AspenTech

Mtell’s autonomous ML agents each monitor 3-5 correlated sensors around an individual mechanical system, learning normal operating patterns without labelled training data and raising an anomaly flag when their pattern diverges from the norm. Precursor detection window of 2-6 weeks before failure provides sufficient lead time for parts procurement and scheduled maintenance, eliminating both emergency repairs and conservative fixed-interval replacements. Mtell Failure Signature Library captures confirmed failure events and automatically updates the agent’s detection boundary, improving sensitivity over the asset lifetime without human re-training. Integration with SAP PM work order system generates advance planned maintenance orders when Mtell raises a high-confidence failure prediction. Proven across Chevron, Shell, SABIC, and major refining customers with documented 90%+ precision on detected failures.