MS Final Oral Exam: Wandi Xiong, Virtual, 2:00PM

Event
Speaker: 
Wandi Xiong
Monday, May 11, 2020 - 2:00pm
Location: 
Virtual
Event Type: 

Design-time Detection of Physical-Unit Changes in Product Lines

Software product lines evolve over time, both as new products are added to the product line and as existing products are updated. This evolution creates unintended as well as planned changes to systems. A persistent problem is that unintended changes are hard to detect. Often they are not discovered until testing or operations. Late discovery is a problem especially in safety-critical, cyberphysical product lines such as avionics, pacemakers, and smart-braking systems, where unintended changes may lead to accidents.

This thesis proposes an approach and a prototype tool to detect unintended changes earlier in development of a new product in the product line. The capability to detect potentially risky, unintended changes at the design stage is beneficial because repair is easier, less costly, and safer in design than when detection is delayed to testing or operations. 

The Product Line Change Detector (PLCD) introduced here analyzes products’ SysML block and parametric diagrams, which are typical project artifacts for cyber-physical systems, in order to detect problematic, unintended changes. The PLCD software automatically detects potential change-related issues, ranks them in terms of severity using the products’ safety-analysis artifacts, and reports them to developers in a graphical format. Developers select and fix the reported issues with the assistance of the tool’s displays, with the tool recording the fixes and updating the SysML diagrams accordingly. 

The evaluation of PLCD’s performance and capabilities uses three product lines, extended from cyber-physical systems in the literature: NASA astronaut jetpack, vehicle dynamics, and low-earth satellite. The evaluation focuses on unintended changes that cause physical unit inconsistencies, such as between meters and feet, since those may lead to accidents in cyber-physical product lines. The evaluation results show that PLCD successfully detects such unintended changes both in a single product and between products in a software product line.

Committee members: Robyn Lutz (major professor), Carl Chang, Andrew Miner