A Post Processing Framework for Analyzing Test Data from Vehicles

Abstract

Today at Volvo Cars, thousands of logs are taken from the test cars containing massive amounts of data, in the Measurement Data Format (MDF). This data should be processed and analyzed. This thesis seeks to design and implement a system that can automatically analyze the vehicles’ logs. There is a lot of prior work in related fields, but we weren’t able to find anything as expressive as this system. Our work focuses on analyzing signal values over time, and correlating multiple simultaneous signals, using a flexible rule-based system based on a custom domainspecific language. This language is designed and developed for this project, and represents most of our contribution. An analysis consists of the user writing a number of rules in this language, representing (for example) "signal X may not remain zero during a consecutive 15-minute period" or "signal X may not be 4 or higher unless signal Y is 6 or higher", and telling the system to run said rules on an MDF file. The language interpreter is written in Python, using a small piece of C++ to interface with an MDF parser. We also implemented a manager process that coordinates the analysis processes, informs the user of the progress, and queues up work if the system has insufficient capacity to run more analysis processes. The system has enjoyed a positive reception at Volvo Cars, exceeding planned problem sizes already in the testing phase; the system could handle the increased load with only small changes, and still offers plenty of potential to increase performance and scalability even further.