Abstract: Owing to the rapid growth of the smart home appliance market, safety risks have become more prominent and require robust assessment methods. Smart appliances integrate hardware, embedded software, connectivity, and user-facing functions within tightly coupled architectures. Interactions between users and devices, along with evolving operating contexts, create diverse and time-varying risk scenarios. Systems theoretic process analysis (STPA) is widely used to identify unsafe control actions and feedback deficiencies, whereas (failure mode and effects analysis, FMEA) is commonly used to evaluate component failure modes and their effects. In practice, the two methods are often applied separately or connected through manual mapping. This separation hampers the integration of evidence from multiple sources and limits traceability across levels, including visualization of risk chains and identification of root causes. This study proposes a knowledge-graph-enhanced fusion framework that integrates STPA and FMEA, denoted as knowledge graph-systems theoretic process analysis-failure mode and effects analysis (KG–STPA–FMEA). A sweeping robot is selected as a representative smart appliance to validate the framework. The framework uses a domain knowledge graph as a semantic layer for unified representation and relational reasoning. The graph organizes information on users, product structure, operating environments, hazards, and accident and recall records. It couples hierarchical control analysis in STPA at the system level with failure mode analysis in FMEA at the component level. Explicit links are maintained among controllers, components, failure modes, hazards, and consequences, thereby enabling coordinated risk identification and quantitative evaluation across layers. Additionally, this design reduces reliance on manual mapping between STPA outputs and FMEA tables. A knowledge graph customized for sweeping robot safety is constructed by integrating heterogeneous sources, including technical documents, standards, user feedback, and accident and recall records. The framework extracts the layered control structure required by STPA and links unsafe control actions to control flaws and feedback issues. In parallel, it identifies component failure modes required by FMEA and maps them to affected functions and physical components. Causal reasoning on the graph links system-level control deficiencies to component failures and downstream hazards and injuries. Evidence is organized as “module–component–failure mode–hazard–injury,” thereby enabling traceable risk-chain queries and visual inspection. For quantitative assessment, the risk priority number (RPN) is extended by introducing coefficients for user impact and environmental impact, yielding an improved risk priority number (IRPN). Expert scoring is combined with fuzzy evaluation to reduce subjectivity and address uncertainty in severity, occurrence, and detectability ratings. Control strategies are derived for prioritized scenarios by strengthening monitoring and feedback. The framework identifies 34 risk scenarios, highlights critical components, and proposes mitigation strategies. In the case study, high-risk scenarios include mechanical entanglement and motor overheating, while key components include the battery system, charging dock, drive motors, and sensors. Results show that knowledge graphs enable the structured integration of safety knowledge from multiple sources and support traceability across levels. KG–STPA–FMEA further reveals systemic risk pathways from control-theory and failure-analysis perspectives. The IRPN, together with fuzzy evaluation, supports risk prioritization that better reflects user behaviors and environmental conditions. This method provides a solid theoretical foundation for improving the safety and risk management of smart home appliances and is applicable to the risk analysis of other intelligent systems and complex devices.