Vehicle Sensor Data Replacement for Model Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning devices for vehicles cannot accurately train learning models when abnormalities occur in sensor values used for training data sets, leading to inadequate model training.
Innovation Solution
A machine learning device that utilizes training data sets from another vehicle with matching detection conditions to replace abnormal sensor values, enabling continued model training through vehicle-vehicle communication or server communication, ensuring accurate data usage even with sensor malfunctions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If training data sets include values detected by sensors in the vehicle, then the learning model can be trained with actual vehicle data, but if an abnormality occurs in a sensor, the training accuracy deteriorates
Solution Approach 1:
The patent introduces another vehicle as an intermediary data source when the primary sensor fails. The parameter value acquiring part obtains state parameter values from another vehicle under matching detection conditions, using this external intermediary data to replace abnormal sensor readings and maintain training reliability without requiring complex sensor redundancy within the same vehicle.
Solution Approach 2:
The system prepares backup data acquisition mechanisms in advance by establishing the capability to receive data from other vehicles. The abnormality detection part and parameter value acquiring part are pre-configured to immediately switch to alternative data sources when sensor abnormalities occur, cushioning against training disruptions before they affect the learning model development.
2Adaptability or versatility
If the vehicle performs all training processing locally, then the learning model can be trained with vehicle-specific data, but the processing power and manufacturing costs increase
Solution Approach 1:
The patent divides the training system into distributed components: individual vehicles collect and pre-process their own training data locally, while a server aggregates data from multiple vehicles and performs centralized model training. This segmentation allows vehicle-specific adaptation through local data collection without requiring full processing power at each vehicle, reducing device complexity while maintaining adaptability.
Solution Approach 2:
The system creates a universal training framework where data from multiple vehicles can be combined to train a single learning model that serves all vehicles. The server performs multi-functional processing by aggregating data from various sources, detecting abnormalities, acquiring alternative parameters, and training models that can be deployed across the vehicle fleet, reducing the processing burden on individual vehicles.
Data Source
AI summary
A machine learning device training a learning model unique to a vehicle is provided with: a processor configured to use training data sets including values of state parameters detected by detectors provided at the vehicle, to train the learning model; and if an abnormality occurs in values of a state parameter detected by a detector, acquire values of the state parameter, where an abnormality has occurred, detected by another vehicle under conditions matching detection conditions when the values of the state parameter included in the training data sets were detected by the detector. If an abnormality occurs in values of the state parameter detected by the detector, the training part uses training data sets including values acquired from another vehicle by the parameter value acquiring part, instead of the values of the state parameter where an abnormality has occurred detected by the detector, to train the leaning model.