Apparatus and method for monitoring abnormal state of vehicle using clustering technique

• US 9,075,701 B2
• Filed: 04/30/2013
• Issued: 07/07/2015
• Est. Priority Date: 05/04/2012
• Status: Expired due to Fees

• Current Assignee: Chungbuk National University Industry Academic Cooperation Foundation (Chungbuk National University), Electronics and Telecommunications Research Institute
• Original Assignee: Chungbuk National University Industry Academic Cooperation Foundation (Chungbuk National University), Electronics and Telecommunications Research Institute
• Inventors: Lee, Shin-Kyung, Lee, Jeong-Woo, Kwon, Oh-Cheon, Yun, Un-Il, Pyun, Gwang-Bum, Ryang, Heung-Mo, Lee, Gang-In
• Primary Examiner(s): Jabr, Fadey
• Assistant Examiner(s): HEINLE, COURTNEY D
• Application Number:
  US13/874,028
• Publication Number:
  US 20130297141A1
• Time in Patent Office: 798 Days
• Field of Search: 701/29.1, 701/31.9, 701/32.1
• US Class Current: 1/1
• CPC Class Codes: B60W 50/0205 Diagnosing or detecting failures; Failure detection models
• CPC Class Codes: G06F 17/00 Digital computing or data processing equipment or methods, specially adapted for specific functions information retrieval, database structures or file system structures therefor G06F16/00
• CPC Class Codes: G06F 18/00 Pattern recognition
• CPC Class Codes: G06F 18/23213 with fixed number of clusters, e.g. K-means clustering
• CPC Class Codes: G06F 18/24137 Distances to cluster centroïds

Abstract

The present invention relates to an apparatus and method for monitoring abnormal state of a vehicle. In the method, CAN data collected from an ECU mounted on the vehicle is transformed into coordinates. The coordinates are applied to a distribution map in a specific space, a number of clusters is calculated based on results of the application, and an initial center point corresponding to the number of clusters is selected. Clustering is performed based on the initial center point, and then clusters are generated. At least one piece of data is extracted from each of the clusters, and a state feature of a corresponding cluster is decided on using a difference between maximum and minimum values of attributes constituting the at least one piece of data. A current state of the vehicle is monitored based on current CAN data of the vehicle and state features of the clusters.

Claims

1. A method for monitoring an abnormal state of a vehicle using a clustering technique, comprising:

transforming, using a hardware computing device, Control Area Network (CAN) data collected from an Electronic Control Unit (ECU) mounted on the vehicle, into coordinates on a distribution map in a multidimensional space, in accordance with a range of the coordinates on the distribution map;

analyzing, using the hardware computing device, the distribution map to thereby determine a number of clusters to be formed, and selecting the determined number of initial center points for forming the clusters;

clustering, using the hardware computing device, the coordinates on the distribution map based on the initial center points, to thereby generate the clusters;

extracting, using the hardware computing device, at least one piece of data from each of the clusters, and calculating, for each cluster, a state feature using a difference between maximum and minimum values of an attribute of the at least one piece of data; and

monitoring, using the hardware computing device, a current state of the vehicle based on current CAN data of the vehicle and the state features of the generated clusters, including

finding, using the hardware computing device, one of the generated clusters that has a state feature identical to a state feature of a cluster corresponding to coordinates transformed from the current CAN data; and

determining, using the hardware computing device, whether the state feature of the found cluster corresponds to a normal state,

upon determining that the state feature of the found cluster corresponds to the normal state, terminating the monitoring; and

upon determining that the state feature of the found cluster corresponds to an abnormal state, notifying a vehicle driver of a danger.

2. The method of claim 1, wherein the CAN data of the vehicle is data corresponding to a battery voltage, a sensed value of a coolant temperature sensor, a sensed value of an air flow sensor, a throttle position voltage, a voltage of an accelerator pedal sensor, an air conditioner pressure, a sensed value of an oxygen sensor, air fuel ratio learning control, a sensed value of a barometric pressure sensor, and fuel pressure of the vehicle.

3. The method of claim 1, wherein the clustering the coordinates comprises, for each of the initial center points,

generating an initial cluster by grouping ones of the coordinates located within a preset radius around said each initial center point;

calculating a new center point based on the coordinates constituting the initial cluster;

generating a new cluster by grouping ones of the coordinates located within a preset radius around the new center point; and

calculating a center point of the new cluster, and repeating the generating and calculating steps unless the center point of the new cluster is identical to the new center point.

4. The method of claim 3, wherein the repeating the generating and calculating steps unless the center point of the new cluster is identical to the new center point comprises:

if the center point of the new cluster is identical to the new center point, analyzing distances from the center point of the new cluster to the coordinates constituting the new cluster; and

removing ones of the coordinates corresponding to noise data based on a result of the analysis of the distances, and then converting the remaining coordinates constituting the new cluster into another new cluster.

5. The method of claim 1, wherein the calculating a state feature for each cluster comprises:

deciding on a representative attribute of said each cluster by determining an attribute of which a difference between the maximum and minimum values falls within a preset threshold range; and

deciding on the state feature of the cluster using the representative attribute and information about an abnormal state of the vehicle.

6. The method of claim 1, further comprising, if there is no cluster having the state feature identical to that of the cluster corresponding to the coordinates transformed from the current CAN data of the vehicle:

searching a multidimensional space in which the coordinates transformed from the current CAN data of the vehicle are located for a normal state cluster and an abnormal state cluster located within a preset radius;

analyzing a state of the vehicle based on a distance from a center point of the normal state cluster to a center point of the abnormal state cluster; and

providing a result of the analysis of the vehicle state to the vehicle driver.

7. An apparatus for monitoring an abnormal state of a vehicle using a clustering technique, comprising:

a hardware computing device; and

a storage medium containing program instructions, the execution of which by the hardware computing device causes the apparatus to provide functions of;

a coordinate transformation unit for transforming Control Area Network (CAN) data, collected from an Electronic Control Unit (ECU) mounted on the vehicle, into coordinates on a distribution map in a multidimensional space, in accordance with a range of the coordinates on the distribution map;

a clustering unit for analyzing the distribution map to thereby determine a number of clusters to be generated, and generating the clusters by clustering the coordinates on the distribution map based on initial center points corresponding to the determined number of clusters;

a feature decision unit for extracting at least one piece of data from each of the clusters, and calculating, for each cluster, a state feature using a difference between maximum and minimum values of an attribute of the at least one piece of data; and

a monitoring unit for monitoring a current state of the vehicle based on current CAN data of the vehicle and the state features of the generated clusters, including

finding one of the generated clusters that has a state feature identical to a state feature of a cluster corresponding to coordinates transformed from the current CAN data; and

determining whether the state feature of the found cluster corresponds to a normal state,

upon determining that the state feature of the found cluster corresponds to the normal state, terminating the monitoring; and

upon determining that the state feature of the found cluster corresponds to an abnormal state, notifying a vehicle driver of a danger.

8. The apparatus of claim 7, wherein the clustering unit is configured to generate, for each of the initial center points, an initial cluster by grouping ones of the coordinates located within a preset radius around said each initial center point, calculate a new center point based on the coordinates constituting the initial cluster, and generate clusters by grouping ones of the coordinates located within a preset radius around the new center point.

9. The apparatus of claim 8, wherein the clustering unit calculates a center point of each generated cluster until the center point of the generated cluster is identical to the new center point.

10. The apparatus of claim 7, further comprising an attribute decision unit for extracting the at least one piece of data from each of the clusters, calculating a difference between maximum and minimum values of each attribute of the at least one piece of data, checking whether the calculated difference falls within a preset threshold range, and deciding on a representative attribute of the cluster based on a result of the checking.

11. The apparatus of claim 10, wherein the feature decision unit decides on a state feature of a cluster using the representative attribute of the cluster and information about an abnormal state of the vehicle.

12. The apparatus of claim 7, wherein if there is no cluster having the state feature identical to that of the cluster corresponding to the coordinates transformed from the current CAN data of the vehicle, a multidimensional space in which the coordinates transformed from the current CAN data of the vehicle are located is searched for a normal state cluster and an abnormal state cluster located within a preset radius, a state of the vehicle is analyzed based on a distance from a center point of the normal state cluster to a center point of the abnormal state cluster, and a result of the analysis of the vehicle state is provided to the vehicle driver.