JP7414588B2 - Abnormality diagnosis device - Google Patents

Description

The present invention relates to an abnormality diagnosis device employed in vehicles and the like.

When diagnosing abnormalities in vehicle parts or vehicle functions, vehicle behavior differs depending on the environment in which the vehicle is traveling (e.g., local roads, expressways, road surface conditions, time of day, weather, etc.), so for example, driving data on general roads is used. Even if an attempt is made to diagnose an abnormality in driving data on an expressway using a diagnostic model for vehicle parts that has been learned based solely on the above, the accuracy of the diagnosis will be low.

In order to improve the accuracy of this diagnosis, there is a technique described in Patent Document 1, for example. In Patent Document 1, a diagnostic model necessary for diagnosing each automobile part is stored in a diagnostic model storage unit in association with a target part name and environmental information. In the brake diagnosis model, "deceleration" and "deceleration distance" are set as feature quantities, in the engine diagnosis model, "engine speed" and "engine coolant temperature" are set as feature quantities, and in the battery diagnosis model, "battery voltage" is set as feature quantities. ” and “acceleration” are set as feature quantities.

Japanese Patent Application Publication No. 2019-123351

In Patent Document 1, it is assumed that diagnostic models for each environment have been learned in advance according to the parts, so there is a problem that abnormality diagnosis cannot be performed if there is no applicable diagnostic model. .

An object of the present invention is to provide an abnormality diagnosis device that can generate a new diagnostic model that can be diagnosed even when there is no diagnostic model that meets the conditions of data regarding the movement of a moving object to be diagnosed. be.

The present invention uses pre-learned movement data of a plurality of existing moving objects and operators when unknown conditions (environment, moving object, operator) are included in the data regarding the movement of the moving object to be diagnosed. From among multiple probability distributions that approximate the distribution and their mixture ratios, a new mixture distribution is determined as a predicted distribution by using the one that is close to the conditions to be diagnosed, and a diagnostic model is generated based on the predicted distribution to detect abnormalities. This is to determine the

In order to achieve the above, the present invention provides an abnormality diagnosis device that determines the possibility of an abnormality in a moving object based on movement data of the moving object. Divide the data for each combination of body and operator, extract the data necessary for diagnosis from each divided data, calculate the features of the parts to be diagnosed from the extracted data, and standardize the calculated features. a data extraction unit that extracts standardized feature data for each combination of a moving object and an operator, and a data extraction unit that divides the feature data extracted by the data extraction unit into an arbitrary number of clusters, and a data distribution approximation unit that calculates a mixture ratio based on the total burden rate to each cluster for each combination and generates a probability distribution for each combination of a moving object and an operator; A probability distribution storage unit that stores the probability distribution; a mixture ratio storage unit that stores the mixture ratio calculated by the data distribution approximation unit; and a probability distribution storage unit that stores the mixture ratio calculated by the data distribution approximation unit; A mixture distribution of a combination of a mobile object and an operator to be diagnosed is generated from the probability distribution stored in the storage unit and the mixture ratio stored in the mixture ratio storage unit, and the mixture distribution of the movement data to be diagnosed is combined with the generated mixture distribution. The present invention is characterized by comprising a diagnosis section that compares distributions and determines an abnormality.

According to the present invention, it is possible to provide an abnormality diagnosis device that can generate a new diagnostic model that can be diagnosed even when there is no diagnostic model that meets the conditions of data regarding the movement of a moving object to be diagnosed. can.

Problems, configurations, and effects other than those described above will be made clear by the following description of the embodiments.

1 is a block diagram showing a configuration related to Example 1 of the present invention. FIG. 2 is a block diagram showing a detailed configuration of a data extraction unit 112 in FIG. 1. FIG. FIG. 3 is a diagram showing an example of data regarding Example 1 of the present invention. FIG. 3 is a diagram showing an example of divided data regarding Example 1 of the present invention. FIG. 3 is a diagram showing a data extraction period regarding Example 1 of the present invention. FIG. 3 is a diagram showing an example of a combination of data for a driver and a vehicle regarding Example 1 of the present invention. 7 is a flowchart showing the operation of the data distribution approximation unit according to the first embodiment of the present invention. 7 is a diagram illustrating an example of approximation of feature data to a probability distribution by the data distribution approximation unit 113 according to the embodiment of the invention. FIG. 7 is a diagram illustrating an example of approximation of feature data to a probability distribution by the data distribution approximation unit 113 according to the embodiment of the invention. FIG. 7 is a diagram illustrating an example of approximation of feature data to a probability distribution by the data distribution approximation unit 113 according to the embodiment of the invention. FIG. 7 is a diagram illustrating an example of approximation of feature data to a probability distribution by the data distribution approximation unit 113 according to the embodiment of the invention. FIG. 5 is a diagram illustrating an example of approximation of feature amount data to a mixed distribution by the data distribution approximation unit 113 according to Example 1 of the invention. FIG. 5 is a diagram illustrating an example of approximation of feature amount data to a mixed distribution by the data distribution approximation unit 113 according to Example 1 of the invention. FIG. FIG. 3 is a diagram illustrating an example of a mixture distribution according to a driving environment of a vehicle according to Example 1 of the present invention. FIG. 3 is a diagram illustrating an example of a mixture distribution according to a driving environment of a vehicle according to Example 1 of the present invention. FIG. 3 is a diagram illustrating an example of a mixture distribution according to a driving environment of a vehicle according to Example 1 of the present invention. 7 is a flowchart showing processing in the data distribution approximation unit 113 when training travel data for a combination of a new vehicle and driver or a combination of a vehicle and a new driver is obtained according to the first embodiment of the present invention. FIG. 3 is a diagram showing an example of a probability distribution regarding Example 1 of the present invention. FIG. 3 is a diagram showing an example of a probability distribution regarding Example 1 of the present invention. FIG. 3 is a diagram showing an example of a probability distribution regarding Example 1 of the present invention. FIG. 2 is a block diagram of the diagnostic unit 12 including detailed blocks of the diagnostic model generating unit 122 according to the first embodiment of the present invention. 12 is a flowchart showing operations in a mixture ratio search unit 1222 and a mixture distribution generation unit 1223 regarding Example 1 of the present invention. FIG. 2 is a block diagram showing a detailed configuration of an abnormality degree calculation unit 123 according to Example 1 of the present invention. FIG. 13 is a diagram showing a predicted distribution 1301 when the feature amount in Example 1 of the present invention is one-dimensional. FIG. 3 is a diagram illustrating an example of data points spread over a feature space according to the first embodiment of the present invention. FIG. 3 is a diagram illustrating an example of extracting a plurality of probability distribution components related to Example 1 of the present invention. It is a figure showing an example of mixture distribution regarding Example 1 of the present invention. It is a figure showing an example of mixture distribution regarding Example 1 of the present invention. FIG. 2 is a block diagram showing the overall configuration for executing online diagnosis of a vehicle according to a second embodiment of the present invention.

Embodiments of the present invention will be described below with reference to the accompanying drawings. Similar components are given the same reference numerals, and similar descriptions will not be repeated.

The various components of the present invention do not necessarily have to exist independently, and one component may be made up of multiple members, multiple components may be made of one member, or a certain component may be different from each other. It is allowed that a part of a certain component overlaps with a part of another component, etc.

FIG. 1 is a block diagram showing the configuration of Example 1 of the present invention. FIG. 1 assumes a configuration in which vehicle abnormality diagnosis is performed in an offline state.

In FIG. 1, the abnormality diagnosis device 1 includes a learning section 11, a diagnosis section 12, a probability distribution storage section 13, and a mixture ratio storage section 14. These functions are realized by loading a program stored in an auxiliary storage device such as a hard disk included in the abnormality diagnosis device 1 into a main storage device such as a semiconductor memory, and executing the program by a calculation device such as a CPU. In the following description, such well-known operations will be omitted as appropriate.

The abnormality diagnosis device 1 is a device that determines the possibility of an abnormality regarding a vehicle from data related to the running of the vehicle, and is owned by, for example, a business that provides a service for diagnosing vehicle abnormalities, and communicates with the business that receives the service. Connect via network, etc.

The learning section 11 includes a learning data input section 111, a data extraction section 112, and a data distribution approximation section 113. Learning travel data 2 that has been accumulated in advance and is a combination of multiple vehicles and multiple drivers is input to the learning data input section 111.

The data extraction unit 112 extracts feature quantities necessary for diagnosing the diagnosis target part from the learning travel data 2 input to the learning data input unit 111. The data distribution approximation unit 113 uses the extracted feature values to approximate the feature space with a plurality of probability distributions and a mixture distribution using the probability distributions, stores the probability distributions in the probability distribution storage unit 13, and mixes the mixture ratio of the probability distributions. It is stored in the ratio storage section 14.

The diagnostic section 12 includes a diagnostic data input section 121 , a diagnostic model generation section 122 , an abnormality degree calculation section 123 , and a diagnostic result output section 124 . The acquired diagnostic travel data 3 is input to the diagnostic data input section 121. The diagnostic model generation unit 122 generates a predicted distribution from the plurality of probability distributions stored in the probability distribution storage unit 13 and the mixture ratio for the plurality of probability distributions stored in the mixture ratio storage unit 14, and generates a predicted distribution based on the predicted distribution. output the diagnostic model. The abnormality degree calculation unit 123 calculates the abnormality degree of the vehicle from the diagnostic model and the diagnostic data. The diagnosis result output unit 124 outputs a determination result of vehicle abnormality based on the calculated abnormality degree and past abnormality degree trends.

The learning driving data 2 and the diagnostic driving data 3 are driving data collected from the vehicle in advance. Here, it is preferable that the learning travel data 2 is data in which there is a high possibility that the vehicle is normal, for example, data in which no parts have failed or emergency maintenance has been performed immediately after data acquisition. Furthermore, the diagnostic driving data 3 is not included in the learning driving data 2. The diagnostic driving data 3 and the learning driving data 2 are stored in a storage device installed in the vehicle, and the data are retrieved from this storage device after the driving is completed.

Next, the configuration of the data extraction section 112 will be explained. FIG. 2 is a block diagram showing the detailed configuration of the data extraction unit 112 in FIG. 1.

In FIG. 2, the data extraction unit 112 includes a driver/vehicle data collection unit 1121, a specific driving mode extraction unit 1122, a feature calculation unit 1123, and a standardization processing unit 1124.

The driver/vehicle specific data collection unit 1121 receives the input data input to the learning data input unit 111, and divides the data for each combination of driver and vehicle. An example of input data is shown in FIG.

FIG. 3A is a diagram showing an example of data regarding Example 1 of the present invention. FIG. 3B is a diagram showing an example of divided data regarding Example 1 of the present invention. FIG. 3C is a diagram showing a data extraction period regarding Example 1 of the present invention. FIG. 3D is a diagram illustrating an example of a combination of data on the driver and the vehicle according to the first embodiment of the present invention.

In Figure 3A, the input data includes data that can distinguish a unique driver (driver name, driver ID, etc.), a unique vehicle or unique car model, and unique vehicle specifications (vehicle name, vehicle ID, etc.). (ID, car model name, car model ID, ID for each vehicle specification, etc.), and sensor data. Sensor data is provided by the vehicle or a data collection terminal mounted on the vehicle, and includes time-series acceleration, speed, and position information, as well as vehicle internal data (engine rotation speed, ) and vehicle operation data (brake operation, etc.).

Furthermore, as shown in FIG. 3A, the input data may include driving data of multiple vehicles even if they are the same driver, or may include driving data of multiple drivers even if the vehicle is the same. There is. In the logistics industry, multiple locations each have multiple vehicles and multiple drivers. Drivers have their own unique driving habits, and experienced drivers and new drivers differ in their driving skills, so they operate the vehicle differently. In this embodiment, in order to create a diagnostic model taking into consideration such differences between drivers and distinguish between vehicle and driver combinations, the driver/vehicle data collection unit 1121 collects information about drivers and drivers as shown in FIG. 3B. Data is divided by vehicle combination.

The specific driving mode extraction unit 1122 extracts driving mode data necessary for diagnosing the vehicle component to be diagnosed. The driving modes are basically idling, acceleration, constant speed, and deceleration. During the travel section from when the vehicle departs to the destination, idling, acceleration, constant speed, and deceleration are repeated, resulting in a plurality of travel modes. When diagnosing vehicle parts, necessary driving data is obtained from these driving modes. For example, in diagnosing a brake system, driving data related to deceleration, that is, brake operation, is required among the driving modes, and other driving data are not required. Therefore, as shown in FIG. 3C, the specific driving mode extraction unit 1122 analyzes the driving mode that repeats running and stopping, and converts one driving mode from start to stop into multiple modes such as acceleration, constant speed, and deceleration. The driving data is divided into driving data (divided driving data 1122a), and only the driving data of deceleration, that is, the data of the braking period 1122b, is extracted from each divided driving data 1122a.

The feature amount calculation unit 1123 calculates a feature amount suitable for the vehicle component to be diagnosed from the extracted time series data of the driving mode. Features include statistical values of time-series data (average value, variance value, maximum value, minimum value, etc.), operation time of vehicle parts to be diagnosed (brake operation time, etc.), and operation results (mileage distance when brake operation is performed). etc.) is desirable.

As shown in FIG. 3D, the standardization processing unit 1124 standardizes the feature quantities calculated by the feature quantity calculation unit for each run, and aggregates them for each combination of driver and vehicle.

The diagnostic model according to this embodiment extracts a plurality of probability distribution parts 42, 43, 44 as shown in FIG. 14B from the distribution of each data point 41 spread in the feature space as shown in FIG. 14A, and extracts these parts. By determining the mixture ratio for each combination, a mixture distribution 45 for each combination of driver and vehicle as shown in FIG. 14C or 14D is created and used as a predicted distribution. By using this predicted distribution, when travel data to be diagnosed is obtained, a probability can be obtained based on each data point obtained. If there is an abnormality in the vehicle component to be diagnosed, the distribution of each data point in the predetermined feature space will be different, so the probability at each data point will be low. Utilizing this property, the diagnostic model calculates the degree of abnormality based on the probability of data points based on the predicted distribution. The data extraction unit 112 specifies the driving mode necessary for the vehicle part to be diagnosed, extracts data for each combination of vehicle and driver, calculates the feature quantity necessary for the vehicle part to be diagnosed from the extracted data, and distributes the data. It is transmitted to the approximation unit 113.

The data distribution approximation unit 113 generates a mixture distribution that becomes the above predicted distribution. As a method for generating a mixture distribution, there is a variational Bayesian estimation method, and by applying this algorithm, it is possible to generate a mixture distribution for each combination of driver and vehicle.

Next, an overview of the operation of the data distribution approximation unit when the variational Bayesian estimation method is applied will be explained using FIG. FIG. 4 is a flowchart showing the operation of the data distribution approximation unit according to the first embodiment of the present invention.

In step S101, the D-dimensional feature data extracted by the data extraction unit 112 is divided into M clusters, and the average vector and covariance matrix of each cluster are calculated. In FIG. 3D, the feature amounts for each driver-vehicle combination are aggregated for each run, but in step S101, the driver-vehicle combination is not particularly concerned and is divided into M pieces. Here, D is the number of features used for diagnosis, and M is the maximum number of probability distributions that are parts of the mixture distribution. For each piece of data divided into M pieces, the average vector and covariance matrix are determined. Although the data is divided into M pieces here without considering the data distribution, it may be divided into M pieces by applying a k-means algorithm or the like.

In step S102, it is assumed that the data distribution is a superposition of Gaussian distributions of each cluster, and for each feature data point, the burden rate for each cluster divided in step S101 is calculated. Here, the burden rate indicates the probability that each data point for each combination of driver and vehicle belongs to each cluster.

In step S103, a mixture ratio is calculated based on the sum of the burden rates to each cluster for each combination of driver and vehicle.

In step S104, the average vector and covariance matrix are updated based on the burden rate. As a result, the center of each cluster moves, and the spread of data in each cluster changes.

In step S105, likelihood is calculated.

In step S106, it is determined whether the likelihood has converged, that is, whether the increase in likelihood has stopped. If the increase has stopped, the calculation ends; if it has not, the process returns to step S102, and a new mixture ratio and average vector are determined. , the processing from step S102 to step S105 is repeated until the likelihood converges under the covariance matrix.

Note that in FIG. 4, the mixture ratio of a specific cluster (probability distribution) may be close to 0. In this case, the mixing ratio may be regarded as 0. Furthermore, if the mixing ratio of a specific cluster is 0 for any combination of driver and vehicle, the number of clusters is M', and the mean vector and covariance matrix of the corresponding cluster need not be stored.

The parameters of the probability distribution of each cluster obtained in FIG. 4 are stored in the probability distribution storage unit 13. For example, in the case of Gaussian distribution as described above, the mean vector and covariance matrix are stored as parameters. Further, the mixture ratio calculated for each combination of vehicle and driver is stored in the mixture ratio storage unit 14.

Next, an example of approximating feature data to a probability distribution will be described using FIG. 5. 5A to 5D are diagrams showing an example of approximation of feature data to a probability distribution by the data distribution approximation unit 113 according to the embodiment of the present invention.

FIG. 5A shows D-dimensional feature data points 1131 plotted with respect to feature m and feature n. On the other hand, the data distribution approximation unit 113 generates probability distributions 1132, 1133, and 1134 as shown in FIGS. 5B, 5C, and 5D, respectively. The center of each probability distribution 1132, 1133, 1134 is determined by the average vector determined in FIG. 4, and the degree of spread and direction of the probability distribution is determined by the covariance matrix determined in FIG.

6A and 6B are diagrams showing an example of approximation of feature amount data to a mixed distribution by the data distribution approximation unit 113 according to the first embodiment of the present invention. 6A and 6B, the data distribution approximation unit 113 sets a mixing ratio for each combination of driver and vehicle for the probability distributions 1132, 1133, and 1134 shown in FIGS. 5B, 5C, and 5D, and Mixture distributions 1135 and 1136 are generated based on the mixture ratio, such as FIG. 6A for a driver and vehicle combination and FIG. 6B for another driver and vehicle combination, respectively.

That is, the data distribution approximation unit 113 generates a probability distribution of the feature values of the parts to be diagnosed obtained from the learning travel data 2 of the vehicle and driver combination, and approximates the probability distribution of the combination of the vehicle and driver to be diagnosed. In order to obtain a mixture distribution, the mixture ratio of the vehicle and driver combination to be diagnosed is calculated.

Furthermore, the data distribution approximation unit 113 generates a mixture distribution according to the driving environment of the vehicle. FIGS. 7A to 7C are diagrams showing an example of a mixture distribution according to the vehicle driving environment according to the first embodiment of the present invention. Assume that a prediction distribution as shown in FIG. 7A is generated when a lot of driving environment data has been acquired for a certain driver and vehicle combination. The state of a vehicle changes depending on the environment in which the vehicle is traveling, even if the combination of vehicle and driver is the same. For example, if the environment in which a vehicle travels is different, such as a general road or a highway, the state of the vehicle will change even if the combination of vehicle and driver is the same. Compared to ordinary roads, on expressways, idling time tends to be shorter, acceleration is greater, constant speed time is longer, and deceleration is greater. Furthermore, the amount of steering wheel operation tends to be smaller on expressways than on general roads. In this embodiment, a mixture distribution is also created depending on the environment in which the vehicle runs. For example, a mixture distribution is generated according to the driving environment, such as FIG. 7B on a general road and FIG. 7C on an expressway. Note that the driving environment can be distinguished based on the average speed, location information, road matching information based on location information, regional information, weather information based on the region, etc. The data distribution approximation unit 113 generates a probability distribution and a mixture ratio corresponding to the environment using learning travel data of a combination of a vehicle and a driver. Then, the data distribution approximation unit 113 stores the probability distribution generated for each environment and the mixture ratio for generating the mixture distribution for each environment in the probability distribution storage unit 13 and the mixture ratio storage unit 14, respectively. The environment in which the vehicle travels can be defined by the data extraction ratio of travel data during brake operation, travel data during accelerator operation, travel data during steering wheel operation, etc. in the travel section. For example, if the driving data for a certain section includes almost no brake operations, there is a high possibility that the vehicle is traveling on an expressway. On the other hand, if the brakes are frequently operated, there is a high possibility that the vehicle is traveling in a congested area. In this embodiment, the environment refers to a predetermined range in which the extraction ratio of at least one of the driving data when operating the brake, the driving data when operating the accelerator, and the driving data when operating the steering wheel in the driving section is determined in advance. The driving condition of the vehicle is that the conditions are as follows.

Note that the environment can also be defined using vehicle position information, map information based on the position information, and weather information. For example, the type of road (expressway, general road) can be determined based on the vehicle's location information, the slope of the road the vehicle is traveling on can be estimated from map information, and the weather information (rainfall, snowfall) can be determined based on the location information. It is possible to define the environment by estimating the road surface condition through acquisition.

Next, in a situation where probability distributions and mixture ratios for multiple driver and vehicle combinations have already been learned, learning driving data for a new unlearned vehicle and driver or vehicle and new driver combination is obtained. The processing of the data distribution approximation unit 113 when the data distribution approximation unit 113 is

FIG. 8 is a flowchart showing processing in the data distribution approximation unit 113 when learning travel data for a combination of a new vehicle and driver or a combination of a vehicle and a new driver is obtained according to the first embodiment of the present invention.

In FIG. 8, in step S201, the D-dimensional feature data of a new vehicle and driver or a combination of a vehicle and a new driver extracted by the data extraction unit 112 is divided into MM' clusters, and each Calculate the cluster mean vector and covariance matrix. Here, M' is the final number of clusters in FIG. If M' has a value greater than or equal to M, the value of M may be increased. In addition, as in the case of FIG. 4, here the feature data is divided into M-M' pieces without considering the distribution of feature data due to the combination of a new vehicle and driver or a combination of a vehicle and a new driver, but k- It may be divided into M pieces by applying a means algorithm or the like.

In step S202, the burden rate for each cluster divided in step S201 is calculated for each feature data point. Here, the burden rate indicates the probability that each data point for each combination of driver and vehicle belongs to each cluster.

In step S203, a mixture ratio is calculated based on the total burden rate for each cluster.

In step S204, the mean vector and covariance matrix of the new cluster are updated based on the burden rate. As a result, the center of the new cluster moves, and the spread of data in each cluster changes.

In step S205, likelihood is calculated.

In step S206, it is determined whether the likelihood has converged, that is, whether the increase in likelihood has stopped. If the increase has stopped, the calculation ends; if it has not, the process returns to step S202, and a new mixture ratio and average The processes from step S202 to step S205 are repeated until the likelihood converges under the vector and covariance matrix.

Through the above processing, the mixture ratio can be generated using a learned probability distribution, so even if there is a small amount of driving data for a new vehicle and driver, or a combination of a vehicle and a new driver, it can be performed with high accuracy. A diagnosable diagnostic model can be generated. That is, the data approximation unit in the first embodiment combines a new vehicle and driver, or a vehicle and a new driver, when learning travel data that combines a new vehicle and driver or a vehicle and a new driver is obtained. Learning combining a new vehicle and driver, or a combination of a vehicle and a new driver, from the probability distribution learned from the combined learning travel data and the mixture ratio generated using the probability distribution stored in the probability distribution storage unit 13. Generates a mixed distribution of driving data.

Specific examples are shown in FIGS. 9A to 9C. 9A to 9C are diagrams showing examples of probability distributions regarding Example 1 of the present invention. In FIG. 9A, data points 1137 and learned probability distributions 1132, 1133, and 1134 for a combination of a new vehicle and driver or a combination of a new driver and a new vehicle are shown simultaneously on the feature space. Upon completion of the processing according to FIG. 8, a new cluster corresponding to data point 1137 is created and a corresponding probability distribution 1138 as shown in FIG. 9B is generated. Ultimately, a mixture distribution 1139 for a combination of a new vehicle and driver or a combination of a vehicle and a new driver as shown in FIG. 9C is generated.

Next, the operation of the diagnostic section 12 will be explained using FIG. 10. FIG. 10 is a block diagram of the diagnostic unit 12 including detailed blocks of the diagnostic model generating unit 122 according to the first embodiment of the present invention.

The diagnostic model generation section 122 includes a new driver/vehicle combination detection section 1221, a mixture ratio search section 1222, a mixture distribution generation section 1223, and a mixture distribution acquisition section 1225.

The new driver/vehicle combination detection unit 1221 estimates the driving environment of the corresponding driving data from the diagnostic driving data of the combination of vehicle and driver input to the diagnostic data input unit 121, and also detects the driving environment corresponding to the combination of the driver and the vehicle. It is determined whether the mixture ratio of the calculated predicted distribution is stored in the mixture ratio storage unit 14. If not stored, it is determined that the combination is a new driver and vehicle or a new driver and vehicle, and the process proceeds to the mixture ratio search unit 1222. On the other hand, if it is stored, the process proceeds to the mixture distribution acquisition unit 1225.

The mixture ratio search unit 1222 selects a new driver and vehicle detected by the new driver/vehicle combination detection unit 1221 from among the environment-specific mixture ratios for each driver and vehicle combination stored in the mixture ratio storage unit 14, or If there is a data that matches the driving environment of the data regarding the combination of the driver and the new vehicle, the mixture ratio is obtained in the priority order of vehicle>driver. If there are multiple items with the same priority, it is possible to take the average of the corresponding mixing ratios, or to classify the environmental conditions in more detail to select a matching item. For example, if there are environmental conditions, the mixture ratio is selected based on the priority order of environment, vehicle, and driver.

The mixture distribution generation unit 1223 generates a mixture distribution (predicted distribution) based on the mixture ratio obtained from the mixture ratio search unit 1222 and the corresponding probability distribution stored in the probability distribution storage unit 13.

The mixture distribution acquisition unit 1225 acquires the mixture ratio corresponding to the existing driver and vehicle combination stored in the mixture ratio storage unit 14 and the probability distribution corresponding to the mixture ratio stored in the probability distribution storage unit 13. , generate a mixture distribution (predicted distribution).

FIG. 11 is a flowchart showing operations in the mixture ratio search unit 1222 and mixture distribution generation unit 1223 according to the first embodiment of the present invention.

In step S301, the driving environment condition of data regarding a new driver and vehicle or a new driver and vehicle combination is included in the mixture ratio for each environment for each combination of driver and vehicle stored in the mixture ratio storage unit 14. It is determined whether there is a match, and if there is, a mixture distribution is generated based on the mixture ratio of the corresponding environment in step S302. If there is no such information, the process advances to step S303.

In step S303, among the mixture ratios for each combination of driver and vehicle stored in the mixture ratio storage unit 14, is there any one that matches the vehicle conditions regarding the combination of a new driver and vehicle or a new driver and vehicle? If yes, a mixture distribution is generated based on the mixture ratio of the vehicle in step S304. If there is no such information, the process advances to step S305.

In step S305, among the mixture ratios for each combination of driver and vehicle stored in the mixture ratio storage unit 14, whether there is one that matches the driver conditions regarding the combination of a new driver and vehicle or a combination of a driver and a new vehicle. If yes, a mixture distribution is generated based on the mixture ratio of the corresponding driver in step S306. If there is no probability distribution, a mixture distribution with an equal mixture ratio is generated in step S307.

In the above, when there are a plurality of corresponding mixing ratios, a mixing distribution is generated by taking the average of each mixing ratio.

The abnormality degree calculation unit 123 calculates the abnormality degree from the mixture distribution generated by the mixture distribution generation unit 1223 or the mixture distribution acquisition unit 1225.

FIG. 12 is a block diagram showing the detailed configuration of the abnormality degree calculation unit 123 according to the first embodiment of the present invention. The abnormality degree calculation section 123 includes a specific driving mode extraction section 1231, a feature quantity calculation section 1232, a standardization processing section 1233, and an abnormality degree calculation section 1234.

The specific driving mode extraction section 1231 and the feature amount calculation section 1232 have the same functions as the specific driving mode extraction section 1122 and the feature amount calculation section 1123 in the data extraction section 112 (see FIG. 2) of the learning section 11.

The mixture distribution generated by the mixture distribution generation unit 1223 or the mixture distribution acquisition unit 1225 and the diagnostic travel data 3 are input to the abnormality degree calculation unit 123.

The standardization processing unit 1233 and the feature calculation unit 1232 hold the average value vector and standard deviation vector when standardized by the standardization processing unit 1124 in the data extraction unit 112 of the learning unit 11. Standardize using this value.

The abnormality degree calculation unit 1234 calculates the abnormality degree a using the following equation, using the mixture distribution generated and acquired by the mixture distribution generation unit 1223 or the mixture distribution acquisition unit 1225 as the predicted distribution p(x).

a=-1/(number of diagnostic data points) x (sum of ln p(x) at each diagnostic data point x)
By taking the logarithm of p(x) as described above, data that deviates from the predicted distribution has a low probability and therefore becomes a large negative value. Furthermore, since the sum of ln p(x) at each point x is taken, the larger the amount of data that deviates from the predicted distribution, the larger the negative value. Furthermore, since it is divided by the number of diagnostic data points, the value is the average value of the degree of abnormality per data point, and by adding a negative sign, the degree of abnormality increases as it deviates from the predicted distribution. It has become.

That is, the abnormality degree calculation unit 123 in the diagnosis unit 12 performs diagnosis from the probability distribution stored in the probability distribution storage unit 13 and the mixture ratio stored in the mixture ratio storage unit 14, based on the combination of the vehicle and driver to be diagnosed. A mixture distribution of vehicle and driver combinations is generated, and an abnormality is determined by comparing the generated mixture distribution with the distribution of diagnostic driving data.

This will be explained using a specific example with reference to FIG. FIG. 13 is a diagram showing a predicted distribution 12231 when the feature amount according to the first embodiment of the present invention is one-dimensional.

In FIG. 13, when the value of the feature quantity x is b, the probability p(b) becomes a large value. On the other hand, when the value of the feature x is a, the probability p(a) is a small value. Therefore, according to the above formula for calculating the degree of abnormality, when the diagnostic data includes many points such as point a that deviate from the predicted distribution, the degree of abnormality becomes high.

The diagnosis result output unit 124 outputs a diagnosis result based on the degree of abnormality calculated by the degree of abnormality calculation unit 1234, that is, determines whether or not there is a possibility of abnormality in the vehicle. In this determination, a threshold value is set for the degree of abnormality or the trend of the degree of abnormality, and when the threshold value is exceeded, it is determined to be abnormal.

Note that the diagnostic result output unit 124 not only outputs a determination result of whether the diagnosis is normal or abnormal, but also informs the driver, vehicle owner, or manager of the possibility of abnormality when a determination result of abnormality is obtained. Additional information may be output in order to gain trust. For example, it is conceivable to show a recent trend of the degree of abnormality or a comparison between the average value of the degree of abnormality during normal times and the degree of abnormality at the time of abnormality determination.

Furthermore, it is also possible to indicate the feature amount that had a large influence on the deterioration of the degree of abnormality. For example, when braking distance is selected as a feature when detecting an abnormality in a braking system, if this feature has a large influence on the deterioration of the abnormality, in addition to abnormality determination, "braking distance It is also possible to specifically indicate what kind of abnormality is occurring, such as "It is growing."

According to the abnormality detection device according to the present embodiment, even if the driving data is a combination of a new vehicle and driver, or a combination of a vehicle and a new driver that does not have a learned diagnostic model, the abnormality detection device can detect it from among the learned diagnostic models. Data that are considered to have similar distribution characteristics can be extracted and used for diagnosis.

Furthermore, according to the abnormality detection device according to the present embodiment, even if the amount of driving data from a new vehicle and driver or a combination of a vehicle and a new driver is small, the learned probability distribution and the newly learned probability Since a diagnostic model can be generated using the distribution, highly accurate diagnosis can be performed.

Next, Example 2 of the present invention will be described. In the second embodiment, a vehicle is diagnosed online.

FIG. 15 is a block diagram showing the overall configuration for executing online diagnosis of a vehicle according to the second embodiment of the present invention. Components having the same functions as those shown in FIG. 1 (Embodiment 1) are designated by the same reference numerals, and detailed description thereof will be omitted.

Vehicle 4 is connected to network 6 via communication base station 5 . Travel data of the vehicle 4 is uploaded from the communication base station 5 via the network 6 to the data collection server 7 at specific timing. Here, specific timing refers to periodic timing such as every second, every 10 seconds, or every minute, every time a trip (from the start of a trip to a stop), every time a trip is made from a starting point to a destination, etc. The timing is right.

The data collection server 7 includes a data acquisition section 71, a data transmission section 72, a diagnosis result acquisition section 73, and a diagnosis result transmission section 74. The data acquisition section 71 converts the driving data uploaded from the vehicle 4 into learning driving data. 2 and outputs it to the data transmitter 72 as online data.

The data transmission section 72 transmits the online data received from the data acquisition section 71 to the abnormality diagnosis device 1 .

The diagnosis result acquisition unit 73 acquires the diagnosis result output from the abnormality diagnosis device 1 and outputs it to the diagnosis result transmission unit 74.

The diagnosis result transmitting unit 74 transmits the diagnosis result to the vehicle 4 corresponding to the received diagnosis result, or to the base of the vehicle 4 or the management source of the vehicle 4 (not shown).

In the abnormality diagnosis device 1, the diagnosis section 12 in FIG. 1 is changed to an online diagnosis section 15.

In the online diagnostic section 15, compared to the diagnostic section 12, the operations of the diagnostic data input section and the diagnostic result output section are changed.

Diagnostic data is input to the diagnostic data input section 151 in accordance with the transmission timing from the data transmitting section 72 of the data collection server 7 .

The diagnosis result output unit 152 transmits the determined diagnosis result (presence or absence of possibility of abnormality) to the diagnosis result acquisition unit 73 of the data collection server 7 .

With the above configuration, according to the second embodiment, even when diagnostic data is acquired online, diagnosis can be performed with high accuracy as in the offline case.

In the present invention, we focused on the fact that the distribution of driving data differs depending on the vehicle, the distribution of driving data differs depending on the driver of the vehicle, and the point that the distribution of driving data also differs depending on the driving environment of the vehicle. Embodiment 2 and Embodiment 2 are described, but the present invention is not limited to vehicles, and can also be applied to construction machines, aircraft, and other machines that have equipment and operators and operate under complex environments. That is, any combination of a moving body and an operator who operates the moving body may be used. In the case of a combination of a moving object and an operator, the learning travel data 2 is used as learning travel data, and the diagnostic travel data 3 is used as diagnostic travel data (movement data to be diagnosed). In the case of the above embodiment, the moving object is a vehicle, and the operator is a driver who operates the vehicle.

Note that the present invention is not limited to the above-described embodiments, and includes various modifications. For example, the embodiments described above are described in detail to explain the present invention in an easy-to-understand manner, and the present invention is not necessarily limited to having all the configurations described. Moreover, it is possible to replace a part of a certain embodiment with the structure of another embodiment, and it is also possible to add the structure of another embodiment to the structure of a certain embodiment. Further, it is possible to add, delete, or replace a part of the configuration of each embodiment with other configurations. Further, each of the above-mentioned configurations, functions, processing units, processing means, etc. may be partially or entirely realized by hardware, for example, by designing an integrated circuit. Furthermore, each of the configurations, functions, etc. described above may be realized by software by a processor interpreting and executing programs for realizing the respective functions. Information such as programs, tables, files, etc. that implement each function can be stored in a memory, a recording device such as a hard disk, an SSD (Solid State Drive), or a recording medium such as an IC card, SD card, or DVD.

DESCRIPTION OF SYMBOLS 1... Abnormality diagnosis device, 2... Travel data for learning, 3... Travel data for diagnosis, 4... Vehicle, 5... Communication base station, 6... Network, 7... Data collection server, 11... Learning section, 12... Diagnosis section, 13...Probability distribution storage unit, 14...Mixing ratio storage unit, 15...Online diagnosis unit, 41...Data points, 42, 43, 44...Probability distribution parts, 45...Mixture distribution, 71...Data acquisition unit, 72...Data transmission Part, 73...Diagnosis result acquisition unit, 74...Diagnosis result transmission unit, 102...Data extraction unit, 111...Learning data input unit, 112...Data extraction unit, 113...Data distribution approximation unit, 121...Diagnosis data input unit, 122 ...Diagnostic model generation section, 123... Abnormality calculation section, 124... Diagnosis result output section, 151... Diagnosis data input section, 152... Diagnosis result output section, 1121... Driver/vehicle specific data collection section, 1122... Specific driving mode extraction Part, 1122a... Driving data, 1122b... Braking period, 1123... Feature quantity calculation unit, 1124... Standardization processing unit, 1131... Feature quantity data points, 1132, 1133, 1134... Probability distribution, 1135, 1136... Mixture distribution, 1137... Data points, 1138...Probability distribution, 1139...Mixture distribution, 1221...New driver/vehicle combination detection section, 1222...Mixture ratio search section, 1223...Mixture distribution generation section, 1225...Mixture distribution acquisition section, 1231...Specific driving mode extraction Section, 1232...Feature quantity calculation unit, 1233...Standardization processing unit, 1234...Anomaly degree calculation unit, 12231...Prediction distribution

Claims (5)

An abnormality diagnosis device that determines the possibility of an abnormality in a mobile object from movement data of the mobile object,
Divide the data for each combination of mobile objects and operators from learning movement data that combines multiple moving objects and multiple operators, extract the data necessary for diagnosis from each divided data, and use the extracted data to a data extraction unit that calculates the feature amount of the diagnosis target component, standardizes the calculated feature amount, and extracts standardized feature amount data for each combination of a moving object and an operator;
The feature data extracted by the data extraction unit is divided into an arbitrary number of clusters, and a mixture ratio is calculated based on the sum of the burden rates on each cluster for each combination of a mobile object and an operator. a data distribution approximation unit that generates a probability distribution for each combination of
a probability distribution storage unit that stores the probability distribution generated by the data distribution approximation unit;
a mixture ratio storage unit that stores the mixture ratio calculated by the data distribution approximation unit;
Based on the combination of the moving object and the operator to be diagnosed, the combination of the moving object and the operator to be diagnosed is determined based on the probability distribution stored in the probability distribution storage section and the mixture ratio stored in the mixing ratio storage section. a diagnosis unit that generates a mixture distribution and compares the generated mixture distribution with the distribution of movement data to be diagnosed to determine an abnormality;
An abnormality diagnosis device characterized by comprising: In claim 1,
The abnormality diagnosis device is characterized in that the mobile object is a vehicle, and the operator is a driver of the vehicle. In claim 2,
The data distribution approximation unit generates the probability distribution and the mixture ratio corresponding to the environment using learning movement data combining the vehicle and the driver,
When the mixture ratio of a combination of a vehicle and a driver is not stored in the mixture ratio storage unit in the movement data to be diagnosed, the diagnosis unit determines the environment, vehicle, etc. from among the mixture ratios stored in the mixture ratio storage unit. An abnormality diagnosis device characterized in that the mixture ratio is selected based on the driver's priority order, and a mixture distribution of vehicle and driver combinations corresponding to movement data to be diagnosed is generated. In claim 3,
In the environment, the extraction ratio of at least one of the driving data when operating the brake, the driving data when operating the accelerator, and the driving data when operating the steering wheel in the driving section is within a predetermined range. An abnormality diagnosis device characterized in that the driving condition of a vehicle is a condition of: In claim 3,
The abnormality diagnosis device is characterized in that the environment is conditioned on at least one of road type information based on position information of the vehicle, a road slope estimation result, and a road surface condition estimation result.

Images