Aspen Mtell

Aspen Mtell is AspenTech’s prescriptive maintenance product, and it is one of the few in this category where the vendor describes a mechanism rather than gesturing at AI. The product is built around agents. Anomaly agents learn normal behaviour and raise an alert when a pattern departs from it; the resulting inspection decides whether that pattern is a genuine failure signature or simply a new normal operating state, and when it is confirmed as a failure a failure agent is created so the same signature is recognised directly in future. AspenTech also describes transferring learned behaviour to other machines of the same type and sensor configuration, and ships industry-specific asset templates to accelerate agent development.

Ownership note: AspenTech is now majority owned by Emerson, and Emerson markets Mtell on its own asset performance pages as part of an integrated APM platform alongside AMS and Aspen Fidelis. The Mtell page itself names integration with Emerson AMS vibration monitoring. The product name has not changed.

Sensor requirement: Mtell is software that learns from data you already have. It does not come with sensors and AspenTech does not require you to buy any. The honest caveat is that the current product page does not state its data prerequisites at all, so you cannot tell from the website how much tag history is needed, at what sample rate, or whether maintenance records are mandatory. Because the method depends on labelled failure events to create failure agents, the quality of your historical maintenance data is likely to decide whether a deployment works, and that question belongs early in the sales process rather than late.

Two capabilities are unusually concrete for this category. First, embedded FMEA: AspenTech states that FMEA automatically prescribes the corrective action, which is what makes the output prescriptive rather than just an alert. Second, work order integration: failure agents are described as sending work orders with failure codes into the EAM system, and the page refers to deep integration with enterprise asset management systems and ERP workflows. For a reliability team, an alert that arrives as a coded work order in the system of record is worth considerably more than one that arrives in a separate dashboard.

What is missing from the public material is equally notable: no deployment statement, no named historian, no OPC or protocol list, no API documentation and no pricing. AspenTech has published accuracy and lead-time figures in older marketing material; those are vendor claims and are deliberately not recorded here.

Fits: process industries with a mature historian and, critically, a usable maintenance history. Refining, chemicals, mining, metals, pulp and paper and upstream oil and gas are the natural fit, and the published case material sits in those sectors. It suits an organisation that already has an EAM system it trusts, because the prescriptive output is designed to land there as a coded work order rather than to replace the EAM. Sites running Emerson AMS vibration monitoring have a named integration path.

Does not fit: plants with thin or unreliable failure history. The agent method needs confirmed failure events to build failure agents, so a site that has never recorded failure codes properly will spend its first phase on data remediation rather than prediction. It is also a poor fit where there is no reliability engineer to adjudicate anomaly alerts, because the method explicitly requires a human inspection decision to separate a real failure signature from a new normal state. Small single-site operations are unlikely to clear the commercial bar.

Typical use cases: rotating equipment such as pumps, compressors and fans; recognising recurring failure signatures across a fleet of like-for-like machines using transferred learning; converting condition alerts into coded EAM work orders; adding a predictive layer to a site that has historian coverage but only calendar-based maintenance.

Buyer caution: ask what proportion of your assets will end up with failure agents rather than anomaly agents only. An estate full of anomaly agents is a better alarm system, not a predictive programme, and the difference is the labelled failure data you can supply.