Prediction and optimization of pure electric vehicle tire/road structure-borne noise based on knowledge graph and multi-task ResNet

Abstract

Pure electric vehicles (PEVs) offer a significant advantage in their lower interior noise levels compared to traditional combustion engines, making them increasingly popular among consumers. However, the absence of engine noise has brought attention to other sources of noise, particularly tire/road structure-borne (TRS) noise, which significantly impacts the sound quality of the PEV interior and passengers’ psychological comfort. To address this issue, numerous studies have employed simulation analysis, data-driven approaches, and experimental test methods for the prediction and optimization of TRS noise. Nevertheless, the optimization and rectification of TRS noise problems present recurring challenges and difficulties due to the involvement of numerous variables and the weak correlation between TRS noise and vibration. In response to these challenges, this paper introduces a novel approach that combines knowledge graph and multi-task ResNet for predicting and optimizing PEV TRS noise. The proposed method has been experimentally validated and has demonstrated superior effectiveness and robustness compared to three commonly used prediction and optimization methods. The paper’s main contributions are twofold: Firstly, it introduces a novel mechanism that fuses mechanism-driven and data-driven approaches by utilizing a knowledge graph to establish dependency and correlation relationships among TRS entities. This fusion, together with the multi-task ResNet, enhances the interpretability of the solution results while improving computational efficiency. Secondly, it introduces a multi-task ResNet prediction method, named TCGC-ResNet, which incorporates a Task-Conditional Gate Control (TCGC) technique. By explicitly separating shared components from task-common components and implementing a gate control mechanism, TCGC-ResNet enables the gradual extraction and separation of task conditional information knowledge. These advancements contribute to a more comprehensive and effective solution for predicting and optimizing PEV TRS noise.