Bridging Expertise Gaps in Safety-Critical System Monitoring: A User-Centred Design of Adaptive Visualization and Explainable AI for Elevator Systems

Abstract

The operational failure of critical infrastructure like elevators directly impacts public safety and mobility. While modern elevators generate vast amounts of sensor data, translating this data into actionable insights for diverse user groups (from maintenance engineers to building managers), remains a significant challenge in the field of Human-Computer Interaction. Existing monitoring dashboards are often hindered by a one-size-fits-all approach, reactive architectures, and interfaces that assume domain expertise, thereby creating an expertise gap that impedes effective collaboration and situational awareness. Our work is a two-phase, user-centred research project that bridges this gap through an adaptive and intelligent monitoring system. In the first phase, a controlled study (n = 20) established an empirical foundation for visualization complexity, demonstrating that effectiveness is task-dependent, not universally simple or complex. These findings directly informed the second phase: the design and evaluation of a novel system that synergistically combines an adaptive visualization interface with an AI Assistant powered by a unified reasoning engine that provides both data-grounded prognostic reasoning and natural-language explanations. A comprehensive evaluation (n = 97) demonstrated the system's effectiveness. It achieved an outstanding usability score (SUS = 98.69, SD = 5.70), provided equitable performance across technical and non-technical users, and enabled a 100% diagnostic accuracy rate for AI-assisted potential fault diagnosis. Qualitative analysis revealed the AI assistant served as an "accessibility bridge" for non-technical users and a "productivity multiplier" for experts. The primary contribution of this work is a set of six empirically validated design principles for building inclusive safety-critical monitoring systems. This research demonstrates that moving beyond static interfaces to strategically adaptive interfaces, and reframing AI as an explainable collaborative partner through transparent prognostic reasoning, provides a viable path to bridging the expertise gap for multi-level users.