JP2014234113A - Vehicular reference value generation apparatus - Google Patents

Abstract

PROBLEM TO BE SOLVED: To enable generation of a reference value making it possible to determine a larger number of categories of abnormality of a vehicle as abnormal.SOLUTION: A generation apparatus includes: a drive scene categorization part 21 for categorizing drive data by drive scene produced by dividing the drive data into units of patterns of changing in the drive data that is obtained by a drive data collection device 1 and is arranged in time-series; a diagnostic characteristic extraction part 22 for extracting a diagnostic characteristic of the drive data by pattern unit; a diagnostic characteristic accumulation part 23 for accumulating, by drive scene, the diagnostic characteristic extracted by the diagnostic characteristic extraction part 22; and a reference value generation part 24 for generating a reference value for determining abnormality of a vehicle from the diagnostic characteristics accumulated in the diagnostic characteristic accumulation part 23.

Description

  The present invention relates to a vehicular reference value generation device that generates a reference value for discriminating a vehicle abnormality.

  Conventionally, in vehicle failure diagnosis, detection of an abnormal state of a vehicle has been studied for early detection of the failure and investigation of the cause. For this purpose, a device that learns a reference value that is a normal value of driving data from normal or abnormal driving data and determines that the vehicle is abnormal when the driving data is not within the range of the reference value has been studied. Yes.

  However, there are various driving scenes according to environmental conditions and driving patterns in the driving situation of the vehicle, and it is considered that the range of reference values that can be taken by various driving data can be changed individually depending on the individual driving scenes. Therefore, in order to realize the above-described device, it is necessary to determine reference values for various operation data under various operation scenes.

  Thus, as means for solving this problem, for example, Patent Document 1 discloses the following reference value generation device. In the reference value generating device disclosed in Patent Document 1, for each driving data such as the engine speed and the vehicle speed, the driving data in an approximated state is clustered into the same group, so that the driving data approximated to the driving scene is grouped. To do. Then, for each group, a reference value is determined from the frequency of operation data values.

Japanese Patent No. 4414470

  However, in the reference value generation device disclosed in Patent Document 1, since the instantaneous value of the operation data is used as the reference value, for example, when the operation data value instantaneously indicates the same “acceleration time”, It was not possible to distinguish between “acceleration” when accelerating from a stopped state and “acceleration” when accelerating after deceleration. Therefore, a criterion for discriminating abnormalities specific to time-series patterns of operation data with a specific history, such as abnormalities that may not occur during acceleration after deceleration but may occur when accelerating from a stopped state, as abnormal The value could not be generated.

  The present invention has been made in view of the above-described conventional problems, and its purpose is to generate a reference value for a vehicle that makes it possible to generate a reference value that can distinguish more types of abnormality of the vehicle as abnormal. To provide an apparatus.

  The vehicle reference value generation device according to the present invention sequentially acquires driving data, which is information relating to a vehicle, including at least one of an operation by a driver of the vehicle, a state of the vehicle, a state of the driver, and a driving environment of the vehicle. The operation data acquired by the operation data acquisition unit (11) and the operation data acquisition unit is divided into time-series operation data pattern units (hereinafter referred to as operation scenes) that repeatedly appear and classified for each operation scene. A classifying unit (21), a feature extracting unit (22) for extracting pattern features which are features expressing driving data in pattern units, and pattern features extracted by the feature extracting unit into driving scenes classified by the classifying unit. And a reference value generation unit (24) for generating a reference value for discriminating a vehicle abnormality from the pattern features stored in the storage unit. It is characterized in Rukoto.

  According to this, the pattern feature extracted by the feature extraction unit is a feature expressing a pattern that repeatedly appears in the operation data arranged in time series. Then, the reference value generation unit generates a reference value for determining a vehicle abnormality from the pattern features accumulated based on the driving scene classified by the classification unit. Therefore, the time series pattern of operation data with a specific history will have a different reference value from the time series pattern of operation data with other history, and an abnormality peculiar to operation data with a specific history will be detected. It becomes possible to generate a reference value for determining an abnormality. As a result, it is possible to generate a reference value that can distinguish more types of abnormality of the vehicle as abnormalities. Further, according to the present invention, it is possible to discriminate more abnormal states than in the past, and to make early detection of failures and automatic diagnosis.

2 is a diagram illustrating an example of a schematic configuration of a vehicle diagnosis unit 100. FIG. 3 is a functional block diagram illustrating an example of a schematic configuration of an operation data collection device 1. FIG. 3 is a functional block diagram illustrating an example of a schematic configuration of a diagnostic device 2. FIG. It is a graph which shows an example of 6 types of driving data arranged in time series. It is a figure for demonstrating double segment structure analysis. It is a figure for demonstrating extraction of the diagnostic feature in the diagnostic feature extraction part 22. FIG. It is a figure for demonstrating the regression coefficient between two driving | running data. It is a figure for demonstrating matrix expression. It is a figure for demonstrating accumulation | storage of the diagnostic feature to the diagnostic feature storage part. It is a figure for demonstrating an example of a reference value production | generation process. It is a figure for demonstrating the other example of a reference value production | generation process. 4 is a flowchart illustrating an example of a flow of reference value generation related processing in the diagnostic device 2. It is a figure for demonstrating an example of an abnormality diagnosis process. It is a figure for demonstrating the other example of an abnormality diagnosis process. 4 is a flowchart illustrating an example of a flow of abnormality diagnosis-related processing in the diagnostic device 2. (A) is a figure for demonstrating an example of the classification of the driving scene by a prior art, (b) is a figure for demonstrating an example of the classification of the driving scene by this invention. (A) is a figure for demonstrating an example of the classification of the driving scene by a prior art, (b) is a figure for demonstrating an example of the classification of the driving scene by this invention. (A) is a figure for demonstrating an example of the classification of the driving scene by a prior art, (b) is a figure for demonstrating an example of the classification of the driving scene by this invention. (A) is a figure for demonstrating an example of the classification of the driving scene by a prior art, (b) is a figure for demonstrating an example of the classification of the driving scene by this invention. (A) is a figure for demonstrating an example of abnormality determination by a prior art, (b) is a figure for demonstrating an example of abnormality determination by this invention. It is a figure which shows an example of a schematic structure of the vehicle diagnostic unit 100a.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings.

(Embodiment 1)
FIG. 1 is a block diagram showing a schematic configuration of a vehicle diagnostic unit 100 to which the present invention is applied. A vehicle diagnosis unit 100 shown in FIG. 1 includes a driving data collection device 1, a diagnosis device 2, and a presentation device 3. The configuration including the driving data collection device 1 and the diagnosis device 2 corresponds to the vehicle reference value generation device in the claims.

  The driving data collection device 1 is mounted on a vehicle and includes a data acquisition unit 11 and a data storage unit 12 as shown in FIG. The data acquisition unit 11 sequentially acquires detection values of various sensors of the host vehicle and stores them in the data storage unit 12. The data acquisition unit 11 corresponds to the operation data acquisition unit in the claims. The data acquisition unit 11 may be configured to acquire the detection value of the sensor of the own vehicle via an in-vehicle LAN compliant with a communication protocol such as CAN (controller area network), or the sensor of the own vehicle via a Zika line. The detected value may be obtained.

The detection values of the various sensors of the vehicle include information related to the driving operation of the vehicle by the driver (hereinafter referred to as driving operation information), information related to the behavior of the vehicle and states other than the behavior appearing as a result of the driving operation (hereinafter referred to as vehicle Status information), information on the status of the driver (hereinafter referred to as driver status information), and the like.
Examples of the sensor that measures driving operation information include an accelerator sensor that measures an accelerator opening, a brake sensor that measures a depression amount of a brake pedal and a brake master cylinder pressure, and a steering sensor that measures a steering angle of a steering wheel. In addition, there are a winker switch for operating the direction indicator, a shift position sensor for detecting a shift position of the transmission, and the like.

Sensors that measure vehicle state information include a speed sensor that measures the traveling speed of the vehicle (hereinafter referred to as a vehicle speed), an acceleration sensor that measures the longitudinal and lateral acceleration of the vehicle, a crank angle sensor that measures the engine speed, A temperature sensor that measures the engine coolant temperature, an air flow meter that measures the intake air amount, and the like can be given.
Examples of sensors that measure driver status information include a heart rate sensor that measures a driver's heart rate, a gaze sensor that measures the driver's line of sight, and a vehicle interior camera that measures the driver's facial expression.

  The data acquisition unit 11 not only simply stores the detection value from the sensor in the data storage unit 12, but also calculates differential value information corresponding to the differential value of the driving operation information, the vehicle state information, and the driver state information. In addition, the data storage unit 12 may be configured to save the data.

  For example, when the detection signal of the steering sensor is acquired, the steering speed that is the differential value information of the steering angle may be calculated and stored together with positive and negative signs representing the steering direction. In addition, when obtaining the detection value of the accelerator sensor or brake sensor, the stepping speed corresponding to the differential value information is calculated from the accelerator opening and the brake master cylinder pressure together with positive and negative signs indicating the stepping-up or stepping-back direction. And a configuration for saving. In addition, the data acquisition part 11 is good also as a structure which calculates not a differential value itself of the detected value by each sensor but a mere difference value as differential value information.

  In this way, by using the driver operation information, the vehicle state information, and the differential value information of the driver state information, it is possible to appropriately represent the tendency of the vehicle state change due to the driver's operation. Regarding the driving operation of the vehicle and the state of the vehicle, not only the size is a problem, but rather the tendency of the change is important.

  For example, when the steering wheel is being steered so that the steering angle in the right direction increases, the vehicle is in a state where the turning to the right is intensified, whereas when the vehicle is returning to the neutral position, the turning state Is weakened and is approaching straight ahead. The operation of the accelerator pedal and brake pedal is the same.The vehicle speed and engine speed increase when the pedal is increased, and the vehicle speed and engine speed decrease when the pedal is returned. It will be completely different. When differential value information is used, such a change tendency can be expressed as information.

  The data acquisition unit 11 may acquire information related to the traveling environment of the own vehicle (hereinafter referred to as traveling environment information) and store the information in the data storage unit 12. Examples of sensors that detect traveling environment information include a surrounding environment sensor, an in-vehicle environment sensor, a GPS receiver, and a road information database.

  As the ambient environment sensor, for example, a temperature sensor that measures the outside air temperature, or a radar device that measures the distance to a vehicle or an obstacle that travels in front of the vehicle can be used. When a preceding vehicle exists, it is used to measure the distance to the preceding vehicle. In addition, as the vehicle interior environment sensor, for example, a microphone that measures the sound in the vehicle interior can be used, and this microphone detects the presence or absence of conversation in the vehicle interior and the content of the conversation when there is a conversation. Used for.

  Further, by using the GPS receiver and the road information database, it is possible to grasp the road situation where the vehicle is traveling. For example, whether the road on which the vehicle is traveling is an expressway or a general road, whether it is a road with multiple lanes or a single lane, and whether there are intersections or branch points ahead of the vehicle's travel position it can.

  Such information obtained from the surrounding environment sensor, the in-vehicle environment sensor, the GPS receiver, and the road information database has a great influence on the vehicle state and the driver state. For this reason, by using these pieces of information, it is possible to improve the accuracy of segmentation in the discretization process described later and the accuracy of segmentation in the pattern extraction process described later.

  Information acquired by the data acquisition unit 11 such as driving operation information, vehicle state information, traveling environment information, driver state information, and differential value information acquired from various sensors is hereinafter referred to as driving data. Further, the data acquisition unit 11 associates the operation data acquired sequentially with the time information (that is, the time stamp) when the operation data is detected or acquired, and stores it in the data storage unit 12.

  Subsequently, the diagnostic device 2 is mainly configured as a microcomputer, and each includes a known CPU, a memory such as a ROM, a RAM, and an EEPROM, an I / O, and a bus connecting them. The diagnostic device 2 is used in a vehicle.

  The diagnostic device 2 acquires the driving data accumulated in the data storage unit 12 of the driving data collection device 1 of the own vehicle. The operation data acquired by the diagnostic device 2 may be deleted from the data storage unit 12.

  And the diagnostic apparatus 2 produces | generates the reference value for determining the presence or absence of abnormality of a vehicle for every driving scene mentioned later by executing the various control programs memorize | stored in ROM based on the acquired driving | running data. Reference value generation related processing is performed. Moreover, the diagnostic apparatus 2 performs abnormality diagnosis related processing for determining whether there is an abnormality in the vehicle.

  As shown in FIG. 3, the diagnostic apparatus 2 includes a driving scene classification unit 21, a diagnostic feature extraction unit 22, a diagnostic feature storage unit 23, a reference value generation unit 24, a reference value storage unit 25, and an abnormality diagnosis unit as functional blocks. 26. Assume that the diagnostic feature storage unit 23 and the reference value storage unit 25 are constructed in an electrically rewritable memory such as a RAM or an EEPROM, or a temporally rewritable storage such as an HDD.

  The driving scene classification unit 21 arranges the driving data acquired from the driving data collection device 1 in time series based on the time stamp described above, and divides the driving data arranged in time series into several groups. The driving scene classification unit 21 corresponds to the classification unit in the claims. In this embodiment, a case where six types of operation data such as accelerator opening, brake master cylinder pressure, steering angle, vehicle speed, differential value of brake master cylinder pressure, and differential value of steering angle are used as operation data is taken as an example. The following explanation will be given.

  The driving data varies greatly depending on environmental conditions and driving patterns. FIG. 4 is a graph showing an example of six types of operation data arranged in time series. As shown in FIG. 4, each operation data takes various values, but as shown by A, a similar pattern repeatedly occurs in a specific region.

  The driving scene classification unit 21 extracts a pattern as described above. For this purpose, the driving scene classification unit 21 divides driving data arranged in time series by double segment structure analysis as an example. This is an automatic operation by a two-stage process that performs a discretization process for discretizing time series data and then a pattern extraction process for extracting patterns that appear repeatedly in the discrete series obtained by the discretization process. The operation data is divided and classified for each scene. The outline of double segment structure analysis is shown below.

  In the discretization process, as shown in FIG. 5, six types of driving data arranged in time series are classified according to the state of the vehicle as a unit of division, and the vehicle data grasped from each of the divided driving data. A symbol string is generated by representing the driving symbol (C1, C2, C3, C4, C5,...) Corresponding to the state.

  In other words, in the discretization process, various vehicle states grasped from each driving data are defined as clusters in a space having the six types of driving data input as dimensions, and any combination of the driving data belongs to any cluster. By statistically processing this, each driving data that changes with time is classified for each vehicle state (that is, for each cluster) as a classification unit. Further, in the discretization process, a symbol string is generated by representing each divided signal by a symbol (C1, C2, C3, C4, C5,...) Corresponding to the cluster to which each combination of operation data belongs.

  In this embodiment, a hierarchical Dirichlet process hidden Markov model (HDP-HMM), which is one of models expressed by a hidden state and a stochastic transition between the states, is used for the discretization process. The HDP-HMM has the flexibility to determine the number of hidden states according to each input signal by assuming an infinite dimensional hidden state (corresponding to a cluster) in the HMM. Therefore, there is an advantage that it is not necessary to design the number of hidden states in advance when using the HMM. In particular, it is preferable to use sticky HDP-HMM as the HDP-HMM. Sticky HDP-HMM adds a bias to the self-transition probability of HDP-HMM. By increasing the self-transition probability, excessive transition of hidden states can be suppressed, and modeling assuming continuity of operation is performed. It becomes possible to carry out efficiently.

  For such HDP-HMM and Sticky HDP-HMM, for example, see “Non-segmental motion imitation learning method using hierarchical Dirichlet process hidden Markov model” (Academic Lecture Meeting of the Society of Instrument and Control Engineers System and Information Division (2010)). Since it is described, further description is omitted.

  Note that the discretization process is not limited to the HDP-HMM described above, and can be performed using another model. For example, it is possible to use a model such as a general HMM, an Nth order Markov model, a hierarchical Markov model, a switching AR model, or a switching Kalman filter. In this case, for example, the number of hidden states of each model is designed in advance, and the parameters of each hidden state and the transition probability between hidden states are calculated in advance or sequentially updated according to the number of hidden states. Then, the most likely hidden state (cluster) can be obtained by calculating the posterior probability of the hidden state based on the parameters of the hidden state and the transition probability between the hidden states.

  Subsequently, in the pattern extraction process, as shown in FIG. 5, the symbol string obtained by the discretization process is segmented for each driving word (W1, W2, W3,...) Meaning a driving scene. That is, in the pattern extraction process, each driving symbol in the symbol string is singly or gathered, and driving scenes (for example, starting, stopping, constant speed driving, turning in a curve, turning right and left at an intersection, right lane and Segmentation to mean left lane change). The driving scene classification unit 21 inputs the result of segmentation by the pattern extraction process to the diagnostic feature extraction unit 22.

  For example, in the example shown in FIG. 5, the driving symbol C1 is replaced by the driving word W1 alone, while the driving symbols C2 and C3 are integrated into the driving word W2. The driving symbols C2, C4, and C5 are the driving word W3. In this way, the arrangement of driving words obtained by segmenting each driving word meaning a driving scene indicates a change in the driving scene.

In this embodiment, Nested Pitman-Yor Language Model (NPYLM), which is an example of an unsupervised chunking technique for discrete character strings using statistical information, is used for pattern extraction processing. This NPYLM has been expanded by incorporating an N-gram model of words into the Hierarchical Pitman-Yor Language Model (HPYLM), and realizes morphological analysis without dictionary data. HPYLM improves the robustness against unknown words and low-frequency words by smoothing the N-gram language model by the Pitman-Yor process, and enables appropriate word segmentation.

  NPYLM and HPYLM are described in detail in “Analysis of Behavior Using Hierarchical Pitman-Yor Language Model” (The 25th Annual Conference of the Japanese Society for Artificial Intelligence, 3B1-OS22c-8. (2011)). Description is omitted.

  Note that the pattern extraction process is not limited to the above-described NPYLM, and may be configured to perform segmentation by other methods. For example, based on a symbol string generated when the vehicle is driven in various driving modes, a predetermined driving word is set in advance to create a driving word dictionary, and the transition probability and generation probability of each driving word Create a database and store it. Then, referring to these driving word dictionaries, transition probability database, and generation probability database, the input symbol string may be segmented for each most probable driving word.

  In the present embodiment, the operation data segmented for each driving word by the double segment structure analysis is described below as a pattern unit, but the present invention is not necessarily limited thereto. For example, the operation data may be classified for each driving symbol that is the result of the discretization processing without performing segmentation for each driving word. When this configuration is adopted, the driving data is classified for each driving symbol. Use operation data as pattern units.

  Returning to FIG. 3, the diagnostic feature extraction unit 22 performs a diagnostic feature extraction process for extracting features of driving data (hereinafter referred to as diagnostic features) in units of patterns classified for each driving scene by the driving scene classification unit 21. The diagnostic feature corresponds to the pattern feature of the claims, and the diagnostic feature extraction unit 22 corresponds to the feature extraction unit of the claims. The diagnostic feature extraction unit 22 accumulates the diagnostic features extracted in the diagnostic feature extraction process in the diagnostic feature storage unit 23 or inputs them to the abnormality diagnosis unit 26. The diagnostic features stored in the diagnostic feature storage unit 23 are used for reference value generation processing described later, and the diagnostic features input to the abnormality diagnosis unit 26 are used for abnormality diagnosis processing described later.

  Here, the extraction of diagnostic features in the diagnostic feature extraction unit 22 will be described with reference to FIG. FIG. 6 is a diagram illustrating an example of driving data classified by the driving scene classification unit 21. A broken line in FIG. 6 is a break of the driving scene classified by the driving scene classification unit 21.

  The driving data in the time zone with the same driving scene classified by the driving scene classification unit 21 has similar statistical properties. For example, taking the driving scene 5 in FIG. 6 as an example, the driving data 3 has a tendency to increase rapidly after the value decreases monotonously (see B in FIG. 6). Moreover, about the driving | running | working data 5, after the value increases gently, the tendency which has fallen rapidly to the minimum value is recognized (refer C of FIG. 6). Furthermore, in the driving scene 5, the time at which the value of the driving data 3 starts to increase and the time at which the value of the driving data 5 suddenly decreases correspond to each other. The diagnostic feature extraction unit 22 quantitatively extracts such fluctuations, amounts, and relationships of operation data as diagnostic features.

  As an example, the diagnostic feature extraction unit 22 extracts a regression coefficient between two pieces of driving data in a time zone in which the driving scene is the same for all combinations of driving data as a diagnostic feature. Thus, it is possible to express the relationship between operation data arranged in time series in a specific operation scene.

  As an example, FIG. 7 shows a plot of the values of the operation data 3 and the operation data 5 in the operation scene 5 in the example of FIG. 6 on a two-dimensional plane. A broken line in FIG. 7 is a regression line, and this slope is a regression coefficient. A matrix representation of regression coefficients for all combinations of each operation data is used as a diagnostic feature. In the example of the present embodiment, since the operation data is 6-dimensional data, it is expressed by a matrix composed of 6 × 6 = 36 cells as shown in FIG. A numerical value is written in each cell of the matrix, but in FIG. 8, for convenience, a numerical value is expressed by shading of each cell.

  The diagnostic feature is not limited to using the above-described regression coefficient, but may be configured to use other values expressing the relationship between operation data arranged in time series. For example, a configuration using a correlation coefficient between two variables, a Granger causal coefficient, a regression residual, or the like may be used. In addition, it is good also as a structure which uses the deviation degree from the regression line in a design model. The design model referred to here expresses a relationship between operation data of various sensors that have been designed in advance by a vehicle manufacturer or the like and confirmed to be normal.

  Moreover, it is not restricted to the structure which uses what was expressed as a matrix as a diagnostic feature. For example, a configuration in which a correlation coefficient between two variables, a regression coefficient, a regression residual, and the like are arranged may be used. In this case, the diagnostic feature may be expressed by a three-dimensional tensor instead of a matrix. Moreover, it is good also as a structure which expresses the one-to-many relationship instead of the one-to-one relationship between driving | operation data. For example, a diagnostic feature may be obtained by arranging the residuals in the multiple regression analysis with a certain driving data as an objective variable and all the remaining driving data as explanatory variables, for each driving data.

  Furthermore, the diagnostic feature may be other than that expressing the relationship between operation data arranged in time series. For example, a configuration may be used in which the values of individual variables in the driving data are quantitatively extracted and the average value of each variable in the same driving scene is used as a vector. In addition, for example, as a means for quantitatively extracting fluctuations in driving data, a configuration in which an average of the differential values of the variable driving data for each driving scene may be used.

  The diagnostic feature extraction unit 22 stores the diagnostic features extracted in the diagnostic feature extraction process in the diagnostic feature storage unit 23. The diagnostic feature storage unit 23 corresponds to a storage unit of claims. In the present embodiment, as shown in FIG. 9, a database for each driving scene is constructed in the diagnostic feature storage unit 23, and the following description will be made assuming that diagnostic features are stored in this database for each driving scene. .

  In addition, it is good also as a structure which accumulate | stores not only in the structure which accumulate | stores a diagnostic feature for every driving scene in the diagnostic feature storage part 23 but in the diagnostic feature storage part 23, without distinguishing for every driving scene. In addition, a configuration in which a plurality of types of driving scenes are divided into the same database may be employed. As an example, an average value of diagnostic features may be calculated for each driving scene, and driving scenes having similar calculated average values may be allocated to the same database.

  Returning to FIG. 3, the reference value generation unit 24 performs a reference value generation process for generating a reference value for determining the presence or absence of abnormality of the vehicle from the diagnostic features stored in the diagnostic feature storage unit 23. In other words, in the reference value generation process, a reference value for determining a normal value and an abnormal value is generated. In the example of this embodiment, a reference value is generated for each driving scene from the diagnostic features stored in the diagnostic feature storage unit 23 for each driving scene classified by the driving scene classification unit 21.

  As shown in FIG. 10, the reference value generation unit 24 discriminates between normal values and abnormal values obtained by a One Class Support Vector Machine (OC-SVM) using diagnostic features accumulated for each driving scene as learning data. A boundary (that is, an identification hyperplane) is generated as a reference value.

  Note that the reference value generation unit 24 may be configured to generate, as a reference value, a normalized deviation in the frequency distribution of diagnostic features stored in the diagnostic feature storage unit 23 (hereinafter, modified example 1). As a first modification, as shown in FIG. 11, for each cell of the diagnostic feature matrix accumulated in the diagnostic feature accumulating unit 23, an identification boundary between a normal value and an abnormal value is used as a reference value from the frequency distribution of values. What is necessary is just to make it the structure to produce | generate. For example, for each diagnostic feature cell, the difference between the average value and the standard deviation of the value is three times the reference value.

  In the above-described embodiment, the configuration in which the reference value is generated by unsupervised learning has been described. However, the configuration is not necessarily limited thereto. For example, as a configuration in which the reference value generation unit 24 generates a reference value based on both the diagnostic characteristics of the driving data when the vehicle is normal and the pattern characteristics of the driving data when the vehicle is abnormal (hereinafter, modified example 2) Also good.

  As a second modification, an identification boundary learned by discrimination by a Support Vector Machine (SVM) between a diagnostic feature of driving data when the vehicle is normal and a diagnostic feature of driving data when the vehicle is abnormal is generated as a reference value. What is necessary is just composition. The driving data when the vehicle is normal and the driving data when the vehicle is abnormal may be provided in advance by actual vehicle tests for when the vehicle is normal and when the vehicle is abnormal.

  The reference value generation unit 24 stores the reference value generated by the reference value generation process in the reference value storage unit 25. In the present embodiment, the reference value storage unit 25 will be described below by taking as an example a case where the reference value for each driving scene classified by the driving scene classification unit 21 is stored.

  Here, the reference value generation related processing in the diagnostic device 2 will be described using the flowchart of FIG. The flowchart of FIG. 12 may be configured to start when the diagnostic device 2 is turned on and end when the diagnostic device 2 is turned off, for example.

  First, in step S1, when the diagnostic device 2 acquires new operation data from the operation data collection device 1 (YES in step S1), the process proceeds to step S2. On the other hand, when the diagnostic device 2 has not acquired new operation data from the operation data collection device 1 (NO in step S1), the flow of step S1 is repeated.

  In step S2, the driving scene classification unit 21 performs the above-described discretization process on the driving data arranged in time series, and the process proceeds to step S3. In step S3, the driving scene classification unit 21 performs the above-described pattern extraction process on the symbol string obtained by the discretization process, and the process proceeds to step S4.

  In step S4, the diagnostic feature extraction unit 22 performs the above-described diagnostic feature extraction processing on the pattern-unit driving data classified for each driving scene in the pattern extraction processing, and proceeds to step S5. In step S5, the diagnostic feature extraction unit 22 accumulates the diagnostic features extracted by the diagnostic feature extraction processing in the diagnostic feature storage unit 23 for each driving scene classified by the driving scene classification unit 21, and the process proceeds to step S6.

  In step S6, the reference value generation unit 24 performs the above-described reference value generation process from the diagnostic features stored in the diagnostic feature storage unit 23, and proceeds to step S7. In step S7, the reference value generation unit 24 stores the reference value generated in the reference value generation process in the reference value storage unit 25 for each driving scene classified by the driving scene classification unit 21, and ends the flow.

  If sufficient diagnostic features are not stored in the diagnostic feature storage unit 23, Steps S6 and S7 may be omitted until they are stored. In that case, if sufficient diagnostic features are not accumulated, step S5 may be performed, and then step S1 may be performed.

  Returning to FIG. 3, the abnormality diagnosing unit 26 uses the reference value stored in the reference value storage unit 25 and the diagnostic feature extracted in the feature extraction process for the driving data acquired from the driving data collection device 1. Abnormality diagnosis processing for performing abnormality diagnosis such as determination of presence / absence of abnormality and determination of an abnormal part is performed. The abnormality diagnosis unit 26 corresponds to an abnormality determination unit in the claims. For example, the abnormality diagnosis unit 26 performs the abnormality diagnosis process by comparing the above-described identification boundary as a reference value with the diagnosis feature extracted by the feature extraction process.

  In the abnormality diagnosis process, as shown in FIG. 13, the degree of deviation from the identification boundary (that is, the reference value) obtained by the above-described OC-SVM is calculated for the diagnostic feature extracted by the feature extraction process. Then, the calculated degree of deviation is set as the degree of abnormality, and when it exceeds the threshold, it is determined as abnormal. Furthermore, when it is determined that there is an abnormality, it is possible to calculate a diagnostic feature that has a large influence on the abnormality from the relationship between the diagnostic feature and the identification boundary. As an example, if the regression coefficient between the brake master cylinder pressure and the differential value of the brake master cylinder pressure in the diagnostic feature matrix is far from the identification boundary, it is determined that the brake master cylinder is abnormal.

  The abnormality diagnosis unit 26 sets the normal vector to the identification boundary obtained by the OC-SVM that is closest to the diagnostic feature as the departure vector. In the example of the present embodiment, since the matrix of diagnostic features is composed of 6 × 6 = 36 cells, this deviation vector is a 36-dimensional vector. The abnormality diagnosis unit 26 rearranges the 36-dimensional departure vector on a 6 × 6 = 36 cell matrix, thereby calculating a departure degree matrix representing the departure degree of each cell of the diagnosis feature matrix.

  When the abnormality diagnosing unit 26 adopts a configuration that generates a standardized deviation as a reference value in the frequency distribution of diagnostic features as in the first modification, as shown in FIG. The degree of deviation is calculated for each cell of the diagnostic feature matrix by comparison with the identification boundary, and the degree of deviation matrix representing the degree of deviation of each cell is calculated.

  In the present invention, the vehicle abnormality means a state in which the driving data is not normal. Therefore, it can be said that the vehicle abnormality is a sign or a candidate of a vehicle failure in which a problem has occurred in the vehicle. Therefore, the abnormality diagnosis unit 26 may be configured to perform from the diagnosis result in the abnormality diagnosis process to further failure diagnosis.

  Here, the abnormality diagnosis-related processing in the diagnostic apparatus 2 will be described using the flowchart of FIG. The flowchart in FIG. 15 may be configured to start every time new operation data is acquired from the operation data collection device 1, for example. In the flowchart of FIG. 15, the case where the abnormality of the vehicle is determined in real time will be described as an example.

  First, in step S21, the driving scene classification unit 21 performs the above-described discretization process on the driving data arranged in time series obtained from the driving data collection device 1, and the process proceeds to step S22. In step S22, the driving scene classification unit 21 performs the pattern extraction process described above on the symbol string obtained by the discretization process, and the process proceeds to step S23.

  In step S23, the driving scene classification unit 21 identifies the current driving scene by performing segmentation for each driving word in the pattern extraction process, and proceeds to step S24. In step S24, the diagnostic feature extraction unit 22 performs the above-described diagnostic feature extraction processing on the pattern unit driving data corresponding to the current driving scene, and the process proceeds to step S25.

  In step S25, the abnormality diagnosis unit 26 reads the reference value corresponding to the current driving scene from the reference value storage unit 25, and proceeds to step S26. In step S26, the abnormality diagnosis unit 26 performs the above-described abnormality diagnosis process from the diagnostic feature extracted in step S24 and the reference value read in step S26, and the process proceeds to step S27. In step S27, the abnormality diagnosis unit 26 outputs the diagnosis result in the abnormality diagnosis process to the presentation device 3, and the flow ends.

  Returning to FIG. 1, the presentation device 3 presents the diagnostic result input from the diagnostic device 2. As the presentation device 3, for example, a configuration using a display device that displays an image or text, a configuration using an audio output device that outputs audio, or a configuration using both of these may be used.

  According to the configuration of the first embodiment, it is possible to generate a reference value for discriminating abnormalities peculiar to the time series pattern of driving data having a specific history as abnormalities, and to detect more types of abnormalities in the vehicle. It becomes possible to determine that there is an abnormality.

  Here, the effect in this invention is demonstrated concretely using FIG. 16 (a)-FIG.20 (b). First, with reference to FIG. 16A to FIG. 18B, a description will be given of the point where it is possible to determine a vehicle abnormality that is difficult to determine with the conventional technology.

  As an example, it is assumed that an abnormality such that the vehicle speed does not normally respond to the accelerator opening occurs only immediately after departure. However, the values of the accelerator opening and the vehicle speed are values that are normally detected. For example, similar values are detected even during acceleration after deceleration. In order to automatically discriminate this abnormality by the device, the relationship between the accelerator opening and the vehicle speed in the driving scene is normal after extracting the acceleration time immediately after departure and the other acceleration time as different driving scenes. It is necessary to determine whether it is abnormal.

  In the prior art as disclosed in Patent Document 1, since clustering is performed based on the instantaneous value of the driving data, when the driving data value instantaneously indicates the same “accelerating”, The acceleration from the stop state and the acceleration after deceleration cannot be distinguished as different driving scenes (see FIG. 16A). Therefore, the abnormality peculiar to the time series pattern of the operation data having the specific history as described above cannot be determined as abnormal.

  On the other hand, in the configuration of the first embodiment, the driving scenes are classified in units of time-series patterns of driving data having a specific history such as “acceleration after departure” that is acceleration following stoppage. "And" acceleration after deceleration "can be distinguished as different driving scenes (see FIG. 16B). Therefore, it is possible to determine that an abnormality peculiar to the time series pattern of operation data having a specific history as described above is abnormal.

  Another example of a driving scene that cannot be distinguished by the conventional technology but can be distinguished by the configuration of the first embodiment is slow driving that is continued when the vehicle is stopped, as shown in FIGS. 17 (a) and 17 (b). There are driving scenes such as “slowing down” and “slowing down in a traffic jam”, which is slowing down following slowing down. FIG. 17A shows an example of the prior art, and FIG. 17B shows an example of the configuration of the first embodiment. In addition, as shown in FIGS. 18 (a) and 18 (b), “start of right turn” which is a right turn after going straight and “right turn” which is a right turn after sudden right turn. There are driving scenes such as “relaxation”. 18A shows an example of the prior art, and FIG. 18B shows an example of the configuration of the first embodiment.

  Next, the difference between the case where classification is performed by clustering according to the prior art and the case where classification is performed as a time series pattern according to the configuration of the first embodiment will be described with reference to FIGS. 19A and 19B. FIG. 19A shows a conventional example, and FIG. 19B shows an example of the configuration of the first embodiment. Here, a description will be given by taking the vehicle speed and the accelerator opening as examples of driving data.

  FIG. 19A shows the classification of driving scenes by the same clustering as in Patent Document 1 with respect to normal values of driving data detected by an actual vehicle test, and the measurement points of driving data belonging to different driving scenes are marked differently. Is displayed. FIG. 19 (b) classifies driving scenes according to the configuration of the first embodiment with respect to normal values of driving data detected by an actual vehicle test, and displays measurement points of driving data belonging to different driving scenes with different marks. It is a thing. In addition, in the example of Fig.19 (a) and FIG.19 (b), the same driving | operation data are used. However, only measurement points of driving data belonging to driving scenes indicated by ellipses are shown.

  As shown in FIG. 19A, when the classification is performed by the clustering of the prior art, the driving data is classified into two driving scenes. On the other hand, as shown in FIG. 19B, when classified according to the configuration of the first embodiment, the driving data is classified into three driving scenes. The driving scenes classified according to the configuration of the first embodiment are more widely distributed in the direction of the accelerator opening. This is because when the driving data history is considered, the change in the vehicle speed is small with respect to the change in the accelerator opening, and the time series of the driving data changes greatly in the accelerator opening direction in FIG. . In conventional clustering, driving data that does not change continuously in time is classified as the same driving scene. Thus, even if the driving data itself is the same, the configuration of the first embodiment can classify driving scenes in more detail according to actual time-series fluctuations.

  Subsequently, the advantages of the abnormality diagnosis in the configuration of the first embodiment will be described with reference to FIGS. 20 (a) and 20 (b). FIG. 20A shows a conventional example, and FIG. 20B shows an example of the configuration of the first embodiment. Here, a description will be given by taking the vehicle speed and the accelerator opening as examples of driving data. Moreover, the white circle of FIG. 20 has shown the measurement point at the time of abnormality, and the black circle has shown the measurement point at the time of normal.

  First, problems of the prior art will be described. In the prior art as shown in Patent Document 1, a range of normal values is defined for each of a plurality of operation data. That is, the range of the normal value is determined for each of the vehicle speed and the accelerator opening. However, even if each of the vehicle speed and the accelerator opening is within a normal value range as a single unit, an abnormality may occur in the vehicle. In such a case, according to the prior art, as shown in FIG. 20A, the measurement point at the time of abnormality cannot be determined as abnormal.

  On the other hand, according to the configuration of the first embodiment, the relationship between a plurality of operation data is modeled by a regression line, and abnormality is determined by the degree of deviation from the reference value of the regression coefficient. That is, the abnormality is determined based on whether the relationship between the vehicle speed and the accelerator opening is within a normal range. Therefore, as shown in FIG. 20 (b), it becomes possible to determine that the measurement point at the time of abnormality that could not be determined as abnormal in the prior art is also abnormal.

  Furthermore, according to the configuration of the first embodiment, as shown in FIG. 19B, the driving scenes are classified according to the time series pattern while leaving the continuity of the driving data. Therefore, the modeling error can be reduced as compared with the case where the relationship between a plurality of driving data is modeled by the regression line with respect to the driving scene classified by the clustering of the prior art. As a result, the reference value for discriminating between the normal value and the abnormal value is calculated in a narrower range, and the ability to discriminate the abnormality is enhanced.

  Although Embodiment 1 showed the structure by which the driving | operation data collection apparatus 1, the diagnostic apparatus 2, and the presentation apparatus 3 were provided separately, it does not necessarily restrict to this. For example, the driving data collection device 1 and the diagnosis device 2 may be integrated, the diagnosis device 2 and the presentation device 3 may be integrated, or the driving data collection device. 1, diagnosis device 2, and presentation device 3 may be integrated.

(Embodiment 2)
The present invention is not limited to the above-described first embodiment, and the following second embodiment is also included in the technical scope of the present invention. Hereinafter, the second embodiment will be described. For convenience of explanation, members having the same functions as those shown in the drawings used in the description of the above-described embodiment are denoted by the same reference numerals, and description thereof is omitted.

  As shown in FIG. 21, the vehicle diagnostic unit 100a of the second embodiment is the same as the vehicle diagnostic unit 100 of the first embodiment except that the diagnostic device 2 and the presentation device 3 are provided outside the vehicle. The diagnosis device 2 according to the second embodiment may be a server provided in, for example, a dealer handling a vehicle inspection service or a maintenance shop. Therefore, the diagnostic device 2 corresponds to the server in the claims, and the operation data collection device 1 corresponds to the in-vehicle device in the claims.

  Moreover, it is preferable that the diagnostic device 2 in Embodiment 2 is configured to acquire driving data from the driving data collection devices 1 of a plurality of vehicles. As a result, the number of operation data samples used in the above-described reference value generation-related processing can be increased, and the accuracy of the reference value generated in the reference value generation-related processing can be improved. The accuracy of abnormality diagnosis can be improved.

  In the second embodiment, the diagnostic device 2 and the driving data collection device 1 are each provided with a wireless communication device, so that the diagnostic device 2 outside the vehicle receives driving data from the driving data collection device 1 used in the vehicle by wireless communication. What is necessary is just to set it as the structure to acquire. The diagnosis device 2 may be configured to acquire operation data from the operation data collection device 1 by wired communication, or the diagnosis device 2 collects operation data via a vehicle-mounted communication module such as a portable terminal or DCM (data communication module). The operation data may be acquired from the device 1.

  In the second embodiment, for example, the above-described abnormality diagnosis is not performed in real time while the vehicle is running, but a test for inspection is performed at a dealer or a maintenance shop, and is acquired by the operation data collection device 1 during the test. The operation data may be used to perform abnormality diagnosis.

  The driving data acquired by the driving data collection device 1 during normal driving of the vehicle is not limited to the above-described configuration, and the driving data acquired by the diagnostic device 2 when the vehicle is brought into a dealer or a maintenance shop. It is good also as a structure which performs abnormality diagnosis using this. In addition, the diagnosis device 2 may acquire the driving data in real time from the driving data collection device 1 of the vehicle that is normally traveling by using the DCM or the like, and the abnormality diagnosis may be performed using the acquired driving data. .

  In the second embodiment, the configuration in which the presentation device 3 is provided outside the vehicle is shown, but the present invention is not necessarily limited thereto. For example, the presentation device 3 may be used in a vehicle, obtain a diagnosis result from the diagnosis device 2 outside the vehicle by communication via DCM or the like, and present the obtained diagnosis result to a driver of the vehicle.

  Furthermore, in the above-described first and second embodiments, the configuration in which the member that generates the reference value and the member that performs abnormality diagnosis is integrally provided is not necessarily limited thereto. For example, an apparatus (hereinafter referred to as a reference value generation device) including a driving scene classification unit 21, a diagnostic feature extraction unit 22, a diagnostic feature storage unit 23, a reference value generation unit 24, and a reference value storage unit 25, and an abnormality diagnosis unit 26 are provided. It is good also as a structure (modification 3) provided with the apparatus (henceforth abnormality diagnosis apparatus) provided separately. In this case, the reference value generating device corresponds to the vehicle reference value generating device.

  In the configuration of Modification 3, the reference value generation device may be provided outside the vehicle, and the abnormality diagnosis device may be used in the vehicle. The reference value generation device and the abnormality diagnosis device may be configured to transmit and receive information to and from each other by wireless communication, wired communication, or the like. According to this, the reference value generated based on the driving data of a plurality of vehicles by the reference value generation device outside the vehicle is transmitted to the abnormality diagnosis device used in the vehicle, and the abnormality diagnosis device uses the reference value. An abnormality diagnosis can be performed.

  When the reference value generating device and the abnormality diagnosing device are provided separately, the abnormality diagnosing device may be provided with the diagnostic feature extracting unit 22 or extracted by the diagnostic feature extracting unit 22 of the reference value generating device. The abnormality diagnosis device may be configured to acquire the diagnosed feature.

  Also, some of the driving scene classification unit 21, the diagnostic feature extraction unit 22, the diagnostic feature storage unit 23, the reference value generation unit 24, and the reference value storage unit 25 are used in the vehicle, while some are outside the vehicle. It is good also as a structure provided. In this case, the member used in the vehicle and the member used outside the vehicle may be configured to transmit / receive information to / from each other by wireless communication, wired communication, or the like.

  The present invention is not limited to the above-described embodiments, and various modifications can be made within the scope of the claims, and the technical means disclosed in different embodiments can be appropriately combined. Such embodiments are also included in the technical scope of the present invention.

DESCRIPTION OF SYMBOLS 1 Driving data collection apparatus (vehicle reference value generation apparatus) 2 Diagnosis apparatus (vehicle reference value generation apparatus) 11 Data acquisition part (driving data acquisition part) 21 Driving scene classification part (classification part), 22 Diagnostic feature Extraction unit (feature extraction unit), 23 diagnostic feature storage unit (storage unit), 24 reference value generation unit, 26 abnormality diagnosis unit (abnormality determination unit)

Claims (17)

A driving data acquisition unit (11) for sequentially acquiring driving data, including at least one of an operation by a driver of the vehicle, the state of the vehicle, the state of the driver, and the driving environment of the vehicle;
A classification unit (21) for classifying the operation data arranged in time series acquired by the operation data acquisition unit for each operation scene divided into pattern units of change of the operation data;
A feature extraction unit (22) for extracting a pattern feature that is a feature of the operation data in units of the pattern;
An accumulator (23) for accumulating the pattern features extracted by the feature extractor based on the driving scene classified by the classifier;
A reference value generation device for a vehicle, comprising: a reference value generation unit (24) for generating a reference value for determining abnormality of the vehicle from the pattern feature stored in the storage unit. In claim 1,
The reference value generating device for a vehicle, wherein the pattern feature extracted by the feature extraction unit includes a representation of a temporal change in the driving data. In claim 2,
The operation data acquired by the operation data acquisition unit is a plurality of types,
The vehicle reference value generation apparatus, wherein the pattern feature extracted by the feature extraction unit expresses a relationship between a plurality of types of driving data. In claim 3,
The reference value generating device for a vehicle, wherein the pattern feature extracted by the feature extraction unit is a matrix representation of a relationship between a plurality of types of driving data. In any one of Claims 1-4,
An abnormality determination that determines whether the vehicle is abnormal from the reference value generated by the reference value generation unit and the pattern feature extracted by the feature extraction unit for the driving data sequentially acquired by the driving data acquisition unit. A vehicle reference value generation device comprising a unit (26). In claim 5,
The abnormality determination unit compares the reference value generated by the reference value generation unit with the pattern feature extracted by the feature extraction unit for the operation data sequentially acquired by the operation data acquisition unit. A vehicle reference value generation device that calculates a deviation degree of the pattern feature from a value and determines that the vehicle is abnormal when the calculated deviation degree exceeds a threshold value. In claim 6,
The operation data acquired by the operation data acquisition unit is of multiple types,
The pattern feature extracted by the feature extraction unit also has a plurality of types corresponding to the plurality of types of operation data acquired by the operation data acquisition unit,
The abnormality determining unit determines that there is an abnormality in a part of the vehicle corresponding to the pattern feature when the pattern feature having a deviation degree exceeding a threshold value is present among a plurality of types of the pattern features. A vehicle reference value generating device. In any one of Claims 1-7,
The classification unit performs discretization on the operation data arranged in time series, and then performs a double segment structure analysis to extract a pattern in the discrete series obtained by the discretization, thereby performing the pattern unit A reference value generating device for a vehicle, characterized in that it is classified into: In any one of Claims 1-8,
The vehicle reference value generation apparatus, wherein the reference value generation unit generates the reference value based on a frequency distribution of the pattern features stored in the storage unit. In claim 9,
The reference value generating unit generates, as the reference value, an identification boundary learned using a One Class Support Vector Machine for the pattern feature stored in the storage unit. Value generator. In claim 9,
The vehicle reference value generation device, wherein the reference value generation unit generates, as the reference value, a normalized deviation in the frequency distribution of the pattern features stored in the storage unit. In any one of Claims 1-11,
The reference value generation unit generates the reference value based on both the pattern feature of the driving data when the vehicle is normal and the pattern feature of the driving data when the vehicle is abnormal. A vehicle reference value generating device. In claim 12,
The reference value generation unit uses the reference value to identify the identification boundary learned by Support Vector Machine between the pattern feature of the driving data when the vehicle is normal and the pattern feature of the driving data when the vehicle is abnormal. A reference value generating device for a vehicle, characterized in that In any one of Claims 1-13,
The operation data acquisition unit acquires operation information that is operation information by the driver as the operation data,
The classification unit classifies the operation information arranged in time series acquired by the driving data acquisition unit for each driving scene classified into pattern units of change of the operation information, and a vehicle reference value Generator. In claim 14,
The vehicle data reference generating device is characterized in that the driving data acquisition unit acquires information on an operation of an accelerator, a brake, and a steering by the driver as the operation information. In any one of Claims 1-15,
An in-vehicle device (1) used in a vehicle and including at least the operation data acquisition unit;
A server (2) that communicates with the in-vehicle device,
The server includes at least one of the classification unit, the feature extraction unit, the storage unit, and the reference value generation unit,
The vehicle-mounted device includes a vehicle that is not included in the server among the classification unit, the feature extraction unit, the storage unit, and the reference value generation unit. . In any one of Claims 5-15,
An abnormality determination that determines whether the vehicle is abnormal from the reference value generated by the reference value generation unit and the pattern feature extracted by the feature extraction unit for the driving data sequentially acquired by the driving data acquisition unit. Part (26),
An in-vehicle device (1) used in a vehicle and including at least the operation data acquisition unit;
A server (2) that communicates with the in-vehicle device,
The server includes at least one of the classification unit, the feature extraction unit, the storage unit, the reference value generation unit, and the abnormality determination unit,
The vehicle-mounted device includes a vehicle that is not included in the server among the classification unit, the feature extraction unit, the storage unit, the reference value generation unit, and the abnormality determination unit. Standard value generator.

Images