Hongpeng Yin

In contemporary industrial settings, the loss of collected data is a common issue, frequently caused by sensor degradation, adverse working conditions, electrical interference during motor startup, and network malfunctions. Consequently, the robust imputation of missing time-series data is a crucial component of any data-driven application. However, a significant gap exists as many state-of-the-art imputation methods are noncausal, leveraging future information to interpolate missing values. This characteristic renders them unsuitable for real-time industrial online monitoring, which demands strict adherence to temporal causality. To address this challenge, this article introduces CondVS4, a novel causal generative model specifically designed for online imputation. CondVS4 guarantees causality by performing posterior inference for missing variables conditioned solely on historical and current observations. Through extensive experiments on three real-world industrial datasets under various missingness patterns, CondVS4 demonstrates superior performance compared to a wide range of baseline methods. Furthermore, we show that our model substantially enhances the reliability of a downstream soft-sensing task, proving its effectiveness and practical value for online industrial applications.

Conveyor rollers are essential components in the hot steel milling process, responsible for transporting products between various stages of production throughout the entire process. Due to the development of Industrial Internet of Things (IIoT), conveyors are now equipped with sensors to collect real-time data, facilitating data-driven fault diagnosis models. While many well-trained data-driven diagnostic models initially demonstrate accurate performance, their effectiveness may decline over time due to unexpected changes, such as the emergence of new faults. This causes the prior distribution of fault data to change over time, and we refer to this as prior drift. In this study, we provide a mathematical definition of prior drift within the context of fault diagnosis and then develop a general learning criterion to guide the incremental updating of fault diagnosis models under such conditions. Building on this criterion, we propose an online active kernel learning-based fault diagnosis model (OAKL) for conveyor rollers. OAKL updates the diagnosis model in an incremental manner and adapts to prior drift. Particularly, it leverages selective querying and labeling of the most informative samples to facilitate early detection of novel faults, avoid performance bias, and reduce the number of incremental samples required. A rigorous theoretical analysis is conducted to demonstrate the theoretical validity of OAKL. The experimental results on a practical industrial case demonstrate the efficacy of our approach in fault diagnosis.

Many real-world data are sequentially collected over time and often exhibit skewed class distributions, resulting in imbalanced data streams. While existing approaches have explored several strategies, such as resampling and reweighting, for imbalanced data stream learning, our work distinguishes itself by addressing the imbalance problem through training modification, particularly focusing on gradient descent techniques. We introduce the harmonized gradient descent (HGD) algorithm, which aims to equalize the norms of gradients across different classes. By ensuring the gradient norm balance, HGD mitigates under-fitting for minor classes and achieves balanced online learning. Notably, HGD operates in a streamlined implementation process, requiring no data-buffer, extra parameters, or prior knowledge, making it applicable to any learning models utilizing gradient descent for optimization. Theoretical analysis, based on a few common and mild assumptions, shows that HGD achieves a satisfied sub-linear regret bound. The proposed algorithm are compared with the commonly used online imbalance learning methods under several imbalanced data stream scenarios. Extensive experimental evaluations demonstrate the efficiency and effectiveness of HGD in learning imbalanced data streams.

Remaining useful life (RUL) prediction plays a pivotal role in prognostics and health management (PHM) systems, which enhances the reliability of operating equipment and reduces maintenance costs. With the advent of Industrial Internet of Things (IIoT) technology, it becomes feasible to obtain performance-degradation data that precisely mirrors the health status of equipment. This facilitates real-time process monitoring of device status and promotes intelligent predictive maintenance methods, thereby achieving more accurate RUL estimations. Nonetheless, IIoT systems often suffer from sensing or communication failures in practical industrial scenarios, leading to fragmented multivariate time-series data with missing observations from partial edge devices, which severely restricts the performance of RUL prediction. To address this issue, a graph-based representation learning (RL) method is skillfully proposed for RUL prediction with missing values. Specifically, the observations and features in multivariate time-series are regarded as two distinct nodes types in a semi-bipartite graph. Furthermore, the observed values are viewed as real edges whereas the dependency between timestamps and the correlation between features as virtual edges. In this scheme, the RL for multivariate time-series is expressed as the graph-level tasks. The acquired representations in the observation missing mode possess the capability to extract meaningful information and unveil underlying patterns from fragmented data, which bolsters the performance of RUL estimations through direct imputation or end-to-end architecture. The effectiveness and robustness of the proposed method are validated through the results of comparative experiments under different missing patterns using the C-MAPSS dataset, demonstrating its capability to enhance predictive maintenance in IIoT-enabled systems while reducing unexpected failures and maintenance costs.