HOLMES PREDICTIVE · CONDITION-BASED MAINTENANCE

Holmes Predictive: the CBM that points to the source, not to the loudest part.

Holmes Predictive is condition-based maintenance (CBM) with continuous causal diagnosis. It watches the same assets as your CBM, but it also tells which component is causing the degradation, which ones suffer it through drag, which will fail first, and when the control system is hiding a loss.

In one sentence for your team: Holmes brings root cause analysis, which today happens after the breakdown, into the plant's day-to-day, before it happens.

Let's talk about your asset See the example

The essentials in 30 seconds

It complements your CBM; it does not replace it. It strengthens the layer where conventional CBM is weakest: diagnosis.
It only reads data. It does not write to PLCs, drives or safety systems: there is no output path to the machine.
Its outputs are diagnoses and inspection recommendations. Intervention decisions stay with your team.
No new sensors to get started. It works with the signals your plant already records.
When it doesn't know, it says so. If the data cannot separate cause from effect, it abstains and asks for the measurement that would settle it.

WHERE WE FIT

By strategy, it is CBM. By function, it strengthens diagnosis.

Placing Holmes does not require inventing a new category. Two standards your reliability team already knows are enough.

Maintenance strategy · EN 13306

The European standard distinguishes corrective maintenance (repair on failure), predetermined preventive maintenance (by calendar or hours) and condition-based preventive maintenance (by measured state). Predictive maintenance is a type of the latter. Holmes decides from the measured state: it is CBM.

Function within CBM · ISO 13374

The reference architecture for condition monitoring divides a CBM system into six blocks. Conventional CBM is strong at detection. Holmes raises the diagnosis block to causal diagnosis across components.

What is new · proactive maintenance

The diagnosis block usually looks at each component on its own. Holmes solves it across the whole machine, continuously. Attacking root causes instead of symptoms is called proactive maintenance; Holmes makes it operational with data.

In short: causal CBM, or continuous root cause analysis.

WHERE TODAY'S CBM FALLS SHORT

Three blind spots that appear when parts are connected.

CBM is very good at detecting that something is wrong, but it looks at each component on its own. In a real machine, components drag one another, and that creates three blind spots.

The victim shouts louder than the culprit

A degrading gearbox drags the chain and the belt it drives. The belt may show more vibration or temperature than the gearbox, and CBM flags it as the sickest. The belt is replaced, the problem returns and the gearbox keeps degrading.

Control hides the loss

When a component loses capacity, the control loop raises the setpoint to keep production up. Performance looks stable and CBM sees nothing abnormal. But performance sustained by compensation is not health: the margin is being used up.

Root cause analysis arrives too late

Separating cause from effect is already done in industry, with 5 whys, cause-and-effect diagrams or fault trees. But it is done after the breakdown, in meetings, with the asset stopped.

THE ANALOGY

The smoke alarm, the fire investigator, and the investigator who arrives in time.

CONVENTIONAL CBM

The smoke alarm

It warns early and well, but it goes off in the room with the most smoke.

ROOT CAUSE ANALYSIS

The fire investigator

They tell you where the fire started, once the house has burned down.

HOLMES PREDICTIVE

The investigator, while the smoke is rising

“The fire is in the kitchen, the living room is only getting the smoke, and the extractor fan is masking it.”

The kitchen is the source component, the living room is the victim and the extractor fan is the control loop covering up the loss. Holmes gives all three answers at once, before the breakdown.

WHAT IT ADDS

Four questions, answered continuously.

Holmes answers four questions, sample by sample. Your CBM also answers the first one; it does not answer the other three.

Comparison between conventional CBM and Holmes Predictive across four diagnostic questions.
Question	Conventional CBM	Holmes Predictive
Is something degrading?	Yes, when the signal crosses its threshold.	Yes, on a health indicator that accumulates progressive degradation and corrects for load.
What is causing it?	No: it points to the component that deviates most.	It separates each component's own loss from the loss it receives from its neighbours.
Is control hiding it?	No.	It measures how much of the controller's action is compensation for a loss.
What will fail first?	Only by extrapolating each signal in isolation.	It combines the source's degradation, how much is transmitted and with what delay (in development).

Early detection. The health indicator compares each component with its own healthy behaviour. On a synthetic gearbox, chain and belt test bench, Holmes raised the alert between 141 and 411 samples earlier than a fixed 5% threshold on the same signal. The comparison with a commercial CBM on the raw signal is in preparation.

EXAMPLE

Gearbox, chain and belt: who causes it and who suffers it.

On a synthetic test bench, all three components of a line were losing capacity at the same time. A CBM would have flagged all three. Holmes separated each component's own loss from the loss it received through drag.

Synthetic test bench. The figures show each component's total loss and how much of it is its own or received from its neighbours.

Holmes's reading on the synthetic gearbox, chain and belt test bench.
Component	Total loss	Own	Received	Holmes's reading	Recommendation
Gearbox	22.6%	22.6%	0.0%	Source	Inspect as a priority.
Chain	21.0%	4.6%	16.3%	Cause not identified	Take a discriminating measurement before replacing.
Belt	9.2%	≈0	11.6%	Victim	Do not attribute a fault of its own; act on the source.

Holmes concludes that the problem is in the gearbox and that the belt has no fault of its own. It also detected that 98.5% of the gearbox controller's action was compensation for that loss: control was hiding it.

What fails first is not always the root. The source tells you where to act, but a victim can reach its limit sooner: because it is more fragile, because it receives from several sources or because transmission worsens with wear. Holmes is adding failure order, shown separately from the causal reading, so your team knows both what to repair and what to protect in the meantime.

When it doesn't know, it says so. If the data cannot separate cause from effect, Holmes does not guess: it says so and asks for the measurement that would settle it. In the example, that is what happens with the chain.

WHAT WE CLAIM TODAY AND WHAT NOT YET

Every capability, with its status and what backs it.

Holmes estimates the time left before a component reaches the limit your team declares. What we do not yet date is when the damage will spread to another component.

Status of each Holmes Predictive capability and the evidence behind it.
Capability	Status	What backs it
Detect degradation earlier than a conventional threshold	Available	Synthetic test benches; comparison with commercial CBM in preparation.
Separate source and victims continuously	Available, in supervised pilot	Synthetic test benches with pre-registered studies.
Detect control compensation	Available, in supervised pilot	Synthetic test benches with a controller.
Time to a component's declared limit	Available in the core	Historical comparison with public data: IMS bearings (13 of 13) and C-MAPSS turbofan (78.8%).
Failure order between components	In development	Study pending.
Date of propagation to another component	In research	Shown today as monitoring, without a date.

Your team declares the limit. Holmes calculates when the threshold you set will be crossed: breakage, loss of function or your intervention criterion. CBM extrapolates the symptom, which carries what the neighbour adds to it; Holmes extrapolates each component's own degradation, already separated, which is why the uncertainty band has physical meaning.

What we always say: there is no validation yet of the date in an installation in operation. Everything above is comparison with historical data or test benches. That is why, in a pilot, Holmes does not issue failure dates or orders: it issues inspection recommendations.

HOW TO GET STARTED

It is installed alongside what you already have.

Holmes is installed without re-instrumenting and without touching the machine's control. Your CBM keeps alerting; Holmes tells your team where each alert comes from.

What does not change in your plant

Your sensors, your SCADA and your CBM stay as they are. Holmes reads the data they already record.
Holmes does not write to PLCs, drives or safety systems: there is no output path to the machine.
Intervention decisions stay with your maintenance team.

What we need for a first case

The machine map: which component drags which, with what approximate delay, and which control loops are involved. We build it with your engineering team.
Your engineers' statements about each connection, including any direct path from the controller. Without them, Holmes abstains on that connection instead of guessing.
Historical data with an initial healthy period of at least 400 samples and, if available, records of past failures.
The tolerable limit of each critical component, if you want an estimate of the time to that limit.

Traceability guarantees

Every engine version carries fingerprints of its modules and parameters, every study is registered before it runs, and the forecast history cannot be rewritten without it showing. For a version to enter official diagnosis, two independent signatures are required: one from engineering and one from computation.

FREQUENT QUESTIONS

What a reliability engineer usually asks.

Is Holmes Predictive the same as my CBM?

No. Your CBM detects that something is wrong in a component. Holmes tells you whether that component is the source or is only receiving damage from another, and whether control is hiding it. It is the ISO 13374 diagnosis layer, which CBM only covers locally.

Does it warn earlier than my CBM?

Against a fixed threshold, yes: on a synthetic test bench it warned between 141 and 411 samples earlier. Against a commercial CBM on the raw signal, we are measuring it and will publish the figures. What Holmes adds to every alert is whether the component is the source or a victim, with its level of evidence.

Why should I trust that the belt has no fault of its own?

Because Holmes separates each component's loss with a reproducible calculation on the machine map, not with a black-box model. Every result states its level of evidence, and when it cannot separate cause from effect it says so and asks for a specific measurement.

What if the machine map is wrong?

That is the point we take most care over. Your engineers validate the map connection by connection. An undeclared connection is not accredited and Holmes abstains on it.

Does it give me the failure date?

Holmes estimates how long a component has until the limit you declare, with a method checked against public historical data. There is no field validation of that date yet, so in a pilot the maximum output is an inspection recommendation.

Do I need to install new sensors?

Not to get started. Holmes works with the signals your plant already records.

Can Holmes touch my machine?

No. There is no write path to PLCs, drives or safety systems. Holmes only reads data and issues diagnoses and recommendations.

Want to see it on one of your assets?

Tell us which machine concerns you and what data it records. We will tell you plainly whether Holmes fits and what a first case would require.

Let's talk about your asset

Holmes Predictive is a platform by Ln2 (Holmes Predictive Space, S.L.) based on the Homeostatic Cycle method, protected by patent applications filed with the Spanish Patent and Trademark Office (OEPM). The figures on this page come from synthetic test benches and public data, as indicated in each case.