Huang, Qixuan (Beihang University), Fang, Jiayue (Key laboratory of information system and technology，Beiji), Suo, Mingliang (Beihang University), Lyu, Yan (Key laboratory of information system and technology,Beijing inst), Zhang, Weiwei (Beihang University), Wu, Yunlong (Beihang University), Li, Bin (Beihang University), Lian, Zhixuan (Beihang University)

Deduction and Evaluation of Key Parameter Quality Using an Adaptive Threshold-Based Algorithm

Scheduled for presentation during the Invited session "Estimation and Prediction" (FrB1), Friday, November 1, 2024, 10:30−10:50, Room T1

7th IFAC Conference on Engine and Powertrain Control, Simulation and Modeling, Oct 30 - Nov 1, 2024, Dalian, China

This information is tentative and subject to change. Compiled on January 2, 2025

Abstract

The stable operation of electromechanical systems is crucial for ensuring the reliability and safety of equipment groups. However, due to complex working conditions and the interacting influences among multiple components, predicting the quality trends of electromechanical products becomes difficult and costly, posing higher demands on health assessment. Against this backdrop, we propose an adaptive threshold-based quality deduction and evaluation algorithm for key parameters. This algorithm innovatively combines multi-smoothing techniques with Convolutional Neural Network (CNN) models to achieve precise prediction and health assessment of electromechanical product quality trends. Specifically, we introduce the HW-CNN model, which utilizes simple exponential smoothing, Holt's Exponential Smoothing, and the Holt-Winters model to perform multi-level smoothing of raw data, significantly enhancing data stability and accuracy. These smoothing techniques effectively reduce noise interference in the data while preserving key trend features. Meanwhile, the CNN models the smoothed data to extrapolate quality deduction. In the health assessment, the HW-CNN incorporates multiple time-domain indicators and an adaptive threshold determination method, considering the weight of different task indicators for precise health state assessment. The integration of these features and thresholds allows the model to dynamically adjust evaluation criteria, adapting to health state assessments under complex working conditions. Experimental tests have verified the effectiveness of the proposed framework.