Jul 2026· International journal of information and communication technology trends· Vol 6, pp. 100-112· 0 citations
TL;DR
An advanced neural network architecture that dynamically adapts to distribution shifts through a continuous domain adaptation mechanism and an adaptive attention module is proposed that significantly outperforms existing state-of-the-art diagnostic models in terms of accuracy, robustness, and generalization capabilities under severe operational transitions.
Abstract
The integration of Internet of Things devices in modern industrial environments has fundamentally transformed paradigms of machine health monitoring and predictive maintenance. However, industrial environments are inherently dynamic, leading to continuous fluctuations in operational conditions such as varying loads, speeds, and environmental temperatures. These fluctuations induce significant nonstationarity in the collected time-series data, rendering conventional assumption of independent and identically distributed data invalid and severely degrading the performance of standard deep learning models for fault diagnosis. This paper presents a comprehensive theoretical and methodological framework for nonstationarity-aware time-series modeling specifically tailored for industrial fault detection and diagnosis. We propose an advanced neural network architecture that dynamically adapts to distribution shifts through a continuous domain adaptation mechanism and an adaptive attention module. By incorporating localized statistical normalization and non-stationary feature alignment, the proposed approach effectively mitigates the adverse effects of concept drift without requiring explicit operational condition labels. Extensive theoretical analysis is provided to formalize the bounds of distribution divergence in degrading machinery. Furthermore, rigorous empirical evaluations on multiple complex industrial datasets demonstrate that the proposed framework significantly outperforms existing state-of-the-art diagnostic models in terms of accuracy, robustness, and generalization capabilities under severe operational transitions. The findings emphasize the critical necessity of embedding nonstationarity awareness directly into the optimization objectives of predictive models, paving the way for more resilient and autonomous industrial health management systems.
Timely anomaly detection in Industrial Internet of Things (IIoT) monitoring requires robust modeling of noisy multivariate sensor streams. Although decomposition-derived residuals provide a potentially useful auxiliary view for multivariate time-series anomaly detection (MTSAD), they are not clean anomaly surrogates in unsupervised settings, as they also contain normal fluctuations, sensor noise, and decomposition artifacts. Therefore, the key challenge is not simply whether residual evidence is useful, but how it can be integrated stably without disturbing the backbone’s native anomaly discrimination process. To address this issue, we propose stable implicit conditioning (SIC), a lightweight and general framework for decomposition-aware unsupervised MTSAD. Instead of reusing residual evidence through explicit downstream intervention, SIC converts compact residual statistics into a bounded sample-dependent channelwise calibration signal for mild representation-level adaptation. When instantiated on the anomaly Transformer (AT), the resulting AT-SIC improves upon the reproduced backbone on four out of five public benchmarks. Additional analyses on decomposition choices, structured disturbances, boundary cases, and cross-backbone transfer show that SIC provides a more reliable evidence utilization path than several intuitive explicit reuse strategies, while introducing only negligible efficiency overhead. These results suggest that decomposition-derived residual evidence is better exploited implicitly than explicitly in this setting.
Guangxia Xu, Zhuo Ye, Xing Huang et al.· IEEE Internet of Things Jour...· 0 citations
Industrial machinery fault detection is a cornerstone of predictive maintenance, directly influencing operational reliability, safety, and production efficiency. Conventional rule-based and machine learning approaches often struggle to handle non-stationary sensor signals, cross-machine variability, and early-stage fault manifestations. This paper proposes a Quantum-Inspired Neuro-Federated Spatio-Temporal Autoencoding Transformer (QNF-STAEFormer) to smartly, scalably and privately diagnose faults in industries. The model incorporates self-adaptive multi-modal sensing, neuromorphic event-driven signal conditioning, physics-directed multi-resolution decomposition, graph-wavelet spatio-temporal encoding, hyperdimensional latent representation learning, and self-supervised predictive modeling. A quantum-inspired reasoning layer facilitates the parallel consideration of various hypotheses on faults and a neuro-federated learning policy supports decentralized collaborative learning at industrial locations. The experimental analysis reveals that the proposed framework has a general fault classification error of 98.4%, F1-score, 98.1%, and AUC, 99.0% which is better than standard CNN, LSTM and transformer-based baselines. The model also has a high early-fault detection ability, where it has a 98.5% detected rate at a 60-minute prediction horizon. These findings prove that the proposed architecture can be used as a strong, precise, and future-proof solution to intelligent monitoring of industrial conditions.
Dr. M. Jayaprakash, P. Sundaram· International Conference Com...· 0 citations
As predictive maintenance transitions from the data-centric paradigm of Industry 4.0 to the sustainable, human-centric framework of Industry 5.0, diagnosing servo motor conditions faces the dual challenges of data scarcity and a profound lack of labeled fault samples. To address this cold-start problem, we present a pseudo-supervised machine learning framework evaluated on a custom five-channel dataset comprising 199 servo motor telemetry samples (current, voltage, temperature, humidity, and vibration). The methodology integrates hard structural partitioning (k-means) and soft posterior confidence estimation (Gaussian Mixture Models) to characterize operating modes without prior annotation. Concurrently, an Isolation Forest model quantifies anomaly intensity and establishes a dynamic quantile-based threshold. A critical innovation of this research is the deterministic risk mapping derived from engineering priors; it defines the "high-risk" (abnormal) state by inversely weighting the physical safety margins of the sensors. This mechanism strictly maps unsupervised clusters to binary pseudo-labels. These labels are subsequently used to supervise downstream discriminators (Random Forest and Support Vector Machine). The final online diagnostic outputs a score-level fusion of the classifier probability and the GMM posterior, gated by the anomaly threshold. Quantitative evaluation demonstrates that the Random Forest model achieved a perfect F1 score of 1.000, while the comparative SVM yielded an F1 score of 0.997, proving the framework to be a robust, interpretable, and highly accurate solution for cold-start industrial health monitoring.
Large-scale spectrum monitoring infrastructures generate high-dimensional spectral time series, providing a critical data foundation for proactive spectrum management, anomaly detection, radio environment awareness, and interference-aware decision-making. In complex electromagnetic environments, real-world deployments are highly nonstationary and frequently affected by unexpected interference, which substantially degrades the predictability of spectrum dynamics and the reliability of downstream spectrum sensing and management systems. Consequently, classical linear forecasting methods and generic deep sequence models often generalize poorly from clean training conditions to interference-corrupted scenarios, as jamming patterns distort the latent representations used for future-spectrum forecasting. This study focuses on multivariate spectrum forecasting, where the objective is to predict multi-step future amplitude or power distributions across all frequency bins from a historical observation window. To address this limitation, we propose DS-SpecIT, a Decomposed Spectral Inverted Transformer for interference-aware spectrum forecasting. Unlike generic long-term forecasting models that mainly minimize average prediction errors, DS-SpecIT is specifically designed to handle structured electromagnetic interference. Its novelty lies in the integration of spectral tokenization, inverted attention over frequency tokens, an interference-aware dual-scale objective, and orthogonality-based latent feature separation. These components enable the model to jointly preserve global spectral trends and reduce local errors inside interference-affected time–frequency regions. Using publicly available spectrum measurements, we establish evaluation protocols under both clean and synthetic-jamming settings. Experiments show that DS-SpecIT maintains competitive clean setting accuracy while achieving stronger global and local robustness under structured interference.
Xi Xiao, Youchen Fan, Ming Lei et al.· Applied Informatics· 0 citations
In modern process industries, data-driven soft sensors have become indispensable for monitoring critical process variables that are inaccessible to direct measurement. Nevertheless, the accurate modeling of industrial processes remains challenging due to their intrinsic complexities, such as time-series behaviors, measurement outliers, redundant variables, and potential concept drift. Existing approaches can address subsets of these challenges but generally lack a unified mechanism that integrates robust offline modeling, variable-importance analysis, and efficient online adaptation. To address these issues, this study proposes an online-updating robust soft sensor framework based on bidirectional long short-term memory (BiLSTM) with integrated gradients (IG) and smoothed quantile loss (SQLoss). During offline modeling, a soft-sensing model is constructed using a BiLSTM, and the proposed SQLoss is introduced to reduce the influence of outliers; the IG method is then employed to evaluate the importance of input variables, enabling input variable selection. During online operation, model parameters associated with significant variables are selectively updated based on IG-derived variable importance, thereby addressing concept drift. Finally, experimental results on an industrial desulfurization process demonstrate that, compared with the best-performing competing basic learner, the proposed SQLoss-BiLSTM-IG reduces the average root mean squared error (RMSE) and mean absolute percentage error by 5.26% and 1.87%, respectively, while increasing the average correlation coefficient by 1.94%; in the online evaluation, the proposed updating strategy achieves a mean RMSE of 2.441, demonstrating its effectiveness in handling concept drift. Moreover, the analysis of key variable importance is consistent with field experience, offering valuable insights for optimizing the desulfurization control system.
Xiuliang Wu, Changchun Pan, Maoyong Cao et al.· Applied System Innovation· 0 citations
The rapid growth of Internet of Things (IoT) systems has generated massive, complex, and highly dynamic time-series data, making anomaly detection essential for system security and operational stability. However, traditional methods based on pointwise representations, statistical thresholds, or simple pairwise associations often struggle to capture complex temporal dependencies and feature interactions in IoT data. To address these challenges, we propose Residual GRU-Attention Anomaly Detector (RGAAD), an unsupervised framework for IoT time-series anomaly detection. RGAAD integrates residual GRU modeling, adaptive self-attention, and gated multi-scale feature fusion to jointly capture temporal dependencies and feature correlations. Extensive experiments on SMD, SWaT, and MSL demonstrate that RGAAD achieves highly competitive performance and consistently outperforms strong baseline methods. These results confirm that explicitly modeling pointwise anomalies and temporal relationships is effective for real-world IoT monitoring.