Cost
The two methods have opposite cost shapes, which is why the right answer changes with fleet size and asset criticality.
Offline:
Cost is per sample. Bottle, shipping and laboratory fee, multiplied by every test point and every interval. Upfront investment is close to zero, and the recurring line grows linearly with the number of assets and the sampling rate. Doubling coverage doubles the recurring cost permanently.
Inline:
Cost is capital. Sensor hardware, installation and integration with the existing monitoring or SCADA system are paid once. After installation, the incremental cost of an additional measurement is typically very small compared with collecting and analyzing another physical sample.
Costs omitted from the offline column
Per-sample pricing understates offline analysis, because the laboratory fee is rarely the largest line.
Labor:
A technician has to reach the sample point and draw the sample correctly. On a large or distributed site that is the dominant cost per data point.
Safety:
Drawing a sample from a live high-pressure circuit is a hazardous task, and industry guidance flags sampling above roughly 350 bar as a specific risk. Some points cannot be sampled at all without shutting the machine down, which converts a routine data point into planned downtime.
Logistics:
Bottles, labeling, chain of custody and shipping, for a population of samples that mostly comes back clean. The waste and carbon cost of transporting fluid to a laboratory that reports no change is real and increasingly accounted for.
Lost oil:
Every sample permanently removes lubricant from the machine. A typical draw is 100 to 250 ml and none of it goes back. On a large circulating system that is negligible. On a small sump, sampled on a schedule, it adds up to a standing consumption of new oil that no one attributes to the monitoring program.
Sampling changes the thing it measures
The oil that leaves in the bottle has to be replaced, and topping up is not a neutral act. Fresh make-up oil dilutes the wear
metal concentration, restores part of the additive package and resets a portion of the trend the program is building. A small
sump sampled frequently can end up measuring its own top-ups as much as its own wear. Opening the system to draw the sample is
also an ingress path for exactly the dirt and moisture the program exists to detect. An inline sensor introduces none of this.
It removes no oil, requires no top-up, and never opens the circuit to take a reading.
The scale of this is easy to underestimate. A 250 ml draw from a 10 liter sump removes 2.5 percent of the charge.
Sampled monthly, that is three liters a year, so roughly a third of the oil in the machine is replaced with fresh oil over twelve
months for no reason other than measurement. On a four-liter gearbox the same schedule turns over three quarters of the charge.
Every one of those top-ups pushes the wear metal concentration down and the additive level up, in a data set whose entire purpose
is to detect the opposite trend.
A program that removes oil in order to measure it is, past a certain sampling rate, measuring its own top-ups.
What a sample actually costs
The cited source reports routine laboratory fees of approximately $10 to $60 per sample, and an in-plant labor cost of $23.63
to draw, label, package and ship it [7]. Using $35 as a representative laboratory fee for the
analysis below gives roughly $59 per data point. That $35 is a modeling assumption, not a surveyed industry average.
Offline, per sample |
Cost |
| Laboratory fee, routine slate |
$10 to $60$35 assumed below |
| Labor to draw and ship |
$23.630.5 h at a loaded rate |
| Total per data point |
~$59 |
| Sample port, one-time |
$20 to $300$150 average installed |
That figure recurs. Sampling one machine monthly costs roughly $700 a year, every year, and sampling it quarterly costs roughly
$235 a year.
What the hardware costs
Inline sensor pricing spans an order of magnitude, and where a system sits in that range changes the payback arithmetic more than
any other variable.
Inline sensor |
Unit price |
Payback vs. monthly sampling |
Payback vs. quarterly sampling |
| KasperAero NZMSMagnetic collection |
$350*At roughly 100 units |
6 months |
18 months |
| Gill WearDetectAnalog or CAN output |
$1,400 to $1,600 |
~2 years |
6 to 7 years |
| Poseidon TridentInductive, per turbine |
$2,000 to $3,000Reported, wind deployments |
3 to 4 years |
9 to 13 years |
Payback assumes $59 per laboratory sample and excludes sensor installation, wiring, data acquisition, power, maintenance and
replacement. These are simple sampling-cost paybacks rather than total lifecycle ROI.
* $350 is a single sensor at an order of roughly 100 units. Unit cost falls substantially with volume and with custom
configurations, reaching the $40 to $50 range at high production quantities. At that price a sensor costs less than one
laboratory sample.
Gill WearDetect pricing is taken from a direct supplier quotation [8]. The Poseidon
figure is a reported range for wind turbine deployments rather than a quoted unit price, and it covers a per-turbine installed
cost that is not strictly like-for-like with a bare sensor [9]. Prices exclude installation,
cabling and any data acquisition hardware.
Where the break-even sits
Payback is the sensor price divided by the recurring sampling cost it displaces, at roughly $59 per sample. On a machine sampled monthly, a $350 sensor is repaid in about six
months, and subsequent measurements carry little incremental sampling cost. A $1,500 sensor takes about two years, and a $2,500
installation three to four. Move to quarterly sampling and every figure roughly triples, which pushes the more
expensive systems past the point where the arithmetic alone justifies them.
Two things sit outside that calculation and usually dominate it. The first is the value of the failure that gets caught, which is
measured in gearboxes and downtime rather than in laboratory fees, and which no payback period captures. The second is that
sampling frequency is not free to increase: an inline sensor produces continuous data at the same price, while matching even a
fraction of that resolution offline means multiplying the recurring cost. Inline sensing is cheapest per data point wherever the
asset is critical, hard to access, or unsafe to sample under load. Offline analysis stays cheaper across large fleets of
lower-criticality equipment sampled infrequently, and it remains the only option when the decision depends on chemistry rather
than on debris. Most sites sit on
both sides of that line at once. For many condition monitoring programs the two methods are complementary rather than
mutually exclusive.