KR20250023119A - Device and method for detecting abnormality of motor, and non-transitory computer-readable storage medium storing program for performing the method - Google Patents

Abstract

A motor abnormality detection device and method, and a non-transitory computer-readable storage medium storing a program for performing the method are disclosed. The motor abnormality detection device according to one aspect of the present invention is a motor abnormality detection device that detects an abnormality in a motor that generates steering assist torque in an EPS (Electric Power Steering) system of a vehicle, the device including: a memory; and a processor; wherein the memory receives state data related to a state of the vehicle and operation data related to a steering operation of a driver of the vehicle and stores an artificial neural network model that estimates a physical quantity related to an output of the motor, and the processor inputs the state data and the operation data into the artificial neural network model and performs an operation to output an estimated value of the physical quantity, and compares the estimated value with an actual measured value of the physical quantity to detect whether the motor is abnormal.

Description

Device and method for detecting abnormality of motor, and non-transitory computer-readable storage medium storing program for performing the method

The present invention relates to a motor abnormality detection device and method, and a non-transitory computer-readable storage medium storing a program for performing the method. More specifically, the present invention relates to a motor abnormality detection device and method for detecting an abnormality in a motor that generates steering assist torque in an EPS (Electric Power Steering) system of a vehicle, and a non-transitory computer-readable storage medium storing a program for performing the method.

Electric Power Steering (EPS) system is a device that assists the steering of a vehicle using a motor. The EPS system analyzes sensing data related to the operation of the steering wheel, vehicle speed, etc., and controls the motor to provide appropriate steering assistance torque. Unlike conventional hydraulic power steering systems, the EPS system can be applied to electric vehicles, and is highly efficient in terms of space and fuel efficiency.

The EPS system has the advantages mentioned above. However, the failure of the EPS system during driving, that is, the Loss-of-Assist (LoA), has a significant impact on the steering performance of the vehicle and can lead to fatal accidents. The malfunction of the motor that generates the steering assistance torque is the most important factor in LoA. Therefore, in order to secure the safety of the vehicle, fail safety is required for the motor that generates the steering torque in the EPS system.

However, conventional approaches to fail-safety of the EPS motor focus on follow-up measures based on failures rather than preventing failures in advance. Meanwhile, some research and development is being conducted to detect EPS motor failures early, but they mainly take a model-based approach that simultaneously estimates the state of the system and the failure. This approach has limitations in that it cannot be extended to embedded applications in vehicles.

Korean Patent No. 1774255 "Electric Steering System", registered on 2017.08.29

The present invention is to solve the above problems, and an object of the present invention is to provide a motor abnormality detection device and method capable of predicting a functional decline of a motor before a failure of the motor of an EPS system occurs, and a non-transitory computer-readable storage medium storing a program for performing the method.

Another object of the present invention is to provide a motor abnormality detection device and method capable of predicting a motor output reduction in advance based on an artificial intelligence-based digital twin algorithm before LoA (Loss-of-Assist) or serious motor torque output reduction occurs in an EPS system, and a non-transitory computer-readable storage medium storing a program for performing the method.

Another object of the present invention is to provide a motor abnormality detection device and method capable of predicting a decrease in the output of an EPS system motor from signals obtained through a CAN (Controller Area Network) of a vehicle, and a non-transitory computer-readable storage medium storing a program for performing the method.

The tasks of the present invention are not limited to the tasks mentioned above, and other tasks not mentioned will be clearly understood by those skilled in the art to which the present invention belongs from the description below.

According to one aspect of the present invention, a motor abnormality detection device for detecting an abnormality in a motor generating steering assist torque in an EPS (Electric Power Steering) system of a vehicle is provided, comprising: a memory; and a processor; wherein the memory receives state data related to a state of the vehicle and operation data related to a steering operation of a driver of the vehicle and stores an artificial neural network model for estimating a physical quantity related to an output of the motor, and the processor inputs the state data and the operation data into the artificial neural network model and performs an operation to output an estimated value of the physical quantity, and compares the estimated value with an actual measured value of the physical quantity to detect whether the motor is abnormal.

In a motor abnormality detection device according to one aspect of the present invention, the status data and the operation data may be composed of signals that can be obtained through a CAN (Controller Area Network) of the vehicle.

In a motor abnormality detection device according to one aspect of the present invention, the physical quantity may be steering torque generated when the driver operates the steering wheel of the vehicle.

In a motor abnormality detection device according to one aspect of the present invention, the status data may include longitudinal velocity of the vehicle, lateral acceleration of the vehicle, yaw rate of the vehicle, and wheel velocity of each wheel of the vehicle.

In a motor abnormality detection device according to one aspect of the present invention, the operation data may include a steering angle and a steering angular velocity.

In a motor abnormality detection device according to one aspect of the present invention, the artificial neural network model may include a generative adversarial network (GAN).

In a motor abnormality detection device according to one aspect of the present invention, the artificial neural network model may include a generator which receives the state data and the operation data and generates the estimated value; and a discriminator which receives the state data, the operation data, and the measured value-related data including the measured value, and outputs a discriminant value for the measured value-related data.

In a motor abnormality detection device according to one aspect of the present invention, the generator may be composed of a multivariate transformer.

In a motor abnormality detection device according to one aspect of the present invention, the discriminator can additionally input estimated value-related data including the state data, the operation data, and the estimated value, and additionally output a discriminant value for the estimated value-related data.

In a motor abnormality detection device according to one aspect of the present invention, the artificial neural network model is constructed by having the generator and the discriminator alternately perform learning, and the state data, the operation data, and the actual measurement values used during the learning may be obtained when the vehicle and the motor are in a normal state.

In a motor abnormality detection device according to one aspect of the present invention, the estimated value can track the actual measured value of the physical quantity obtained when the vehicle and the motor are in a normal state.

In a motor abnormality detection device according to one aspect of the present invention, the memory further stores an abnormality detection model, and the processor can input error data related to the difference between the estimated value and the actual value into the abnormality detection model to determine whether the motor is abnormal.

In a motor abnormality detection device according to one aspect of the present invention, the abnormality detection model may include an OCSVM (One-Class Support Vector Machine) algorithm.

In a motor abnormality detection device according to one aspect of the present invention, there are a plurality of data sets including the state data, the operation data, the actual measurement value, and the estimated value, and the error data may include an average and a standard deviation of errors between the actual measurement value and the estimated value obtained from each of the plurality of data sets, a maximum absolute error between the actual measurement value and the estimated value of the plurality of data sets, and a discriminant value of the discriminator for data related to the actual measurement value of the plurality of data sets.

According to another aspect of the present invention, a motor abnormality detection method for detecting an abnormality in a motor that generates steering assist torque in an EPS (Electric Power Steering) system of a vehicle is provided, the method including: a step in which a processor receives state data related to a state of the vehicle, operation data related to a steering operation of a driver of the vehicle, and an actual value of a physical quantity related to an output of the motor; a step in which the processor inputs the state data and the operation data into an artificial neural network model and performs an operation to output an estimated value of the physical quantity; and a step in which the processor compares the estimated value with the actual value to detect whether the motor is abnormal.

In a motor abnormality detection method according to another aspect of the present invention, the status data and the operation data may be composed of signals that can be obtained through a CAN (Controller Area Network) of the vehicle.

In a motor abnormality detection method according to another aspect of the present invention, the physical quantity may be steering torque generated when the driver operates the steering wheel of the vehicle.

In a motor abnormality detection method according to another aspect of the present invention, the status data may include longitudinal velocity of the vehicle, lateral acceleration of the vehicle, yaw rate of the vehicle, and wheel velocity of each wheel of the vehicle.

In a motor abnormality detection method according to another aspect of the present invention, the operation data may include a steering angle and a steering angular velocity.

In a motor abnormality detection method according to another aspect of the present invention, the artificial neural network model may include a generative adversarial network (GAN).

In a motor abnormality detection method according to another aspect of the present invention, the artificial neural network model may include a generator that receives the state data and the operation data and generates the estimated value, and the step of outputting the estimated value of the physical quantity may include a step in which the processor inputs the state data and the operation data into the generator and outputs the estimated value generated by the generator.

In a motor abnormality detection method according to another aspect of the present invention, the artificial neural network model may further include a discriminator which receives the state data, the operation data, and the measured value-related data including the actual measurement value, and outputs a discriminator value for the measured value-related data, and the step of outputting the estimated value of the physical quantity may further include a step in which the processor inputs the measured value-related data including the state data, the operation data, and the measured value of the physical quantity into the discriminator, and outputs a discriminator value for the measured value-related data generated by the discriminator.

In a motor abnormality detection method according to another aspect of the present invention, the step of detecting whether the motor is abnormal may include a step in which the processor calculates error data related to a difference between the estimated value and the actual measured value; and a step in which the processor inputs the error data into an OCSVM (One-Class Support Vector Machine) algorithm to perform an operation.

In a motor abnormality detection method according to another aspect of the present invention, there are a plurality of data sets including the state data, the operation data, the estimated value, and the actual value, and the error data may include an average and a standard deviation of errors between the actual values and the estimated values obtained from each of the plurality of data sets, a maximum absolute error between the actual values and the estimated values of the plurality of data sets, and a discriminant value of the discriminator for data related to the actual values of the plurality of data sets.

In a motor abnormality detection method according to another aspect of the present invention, the artificial neural network model is constructed by having the generator and the discriminator alternately perform learning, and during the learning, the state data, the operation data, and the actual measurement values may be obtained when the vehicle and the motor are in a normal state.

According to another aspect of the present invention, a non-transitory computer-readable storage medium storing a program including at least one instruction for performing the motor abnormality detection method is provided.

According to the above configuration, a motor abnormality detection device and method according to one aspect of the present invention, and a non-transitory computer-readable storage medium storing a program for performing the method, can predict a functional decline of a motor before a failure of the motor of an EPS system occurs based on an artificial intelligence-based digital twin algorithm.

In addition, a motor abnormality detection device and method according to one aspect of the present invention, and a non-transitory computer-readable storage medium storing a program for performing the method, can predict a decrease in the output of a motor of an EPS system from signals obtainable through a CAN (Controller Area Network), thereby not requiring the placement of additional sensors in a vehicle, and providing economy and efficiency.

It should be understood that the effects of the present invention are not limited to the effects described above, but include all effects that can be inferred from the composition of the invention described in the detailed description or claims of the present invention.

FIG. 1 is a drawing showing the configuration of a motor abnormality detection device according to one embodiment of the present invention.
Figure 2 is a diagram showing the vehicle's EPS (Electric Power Steering) system and status data and operation data.
FIG. 3 is a diagram showing models stored in the memory of a motor abnormality detection device according to one embodiment of the present invention.
Figure 4 is a diagram showing the detailed configuration of the artificial neural network model shown in Figure 3.
Figure 5 is a diagram showing the detailed configuration of the generator of the artificial neural network model shown in Figure 4.
FIG. 6 is a diagram showing the operation of a motor abnormality detection device according to one embodiment of the present invention.
Figure 7 is a table showing the effectiveness of a motor abnormality detection device according to one embodiment of the present invention.
Figure 8 is a flowchart of a motor abnormality detection method according to one embodiment of the present invention.
FIG. 9 is a detailed flowchart of a step of outputting an estimated value of a physical quantity in a motor abnormality detection method according to one embodiment of the present invention.
FIG. 10 is a detailed flowchart of a step for detecting whether a motor is abnormal in a method for detecting motor abnormality according to one embodiment of the present invention.

Hereinafter, with reference to the attached drawings, embodiments of the present invention will be described in detail so that those with ordinary skill in the art can easily implement the present invention. The present invention may be implemented in various different forms and is not limited to the embodiments described herein. In order to clearly describe the present invention, parts that are not related to the description are omitted in the drawings, and the same reference numerals are assigned to the same or similar components throughout the specification.

The words and terms used in this specification and claims should not be construed as limited to their usual or dictionary meanings, but should be interpreted as having meanings and concepts consistent with the technical idea of the present invention, in accordance with the principles by which the inventor can define terms and concepts in order to best describe his or her invention.

In this specification, it should be understood that terms such as “include” or “have” are intended to describe the presence of a feature, number, step, operation, component, part or combination thereof described in the specification, but do not exclude in advance the possibility of the presence or addition of one or more other features, numbers, steps, operations, components, parts or combinations thereof.

FIG. 1 is a drawing showing the configuration of a motor abnormality detection device according to one embodiment of the present invention.

A motor abnormality detection device (1) according to one embodiment of the present invention can detect an abnormality in a motor that generates steering assistance torque in an EPS (Electric Power Steering) system of a vehicle. More specifically, a motor abnormality detection device (1) according to one embodiment of the present invention can predict a functional deterioration of a motor before a failure of the motor occurs through a digital twin algorithm for a vehicle based on artificial intelligence. Specifically, the digital twin algorithm can virtually construct an EPS system of a vehicle.

A motor abnormality detection device (1) according to one embodiment of the present invention detects a functional degradation of the motor before a LoA (Loss-of-Assist) of the EPS system of a vehicle occurs due to a failure of the motor. In particular, the motor abnormality detection device (1) according to one embodiment of the present invention can sensitively determine even a slight performance degradation (deterioration) of the motor.

Accordingly, according to one embodiment of the present invention, it is possible to predict in advance the occurrence of a motor failure in the EPS system. As a result, it is possible to take preventive measures before the occurrence of a motor failure, and the failure of the motor can be prevented.

Referring to FIG. 1, a motor abnormality detection device (1) according to one embodiment of the present invention may include a memory (100) and a processor (200).

The memory (100) can store data, models, etc. required for detecting abnormalities of the motor. For example, the memory (100) can include hardware devices configured to store and execute program instructions, such as ROM, RAM, flash memory, etc. In addition, the memory (100) can include magnetic media such as floppy disks and magnetic tapes, optical media such as CD-ROMs (Compact Disk Read Only Memory) and DVDs (Digital Video Disk), magneto-optical media such as floptical disks, etc.

The processor (200) can input the data into the model and perform calculations. For example, the processor (200) can be a hardware unit that performs calculations and control within a computer. The processor (200) can include at least one ALU (Arithmetic Logic Unit) and a register.

The memory (100) receives state data related to the state of the vehicle and operation data related to the steering operation of the driver of the vehicle and stores an artificial neural network model that estimates a physical quantity related to the output of the motor.

The processor (200) inputs the state data and the operation data into the artificial neural network model, performs an operation, outputs an estimated value of the physical quantity, and compares the estimated value with an actual measured value of the physical quantity to detect whether the motor is abnormal.

In relation to detecting whether the motor is abnormal, the memory (100) can further store an abnormality detection model. In addition, the processor (200) can input error data related to the difference between the estimated value and the actual measured value into the abnormality detection model to determine whether the motor is abnormal.

FIG. 2 is a diagram showing an EPS (Electric Power Steering) system of a vehicle and status data and operation data. Specifically, FIG. 2 exemplarily shows an EPS system (10) including a motor (18) that is a detection target of a motor abnormality detection device (1) according to one embodiment of the present invention, and status data (D _vehicle ) and operation data (D _steering ).

Referring to FIG. 2, an EPS system (10) of a vehicle may include a steering wheel (11) operated by a driver, a steering shaft (12) connected to the steering wheel (11) and transmitting steering force to the wheels of the vehicle, a steering angle sensor (13) disposed on the steering shaft (12) and measuring a steering angle and steering angular velocity of the steering wheel (11), a steering torque sensor (14) disposed on the steering shaft (12) and measuring a steering torque of the steering wheel (11), and a vehicle status measurement module (15) including one or more sensors that measure information related to the status of the vehicle.

The vehicle condition measurement module (15) can measure the longitudinal velocity (V _x ), lateral acceleration ( _ay ), yaw rate (θ' _z ), and wheel velocity (V _{wheel ) of each of the multiple wheels of the vehicle. Generally, since the vehicle has four wheels, the wheel velocity (V wheel )} can be four types of data. In other words, the vehicle condition measurement module (15) can measure a total of nine types of data.

The steering angle (θ _steering ) and steering angular velocity (θ' _steering ) measured by the steering angle sensor (13), the steering torque (T _steering ) measured by the steering torque sensor (14), the vehicle speed (V _x ), lateral acceleration ( _ay ), yaw rate (θ' _z ) measured by the vehicle state measurement module (15), and the wheel speed (V _wheel ) of each of the vehicle's multiple wheels are transmitted to the electronic control unit (ECU, 17) through the CAN (Controller Area Network) bus (16).

The electronic control unit (17) receives data transmitted through the CAN bus (16), performs calculations to generate a command for the motor (18) that generates steering assistance torque, and transmits the command to the motor (18). At this time, the command can be provided to the motor (18) in the form of current. The electronic control unit (17) can perform feedback control for the motor (18).

Torque input through the steering wheel (11) and torque generated by the motor (18) are transmitted to a steering mechanism (19) connected to the wheels of the vehicle. The direction of the wheels of the vehicle can be adjusted and steering of the vehicle can be performed through the steering mechanism (19) driven by the torque input through the steering wheel (11) and torque generated by the motor (18).

As described above, the memory (100) receives state data related to the state of the vehicle and operation data related to the steering operation of the driver of the vehicle, and stores an artificial neural network model that estimates a physical quantity related to the output of the motor. In addition, the processor (200) inputs the state data and the operation data into the artificial neural network model, performs a calculation, outputs an estimated value of the physical quantity, and compares the estimated value with an actual measured value of the physical quantity to detect whether the motor is abnormal. In relation to detecting whether the motor is abnormal, the memory (100) may additionally store an abnormality detection model.

In one embodiment of the present invention, the state data may include longitudinal velocity (V _x ), lateral acceleration (a _y ), yaw rate (θ' _z ) of the vehicle, and wheel speed (V _wheel ) of each of a plurality of wheels of the vehicle. When the vehicle has four wheels, the state data may include seven types of data.

Additionally, the manipulation data may include a steering angle (θ _steering ) and a steering angular velocity (θ' _steering ). In this way, the manipulation data may include two types of data.

In this way, in the motor abnormality detection device (1) according to one embodiment of the present invention, the status data and the operation data may be composed of signals that can be obtained through the CAN (Controller Area Network) of the vehicle.

Meanwhile, the above physical quantity may be the steering wheel torque (T _steering ) generated when the driver of the vehicle operates the steering wheel (11). In other words, in one embodiment of the present invention, the processor (200) outputs an estimated value of the steering wheel torque through calculation, and compares the estimated value with the actual value of the steering wheel torque to detect whether the motor (18) is abnormal. At this time, the actual value may be the steering torque (T _steering ) measured by the steering torque sensor (14).

Meanwhile, the Dugoff tire model can be utilized for estimation of steering wheel torque. Since the Dugoff tire model is a well-known model in the field of automobile engineering, a detailed description is omitted.

FIG. 3 is a diagram showing models stored in a memory of a motor abnormality detection device according to one embodiment of the present invention. Referring to FIG. 3, the memory (100) can store an artificial neural network model (110) and an abnormality detection model (120).

Below, the artificial neural network model (110) that receives the above-mentioned state data and the above-mentioned operation data and outputs an estimate of the steering wheel torque, the abnormality detection model (120) that compares the estimated value of the steering wheel torque with the actual measured value to detect an abnormality in the motor (18), and the operation of the processor (200) using the above-mentioned models will be examined in detail.

The artificial neural network model (110) receives the state data and the operation data and outputs an estimate of the steering torque. In one embodiment of the present invention, the artificial neural network model (110) may include a generative adversarial network (GAN).

The above generative adversarial network is constructed by having the generator and discriminator alternately perform learning, and the state data, the operation data, and the actual measurement values used during the learning may be obtained when the vehicle and motor (18) are in a normal state. Accordingly, the estimated value may follow the actual measurement values of the physical quantities obtained when the vehicle and motor (18) are in a normal state.

In other words, the artificial neural network model (110) can be a virtual twin model of the EPS system of the vehicle. More specifically, the artificial neural network model (1110) can be a deep learning-based neural twin model for the EPS system.

In one embodiment of the present invention, an artificial neural network model (110) receives nine types of data including the status of a steering wheel and a vehicle and estimates steering torque. At this time, the steering torque estimated by the artificial neural network model (110) can reflect normal operating conditions.

Figure 4 is a drawing showing a detailed configuration of the artificial neural network model shown in Figure 3. Referring to Figure 4, the artificial neural network model (110) may include a generator (111) and a discriminator (112).

The generator (111) receives the state data and the operation data and generates the estimated value. The generator (111) may include a neural network. In one embodiment of the present invention, the generator (111) serves as a virtual twin model of the EPS system of the vehicle.

For example, the generator (111) may be composed of a multivariate transformer. That is, the artificial neural network model (110) may be a GAN based on a multivariate transformer.

In one embodiment of the present invention, the generator (111) can receive a total of nine types of data, including longitudinal velocity (V _x ), lateral acceleration (a _y ), yaw rate (θ' _z ) of the vehicle, wheel speeds of each of a plurality of wheels of the vehicle (V _wheel ) (when the vehicle has four wheels, four are obtained), steering angle (θ _steering ), and steering angular velocity (θ' _steering ). In addition, the generator (111) calculates the input data and outputs an estimate of steering wheel torque.

Fig. 5 is a diagram showing a detailed configuration of a generator of an artificial neural network model shown in Fig. 4. Referring to Fig. 5, the generator (111) may include an encoder (111a) and a decoder (111b).

The encoder (111a) and the decoder (111b) constitute a network for estimating the steering torque. In other words, the generator (111) may be configured as an encoder (111a)-decoder (111b) structure network trained in an end-to-end manner for estimating the steering torque.

The encoder (111a) can be designed to extract high-level features from input data using a transformer block and linear operation. At this time, the transformer block can include an embedding layer, a self-attention layer, and a feed-forward layer.

First, in the transformer block, the embedding layer can encode spatiotemporal information into input sequences using a trainable embedding matrix.

Next, deep features can be extracted through a self-attention layer, and refinement can be performed using a feed-forward layer.

At this time, the self-attention layer can utilize the scaled dot-product attention mechanism. The self-attention layer can be the most important component in the transformer-based DNN (Deep Neural Network) architecture that enables the model to process and understand dependencies within the input sequence.

Scaled inner product attention first applies layer normalization to the input sequence, and the result can be processed into three unique vectors: query (Q), key (K), and value (V) through linear operation.

Next, the attention score (AS) for each position of the input sequence can be calculated based on the derived vectors. The calculation formula for the attention score (AS) can be given as Equation 1 below.

[Formula 1]

(σ(·) is the softmax operation, ∥x _input ∥ _dim is the feature dimension of the input, (·) means the inner product)

Attention scores quantify the relative importance of each position in an input sequence for a given query, i.e., more relevant positions are assigned higher scores.

The output of the transformer block can be fed to two different linear operations to compute the query, key vector and initial input value (?* ₁ ) of the decoder (111b).

The decoder (111b) can estimate the steering torque using the acquired values in an autoregressive manner similar to sequence-to-sequence recurrent neural networks.

Like the encoder (111a), the decoder (111b) may include a transformer block and linear operations. Additionally, the decoder (111b) may include a cross-attention layer within the transformer block.

The cross-attention layer can perform scaled inner product attention using the query and key vectors of the encoder (111a) to learn the dynamic correlation between the input signal and the estimated signal as mentioned above.

The discriminator (112) receives the state data, the operation data, and the actual measurement-related data including the actual measurement value, and outputs a discriminant value for the actual measurement-related data. In addition, the discriminator (112) can additionally receive the state data, the operation data, and the estimated value-related data including the estimated value, and additionally output a discriminant value for the estimated value-related data.

For example, the discriminator (112) may be configured as a discriminator network used for adversarial training of GAN. In other words, the generator (111) and the discriminator (112) may perform training in an adversarial training manner to improve estimation performance and set up a GAN model. More specifically, the discriminator (112) may perform optimization alternately with the generator (111) to solve the Wasserstein min-max problem as in Equation 2.

[Formula 2]

(xD,Real represents a real data set consisting of input data and measured values of steering torque, and xD,Fke represents a virtual data sample consisting of input data and estimated values of steering torque.)

In the artificial neural network model (110), the generator (111) is trained to deceive the discriminator (112) that distinguishes the spatiotemporal characteristics of xD,Real and xD,Fke. Accordingly, the generator (111) can generate an estimate of steering wheel torque that follows the actual measured value of steering torque in a vehicle EPS system in a normal state.

In this regard, the loss function for GAN training can be defined as in Equation 3 below.

[Formula 3]

(N is the batch size, L _G is the loss function of the generator, L _D is the loss function of the discriminator)

The first term of the loss of the generator (111) corresponds to the mean absolute error loss (L _MAE ) of supervised training. The remaining terms of the loss of the generator (111) together with the loss of the discriminator (112) constitute the GAN loss function (L _GAN ).

Meanwhile, the deterioration level of the motor can be predicted based on the estimated value of the physical quantity generated by the generator (111) and the actual measured value of the physical quantity.

In one embodiment of the present invention, the level of deterioration of the motor can be predicted through an abnormality detection model (120).

The abnormality detection model (120) can perform calculations by receiving error data related to the difference between the estimated value and the actual measurement value. Whether the motor (18) is abnormal can be determined through the abnormality detection model (120). In one embodiment of the present invention, the abnormality detection model (120) can include an OCSVM (One-Class Support Vector Machine) algorithm.

In addition, there may be a plurality of data sets including the state data, the operation data, the actual measurement value, and the estimated value, and the error data may include an average and a standard deviation of errors between the actual measurement value and the estimated value obtained from each of the plurality of data sets, a maximum absolute error between the actual measurement value and the estimated value of the plurality of data sets, and a discriminant value of a discriminator (112) for data related to the actual measurement value of the plurality of data sets.

In other words, a plurality of data sets including the above state data, the above operation data, the above measured values, and the above estimated values can be treated as one batch. For example, 100 to 150 data sets (as a specific example, 128 data sets) can be treated as one batch, and the above error data can be obtained for each batch.

The error data obtained for one batch can be input into an abnormality detection model (120). In addition, the level of abnormality (motor performance degradation) of the motor can be detected based on the output (feature) obtained from the abnormality detection model (120) that has received the error data.

Regarding the anomaly detection metric, the F1 score given in Equation 4 below can be considered.

[Formula 4]

(TP is true positive, FP is false positive, FN is false negative, positive indicates motor degradation)

Meanwhile, the discriminator (112) of the artificial neural network model (110) can produce not only a discriminant value for the actual measurement-related data of the data set but also a discriminant value for the estimated value-related data of the data set. The discriminant value of the discriminator (112) for the estimated value-related data of the data set can be provided as feedback to the generator (111) of the artificial neural network model (110) and utilized for training.

So far, the configuration of the motor abnormality detection device (1) according to one embodiment of the present invention has been described in detail. Hereinafter, the operation of the motor abnormality detection device (1) according to one embodiment of the present invention will be examined in detail.

Fig. 6 is a drawing showing the operation of a motor abnormality detection device according to one embodiment of the present invention. Referring to Fig. 6, a motor abnormality detection device (1) according to one embodiment of the present invention can operate as follows.

First, the memory (100) stores status data (D _vehicle ), operation data (D _steering ), and actual values of steering torque (T _steering ) obtained from the EPS system of the vehicle. As described above, the status data (D _vehicle ) may include longitudinal velocity (V _x ), lateral acceleration (a _y ), yaw rate (θ' _z ) of the vehicle, and wheel speeds (V _wheel ) of each of the plurality of wheels of the vehicle. In addition, the operation data (D _steering ) may include a steering angle (θ _steering ) and a steering angular velocity (θ' _steering ).

Meanwhile, the state data (D _vehicle ), the operation data (D _steering ), and the actual value of the steering torque (T _steering ) may be handled as a plurality of batches. In addition, in one embodiment of the present invention, the processing and calculation of data may be performed in units of batches of data.

Next, the processor (200) inputs state data (D _vehicle ) and operation data (D _steering ) into the artificial neural network model (110) and performs calculations to obtain an estimate of the steering wheel torque ( ) is printed.

Next, the processor (200) outputs the error data for each data batch. As described above, the error data may include the average and standard deviation of the errors between the actual values and the estimated values obtained from each of the plurality of data sets, the maximum absolute error between the actual values and the estimated values of the plurality of data sets, and the discriminant value of the discriminator (112) for the data related to the actual values of the plurality of data sets.

Finally, the processor (200) inputs the error data into the abnormality detection model (120) and outputs the result. As described above, the abnormality detection model (120) may include an OCSVM (One-Class Support Vector Machine) algorithm. Based on the output (feature) obtained from the abnormality detection model (120) that inputs the error data, the level of abnormality (motor performance degradation) of the motor can be detected.

Figure 7 is a table showing the effectiveness of a motor abnormality detection device according to one embodiment of the present invention.

FIG. 7 shows the F1 score (abnormality detection index) calculated for each technique and model after predicting the motor deterioration level using various techniques and models when the vehicle speed is 20 km/h and the motor deterioration levels of the EPS system of the vehicle are 10%, 25%, and 40%, respectively; when the vehicle speed is 30 km/h and the motor deterioration levels of the EPS system of the vehicle are 10%, 25%, and 40%, respectively; and when the vehicle speed is 40 km/h and the motor deterioration levels of the EPS system of the vehicle are 10%, 25%, and 40%, respectively.

In Fig. 7, the “feature-based” technique refers to the case where a neural twin structure is not used, and “Neural twin-based” refers to the case where a deep learning-based neural twin structure is used.

Here, the "feature-based" techniques include the empirical cumulative distribution-based outlier detection (ECOD) algorithm, the deep one-class classification (DOCC) model, the beta-variational autoencoder (β-VAE) model, and generative adversarial active learning (GAAL).

Additionally, the "Neural twin-based" technique includes a gated recurrent unit (GRU) model, an LSTM model, their attention-based versions (A-GRU, A-LSTM), a simple multivariate transformer model (MVT) without GAN training, and a combination of an artificial neural network model (110) and an anomaly detection model (120) stored in a motor anomaly detection device (1) according to one embodiment of the present invention (MVT-GAN).

Meanwhile, the technique and model shown in Fig. 7 are applied to data acquired from the CAN bus through an in-vehicle experiment. For example, the data can be acquired at a sampling rate of 100 Hz.

Looking at Fig. 7, it can be confirmed that the combination (MVT-GAN) of the artificial neural network model (110) and the abnormality detection model (120) stored in the motor abnormality detection device (1) according to one embodiment of the present invention shows the highest F1 score in most cases. Specifically, in the case of one embodiment of the present invention, it can be predicted with a performance of an F1 score of 0.85 or higher up to a 10% decrease compared to the maximum output of the motor.

In other words, it can be confirmed that the motor abnormality detection device (1) according to one embodiment of the present invention accurately predicts the level of deterioration of the motor before the motor in the EPS system of the vehicle completely fails. In particular, it can be seen that one embodiment of the present invention can accurately predict even a slight output decrease of the motor.

In this way, the present invention enables estimation of the future operation and remaining useful life (RUL) of the system, and predictive diagnosis suitable for predictive maintenance (PdM) applications. Accordingly, preventive measures and maintenance can be effectively induced before a motor failure of the EPS system of a vehicle occurs.

Above, a motor abnormality detection device according to one embodiment of the present invention has been described in detail. Below, a motor abnormality detection method according to one embodiment of the present invention will be described.

Figure 8 is a flowchart of a motor abnormality detection method according to one embodiment of the present invention.

A motor abnormality detection method (S100) according to one embodiment of the present invention is a motor abnormality detection method for detecting an abnormality in a motor that generates steering assistance torque in an EPS (Electric Power Steering) system of a vehicle, and can be performed by a motor abnormality detection device (1) according to one embodiment of the present invention.

Referring to FIG. 8, a motor abnormality detection method (S100) according to one embodiment of the present invention can be performed as follows.

First, the processor (200) receives status data related to the status of the vehicle, operation data related to the steering operation of the driver of the vehicle, and actual measurements of physical quantities related to the output of the motor (S110).

In one embodiment of the present invention, the state data may include longitudinal velocity (V _x ), lateral acceleration (a _y ), yaw rate (θ' _z ) of the vehicle, and wheel speed (V _wheel ) of each of the plurality of wheels of the vehicle. As described above, when the vehicle has four wheels, the state data may include seven types of data.

Additionally, the manipulation data may include a steering angle (θ _steering ) and a steering angular velocity (θ' _steering ). That is, the manipulation data may include two types of data.

The data listed as examples of the above status data and the above operation data are signals that can be obtained through the CAN (Controller Area Network) of the vehicle. In this way, if the above status data and the above operation data are composed of signals that can be obtained through the CAN (Controller Area Network) of the vehicle, there is no need to place a separate sensor in the vehicle for data collection, so the efficiency of data collection can be increased.

In addition, when data transmission to the outside of the vehicle is required, data transmission can also be performed efficiently. For example, when a motor abnormality detection device (1) according to one embodiment of the present invention is placed at a remote location outside the vehicle and a motor abnormality detection method according to one embodiment of the present invention is performed at the remote location, the status data and the operation data can be efficiently received through communication while the vehicle is driving.

Meanwhile, the above physical quantity may be the steering wheel torque (T _steering ) generated when the driver of the vehicle operates the steering wheel (11). In other words, in one embodiment of the present invention, the processor (200) outputs an estimated value of the steering wheel torque through calculation, and compares the estimated value with the actual value of the steering wheel torque to detect whether the motor (18) is abnormal. At this time, the actual value may be the steering torque (T _steering ) measured by the steering torque sensor (14).

At this time, a plurality of data sets including the state data, the operation data, the actual measurement value, and the estimated value may be treated as one batch. In other words, the processor (200) may receive a plurality of data sets, organize them into one batch in units of a certain number, and perform the operations below in units of batches.

Next, the processor (200) inputs the state data and the operation data into an artificial neural network model and performs a calculation to output an estimate of the physical quantity (S120). As described above, the artificial neural network model (110) may include a generative adversarial network (GAN).

The artificial neural network model (110) receives the state data and the operation data as input and outputs an estimate of the steering torque. In one embodiment of the present invention, the artificial neural network model (110) may include a generative adversarial network (GAN). In addition, the artificial neural network model (110) may include a generator (111) and a discriminator (112).

FIG. 9 is a detailed flowchart of a step of outputting an estimated value of a physical quantity in a motor abnormality detection method according to one embodiment of the present invention.

Referring to Fig. 9, the step of outputting an estimate of the physical quantity can be performed as follows.

First, the processor (200) inputs the state data and the operation data to the generator (111), and outputs the estimated value generated by the generator (111) (S121). Here, the generator (111) receives the state data and the operation data and generates the estimated value (S121).

For example, the generator (111) may be composed of a multivariate transformer. In other words, the artificial neural network model (110) may be a GAN based on a multivariate transformer. In one embodiment of the present invention, the generator (111) serves as a virtual twin model of the EPS system of the vehicle.

The generator (111) can receive a total of nine types of data, including longitudinal velocity (V _x ), lateral acceleration (a _y ), yaw rate (θ' _z ) of the vehicle, wheel speeds of each of the multiple wheels of the vehicle (V _wheel ) (when the vehicle has four wheels, four are obtained), steering angle (θ _steering ), and steering angular velocity (θ' _steering ). In addition, the generator (111) calculates the input data and outputs an estimate of steering wheel torque.

Next, the processor (200) inputs the measured value-related data including the state data, the operation data, and the measured value of the physical quantity into the discriminator (112), and outputs a discriminator value for the measured value-related data generated by the discriminator (112) (S122).

Meanwhile, the processor (200) inputs the estimated value-related data including the state data, the operation data, and the estimated value of the physical quantity into the discriminator (112), and outputs a discriminator value for the estimated value-related data generated by the discriminator (112) (S123).

In the step (S122) of outputting a discriminant value for data related to actual measurements and the step (S123) of outputting a discriminant value for data related to estimated values, the discriminator (112) receives data related to actual measurements including the state data, the operation data, and the actual measurements, and outputs a discriminant value for the data related to actual measurements. The discriminant value for data related to actual measurements generated by the discriminator (112) can constitute error data input to the anomaly detection model (120), as will be described later.

In addition, the discriminator (112) can additionally receive the state data, the operation data, and the estimate-related data including the estimate, and output a discriminant value for the estimate-related data. The discriminant value for the estimate-related data generated by the discriminator (112) can be utilized for training the generator (111) of the artificial neural network model (110).

The step (S122) of outputting the discriminant value for the measured data and the step (S123) of outputting the discriminant value for the estimated data may be performed simultaneously. In addition, either the step (S122) of outputting the discriminant value for the measured data or the step (S123) of outputting the discriminant value for the estimated data may be performed first, and the other may be performed later.

In a motor abnormality detection method according to one embodiment of the present invention, the artificial neural network model (110) may include a generative adversarial neural network. At this time, the generative adversarial neural network is constructed by having a generator (111) and a discriminator (112) alternately perform learning, and during the learning, the state data, the operation data, and the actual measurement values may be obtained when the vehicle and the motor are in a normal state.

In other words, the artificial neural network model (110) can be a virtual twin model of the EPS system of the vehicle. More specifically, the artificial neural network model (1110) can be a deep learning-based neural twin model for the EPS system.

This artificial neural network model (110) receives nine types of data including the status of the steering wheel and the vehicle and estimates the steering torque. At this time, the steering torque estimated by the artificial neural network model (110) can reflect normal operating conditions.

Finally, after the step (S120) of outputting an estimate of the above physical quantity, the processor (200) compares the estimate with the actual measured value to detect whether the motor is abnormal (S130).

Figure 10 is a detailed flowchart of a step for detecting whether a motor is abnormal in a motor abnormality detection method according to one embodiment of the present invention.

Referring to Fig. 10, the step of detecting whether the motor is abnormal can be performed as follows.

First, the processor (200) calculates error data related to the difference between the estimated value and the actual value (S131).

As described above, there are a plurality of data sets including the state data, the operation data, the actual values, and the estimated values, and the error data may include the average and standard deviation of errors between the actual values and the estimated values obtained from each of the plurality of data sets, the maximum absolute error between the actual values and the estimated values of the plurality of data sets, and the discriminant value of the discriminator (112) for the actual value-related data of the plurality of data sets.

A plurality of data sets including the above state data, the above operation data, the above measured values, and the above estimated values can be treated as one batch. For example, 100 to 150 data sets (as a specific example, 128 data sets) can be treated as one batch, and the above error data can be obtained for each batch.

Next, the processor (200) inputs the error data into the anomaly detection model (120) and performs a calculation (S132). At this time, the anomaly detection model (120) may include an OCSVM (One-Class Support Vector Machine) algorithm. In other words, the processor (200) may input the error data into the OCSVM (One-Class Support Vector Machine) algorithm and perform a calculation.

In more detail, the error data obtained for one batch can be input into an abnormality detection model (120). In addition, the level of abnormality (motor performance degradation) of the motor can be detected based on the output (feature) obtained from the abnormality detection model (120) that has received the error data.

Meanwhile, the present invention additionally provides a non-transitory computer-readable storage medium storing a program for performing an operation method of a motor abnormality detection method. Specifically, the present invention can provide a non-transitory computer-readable storage medium storing a program including at least one instruction for performing an operation method of a motor abnormality detection method according to an embodiment of the present invention.

At this time, instructions may include not only machine language code generated by a compiler, but also high-level language code executable by a computer.

The above recording medium may include a hardware device configured to store and execute program instructions, such as a magnetic media such as a hard disk, a floppy disk, and a magnetic tape, an optical media such as a compact disk read only memory (CD-ROM), a digital video disk (DVD), a magneto-optical media such as a floptical disk, a ROM, a RAM, a flash memory, etc.

Although the embodiments of the present invention have been described, the spirit of the present invention is not limited to the embodiments presented in this specification, and those skilled in the art who understand the spirit of the present invention will be able to easily propose other embodiments by adding, changing, deleting, or adding components within the scope of the same spirit. However, this will also be considered to fall within the spirit of the present invention.

1: Motor abnormality detection device
100: Memory
200: Processor

Claims (26)

A motor abnormality detection device that detects an abnormality in a motor that generates steering assistance torque in a vehicle's EPS (Electric Power Steering) system.
memory; and
a processor; including;
The above memory receives state data related to the state of the vehicle and operation data related to the steering operation of the driver of the vehicle and stores an artificial neural network model that estimates physical quantities related to the output of the motor.
A motor abnormality detection device in which the processor inputs the state data and the operation data into the artificial neural network model, performs calculations to output an estimate of the physical quantity, and compares the estimate with an actual measured value of the physical quantity to detect whether the motor is abnormal. In paragraph 1,
A motor abnormality detection device in which the above status data and the above operation data are composed of signals that can be obtained through the CAN (Controller Area Network) of the vehicle. In paragraph 1,
The above physical quantity is a motor abnormality detection device which is a steering torque generated when the driver operates the steering wheel of the vehicle. In paragraph 1,
A motor abnormality detection device wherein the status data includes longitudinal velocity of the vehicle, lateral acceleration of the vehicle, yaw rate of the vehicle, and wheel velocity of each wheel of the vehicle. In paragraph 1,
The above operation data is a motor abnormality detection device including steering angle and steering angular velocity. In paragraph 1,
The above artificial neural network model is a motor abnormality detection device including a generative adversarial network (GAN). In paragraph 6,
The above artificial neural network model is,
A generator for receiving the above state data and the above operation data and generating the above estimate; and
A motor abnormality detection device including a discriminator that inputs data related to actual measurements including the above-mentioned status data, the above-mentioned operation data, and the above-mentioned actual measurements, and outputs a discriminator value for the data related to the actual measurements. In paragraph 7,
The above generator is a motor abnormality detection device consisting of a multivariate transformer. In paragraph 7,
A motor abnormality detection device in which the above-mentioned discriminator additionally receives the state data, the operation data, and the estimated value-related data including the estimated value, and additionally outputs a discriminant value for the estimated value-related data. In Article 9,
The above artificial neural network model is constructed by having the generator and the discriminator alternately perform learning.
A motor abnormality detection device wherein the state data, the operation data and the actual measurement values used during the above learning are obtained when the vehicle and the motor are in a normal state. In Article 10,
The above-mentioned estimated value is a motor abnormality detection device that follows the actual measured value of the above-mentioned physical quantity obtained when the above-mentioned vehicle and the above-mentioned motor are in a normal state. In paragraph 7,
The above memory further stores the anomaly detection model,
The above processor is a motor abnormality detection device that inputs error data related to the difference between the estimated value and the actual measured value into the abnormality detection model to determine whether the motor is abnormal. In Article 12,
The above abnormality detection model is a motor abnormality detection device including an OCSVM (One-Class Support Vector Machine) algorithm. In Article 12,
There are multiple data sets including the above state data, the above operation data, the above measured value and the above estimated value,
A motor abnormality detection device, wherein the above error data includes an average and standard deviation of errors between actual measurements and estimates obtained from each of the plurality of data sets, a maximum absolute error between actual measurements and estimates of the plurality of data sets, and a discriminant value of the discriminator for data related to actual measurements of the plurality of data sets. A motor abnormality detection method for detecting an abnormality in a motor that generates steering assistance torque in an EPS (Electric Power Steering) system of a vehicle,
A step in which a processor receives state data related to the state of the vehicle, operation data related to the steering operation of a driver of the vehicle, and actual measurements of physical quantities related to the output of the motor;
A step in which the processor inputs the state data and the operation data into an artificial neural network model and performs a calculation to output an estimate of the physical quantity; and
A method for detecting motor abnormalities, comprising: a step in which the processor compares the estimated value with the measured value to detect whether the motor is abnormal; In Article 15,
A motor abnormality detection method wherein the above status data and the above operation data are composed of signals that can be obtained through the CAN (Controller Area Network) of the vehicle. In Article 15,
A method for detecting motor abnormality, wherein the above physical quantity is a steering torque generated when the driver operates the steering wheel of the vehicle. In Article 15,
A motor abnormality detection method wherein the status data includes longitudinal velocity of the vehicle, lateral acceleration of the vehicle, yaw rate of the vehicle, and wheel velocity of each wheel of the vehicle. In Article 15,
A motor abnormality detection method wherein the above manipulation data includes steering angle and steering angular velocity. In Article 15,
The above artificial neural network model is a method for detecting motor abnormalities including a generative adversarial network (GAN). In Article 20,
The artificial neural network model includes a generator that receives the state data and the operation data as input and generates the estimate,
A method for detecting motor abnormalities, wherein the step of outputting an estimate of the above physical quantity includes a step of the processor inputting the state data and the operation data into the generator and outputting an estimate generated by the generator. In Article 21,
The artificial neural network model further includes a discriminator that receives data related to actual measurements including the state data, the operation data, and the actual measurements, and outputs a discriminant value for the data related to the actual measurements.
A method for detecting motor abnormalities, wherein the step of outputting an estimate of the physical quantity further includes a step of the processor inputting actual measurement-related data including the state data, the operation data, and actual measurements of the physical quantity into the discriminator, and outputting a discriminator value for the actual measurement-related data generated by the discriminator. In paragraph 22,
The step of detecting whether the above motor is abnormal is as follows:
a step in which the processor produces error data related to the difference between the estimated value and the measured value; and
A method for detecting motor abnormalities, comprising: a step in which the processor inputs the error data into an OCSVM (One-Class Support Vector Machine) algorithm to perform a calculation. In paragraph 23,
There are multiple data sets including the above state data, the above operation data, the above estimated value and the above measured value,
A method for detecting motor abnormalities, wherein the above error data includes an average and standard deviation of errors between actual measurements and estimates obtained from each of the plurality of data sets, a maximum absolute error between actual measurements and estimates of the plurality of data sets, and a discriminant value of the discriminator for data related to actual measurements of the plurality of data sets. In paragraph 22,
The above artificial neural network model is constructed by having the generator and the discriminator alternately perform learning.
A motor abnormality detection method wherein the state data, the operation data and the actual measurement values during the above learning are obtained when the vehicle and the motor are in a normal state. A non-transitory computer-readable storage medium having stored thereon a program comprising at least one instruction for performing a motor abnormality detection method according to any one of claims 15 to 25.

Images