arXiv:2606.04073v1 Announce Type: new Abstract: This paper proposes a two-stage pseudo anomaly-guided anomaly detection method (\textbf{T}wo-stage \textbf{P}seudo \textbf{A}nomaly-guided \textbf{A}nomaly \textbf{D}etection, \textbf{TPA-AD}) for axle-box bearing time-series anomaly detection (time series anomaly detection, TSAD) under the setting where only normal samples are available for training. The method first generates pseudo-anomalous windows near the normal boundary using a reconstruction model and per-feature target-error control. It then learns anomaly-sensitive representations throu
The paper presents a new method for anomaly detection in time-series data, specifically for machinery like axle-box bearings, addressing a common challenge in industrial AI applications where only normal training data is available.
Improved anomaly detection methods can lead to more robust predictive maintenance, reducing downtime and operational costs in critical infrastructure and industrial settings.
The proposed TPA-AD method offers a more accurate and reliable way to identify anomalies in machinery, potentially enhancing the efficiency and safety of industrial operations.
- · Industrial automation sector
- · Predictive maintenance providers
- · Manufacturers of heavy machinery
- · AI/ML model developers
- · Companies relying on traditional, less efficient anomaly detection methods
Increased operational efficiency and reduced unplanned downtimes due to earlier detection of machinery failures.
Broader adoption of AI-driven predictive maintenance across various industrial sectors, standardizing its implementation.
Consolidation in the industrial AI market as superior anomaly detection techniques become a key differentiator, influencing hardware design.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.LG