Load-sharing redundancy: when losing a unit stresses the rest
In a load-sharing arrangement, several units carry a total load together, and when one fails the survivors must carry more — so they age faster. Think of three pumps sharing a duty, bolts in a flange, strands in a cable, or power supplies feeding a rail. As long as all are healthy each carries its share; the moment one drops out, the rest are stressed harder and their remaining life shortens. The failures are not independent, and they're not identical over time either — the danger climbs as the group thins out.
Ordinary parallel-redundancy math assumes each unit ages at a fixed rate regardless of how its siblings are doing. For load-sharing systems that assumption is optimistic, sometimes badly so, because it ignores the very effect that makes these systems fail — the cascade, where each failure raises the load and hastens the next.
The load-life mechanism
Load-sharing couples two ideas you can model separately and then combine:
A load-life relationship. How does a single unit's life depend on the load it carries? More load means shorter life, captured by an accelerated-failure-time (AFT) model with load as the stress — often an inverse-power relationship,
life ∝ load^(−n), the same family used in accelerated life testing.Load redistribution. A total load L shared by s working units means each carries L / s. Start with n units at L/n each; when one fails the remaining n−1 jump to L/(n−1); and so on. Each step up the load ladder accelerates the survivors according to the load-life relationship.
The group is considered working while at least k of the n units survive (k-of-n). Put those pieces together and you get the system's reliability — including the accelerating hand-off that plain parallel redundancy can't see.
A worked example
Three identical pumps share a duty; the group works as long as any one of them runs (k = 1 of 3). Each pump follows an inverse-power load-life law, so carrying double the load cuts its characteristic life by 2ⁿ.
- All three healthy: each carries a third of the load and ages slowly.
- One fails: the two survivors now carry half the load each — noticeably faster wear.
- Two failed: the last pump carries the entire duty and burns through its remaining life quickly.
The result is a reliability curve that starts out looking like ordinary triple redundancy but falls away faster in the tail, because the last survivors are running far harder than the design assumed. Requiring more survivors (say k = 2 of 3, so the group fails on the second unit lost) lowers reliability further — there's less slack before the accelerating cascade bites.
Load-sharing vs standby vs independent parallel
Three redundancy patterns that are easy to conflate but behave differently:
- Independent parallel — every unit runs from the start and ages at a fixed rate, unaffected by the others. The most optimistic model; correct only when units genuinely don't share load.
- Standby — spare units are dormant (cold) or lightly loaded (hot) until switched in on a failure. Redundancy here is about switching successfully, not about survivors being stressed.
- Load-sharing — all units are active and share the load, and survivors are stressed more as siblings fail. The realistic model for pumps, cables, fasteners, and power supplies that split a common duty.
Choosing the wrong one of these silently changes your answer. If your units split a load, model load-sharing; using independent parallel there will credit you with reliability the physics won't deliver.
What you need to model it
Because the effect is driven by how life responds to load, the input is a load-life model — an accelerated-failure-time fit with load as its single covariate — plus the total shared load, the number of units, and the k-of-n survival requirement. Fit the load-life relationship from failures gathered at a few different loads (the same data you'd use for a load-based accelerated life test), then drop it into the diagram as a load-sharing block.
Frequently asked questions
What is load-sharing redundancy?
A redundant arrangement where all units are active and share a common load, so when one fails the survivors carry more and fail faster. Unlike independent parallel redundancy, the units' failure rates rise as the group thins, producing a cascade.
Why doesn't ordinary parallel math work for load-sharing?
Parallel redundancy assumes each unit ages at a constant rate no matter what its neighbours do. In a load-sharing system the survivors are stressed harder after each failure, so their remaining life shortens — an acceleration the constant- rate model ignores, making it optimistic in the tail where it matters.
What's the difference between load-sharing and standby redundancy?
In standby, spares are dormant or lightly loaded until switched in, and the risk is switching reliability. In load-sharing, all units are working and sharing the load from the start, and the risk is the survivors being over-stressed as others fail. Different mechanisms, different math.
What input does a load-sharing model need?
A load-life relationship — how a single unit's life depends on the load, expressed as an accelerated-failure-time model with load as the covariate — plus the total shared load, the number of units, and how many must survive (k-of-n).