KR102160140B1 - Self-diagnosis System for Autonomous Vehicle based Deep Learning - Google Patents

Abstract

The embodiment relates to a self-diagnosis method for an autonomous vehicle based on deep learning. Specifically, this self-diagnosis method sets and registers a vehicle part diagnosis format as back-propagation by a hidden layer determined as a dynamic number according to the number of input sensor information and the number of output parts according to the association between vehicle parts. Therefore, by learning the sensor information of the vehicle according to the vehicle parts diagnosis format, the risk of each of the vehicle parts and consumables and the risk according to the dependency relationship of other parts due to the defective part is inferred. Therefore, by calculating the correlation between the parts of the vehicle, the state of the current parts is clearly diagnosed and transmitted to the user. And, according to the state of these parts, the overall condition or failure of the vehicle is accurately and quickly diagnosed.

Description

Integrated diagnosis system for autonomous vehicles based on deep learning {Self-diagnosis System for Autonomous Vehicle based Deep Learning}

The content disclosed in the present specification relates to a self-diagnosis method for an autonomous vehicle that runs by itself even if the driver does not manipulate the vehicle.

Unless otherwise indicated herein, the content described in this section is not prior art to the claims of this application, and inclusion in this section is not admitted to be prior art.

In general, an autonomous vehicle is a vehicle that recognizes the environment and is capable of autonomous driving and working for an autonomous tractor.

However, in the light of recent technologies, autonomous vehicles are not a distant future. As the QoS of V2X increases through various communication methods such as WAVE and LTE, the vehicle is informed of the current traffic conditions around itself in real time, and accordingly, the user is warned or the vehicle solves the problem by itself.

As vehicle driving is automated, it is difficult for vehicle drivers to pay close attention to the state of the vehicle. This is because when the vehicle runs autonomously, the driver himself can no longer feel the state of the vehicle that he feels while operating the vehicle. Therefore, the vehicle must accurately and quickly inform the driver of not only the surrounding danger situation, but also the danger inside the vehicle.

Prior literature that is the technology of this background is the following patent literature.

(Patent Document 1) KR10-2000-0054385 A

(Patent Document 2) KR10-1117168 Y1

However, most of the tools for performing diagnosis of the current vehicle, including these patent documents, are installed on the OBD and diagnose the overall condition of the vehicle using the pin number of the OBD. However, no diagnostic tool has been developed that clearly informs the user of the condition of the part by calculating the overall condition of the vehicle or the relationship between the faulty parts.

The disclosed content provides a self-diagnosis method for autonomous vehicles based on deep learning to support rapid communication of sensor information, accurately and quickly diagnose vehicle status, and notify the result of diagnosis to the cloud and surrounding vehicles. I want to.

A self-diagnosis method for an autonomous vehicle based on deep learning according to an embodiment,

According to the association between vehicle parts, the vehicle parts diagnosis format is set as back-propagation by a hidden layer determined as the number of input sensor information and the dynamic number according to the number of output parts. So, according to this vehicle parts diagnosis format, it learns the sensor information of the vehicle in real time and infers the risk of each of the vehicle parts and consumables and the risk according to the dependency relationship of other parts due to the defective parts. To do.

According to the embodiments, the relationship between the parts of the vehicle is calculated, the state of the current parts is clearly diagnosed and transmitted to the user.

And, according to the state of these parts, the overall condition or failure of the vehicle is accurately and quickly diagnosed.

In addition, when such an accident or failure is detected, a vehicle with a high risk of an accident is quickly detected by nearby vehicles. And, furthermore, it prevents continuous accidents by notifying the surrounding vehicles and infrastructure of danger.

In addition, it supports communication of all major protocols in the vehicle and guarantees fast speed. Accordingly, further, real-time communication between vehicle networks is performed in an integrated manner. And, accordingly, the transmission delay is minimized.

1 is a diagram showing the configuration of a system to which a self-diagnosis method for an autonomous vehicle based on deep learning is applied according to an embodiment
2 is a diagram conceptually showing the configuration of a self-diagnosis module according to the vehicle part diagnosis format
3 is a flow chart sequentially showing a self-diagnosis method for an autonomous vehicle based on deep learning according to an embodiment;

1 is a diagram showing the configuration of a system to which a self-diagnosis method for an autonomous vehicle based on deep learning is applied according to an embodiment.

As shown in FIG. 1, the system of an embodiment includes an OBD 100 equipped with a self-diagnosis module for self-diagnosing vehicle parts through vehicle sensors and sensor information, and an example connected to the OBD 100 and IoT. For example, it includes an external information processing device such as a vehicle repair shop.

Additionally, the system of one embodiment includes a self-diagnostic integrated lightweight edge gateway that supports communication of all major protocols in the vehicle within its OBD 100, ensures fast speeds, and provides real-time communication between vehicle networks.

In this case, the sensor of the vehicle provides sensing information for self-diagnosis of the OBD 100. Thus, the OBD 100 performs self-diagnosis of, for example, various parts of a vehicle based on the sensing information, and transmits the result to an external information processing device such as a vehicle repair shop through IoT. Here, the sensor of the vehicle is an ABS sensor, an airbag sensor, etc. through a FlexRay bus, a temperature sensor, a steering wheel sensor, an instrument sensor, etc. through a CAN bus, and an image sensor through a MOST ring.

First, the OBD 100 sets and registers a vehicle part diagnosis format as back-propagation by a hidden layer determined as a dynamic number according to the number of input sensor information and the number of output parts according to the association between vehicle parts. Through this, in this state, so this OBD 100 receives and registers the sensor information of the vehicle, learns according to the vehicle part diagnosis format, and learns the risk of the vehicle parts and consumables and other parts due to the defective parts. Infer the risk according to the subordinate relationship of So, by calculating the relationship between the parts of the vehicle, the condition of the current parts is accurately and quickly diagnosed. Hereinafter, an operation of inferring the degree of risk for parts and consumables of a vehicle according to the vehicle parts diagnosis format will be described in detail.

In this case, the vehicle component diagnosis format made according to an equation according to an embodiment is applied to the operation of inferring the risk of the vehicle components and consumables.

Specifically, the vehicle parts diagnosis format is according to the following [Equation 1].

[Equation 1]

NET _h1j = ∑x _i W _ij (i = from 0 to n), h _1j = max(0, NET _h1j )

NET _yk = ∑h _mi Z _ik (from i = 1 to l), y _k = 1/1+e ^{-NET yk}

Here, x _i is the value of the i-th input node of the vehicle's sensor information, W _ij is the connection strength between the i-th input node and the j-th hidden node, and h _mi is the hidden layer in the order determined according to the number of hidden layers. The value of the layer, Z _ik is the strength of the connection between the ith input node and the kth hidden layer, y _k is the risk of vehicle parts and consumables.

In this case, the sensor information of the vehicle is speed, starting voltage, mileage, tire pressure, and the like. And, the hidden layer is a modeling that infers the risk of parts according to the association between parts by inputting sensor information of the vehicle. For example, the association between these parts is determined based on the speed, driving, etc. that are important in the vehicle, and specifically, as an example, is made to be highly related in the order of engine, tire, headlight, and generator. Such information is determined in correspondence with the dependency relationship of other parts by, for example, defective parts, and the dependency relationship is established. In addition, it is collected from real-time sensor information of the vehicle or batch sensor information of a vehicle service center registered in advance to receive the input of the sensor information, and the operation of risk inference is operated based on the cloud.

Therefore, when inferring the risk of such a vehicle component and the risk of other components due to the defective component, the influence of the components is calculated according to the relationship between the vehicle components according to [Equation 1] to infer the risk.

Next, the configuration of the self-diagnosis module according to the vehicle part diagnosis format applied to the OBD of the system according to the embodiment of FIG. 1 will be described in detail.

2 is a diagram conceptually showing a configuration of a self-diagnosis module according to the vehicle parts diagnosis format.

As shown in FIG. 2, the self-diagnosis module according to the vehicle parts diagnosis format according to an embodiment is a self-diagnosis module for an autonomous vehicle according to an embodiment, which operates in a cloud environment and is composed of two sub-modules. Hereinafter, such a module will be described in detail.

The first module is a vehicle component diagnosis sub-module according to a vehicle component diagnosis format according to an embodiment. Hereinafter, such a vehicle parts diagnosis sub-module is abbreviated as VPDS. This VPDS diagnoses the condition of each part by considering the association of each part. In addition, the input/output of the VPDS, the number of hidden layers, and a modeling method in order to perform such VPDS diagnosis are described.

First, the VPDS uses information such as speed, starting voltage, mileage, tire pressure, RPM, coolant, engine oil, and rear sensors that can be collected from the vehicle's sensors, and considers the relationship between each part. Diagnose the condition of

In this case, the VPDS uses the information received from the vehicle as input. The information collected from the in-vehicle sensor is digital numerical information and indicates the status of each part. The input of VPDS is defined by combining the set of sensors, the information of each sensor, and the number of sensors. And, the output value of VPDS indicates the state of each part. Each component is diagnosed with three output nodes, for example. The output of this VPDS includes a set of output nodes, an output value of each node, and three times the number of parts to be diagnosed. The three nodes corresponding to each part represent "normal", "warning" and "danger" respectively.

In addition, the hidden layer of this VPDS uses back-propagation. In this case, the hidden layer dynamically determines the number of hidden layers according to the number of input sensor information and the number of output components. For example, the number of these hidden layers is fixed at 5. The number of nodes for each layer of the hidden layer is set equal to the number of input information. So, it allows you to calculate the influence of parts on each other through hidden layers.

In addition, the mathematical design for modeling the VPDS and its structure are made by [Equation 1] described with reference to FIG. 1.

And, the second module is a sub-module according to the overall vehicle condition diagnosis format. Hereinafter, such a vehicle overall condition diagnosis sub-module is abbreviated as TDS. The TDS uses the output of the above-described VPDS as input information to diagnose the state of the entire vehicle and notify the user. The mathematical design of this TDS and its structure are made according to the following [Equation 2]. So, it demonstrates with reference to the [expression 2].

Thus, the deep learning model according to an embodiment learns through such a training data set, and when the learning is completed, self-diagnosis of the vehicle is performed. The cloud is designed and configured to transmit the current sensor information and the difference between the normal value of the sensor information and the result of self-diagnosis to the vehicle manager.

Meanwhile, the OBD of the system of FIG. 1 according to an embodiment diagnoses the overall state or failure of the vehicle according to an embodiment.

In this case, the OBD is the overall condition or failure of the vehicle according to the risk of each of the vehicle parts and consumables according to the relationship between the vehicle parts according to an embodiment and the risk according to the dependency relationship of other parts due to the defective parts. Make a diagnosis of the back.

Specifically, first, according to the following [Equation 2], a format for diagnosing the overall condition of the vehicle using a preset step function according to the association between vehicle components and the risk of vehicle components is set and registered.

[Equation 2]

NET _n = ∑x _i W _in (i = 0 to g), f(NET _n )(0(NET <0) 1(NET≥0)

Here, x _i is the value of the i-th input node of the vehicle's part information, W _in is the connection strength between the i-th input node and the n-th input node, and g is the number of combinations of the vehicle's parts using input information. A fixed value

Additionally, the overall vehicle condition diagnosis format is a simple neural network model that diagnoses the overall condition of the vehicle by using the diagnosis result for each autonomous vehicle component calculated by the vehicle component diagnosis format as an input. The whole vehicle condition diagnosis format is a simple structured neural network that does not have a hidden layer and uses a unipolar step function as an activation function. The input information is the output value of the vehicle part diagnosis format, and the output information is classified into'normal','check', and'danger', respectively. So, accordingly, it is set as 3 nodes. In this case, the'normal' means that the vehicle has no problem at all in its full condition. And, the'check' means that an abnormality has occurred in some parts that have a low impact on the accident. In addition,'danger' means that an abnormality has occurred in some parts that have a high impact on the accident.

So, after inferring the degree of risk of the vehicle part, etc., the overall condition of the vehicle is learned by learning the risk of the vehicle parts and consumables in real time and the degree of risk according to the dependency relationship of other parts due to the defective part according to the overall condition diagnosis format. Diagnose

For example, by integrating the results of each analysis of engine, tire, headlight and generator, the overall condition diagnosis format of the vehicle determines the overall condition. Specifically, in the case of engine-normal_130kg/c, tire-check_28_PSI, headlight-danger_92,500cd, and generator-check_12v, the overall condition of the vehicle is diagnosed by'checking'.

Thus, accordingly, the overall condition or failure of the vehicle is accurately and quickly diagnosed. This clearly informs the user.

In addition, in this case, the OBD of the system of FIG. 1 according to an embodiment enables surrounding vehicles to quickly detect a vehicle with high risk of an accident according to the diagnosis result when the vehicle is diagnosed.

To this end, the OBD according to this embodiment notifies the cloud and surrounding vehicles of the diagnosis result by performing V2X communication based on edge computing.

Specifically, the vehicle parts diagnosis format model is created by modeling the vehicle parts diagnosis format by the association between the vehicle parts according to the risk of each of the parts and consumables of the vehicle and the risk according to the dependency relationship of other parts by the defective parts. .

In addition, the overall vehicle condition diagnosis format is modeled by modeling the overall condition diagnosis format by modeling the relationship between the vehicle parts according to the degree of risk according to the degree of risk of the parts and consumables of the vehicle and the dependence of other parts due to the defective parts.

Next, the vehicle component diagnostic format model and the vehicle overall state diagnostic format model are transmitted to a main cloud server that provides a vehicle component diagnostic format model and a vehicle overall state diagnostic format model corresponding to each vehicle device. For example, the vehicle device is differentiated according to the type and number of sensors in the vehicle, and the like. In this case, the main cloud server continuously updates the vehicle parts diagnosis format model and the entire vehicle condition diagnosis format model and transmits it to the vehicle, so that the learning accuracy of the vehicle parts diagnosis format and the overall vehicle condition diagnosis format according to an embodiment Improves. In addition, accordingly, the vehicle improves the self-diagnosis speed by using the vehicle parts diagnosis format model and the overall vehicle condition diagnosis format model transmitted from the main cloud server.

In this state, therefore, from the main cloud server, in response to the vehicle device, a self-contained vehicle part diagnosis format model and an overall vehicle condition diagnosis format model are provided.

Then, by first learning the sensor information of the vehicle in real time by the vehicle parts diagnosis format model provided in this way, the risk of each of the vehicle parts and consumables and the risk according to the dependency relationship of other parts due to the defective part is inferred.

Next, the risk of each of the parts and consumables of the vehicle inferred by the vehicle overall condition diagnosis format model and the degree of risk according to the dependency of other parts due to the defective parts are learned.

So, if the risk according to this learning is the risk that has been diagnosed as dangerous, a risk warning message is generated and responded to, and transmitted to adjacent vehicles within a preset communication radius. In this case, the danger warning message is generated by the V2X-based accident notification service format, and when the danger warning message is generated, the hop field of the danger warning message is set to 2. And, when the danger warning message is transmitted to an adjacent vehicle, the hop field of the danger warning message is set to 1.

Therefore, accordingly, a vehicle with a high risk of an accident is quickly detected by the surrounding vehicle.

Through this, when an accident or breakdown is detected, the danger is notified to nearby vehicles and infrastructure to prevent continuous accidents.

In more detail, in this case, additionally, in this case, the surrounding vehicle that has received the danger warning message transmits the message to another adjacent vehicle through V2V communication.

Then, the other adjacent vehicle knows the state of the original vehicle by receiving the danger warning message of the original vehicle. In this case, since the hop field of the danger alert message is 0, other adjacent vehicles do not transmit it to other vehicles.

In addition, in response to the occurrence of a risk alert message according to the V2X-based accident notification service format, the diagnosis result is notified to an adjacent infrastructure within a preset communication radius.

If this is followed, the first vehicle transmits a danger warning message to both the adjacent vehicle and the adjacent infrastructure at the same time. At this time, since the risk alert message must be transmitted only to the adjacent infrastructure, hops are not used. Accordingly, the infrastructure receiving the first vehicle's danger warning message does not transmit the state of the initial vehicle to the surrounding infrastructure, but only to the surrounding vehicles. The first vehicle transmits sensing information and analysis results based on the vehicle parts diagnosis format and the entire vehicle condition diagnosis format to the main cloud server.

Therefore, when an accident or failure is detected, the danger is notified to surrounding vehicles and infrastructure to prevent continuous accidents.

3 is a flowchart sequentially showing a self-diagnosis method for an autonomous vehicle based on deep learning according to an embodiment.

As shown in FIG. 3, the self-diagnosis method for an autonomous vehicle based on deep learning according to an embodiment first uses an autonomous vehicle that self-diagnoses the entire state of the vehicle or parts of the autonomous vehicle according to OBD. It is about self-diagnosis method for

In the self-diagnosis method according to this embodiment, sensor information of a vehicle is received and registered (S301). In this case, sensor information of the vehicle is input by the lightweight edge gateway method according to an embodiment. And, the main protocols in the vehicle communicate with each other. To this end, for example, by transmitting and receiving as a message between the main internal communication protocol buses of the vehicle, sensor information of the vehicle is received, and the sensor information of the vehicle as the message is registered in a preset message queue to be processed. To this end, first, an address mapping table that maps preset address information including a destination address, a destination bus number, a source ID, and a cycle of each header of a message is set and registered to be required for message conversion between main internal communications of the vehicle. For example, this address mapping table consists of cycle, ID, priority, source protocol, source bus number, source address, destination protocol, destination bus number, and destination address. Therefore, each time sensor information of the vehicle is input, the address information of each header is extracted to reduce the weight of the message header, so that main internal communications of the vehicle are mapped using the address mapping table. Then, the header information according to the address mapping table and the content of the original message in the message queue are combined, and the message is converted and transmitted. Therefore, for real-time communication between vehicle networks, the message header is lightened and the message is converted using a mapping table. Through this, transmission delay caused by message conversion is minimized through weight reduction of data. Therefore, the message format of different protocols inside the vehicle is integrated. In this way, the message conversion transmission delay time incurred during transmission is reduced, and the transmission time is improved even in the case of traffic overload transmission delay. Through this, unlike a hardware vehicle gateway, it is a software gateway applied to OBD and has high compatibility with existing vehicles. In addition, transmission errors and overloads due to message segmentation are reduced by using a lightweight message header.

Next, and in that state, first, according to the association between vehicle parts, the vehicle parts diagnosis format is set and registered by back-propagation by the hidden layer determined as the number of input sensor information and the dynamic number according to the number of output parts ( S302).

In this case, the vehicle parts diagnosis format is according to the following [Equation 3].

[Equation 3]

NET _h1j = ∑x _i W _ij (i = from 0 to n), h _1j = max(0, NET _h1j )

NET _yk = ∑h _mi Z _ik (from i = 1 to l), y _k = 1/1+e ^{-NET yk}

Here, x _i is the value of the i-th input node of the vehicle's sensor information, W _ij is the connection strength between the i-th input node and the j-th hidden node, and h _mi is the hidden layer in the order determined according to the number of hidden layers. The value of the layer, Z _ik is the strength of the connection between the ith input node and the kth hidden layer, y _k is the risk of vehicle parts and consumables.

So, according to the vehicle parts diagnosis format, the influence between parts is calculated by learning the sensor information of the vehicle in real time, so the risk of each of the parts and consumables of the vehicle and the risk according to the dependency of other parts due to the defective parts are inferred. Do (S303).

Accordingly, by calculating the correlation between the parts of the vehicle accordingly, the state of the current parts is accurately and quickly diagnosed.

As described above, according to an exemplary embodiment, the vehicle component diagnosis format is set as back-propagation by a hidden layer determined as the number of input sensor information and a dynamic number according to the number of output components according to the association between vehicle components. Thus, according to the vehicle component diagnosis format, the vehicle sensor information is learned in real time to infer the risk of each vehicle component and consumables and the risk according to the dependency relationship of other components due to the defective component.

Through this, the relationship between the parts of the vehicle is calculated and the state of the current parts is clearly diagnosed so that they can be delivered to the user.

Meanwhile, according to an exemplary embodiment, the overall condition or failure of the vehicle can be diagnosed according to the risk of each of the components and consumables of the vehicle and the risk according to the dependency relationship of other components due to the defective component.

To this end, according to the following [Equation 4], an exemplary embodiment first sets and registers a format for diagnosing the overall condition of the vehicle using a preset step function according to the association between vehicle components and the risk of vehicle components.

[Equation 4]

NET _n = ∑x _i W _in (i = 0 to g), f(NET _n )(0(NET <0) 1(NET≥0)

Here, x _i is the value of the i-th input node of the vehicle's part information, W _in is the connection strength between the i-th input node and the n-th input node, and g is the number of combinations of the vehicle's parts using input information. A fixed value

Thus, the overall condition of the vehicle is diagnosed by learning the risk of the vehicle parts and consumables in real time and the risk according to the dependency relationship of other parts due to the defective parts by this format for diagnosing the overall condition of the vehicle.

Through this, the overall condition or failure of the vehicle is diagnosed.

Therefore, the condition of the vehicle is accurately and quickly diagnosed.

On the other hand, an embodiment enables surrounding vehicles to quickly detect a vehicle with a high risk of an accident in the event of such a vehicle failure.

To this end, one embodiment first performs V2X communication based on edge computing to inform the cloud and surrounding vehicles of the diagnosis result.

Specifically, the vehicle parts diagnosis format model is created by modeling the vehicle parts diagnosis format by the association between the vehicle parts according to the risk of each of the parts and consumables of the vehicle and the risk according to the dependency relationship of other parts by the defective parts. .

In addition, the overall vehicle condition diagnosis format is modeled by modeling the overall condition diagnosis format by modeling the relationship between the vehicle parts according to the degree of risk according to the degree of risk of the parts and consumables of the vehicle and the dependence of other parts due to the defective parts.

Next, the vehicle parts diagnosis format model and the entire vehicle condition diagnosis format model are transmitted to the main cloud server that provides the vehicle parts diagnosis format model and the entire vehicle condition diagnosis format model corresponding to each vehicle device.

In addition, from the main cloud server, the vehicle component diagnosis format model and the overall vehicle condition diagnosis format model suitable for the vehicle device are provided from the main cloud server.

Therefore, by learning the vehicle sensor information in real time by the vehicle parts diagnosis format model provided in this way, the risk of each of the parts and consumables of the vehicle and the risk according to the dependency relationship of other parts due to the defective part is inferred.

Then, according to the overall condition diagnosis format model of the vehicle, each risk of parts and consumables of the vehicle and the degree of risk according to the dependency relationship of other parts due to the defective part are learned.

Accordingly, in the case of a risk that has been diagnosed as dangerous according to this learning, a risk warning message is generated and transmitted to a vehicle or infrastructure adjacent to a preset communication radius.

Accordingly, the surrounding vehicle can quickly detect a vehicle with a high risk of an accident.

Through this, when an accident or breakdown is detected, the danger is notified to nearby vehicles and infrastructure to prevent continuous accidents.

* Description of the symbols for the main parts of the drawing *
100: OBD

Claims (5)

In the self-diagnosis method for an autonomous vehicle that self-diagnoses parts of an autonomous vehicle according to OBD or the overall condition of the vehicle,
A first step of setting and registering a vehicle part diagnosis format as back-propagation by a hidden layer determined as a dynamic number according to the number of input sensor information and the number of output parts according to the association between vehicle parts;
A second step of receiving and registering sensor information of a vehicle; And
And a third step of inferring a risk according to a dependency relationship between a vehicle component and a consumable component and another component due to a defective component by learning sensor information of the vehicle according to the vehicle component diagnosis format,
The vehicle parts diagnosis format is
[Equation 5]
NET _h1j = ∑x _i W _ij (i = 0 to n), h _1j = max(0, NET _h1j )
NET _yk = ∑h _mi Z _ik (from i = 1 to l), y _k = 1/1+e ^{-NET yk}
Follow,
Wherein, xi is the value of the i-th input node of the vehicle sensor information, Wij is the connection strength between the i-th input node and the j-th hidden node, hmi is the value of the hidden layer in an order determined according to the number of hidden layers, Zik is the strength of the connection between the ith input node and the kth hidden layer, yk is the risk of vehicle parts and consumables.
After the third step,
Step 4-1 of setting and registering the entire vehicle condition diagnosis format using a preset step function according to the association between vehicle components and the risk of vehicle components according to the following [Equation 6]; And
[Equation 6]
NET _n = ∑x _i W _in (i = 0 to g), f(NET _n )(0(NET <0) 1(NET≥0)
Here, xi is the value of the i-th input node of the vehicle part information, Win is the connection strength between the i-th input node and the n-th input node, and g is a value determined in response to the number of combinations of the vehicle parts' input information. ego,
A 4-2 step of diagnosing the overall condition of the vehicle by learning the risk of the vehicle parts and consumables in real time and the risk according to the dependency relationship of other parts due to the defective parts according to the overall vehicle condition diagnosis format,
The autonomous vehicle includes an integrated lightweight edge gateway of self-diagnosis that supports communication of all major protocols within the vehicle within the OBD, guarantees fast speed, and communicates in real time between vehicle networks, and the sensor of the autonomous vehicle is self-contained by OBD. Provides sensing information for diagnosis and transmits the self-diagnosis result to external information processing devices including vehicle repair shops through IoT,
The sensor includes an ABS sensor through a FlexRay bus, an airbag sensor, a temperature sensor through a CAN bus, a handle sensor, an instrument sensor, and an image sensor through a MOST ring,
The self-diagnosis module according to the vehicle parts diagnosis format is an autonomous vehicle self-diagnosis module, and this module consists of two sub-modules operating in a cloud environment.
The first sub-module is VPDS, which is a vehicle parts diagnosis sub-module according to the vehicle parts diagnosis format, and diagnoses the state of each part by considering the association of each part.
The VPDS uses the speed, starting voltage, mileage, tire pressure, RPM, coolant, engine oil, and rear sensor information that can be collected from the vehicle's sensors, and diagnoses the condition of each part by considering the relationship between each part. and
The VPDS uses the information received from the vehicle as input data, the information collected from the in-vehicle sensor is digital numerical information and indicates the state of each part, and the input data of the VPDS is a set of sensors, information of each sensor, Self-driving vehicles based on deep learning are defined by combining the number of sensors, and the output value of the VPDS represents the status of each part including "normal", "warning", and "danger" Diagnostic method.
delete delete The method of claim 1,
The first step
Deep learning, characterized in that it is collected from the real-time sensor information of the vehicle or the arrangement sensor information of the vehicle service center registered in advance to receive the input of the sensor information, and the operation of the risk inference according to the third step is operated based on the cloud. Self-diagnosis method for self-driving vehicles based on. The method of claim 1,
After the third step,
The fourth'-, which creates a vehicle parts diagnosis format model by modeling the vehicle parts diagnosis format according to the relationship between vehicle parts according to the risk of each of the parts and consumables of the vehicle and the risk according to the dependency relationship of other parts by defective parts. Including step 1,
After step 4-2,
Part 4'-2, which generates an overall vehicle condition diagnosis format model by modeling the overall condition diagnosis format by modeling the relationship between vehicle parts according to the degree of risk according to the degree of risk due to the dependency relationship between the parts and consumables of the vehicle and other parts due to defective parts. Including steps,
Step 4'-3 of transmitting the vehicle component diagnosis format model and the vehicle overall state diagnosis format model to a main cloud server that provides a vehicle component diagnosis format model and an overall vehicle condition diagnosis format model corresponding to each vehicle device;
A 4'-4 step of receiving a vehicle part diagnosis format model and an overall vehicle condition diagnosis format model from the main cloud server corresponding to the vehicle device;
By learning the vehicle sensor information in real time by the vehicle parts diagnosis format model provided by step 4'-4, the risk of each of the parts and consumables of the vehicle and the risk according to the dependency relationship of other parts due to the defective parts are determined. Step 4'-5 inferring;
According to the risk of each part and consumables of the vehicle inferred in the step 4'-5 by the overall vehicle condition diagnosis format model in step 4'-4 and the dependency relationship of other parts due to the defective part Step 4'-6 of learning the degree of risk; And
In the case of a risk level diagnosed as dangerous according to the learning by the 4′-6 steps, a risk warning message is generated and transmitted to an adjacent vehicle within a preset communication radius. Self-diagnosis method for autonomous vehicles based on learning.

Images