The record becomes evidence
The free text is where the diagnosis is. Reading it across every work order changes which assets rank as your worst, and everything downstream of that.
IndustriesAsset reliability
Half your failure codessay “other”.
Your prediction platform is trained on a field your technicians do not fill in. The diagnosis is one column over, in text nothing reads. Nexyron reads it.
Twenty years of investment went into one half of the question.
Asset performance management predicts failure from sensor data, and it does that well. Anomalies, risk ranking, remaining useful life. Solved, well funded, and not what this page is about.
Everything downstream of that prediction is computed from the failure code. Failure codes are entered under time pressure, at the end of a shift, on a form with a dropdown. Ask anybody who has cleaned this data what fraction are absent, defaulted, or set to “other”. In most organisations the honest answer is close to half.
Every statistic your reliability programme produces is computed from a field the workforce does not reliably complete.
The actual diagnosis is one column over. The note saying the seal failed because the housing was misaligned during the previous repair. The engineering assessment from a sister site three years ago that names the same root cause and the remedy that fixed it. Both were written down. Neither is indexed, so neither is knowledge.
Your historian holds signals. Your maintenance system holds records. The explanation is in prose, and the entire category treats prose as an attachment.
What Nexyron does instead
Read the record as evidence, not as a code.
Nexyron takes work orders, inspection findings, engineering assessments, sensor history, asset registers and the network the assets sit in, and derives one connected model across all of it. The free text is read at population scale, in the technicians' own words, and indexed beside the records it describes.
Manufacturer, model, vintage, installer, commissioning crew, supplier batch, duty and environment stop being columns and become relationships. Which is what allows a concentration in a shared attribute to be found by computation rather than by an engineer's hunch.
Embeddings run locally, in-process, on the machine doing the work. No inference call leaves the perimeter to read a maintenance note.
Your asset performance management platform keeps predicting. Nexyron answers the three questions underneath it that nothing in your estate has ever held.
A simulated review
Constructed to show the method. No customer, no benchmark, no performance claim.
- 01
Which assets actually fail the most
Ten years of work orders read as text rather than grouped by code, to find the true bad actors, the assets that keep coming back. A failure mode appearing in 4% of coded records appears in 19% of the described ones, consistently, in the technicians' own words. It was never a dropdown option, so no model has ever been trained on it.
- 02
Has this happened before
The same search across all 210 substations by description rather than code. Eleven events, four sites, nine years. Two carry an engineering assessment with a root cause and a remedy. One of those sits in a document store the maintenance system does not index.
The organisation solved this in 2019 and forgot.
- 03
Is it a fleet problem
The affected units against manufacturer, model, vintage, installer, commissioning crew, supplier batch, duty and environment. Failures concentrate in one vintage, one manufacturer, one contractor and an eighteen-month window. The same model elsewhere does not show it.
- 04
What else goes down
For each candidate unit, what loses supply, which are the only path to a downstream section, which pairs cannot be worked in the same window. Two units near the top of the risk list are the only feed to a hospital.
- 05
Did pulling work forward help
Last year 180 units were brought forward and failures fell. The units chosen were not chosen at random, so comparing them to the rest compares two populations that were never alike. For 119, comparable units exist and the effect is estimated. For 61 they do not, and Nexyron says so.
- 06
Were we warned
For every unit that failed, which inspection findings were open at the time. In two cases a recommendation naming the eventual failure mode had been open for more than a year and was closed administratively when the work order was rescheduled.
Five things, from data already sitting in your estate.
Institutional memory becomes reachable
“Has this happened before and what fixed it” is the most common question in reliability engineering and the worst served in the entire toolchain.
Fleet causes become visible
Manufacturer, vintage, installer, crew and batch are relationships. Concentration in a shared attribute turns an interval argument into a targeted campaign.
Criticality follows the network, not the register
What else loses supply is a path question over the topology, answered in the same query as the risk score.
Open findings at failure become auditable
After a serious incident this is asked with urgency, and it should not take three weeks and four people.
The dimension nobody in this category occupies
Did the maintenance we did prevent anything?
Read the public material of every major platform in this field. Detection, prediction, scoring, prioritisation, work management. Not one of them describes measuring whether a maintenance action changed the outcome.
There is a reason. A failure that was prevented leaves no trace and there is no event to count, so answering it means constructing what would have happened otherwise. That is a different discipline from prediction and a different piece of mathematics.
Nexyron carries 42 causal and decision procedures on the same graph the assets live in. Uplift instead of association, overlap tested before a number is believed, policies and scenarios compared against each other before anybody commits a budget.
Every reliability budget in the world is defended annually with anecdote. This is the instrument that ends that, including where it has to answer that the data cannot support a claim.
Running it
Inside the perimeter, including air-gapped.
Operational technology environments are conservative, segregated and frequently air-gapped by design, and the frameworks governing critical infrastructure hold the operating entity responsible for whatever runs inside its perimeter.
Nexyron installs as an application and runs the whole engine locally. No telemetry leaves, no cloud dependency, no shared tenancy, and the same engine scales into a cluster on infrastructure your own team governs.
Take your last unplanned failure.
Two questions, both answerable this week by hand. What did the free text say, and what did the failure code say. Then ask what was open on that asset when it failed.