The thesis
Physical degradation is a process, not an event. A bearing does not go from healthy to seized in an instant. Its lubricant thins or becomes contaminated. Microscopic damage forms on a raceway. Each time a rolling element passes the damage, it produces a small impact. The impacts are faint at first and detectable only at particular frequencies. Over time they grow, heat rises, and eventually the damage is audible to a person standing nearby. By then the remaining life is short.
The same structure appears across very different equipment. Insulation in a transformer deteriorates over years before it fails. A pump operating away from its design point begins to cavitate before its impeller is damaged. A loose electrical connection warms before it arcs. In each case there is a period, sometimes long, in which the machine is still working and is already telling anyone equipped to listen that it is changing.
That is what we mean when we say failure has a signature. The claim is not that every failure can be predicted. Some are sudden. The claim is that a large share of costly failures are preceded by physical evidence, and that evidence can be captured and interpreted.
From monitoring to understanding
Condition monitoring is an established discipline, and the physics behind it is well understood. Vibration analysis, thermography and acoustic inspection have been used by reliability engineers for decades. What has limited their reach is cost and expertise. A skilled analyst with a handheld instrument can assess a machine thoroughly, and can visit only so many machines in a day.
Machine Health Intelligence is our attempt to extend that kind of judgment to assets that have never justified a specialist's time. It rests on three ideas.
Each asset is its own baseline
Two identical motors installed in different places behave differently. Mounting, load, ambient conditions and maintenance history all leave their mark. A fixed threshold applied to both will be wrong for at least one of them. A model that learns what normal looks like for each individual asset can notice a change that a universal limit would miss.
Agreement between signals is evidence
A rise in vibration could be a developing fault or a change in load. A rise in vibration, accompanied by a new acoustic component and a slow increase in temperature at the same location, is a much more specific statement. Combining modalities is how a system distinguishes a real change from an incidental one. Read more about multimodal sensing.
Patterns recur across machines
If a particular kind of compressor tends to degrade in a particular way, then the tenth time the pattern appears it should be recognized faster and with more confidence than the first. This is the part of the thesis that depends on scale, and it is the part we are most careful about. It becomes true only as deployments accumulate.
What a useful output looks like
The end of the process should be a statement a maintenance planner can use. We think that statement has five parts: the asset concerned, what has changed, an estimate of how likely failure is, an estimate of the time window, and a recommended first action. It should also carry an honest expression of confidence, because a planner deciding whether to take a production line down needs to know how much weight a finding will bear.
Explanation matters as much as prediction. A finding that cannot be explained will not be trusted, and a finding that is not trusted will not be acted on. We intend for every recommendation to be traceable to the signals that produced it.
What we do not claim
Machine Health Intelligence is a thesis and a development program. AFFERENT does not publish prediction accuracy figures, because we have not yet validated them at a scale that would make them meaningful. We do not describe the size of our dataset, because it is early. Illustrative interfaces and waveforms on this website are labeled as illustrations.
We state the thesis with confidence because the underlying physics is established. We describe our progress with restraint because that is what the facts support.
