Percussion-based methods have attracted growing interest in the assessment of bolt looseness. Nevertheless, their suitability for field applications is constrained by the irregularity in manual percussion force. Variabilities in percussion forces can distort the characterization of signals, resulting in an insufficient assessment of bolt looseness. In response to this challenge, the paper introduces a force-adaptive percussion method that utilizes sound phase as a feature, theoretically demonstrating its resilience to percussion force irregularities for the first time. Verification experiments were conducted on a standard beam-column bolted joint. Experimental results showed that phase features of varied percussion signals under identical preload conditions exhibit good consistency, in contrast to the Mel-frequency cepstral coefficients (MFCCs), another prevalent characteristic feature. To assess the effectiveness of the proposed strategy, a residual structure-integrated network was applied for bolt looseness assessment using both phase features and the MFCCs. The results indicated that the model trained with phase features attained higher classification accuracy and superior generalization capability compared to another model trained with MFCCs. These findings substantiated the validity and superiority of the proposed method, indicating its potential to substantially enhance the applicability of field bolt looseness assessment.

基于冲击的方法引起了人们对螺栓松动评估越来越多的兴趣。然而，它们在现场应用的适用性受到手动敲击力不规则性的限制。冲击力的变化会扭曲信号的特征，导致螺栓松动的评估不充分。针对这一挑战，本文提出了一种以声相为特征的力自适应打击方法，首次从理论上证明了其对打击力不规则性的恢复能力。在标准梁柱螺栓连接上进行了验证实验。实验结果表明，与另一个普遍的特征特征梅尔频率倒谱系数（MFCC）相比，相同预载条件下不同打击信号的相位特征表现出良好的一致性。为了评估所提出策略的有效性，利用相位特征和 MFCC，应用残差​​结构集成网络进行螺栓松动评估。结果表明，与使用 MFCC 训练的另一个模型相比，使用相位特征训练的模型获得了更高的分类精度和优异的泛化能力。这些发现证实了所提出方法的有效性和优越性，表明其有可能大大提高现场螺栓松动评估的适用性。