Graphen Artificial Intelligence
Massive Automotive Knowledge Platform to provide Innovative Solutions
Graphen Aica predicts car fixes by looking at sensor data as well as the error code generated by the car computer. There are about 46 sensors in the car and close to 3,000 error codes for powertrain, which makes it a totally of 17 billion combinations of error codes and sensors - a huge challenge to coorelate to car fixes. The conditions of a car can also vary based on engines, manufactures, milage date, and etc, which makes the prediction of car repair extremely difficult with traditional methods.
Combining advanced Machine Learning and Baysian Network, Graphen Aica is capable of analyzing all 17 billion combinations of error codes with high accuracy.
To improve the prediction of the system, several subsystems were implemented to take into consideration of local geo and weather conditions of the car.
Weather was clustered with zip to find relation between issues and geolocation.
For example, it was discovered that there is a correlation of significantly higher likelihood of camshaft, throttle and coolant issues found in cars regularly driving around high hilly areas.
The analysis of local geo and weather conditions help to improve the accuracy of car fix and mainteinance.
Graphen's system successfully achieved the prediction accuracy far beyond expectations. Built with advanced AI to perform advanced tasks including Massive Data Cleaning and Crawling, Domain Knowlwedge Graph Analytics, Machine Learning, Deep Learning, Bayesian Network, etc., the system keeps self-improving. The technology is proven to be ideal and effective for establishing solutions to enter a new market or apply to the ever growing new car models.
The entire fault reasoning graph was created and visualized in Ardi, Graphen's advanced AI foundational platform, utlizing its Database, Reasoning, etc., and feeding the Ardi Machine Learning.
The entire car was divided into subsystems and analyzed where root and intermediate conditions were analyzed. Behavior Analysis was performed based on combining Machine Learning and Bayesian Network.
Car logic of error code prioritizing of each manufacturer was reverse-engineered with a model. As well as getting associated (co-occurring) error codes for a given primary code which helps the AI reason the inner working of the car. 86% of primary error codes can be explained using this emulator where a priority graph of error codes is created.
Built upon Ardi, the system is extremely helpful in continuous monitoring and predictive maintenance as well as care diagnostics and fixes.
