ABSTRACT Limited‐sample rolling bearing fault diagnosis is challenging because only a small number of labeled fault samples are available for model training. Local transient impulses and subtle fault‐induced variations can be obscured by background fluctuations and redundant components, while these responses may be further attenuated during feature propagation. To improve the utilization of such fault‐related information, this paper proposes DPFM‐Mamba. Before temporal modeling, the Dual‐Path Feature Modulation module (DPFM) constructs a local trend reference and its corresponding variation component to jointly generate adaptive modulation weights, thereby strengthening insufficiently salient fault‐related responses. The modulated features are then processed by Mamba, whose input‐dependent selective state space mechanism captures temporal dependencies across different feature patches with linear complexity. During temporal downsampling and channel mapping, the Residual‐Preserved Transition Module (RPTM) combines parallel transition and residual paths with channel‐wise adaptive fusion to reduce response attenuation, after which one‐dimensional deformable convolution captures local abnormal responses with temporal shifts. Experiments on the Jiangnan University bearing dataset show that DPFM‐Mamba achieves an average diagnostic accuracy of 93.47% at a 10% training‐sample ratio, outperforming all compared methods.
ABSTRACT Verifying product reliability to ensure that its lifetime meets the warranty period is critical for manufacturers to maintain competitiveness and control warranty costs. Reliability acceptance sampling plans (RASP) assist practitioners in limiting the risk of incorrect decisions while efficiently verifying product reliability. Recent literature has discussed the application of a modified warranty‐oriented lifetime performance index, incorporating accelerated life testing (ALT) and failure‐censoring scheme, to develop a modified RASP (MRASP). However, we note that MRASP decides based on a single sampling inspection of the lot, which may be overly arbitrary. Furthermore, the method requires a relatively large number of test samples to obtain information on product reliability, leading directly to increased testing costs. Therefore, this study develops a novel sampling system, termed QSSR, which combines a quick‐switching sampling decision‐rule switching mechanism with a resubmission‐based dual‐inspection mechanism to address the limitations of the existing MRASP. Through detailed investigation and comparison, we find that the proposed QSSR significantly reduces AFN compared with MRASP, indicating improved AFN‐based test efficiency, and offers superior discriminatory power in assessing lot reliability. Finally, a case study is presented to demonstrate the practical applicability of the proposed method.
ABSTRACT Monitoring time between events and amplitude (TBEA) data is critical in many applications, particularly in manufacturing, reliability engineering, and service operations. Yet, most existing TBEA control charts focus on a single quality characteristic and are typically developed under restrictive distributional assumptions. In practice, each event is characterized by two independent variables: the waiting time since the previous event and the magnitude (amplitude) of the event itself, whose underlying distributions are often unknown. This paper proposes a novel rank‐based one‐sided cumulative sum (CUSUM) control chart for univariate TBEA monitoring, designed to jointly monitor shifts in both the interarrival time and the amplitude. The run‐length properties of the proposed nonparametric CUSUM‐TBEA chart are derived via a Markov chain approximation, enabling computation of in‐control (IC) and out‐of‐control (OOC) performance measures using closed‐form expressions. Monte Carlo simulations confirm the accuracy of the Markov chain results and provide guidance on the required simulation size for reliable performance estimation. Under identical IC conditions, the proposed CUSUM–TBEA chart generally exhibits faster detection and reduced run‐length variability compared to the classical exponentially weighted moving average (EWMA) chart, particularly for moderate to large shifts. The effects of Phase I sample size, unknown shift magnitudes, and correlation are also investigated. The results reveal robust performance across a wide range of monitoring scenarios. An illustrative example involving wildfire occurrence data further highlights the practical applicability of the proposed method. Overall, the results show that the proposed chart is a powerful distribution‐free and a practical tool for monitoring event‐driven processes with concurrent changes in event timing and magnitude.
ABSTRACT Storage reliability plays a critical role in systems that remain dormant throughout their entire life‐cycle, particularly for mission‐critical infrastructure such as radiation detection warning systems and various military defense applications. Typically, operational reliability is not always 1 at the onset of storage, and the exact causes of survival for a parallel system are often unknown. This paper investigates a storage reliability model that considers potential initial failures and masked data from parallel systems. To make full use of inspection and masked data, a measure based on the least squares (LS) method, combined with an Expectation‐Maximization (EM)‐like algorithm, is proposed for parallel systems. An LS‐based parametric estimation procedure is developed to update the testing data and the initial reliability and failure rates of the elements in the parallel system are subsequently estimated using the LS method. The effectiveness of the proposed methods and algorithms is verified through simulations using data from a Weibull lifetime distribution. The findings presented in this paper are expected to facilitate accurate evaluation of production reliability and identification of production quality. The technique for mining masked data from parallel systems is conducive to accurately evaluating product quality and preventing critical system failures.
ABSTRACT In order to enable factor screening in large‐scale mixture‐order‐of‐addition (MOofA) experiments, this paper develops a new methodology. Based on the theory of complete pairwise partitioning, we decompose the full MOofA model into several sub‐models and construct an orthogonal balanced design for these sub‐models. The orthogonal balanced design ensures both mutual orthogonality among order terms within each sub‐model and elimination of confounding between order terms and mixture main effects. This approach effectively reduces the number of experimental runs while enabling efficient and accurate identification of significant factors.
