to obtain an expected probability indicating a probability that a quiet sound of the brake device is generated; determining that the quiet sound is generated in the brake device based on the expected probability being greater than a predetermined threshold; and determining an oil pressure control mode of the vehicle based on one or a combination of a stability control mode of the vehicle, an outside air temperature measured by the external sensor, and a travel time of the vehicle based on the quietness being generated in the brake device.The quietness prediction method of claim 16, wherein the determining the oil pressure control mode of the vehicle comprises determining the oil pressure control mode based on a first sub-condition for comparing the outside air temperature to a first threshold, based on a second sub-condition for comparing the travel time to a second threshold, and based on the stability control mode being disabled.The quiet sound prediction method according to claim 17, wherein the determining the oil pressure control mode of the vehicle further comprises: determining the oil pressure control mode as a first oil pressure control mode based on the first sub-condition being satisfied and the second sub-condition being satisfied; determining the oil pressure control mode as a second oil pressure control mode based on the first sub-condition being satisfied and the second sub-condition being not satisfied; and determining the oil pressure control mode as a third oil pressure control mode based on the first sub-condition being not satisfied and the second sub-condition being not satisfied.The quietness prediction method according to claim 17, further comprising determining a torque compensation amount corresponding to the determined oil pressure control mode based on an oil pressure reduction amount corresponding to the determined oil pressure control mode, a friction coefficient between a brake disk and a brake pad included in the brake device, and a piston surface of the brake disk included in the brake device.The quiet sound prediction method according to claim 19, further comprising applying the torque compensation amount to one or a combination of a regenerative braking force motor for generating a regenerative braking force by regenerative braking in the vehicle, an electronic parking brake motor for generating a force of the electronic parking brake by an electronic parking brake of the vehicle, or a transmission for generating a transmission braking force by an engine brake of the vehicle in the vehicle to generate a compensation braking force lost by the oil pressure reduction amount for braking the vehicle.

Images