Premium
Challenges in Detecting Emergent Behavior in System Testing
Author(s) -
Kjeldaas Kent Aleksander,
Haugen Rune André,
Syverud Elisabet
Publication year - 2021
Publication title -
incose international symposium
Language(s) - English
Resource type - Journals
ISSN - 2334-5837
DOI - 10.1002/j.2334-5837.2021.00896.x
Subject(s) - anomaly detection , aerospace , computer science , robustness (evolution) , reliability engineering , root cause , integration testing , root cause analysis , risk analysis (engineering) , dependency (uml) , fault detection and isolation , systems engineering , engineering , data mining , artificial intelligence , medicine , biochemistry , chemistry , actuator , software , gene , programming language , aerospace engineering
System integration testing in the defense and aerospace industry is becoming increasingly complex. The long lifetime of the system drives the need for sub‐system modifications throughout the system life cycle. The manufacturer must verify that these modifications do not negatively affect the system's behavior. Hence, an extensive test regime is required to ensure reliability and robustness of the system. System behaviors that emerge from the interaction of sub‐systems can be difficult to pre‐define and capture in a test setup using acceptance criteria. Typical challenges with current test practice include late detection of unwanted system behavior, high cost of repetitive manual processes, and risk of release delays because of late error detection. This paper reviews the state of practice at a case company in the defense and aerospace industry. We use an industry‐as‐laboratory approach to explore the situation in the company. The research identifies the challenges and attempts to quantify the potential gain from improving the current practice. We find that the current dependency on manual analysis generates resources ‐and scheduling constraints and communication issues that hinder efficient detection of system emergent behavior. We explore two approaches to automate anomaly detection of system behavior from test data. The first approach looks at anomaly detection in a top‐down approach to give an indication of the system integrity. The second approach uses anomaly detection on system parts, resulting in the ability to localize the root causes. The work lays the foundation for further research of automated anomaly detection in system testing.