An Entropy-Based Car Failure Detection Method Based on Data Acquisition Pipeline.

Area of Science:

• Automotive Engineering
• Data Science
• Machine Learning

Background:

• Modern vehicles utilize Electronic Control Units (ECUs) to collect and process diagnostic data from various components.
• Current vehicle systems communicate limited fault information to drivers, potentially overlooking critical issues.
• Advanced data mining techniques offer a way to extract deeper insights from ECU data.

Purpose of the Study:

• To develop an environment for collecting diagnostic data from automotive ECUs.
• To process collected ECU data using parameterized entropies and data mining algorithms.
• To create a classifier capable of identifying hidden vehicle malfunctions.

Main Methods:

• Implementation of a dedicated data collection environment for automotive ECUs.
• Application of parameterized entropies for data analysis.
• Utilization of data mining algorithms for pattern recognition.
• Development of a machine learning classifier for fault detection.

Main Results:

• Successful collection and processing of diagnostic data from ECUs.
• Identification of previously undetected vehicle faults through data mining.
• A trained classifier demonstrated the ability to detect a malfunctioning thermostat without standard equipment alerts.

Conclusions:

• Data mining significantly enhances the diagnostic capabilities of automotive ECUs beyond standard reporting.
• The developed system provides a valuable tool for proactive vehicle maintenance and safety.
• This approach can potentially identify a wider range of subtle component failures.