Self-diagnosis for in-vehicle networks

Patent metadata

Field	Value
Publication number	US-12154387-B2
Application number	US-202117396710-A
Country	US
Kind code	B2
Filing date	Aug 8, 2021
Priority date	Aug 7, 2020
Publication date	Nov 26, 2024
Grant date	Nov 26, 2024

Abstract

Methods and systems are provided for fault diagnosis in a vehicular communication network. The methods and systems utilize a trained neural network model which is downloaded to a local computer associated with the vehicular communication network of a given vehicle and which applies inputs from the given vehicle to output maintenance recommendations for the given vehicle.

First claim

The invention claimed is:

1. A method for fault diagnosis in a vehicular communication network, the method comprising:
downloading a neural network trained model to a local computer associated with the vehicular communication network of a first vehicle, the vehicular communication network comprising communication links, each communication link comprising one or more cables, and the trained neural network model having been trained based on: 1) collected diagnostic information from components of the vehicular communication network of a plurality of second vehicles over a period of time; and 2) collected failure information of the components of the vehicular communication network in the plurality of second vehicles over the period of time; and the trained neural network model having generated, based on the collected diagnostic information, a set of health metrics indicating a likelihood of failure of the components to perform in a vehicular communication network in a specified manner;
deploying the trained neural network model by the local computer;
collecting diagnostic information from the components of the vehicular communication network of the first vehicle, the collected diagnostic information including channel quality metrics, including at least one loss metric with respect to a strength of signals transmitted over the cables, and link quality metrics, including a link error rate with respect to data conveyed over the communication links;
for a given link, which comprises a given cable, inputting the collected diagnostic information, including the at least one loss metric with respect to the given cable and the link error rate with respect to the given link, into the trained neural network model;
generating, by the trained neural network model, the set of health metrics to predict a state of health of the components of the vehicular communication network, including at least a link health indication with respect to the given link based on both the at least one loss metric with respect to the given cable and the link error rate with respect to the given link,
wherein the neural network model is trained using an objective function that is a weighted sum of a cross-entropy and a mean-square error with respect to the link health indication and the at least one loss metric; and
outputting maintenance recommendations based on the set of health metrics generated by the trained neural network model.

2. The method of claim 1 , wherein collecting the diagnostic information comprises collecting information from sensors associated with the components of the vehicular communication network in the first vehicle.

3. The method of claim 1 , wherein the collected diagnostic information and the output from the first vehicle are periodically provided to the trained neural network model to further train the trained neural network model.

4. The method of claim 1 , wherein the failure information from one or more of the first vehicle or the plurality of second vehicles is periodically provided to the trained neural network model to further train the trained neural network model.

5. The method of claim 1 , wherein the diagnostic information further comprises at least one input that is selected from a group of inputs including a temperature, a cable length, an accumulated time, a communication speed, a measure of interference, cable type, and chip voltage.

6. The method of claim 1 , wherein the at least one loss metric includes one or more parameters selected from a group of parameters including an overall insertion loss (IL), an overall return loss (RL), a near end RL and a far end RL.

7. The method of claim 1 , wherein the link quality metric of a communication link in the network further includes a signal to noise ratio (SNR) of the link.

8. The method of claim 5 , wherein collecting the diagnostic information for the temperature includes measuring one or more parameters selected from a group of parameters including core temperature and junction temperature of an integrated circuit (IC) chip at one or more locations on the IC chip and ambient temperature.

9. The method of claim 5 , wherein collecting the diagnostic information for the accumulated time includes measuring accumulated power up time and system time.

10. The method of claim 1 , and comprising outputting from the trained neural network model one or more further outputs selected from a group of outputs including a system reliability indication, one or more warnings, a cable fault indication, a cable fault location, and an IC chip fault indication, which are members of the set of health metrics generated by the trained neural network model.

11. The method of claim 10 , wherein the one or more warnings are selected from a group of warnings, which include channel warnings, communication quality warnings, and chip warnings.

12. A system for fault diagnosis in a vehicular communication network comprising:
a memory of a local computer associated with the vehicular communication network of a first vehicle, the vehicular communication network comprising communication links, each communication link comprising one or more cables, the memory configured for storing a neural network trained model:
the neural network trained model having been trained based on: 1) collected diagnostic information from components of the vehicular communication network of a plurality of second vehicles over a period of time; and 2) collected failure information of the components of the vehicular communication network in the plurality of second vehicles over the period of time; and
the neural network trained model having generated, based on the collected diagnostic information, a set of health metrics indicating a likelihood of failure of the components to perform in a vehicular communication network in a specified manner; and

a processor configured to:
deploy the trained neural network model stored in the memory;
collect diagnostic information from the components of the vehicular communication network of the first vehicle, the collected diagnostic information including channel quality metrics, including at least one loss metric with respect to a strength of signals transmitted over the cables, and link quality metrics, including a link error rate with respect to data conveyed over the communication links;
for a given link, which comprises a given cable, input the collected diagnostic information, including the at least one loss metric with respect to the given cable and the link error rate with respect to the given link, into the trained neural network model;
generate, using the trained neural network model, the set of health metrics to predict a state of health of the components of the vehicular communication network, including at least a link health indication with respect to the given link based on both the at least one loss metric with respect to the given cable and the link error rate with respect to the given link,
wherein the neural network model is trained using an objective function that is a weighted sum of a cross-entropy and a mean-square error with respect to the link health indication and the at least one loss metric; and
output maintenance recommendations for the components of the vehicular communication network of the first vehicle, based on the set of health metrics generated by the trained neural network model.

13. The system of claim 12 , wherein the components of the vehicular communication network in the first vehicle comprise one or more sensors from which the diagnostic information is obtained.

14. The system of claim 12 , wherein the components of the vehicular communication network of the first veh

Assignees

• Marvell Asia Pte Ltd

Inventors

• Farjadrad Ramin
• Wu Dance
• Wu Xing
• Dai Shaoan
• Sun Wensheng

Classifications

• G06N3/0499

  Feedforward networks · CPC title

• G06N3/09

  Supervised learning · CPC title

• G07C5/0808

  Diagnosing performance data (testing of vehicles G01M17/00; testing of electrical installation on vehicles G01R31/005) · CPC title

• G06N3/04

  Architecture, e.g. interconnection topology · CPC title

• G07C5/006 Primary

  Indicating maintenance · CPC title