Loose particles inside sealed electronic equipment pose a serious threat to their reliable operation. Particle impact noise detection (PIND) method can identify their presence (shallow information), while obtaining their deep information (i.e., material and location) provides key basis for accurate management and cleaning of loose particles. Current research works focus on the pulses in loose particle signals, constructing feature vectors or spectrograms, and training classification models for loose particle material identification and localization. However, they ignore the complete motion state contained in the entire signal, which is precisely the key to localization. In this study, the authors first proposed and demonstrated the complementarity and applicability of signal perspective and pulse perspective in detecting deep information of loose particles. Specifically, the entire signal provides feedback on the motion process of “contact, bounce suspension, and recontact” of loose particles and the corresponding phase, which is suitable for loose particle localization. The pulse directly carries the contact energy of loose particles, which is suitable for loose particle material identification. On this basis, the authors proposed a deep information detection method for loose particles from signal and pulse perspectives. It systematically compared the classification effect of classification models in material identification and localization tasks, which were trained on datasets and image sets constructed from two perspectives. Finally, experiments validated and determined the optimal solution, i.e., the combination of pulse perspective and spectrogram technology is the optimal way to achieve loose particle material identification, while the combination of signal perspective and feature engineering is the optimal way to achieve loose particle localization. Experimental results in real application scenarios fully confirmed the feasibility, practicality, and superiority of the proposed method, ensuring the reliability of the deep information detection results of loose particles.

Loose particle is an important factor leading to the failure of sealed electronic components, and conducting loose particle detection is crucial. The presence of component signals seriously affects loose particle detection, thus accurately identifying detected signals becomes the key. Existing studies ignored various types of detected signals, and only considered training single-structure classifiers from the vector perspective, resulting in limited classification effect and application ability. Based on this, an integrated model construction method from perspectives of multiple information carriers was proposed. Four types of detected signals in real scenarios were considered for the first time, and techniques such as feature engineering, image processing, and audio processing were used to convert detected signals into vectors, time-domain images, spectrograms, and audios, which belong to the information carrier of vectors, images, and audios. Then, the dataset, time-domain image set, spectrogram image set, and audio set were created. Classifiers or neural networks suitable for vectors and images were trained, parameter optimization was performed to obtain the optimal vector-classifier, time-neural network, and spectrogram-neural network. One neural network suitable for audios was proposed and optimized, and audio-neural network was obtained. On this basis, decision-level fusion technology was referenced, and applicable fuzzy rules and weight calculation formulas were newly proposed, from which an integrated model was constructed. Experiments show that the integrated model achieves a significant classification accuracy of 94.94%. Applications in real scenarios show that the integrated model achieves the highest and most stable classification accuracy of 92.31%. Ablation experiments and robustness validations fully demonstrate the feasibility, practicality, and superiority of the proposed method. This study is an important supplement to existing research on loose particle detection, providing reference for multisource information fusion in similar fields.

Component signal seriously affects the loose particle detection results. The existing research focused on pure loose particle and component signals, training suitable classifiers to classify the data of two labels from two signals. However, in real application scenarios, pure signals rarely appear, and the data classification results are not the required signal recognition or loose particle detection results. The feasibility and practicality of the existing research are limited. In this article, the authors proposed a loose particle detection method based on the recognition of overlapping signals. By obtaining the optimal recognition model and standard confidence probability, the pure and overlapping signals can be accurately recognized, and the loose particle detection can be realized in a comprehensive manner. Multiple detection results in real application scenarios indicated that the obtained overlapping signal recognition and loose particle detection results were stable and reliable. Compared with the existing research, the loose particle detection sensitivity has been significantly improved.

Gaussian processes (GPs) and neural networks (NNs) are two kinds of function approximation models that are widely used for health prognostics. As an emerging meta-learning method, neural processes (NPs) combine the advantages of GPs and NNs. Similar to GPs, NPs can quickly adapt to new observations and can estimate the uncertainty in their prediction. Similar to NNs, NPs are scalable for complex functions and large datasets. However, NPs do not have the capability to handle the prognostics in the time series prevailing in health prognostics. Inspired by these, an NP-based health prognostic toward prognostic uncertainty is presented. The method is validated by datasets that are sourced from real industrial case. Meanwhile, the proposed method is compared with the state-of-the-art NN- and GP-based health prognostic methods, which are also developed for prognostics with uncertainty estimations. The results show that the proposed method performs the prognostics with superior accuracy and quality.

Dependent degradation processes of performance characteristics are ubiquitous in engineered systems. The initial state of each degradation process is usually random and relates to the degradation rate. Unfort unately, most existing Wiener-based bivariate degradation models fail to consider these two features, which may limit the accuracy of system reliability assessment. In order to surmount this limitation, in this article, we develop a new bivariate degradation model, which involves a generalized Wiener process-based marginal degradation model that considers the random initial state and the correlation between the initial state and the degradation rate and a corresponding degradation increment-based dependence structure (DS). The proposed bivariate degradation model and its reliability function are derived first. Then, a two-stage statistical inference is introduced using the maximum likelihood estimation method. Furthermore, a simulation study investigates the performance of the statistical inference and the misspecification effects of DSs and marginal degradation models. Finally, an illustrative example demonstrates the effectiveness of the proposed model.