3 August 2024 in Knowledge-Based Systems
6 February 2024 in Journal of the Taiwan Institute of Chemical Engineers
22 May 2025 in Chinese Journal of Chemical Engineering
5 March 2025 in IEEE Transactions on Instrumentation and Measurement
Detecting anomalies in multivariate time series (MTS) data is crucial for maintaining the stability of industrial manufacturing processes and biochemical operations. However, current methods often focus on capturing the normal patterns of training data while overlooking the potential of latent representations. This research introduces the Hypersphere Constraint (HSC), an innovative self-supervised model for anomaly detection in MTS. This approach uniquely integrates a one-class classification framework to regulate latent distribution. First, the HSC employs a Temporal Convolutional Network (TCN) and a Multi-layer Perceptron (MLP) to extract latent representations of input data, imposing constraints on the latent distribution to achieve a one-class loss. Second, a self-attention mechanism is applied to reconstruct the input data and calculate the reconstruction loss. Anomalies are identified by combining the one-class loss with the reconstruction loss. By integrating one-class classification with a reconstruction-based method, the HSC significantly increases sensitivity to anomalous data, enhancing the distinction between normal and abnormal data. Evaluations on three real-world datasets and a simulated dataset demonstrate that the HSC model outperforms existing state-of-the-art methods in anomaly detection.
23 December 2025 in Chinese Journal of Chemical Engineering
2 June 2026 in IEEE Transactions on Industrial Informatics
Soft sensing is critical for real-time state awareness and quality control in biomanufacturing processes. However, existing soft sensing methods often underperform because they overlook the intrinsic multistage characteristics of these processes. Moreover, interpretability based on global variable contributions cannot provide actionable guidance for process optimization. To address these issues, we proposed the Bayesian optimization-driven multimetric partitioning with generalized additive models (BOMP-GAM). First, the framework partitions process data into different stages by fusing spatiotemporal and causal similarity through Bayesian optimization. Second, a node-level generalized additive model is used to yield stage-specific predictions and shape function of measured variables. Third, the final predictions were obtained by dynamically weighting the stagewise predictions based on multimetric similarity. Experiments on both simulated and real biomanufacturing processes show that BOMP-GAM achieves state-of-the-art predictive accuracy while uncovering stage-specific shape functions of measured variables. These insights enable targeted operational adjustments and ultimately improve production efficiency.