Bearing fault diagnosis supports predictive maintenance by detecting mechanical failures early and reducing downtime in industrial monitoring systems. Recent deep learning models have achieved strong performance in classifying fault types. In practical industrial applications, however, certain training data may need to be removed due to privacy regulations, contractual obligations, or data quality issues. Machine unlearning has therefore become increasingly relevant, although prior work has rarely studied it in bearing fault diagnosis. This paper provides a systematic analysis of existing unlearning methods for fault diagnosis systems and shows their poor performance on time-series vibration data typical of bearing faults. To overcome this limitation, a novel structural embedding preserving unlearning method, Soft-directed Embedding Unlearning (SEU) is proposed, which directly manipulates the embedding space by pushing forget samples away from original class clusters while preserving the structure of retain samples using soft examples near decision boundaries. We use a manifold similarity measure to quantify distance-based separation between forget and retain samples and show that SEU provides a stable approximation of the ideal manifold structure. Experiments on both public and private data show that SEU achieves more effective machine unlearning than the evaluated baselines.
Air taxis are progressing from technological demonstration to early commercialization, but limited research has examined how prospective users translate evaluations of future service attributes into ride acceptance and willingness to pay a premium. Drawing on the cognition, affect, and behavior framework, this study develops a reflective second order model in which expected service quality (ESQ) captures the importance potential users assign to anticipated attributes related to service quality. ESQ is associated with utilitarian, hedonic, and social value, which are subsequently related to ride intention and willingness to pay a premium. Survey data from 623 respondents in China were analyzed using structural equation modeling. The model was also estimated separately for respondents who selected commuting or tourism as their most likely use context. Risk aversion and digital literacy were examined as moderators. ESQ was positively associated with all three value dimensions. Utilitarian and hedonic value were positively associated with both behavioral outcomes, whereas social value was associated with willingness to pay a premium but not positively associated with ride intention. Bootstrap analyses showed significant indirect associations between ESQ and ride intention through utilitarian and hedonic value. ESQ was also indirectly associated with willingness to pay a premium through all three value dimensions. Among respondents who selected commuting, utilitarian value accounted for the largest share of the indirect association with ride intention, whereas hedonic value accounted for the largest share among those who selected tourism. For willingness to pay a premium, hedonic value accounted for the largest share in the commuting group, while social value accounted for the largest share in the tourism group. Moderating effects were limited and varied across contexts. These findings show that experiential and symbolic value relate differently to basic ride acceptance and premium valuation, and that these relationships vary across preferred use contexts.
