Mechanical failure risks in complex thermal systems lead to catastrophic downtime, which is mitigated by engineering cross-domain attention mechanisms to isolate subtle signal anomalies. This approach transforms raw sensor streams into high-fidelity diagnostic levers for autonomous self-healing.