Authors:
T. Karpagam, S. Tejas

Addresses:
Department of Artificial Intelligence and Data Science, R.M.K. College of Engineering and Technology, Puduvoyal, Thiruvallur, Tamil Nadu, India. Department of Data Science, Analytics and Engineering, Arizona State University, Tempe, Arizona, United States of America.

Abstract:

In this study, researchers focus on a key challenge in advanced manufacturing for the continuous monitoring and predictive maintenance of high-velocity industrial data streams, namely anomaly detection. detection and forecasting of rare failure events. At a conceptual level, the more widely used deep learning approaches often perform poorly in this area due to extreme class imbalance and lack of interpretability within reasonable analytical limits. To overcome this problem, researchers introduce a Neuro-Symbolic Active Learning framework. In many observed contexts, this architecture fuses the pattern recognition power of Temporal Convolutional Networks with the reasoning capacity of a symbolic logic module. In many observed contexts, it employs a predefined collection of physics-based constraints to check neural predictions and prevent false positives., depending on contextual factors Moreover, an uncertainty-based Active Learning loop is implemented to discard only the most uncertain samples for labelling and refine the training process on unlabelled streams., as reflected in earlier discussions This study uses a subset of the UCI SECOM dataset with 441 instances to model rare-event case., to some extent By employing Python and PymTorch researchers experimentally validate that, in terms of precision-recall, this hybrid model consistently outperforms various baselines. The outcomes suggest that the well-posed combination of neural networks and data-driven knowledge, with an active sampling mechanism, is a powerful and interpretable approach for predicting maintenance in Industry 4.0, to some extent.

Keywords: Neuro Symbolic AI; Active Learning; Rare Event Forecasting; Predictive Maintenance; Industrial IoT; Internet of Things; Artificial Intelligence; Recurrent Neural Networks; Gated Recurrent Units.

Received on: 20/04/2025, Revised on: 23/06/2025, Accepted on: 14/08/2025, Published on: 03/03/2026

DOI: 10.69888/FTSNL.2026.000641

FMDB Transactions on Sustainable Neuroscience Letters, 2026 Vol. 1 No. 1, Pages: 13–21

• Views : 33
• Downloads : 9

Download PDF