Vehicle Anomaly Detection With Adaptive Threshold Model Updating
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current predictive maintenance models for vehicle components fail to continuously update anomaly detection thresholds and algorithms with new data, leading to inaccurate predictions and high rates of false positives and negatives, as they rely on outdated averaging processes and single-factor analysis, neglecting variance and stationarity changes in data.
Innovation Solution
Implement a method for continuous health monitoring using machine learning to adapt anomaly detection thresholds and root cause analysis based on aggregated connected data, updating and ranking statistical and machine learning models to select the best performing model, which estimates marginal and conditional probabilities for more accurate predictions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If traditional averaging processes and single-factor analysis are used to establish anomaly detection thresholds, then the model is simpler to implement, but the prediction accuracy decreases and false positive/negative rates increase
Solution Approach 1:
The patent implements dynamic anomaly detection thresholds that continuously adapt to changing data distributions and operational conditions. Instead of static thresholds based on simple averaging, the system uses machine learning models that evolve with incoming data, adjusting thresholds to account for variance and non-stationarity in vehicle component data streams.
Solution Approach 2:
The system transforms the anomaly detection approach by changing from single-factor analysis to multi-factor analysis, incorporating multiple variables and their interactions. The patent modifies detection parameters dynamically based on data characteristics, using probability density functions and conditional probabilities to capture complex relationships between multiple component factors.
2Ease of operation
If fixed detection thresholds are used, then the model is easier to calibrate, but it cannot adapt to new information and latent effects discovered from connected data streams
Solution Approach 1:
The patent implements continuous feedback loops where anomaly detection models are regularly updated with new data from connected vehicle fleets. The system uses feedback from actual component failures and operational data to refine probability density functions, adjust detection thresholds, and improve conditional probability estimates, enabling the model to adapt to newly discovered latent effects while maintaining operational simplicity.
Solution Approach 2:
The system performs preliminary analysis of data distributions and identifies potential latent effects before they become significant anomalies. By proactively updating models with connected data and calculating probability density functions in advance, the system prepares adaptive thresholds that anticipate future anomaly patterns rather than reacting to them after occurrence.
3Power
If simple averaging processes are used to update anomaly detection models, then the updating process is computationally simpler, but it fails to capture variance in the data and provides inadequate learning
Solution Approach 1:
The patent replaces simple mechanical averaging processes with machine learning-based probability density function estimation. Instead of computationally lightweight but unreliable averaging, the system uses statistical modeling and conditional probability calculations that capture data variance and relationships, substituting computational complexity for improved prediction reliability through connected vehicle data learning.
4Adaptability or versatility
If aggregated general models are used for anomaly detection, then the model covers broader conditions, but it loses information on individual root causes of component deterioration
Solution Approach 1:
The patent segments the anomaly detection model into hierarchical levels: general probability density functions that capture broad operational patterns and component-specific conditional probability models that identify individual root causes. This segmentation allows the system to maintain generalized knowledge across vehicle fleets while preserving detailed information about specific deterioration mechanisms through separate cause-effect modeling for different component types.
Data Source
AI summary
Methods and systems are provided for increasing an accuracy of anomaly detection in assets such as vehicle components. In one example, a method provides for continuous health monitoring of connected physical assets, comprising adapting thresholds for anomaly detection and root cause analysis algorithms for the connected assets based on an aggregation of new connected data using machine learning; updating and ranking advanced statistical and machine learning models based on their performance using connected data until confirming a best performing model; and deploying the best performing model to monitor the connected physical assets.