Work

SGHA: A score-guided heuristic hyperparameter optimization algorithm for real-time performance monitoring requirements of aero engines

Year: 2026

Type: article

Abstract: Context: Aero engine Prognostics and Health Management (PHM) demands strict real-time performance and reliability, but traditional hyperparameter optimization treats parameter combinations as isolated entities and discards valuable historical evaluation knowledge. This leads to excessive computational overhead that severely bottlenecks the practical deployment of PHM systems in real-world aviation scenarios. Objectives: To develop an efficient hyperparameter optimization algorithm that balances search speed and optimization accuracy, thereby addressing the dual requirements of real-time responsiveness and prediction reliability for aero engine PHM systems. Methods: A Score-Guided Heuristic Algorithm (SGHA) is proposed: it constructs a knowledge-reusable framework, establishes independent performance profiles for individual hyperparameters to quantify their contribution to model performance, adopts a targeted rotation optimization strategy to eliminate inter-parameter coupling interference, and employs a score-weighted heuristic sampling mechanism to dynamically adjust the exploration-exploitation trade-off in high-dimensional hyperparameter search spaces. Results: On the C-MAPSS FD004 subset (a standard Remaining Useful Life (RUL) prediction benchmark with variable operating conditions and fault modes), SGHA reduces RMSE by 18.06% compared to GA-Transformer and 23.12% compared to HyperOpt-LSTM. For thrust prediction using ground test data of a certain type of aero engine, it achieves error reductions of approximately 90.2% and 60.7% relative to the baseline LSTM and genetically-optimized GA-LSTM, respectively. Conclusion: SGHA demonstrates superior computational efficiency, optimization precision, and environmental adaptability, providing a practical and reliable parameter optimization solution for mission-critical aero engine real-time performance monitoring and PHM system enhancement.

Source: Array

DOI: 10.1016/j.array.2026.100776

Authors: Mingyang Zhou, Yafeng Shen, Weiqun Fan, Keqiang Miao, Chunyan Hu

Institutions: University of Chinese Academy of Sciences, Chinese Academy of Sciences

Language: English

Topic: Machine Fault Diagnosis Techniques

Subfield: Control and Systems Engineering

Field: Engineering

Domain: Physical Sciences

SDG: Industry, innovation and infrastructure

Open Access status: gold

Estimated APC paid: $1,680