A Multisource Hardware Sensing Signal Fusion Network for Robust State Prediction and Anomaly Perception

Area of Science:

• Intelligent Manufacturing
• Edge Computing
• Digital Systems

Background:

• Complex industrial environments generate vast multisource hardware sensing signals.
• Signal heterogeneity, noise, and non-stationarity challenge conventional prediction methods.
• Existing approaches struggle with dynamic relationships and stable prediction under disturbances.

Purpose of the Study:

• To propose a multisource hardware sensing signal fusion network for robust state prediction and anomaly perception.
• To address challenges in intelligent manufacturing and digital production test scenarios.
• To improve the performance of complex digital systems in Hebei Province, China.

Main Methods:

• Uniformly modeled multisource engineering sensing signals (environmental, power, edge, vibration, network, output).
• Developed an end-to-end prediction framework with cross-source signal alignment for temporal coherence.
• Incorporated disturbance-aware residual correction and context-adaptive fusion.

Main Results:

• Achieved superior state prediction performance (MAE: 0.0968, RMSE: 0.1457, MAPE: 8.12%, R2: 0.9416).
• Demonstrated enhanced disturbance robustness with minimal performance drop (Avg. Drop: 28.98%).
• Achieved high accuracy in abnormal-state recognition (Accuracy: 94.32%, F1-score: 93.30%).

Conclusions:

• The proposed method effectively improves state prediction accuracy and disturbance robustness.
• It enhances anomaly warning capabilities in complex industrial and financial-industrial digital systems.
• Provides an effective modeling scheme for AI-driven industrial and financial sensing and decision-making.