QMine: A Framework for Mining Quantitative Regular Expressions from System Traces

Abstract

Dynamic behavior of real-time systems and the ability to distinguish between normal and abnormal behavior is critical in safety-critical systems. Temporal patterns define the order of occurrence of events. Temporal properties help draw insights over system specifications. However, given the complexity of modern-day software in cyber-physical systems, the specifications are either not specified or loosely specified. We propose a framework for automating the task of mining temporal specifications from system traces with both events and quantitative values. Our framework, QMine, is an online property mining framework that extracts properties specified in the form of Quantitative Regular Expression (QRE) templates. QMine is shown to be sound and complete. Moreover, we evaluate our framework using real-world industry-standard traces such as Arrhythmia dataset.