Abstract Developing intelligent fault diagnosis (IFD) methods based on deep neural networks has attracted considerable attention and brought successful breakthroughs for industrial applications in recent years. However, the historical IFD methods often concentrate on single diagnosis tasks, such as emerging fault detection or compound fault decoupling (CFD), leading to potential misdiagnosis and missed diagnosis, and also necessitate sufficient labeled data for model training in advance. Aiming to address such problems, a collaborative fault diagnosis framework, named the Open-Set Capsule Network (OpensetCapsNet), is proposed to mitigate the misdiagnosis of both emerging and compound faults. First, the OpensetCapsNet is constructed with four parts: a feature extractor for feature learning, an emerging fault detector for unknown fault detection, a compound fault classifier for intelligent CFD, and a domain discriminator for domain adaptation. Second, a collaborative training strategy is proposed based on adversarial learning and multitask learning. This strategy enables training the OpensetCapsNet with datasets that include both health and known fault samples collected from one working condition, alongside unlabeled samples from varying conditions. Finally, the cross-validation experiments were conducted on an automobile transmission, demonstrating the OpensetCapsNet’s capacity for collaborative execution of diagnosis tasks, encompassing known fault classification, emerging fault detection, and CFD. The proposed framework significantly reduces the risk of misdiagnosis and missed diagnosis, offering a substantial advancement for IFD in industrial applications.
Abstract Industrial equipment poses significant challenges to industrial operations and predictive maintenance due to its high failure rate under complex operating conditions. Existing AI-based diagnostic methods ignore endpoint effects in nonstationary signals and fail to comprehensively quantify the semantic correlation features extracted by the model, resulting in a lack of fine-grained features. To address these issues, a diagnostic framework integrating global dynamic perception and collaborative optimization is proposed, aiming to develop a synergistic optimization framework combining adaptive exponential decay particle swarm optimization with empirical mode decomposition. Furthermore, a multiscale entropy-based attention mechanism is employed to enhance perception of fine-grained feature information, enabling predictive diagnostics. Specifically, by minimizing endpoint fluctuation metrics and rationalizing kurtosis indicators as objective functions, the collaborative optimization framework adaptively adjusts the inertia weight and learning factor combination of the regression model to suppress endpoint effects in signal decomposition. Additionally, a global dynamic perception attention mechanism is proposed by integrating wavelet entropy and sample entropy to construct the extraction of feature semantic association information. Experiments demonstrate that the proposed method achieves an accuracy rate of 99.2% in health diagnosis tasks, showcasing the advantages of AI-based diagnostics.
Abstract Belt conveyor support structures are often exposed to corrosive environments due to dust accumulation, making them vulnerable to corrosion damage and posing significant safety risks. Particularly, the minimum cross-sectional thickness of a lower-chord member is critical for structural safety assessment, because it directly determines the member’s ultimate tensile capacity. This study presents a framework for evaluating the minimum cross-sectional thickness of such members. In this approach, the target member is divided into multiple sections, and the average thickness of each section is estimated using the transitional Markov Chain Monte Carlo (TMCMC) algorithm, based on the member’s cross-sectional modes. A linear regression model is then developed from real corroded members to relate the average cross-sectional thickness to the minimum cross-sectional thickness. Finally, the global minimum cross-sectional thickness is inferred by combining the estimated thicknesses with the regression model. Based on a numerical example, this study investigates several factors that may influence the performance of the TMCMC algorithm. It is found that extending the Markov Chain length significantly improves the identification accuracy and consistency, even when the chain number is limited. This strategy is subsequently validated on five real specimens, including a naturally corroded member. The experimental validations also indicate that the average cross-sectional thicknesses are more identifiable than the section thicknesses, exhibiting lower estimation uncertainty, error, and higher consistency. For the naturally corroded member, the proposed framework effectively estimates the damage level and location of the global minimum cross-sectional thickness, highlighting its practical applicability.
Abstract Marine reinforced concrete (RC) structures suffer from corrosion-induced deterioration caused by the surrounding environment, which dramatically increases structural failure risk and reduces reliability. Although numerous studies have focused on the deterioration modeling and reliability assessment of existing RC structures using inspection data, many of these models still fail to adequately incorporate the spatiotemporal variability inherent in such data. Therefore, this paper investigates the influence of inherent variability in inspection data with different patterns on structural safety evaluation using two proposed spatiotemporal deterioration modeling methods: the data augmentation modeling (AUM) and the direct modeling (DM). It is found that the smoothness parameter of the gamma process and the correlation distance of the Gaussian random field, as well as the correlation among environmental and structural parameters, are not only time-variant but also exhibit spatial variability. The AUM method outperforms the DM method in terms of accuracy and effectiveness in spatiotemporal reliability analysis by better accounting for the spatiotemporal correlation of inspection data. Moreover, generating a spatiotemporal stochastic field of model parameters enhances the adaptability of the deterioration model to inspection data, leading to improved accuracy in reliability assessment.
Abstract Long-term hazard and risk analyses can rely on ensembles of multiyear event sequences to represent uncertainty in hazard occurrence, event timing, and cumulative loss. These ensembles capture temporal dependence and long-horizon loss behavior, but they can be computationally demanding to propagate through hazard, exposure, and loss models when repeated evaluation is required. This paper presents a framework for reducing ensembles of long-term hazard-loss scenarios to a compact, weighted subset while preserving statistical characteristics relevant to engineering risk assessment. The approach operates on multidecade event sequences rather than individual hazard events. The reduction preserves the distribution of cumulative loss, the temporal spacing of events, the annual occurrence rates of hazard events, and the first and second moments of annual loss. These characteristics are computed from the full scenario set and enforced as fidelity targets in the reduced representation. The framework is demonstrated using simulated 30-year hurricane-loss sequences for eastern North Carolina. Results show that a reduced set of representative sequences reproduces the temporal, probabilistic, and spatial characteristics of the full set with limited deviation. Sensitivity analyses examine how accuracy varies with the number of retained sequences and the weighting structure. Although illustrated for hurricanes, the framework is applicable to other hazards characterized by long-term event sequences.
Abstract The identification and control of key risk factors in the coal mine ventilation system is an important part and necessary means in preventing a series of accidents such as gas accumulation, gas outburst, and gas explosion. Risk assessment plays a positive role in eliminating the risks of the ventilation system. To accurately quantify the risks of the coal mine ventilation system and its indicators, this study proposes a new method for analyzing the risks of the coal mine ventilation system by integrating fault tree analysis and fuzzy multistate Bayesian networks. Firstly, a risk indicator system is constructed from the four dimensions of human, machine, environment, and management. Secondly, the risk factors are classified into multiple states, and a fuzzy multistate Bayesian network model is constructed. Finally, the method is validated using the H coal mine as a case study. The results showed that the probability of the coal mine ventilation system being in a high-risk state is 29%. Among the risk factors, work experience and adequacy of safety training are the main causes, and the sensitivity value of safety investment is the highest. The prevention and control of key factors can significantly reduce the risk. Furthermore, through a comparative analysis of multistate and two-state Bayesian networks, it was found that the multistate Bayesian network can better represent the actual risk state. The findings reveal the critical risk factors that require focused attention in coal mine ventilation systems to enhance system safety. These insights provide managers with actionable prevention and control strategies, along with theoretical support, to help effectively reduce the occurrence of coal mine accidents.
