Networked Vehicle Anomaly Detection Using Graph Trace Embeddings
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The challenge in networked vehicles is to efficiently and reliably identify software errors during development, especially in a distributed system with limited field data recording and remote updates, where errors can impair vehicle connectivity, necessitating rapid and flexible anomaly detection.
Innovation Solution
A method using graph neural networks and deep neural networks to translate trace data into a graph representation, analyze node similarities and dependencies, and determine software errors through temporal evaluation and classification or regression, leveraging a sliding window approach for efficient and accurate error identification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional manual testing methods are used with test fleets and trained testers, then test coverage can be obtained, but the process is time-consuming and costly, and errors cannot be identified quickly enough in the short development phase
Solution Approach 1:
The patent replaces manual mechanical testing processes with an automated anomaly detection system that uses machine learning models to analyze trace data. The system automatically translates trace lines into graph representations and processes them through neural networks, eliminating the need for manual testing by trained testers and significantly reducing the time required for error detection while maintaining or improving detection accuracy
Solution Approach 2:
The patent introduces an intermediary automated analysis system that acts as a mediator between the generated trace data and the developers. The anomaly detection system processes trace data, identifies anomalies, and provides structured error information to developers, replacing the intermediary role previously filled by manual testers and enabling faster error identification within the constrained development phase
2Reliability
If the separation between development phase and application phase is maintained strictly, then test validation can be performed thoroughly, but flexibility to react to new requirements during service life is reduced
Solution Approach 1:
The patent applies preliminary action by performing anomaly detection on trace data generated during the development phase, enabling early identification of errors before the software is deployed. This allows thorough validation to be completed within the development phase itself, eliminating the need for strict separation between development and application phases while maintaining software quality and enabling flexibility for future updates
Solution Approach 2:
The patent segments the software testing process into distinct automated stages: trace data generation, translation into graph representations, anomaly detection through machine learning models, and error classification. This segmentation enables thorough validation to be completed efficiently within the development phase, allowing the system to maintain high software quality while being flexible enough to accommodate continuous updates and new requirements
3Quantity of substance
If trace data recording in the field is expanded, then more error data can be collected for analysis, but the scope for recording is limited in customer vehicles
Solution Approach 1:
The patent extracts the essential error detection functionality from the vehicle itself and moves it to a centralized server-based anomaly detection system. The vehicle only needs to generate and transmit trace data, while the complex analysis, graph translation, and anomaly detection processes are performed externally on the server, reducing the recording equipment requirements in the vehicle while still enabling comprehensive error analysis
Solution Approach 2:
The patent creates a virtual representation of the vehicle's trace data by translating trace lines into graph representations on the server. This copying approach allows the complex analysis to be performed on replicated data structures without requiring additional physical recording equipment in the vehicle, enabling expanded trace data analysis while maintaining simplicity in the vehicle's hardware configuration
4Adaptability or versatility
If remote software updates are used to install new components, then flexibility is improved, but errors in network connection components prevent reliable updates
Solution Approach 1:
The patent applies preliminary action by detecting anomalies in network connection-related trace data before remote software updates are attempted. The anomaly detection system identifies potential connection errors in advance, allowing developers to fix issues before they prevent successful updates, thereby maintaining both the flexibility of remote updates and the reliability of update delivery
Solution Approach 2:
The patent implements a feedback mechanism where the anomaly detection system continuously monitors trace data for network connection errors and provides information back to developers. This feedback loop enables early identification of connection-related issues that could prevent successful remote updates, allowing corrective actions to be taken before updates fail, thus maintaining both update flexibility and reliability
Data Source
AI summary
Systems, methods, and apparatuses are provided for learning-based anomaly detection to determine a software error in a networked vehicle. Trace lines are translated using a controller of the vehicle. A node list with weighted links is input into a graph neural network. Similarities and dependencies of each node are output as embedded features in a floating-point format with respect to other nodes of the node list in an embedded representation and for each node of the node list. Embedded features of nodes are sorted into a temporal sequence based on a timestamp of each translated trace line. The embedded features of the nodes are augmented with similar embedded features of nodes determined using a distance metric. Similar embedded features are input into a deep neural network together with the embedded features. A time of an error probability and/or error class of an anomaly is output to determine the software error.