When a custody gas chromatograph analyzes a sample, it computes a mole percent for each component from that component's peak area and response factor, and then it usually normalizes the results so they sum to exactly one hundred percent before reporting them. The unnormalized total is the raw sum of those component mole percents before that normalization step is applied, and in a healthy analyzer it should land very close to one hundred on its own. That single number is arguably the best real-time health indicator a custody GC produces, because normalization deliberately hides the very error the unnormalized total reveals. A drifting unnormalized total exposes calibration and column problems that the final, normalized composition papers over, which is why it deserves its own limit alarm.
GC unnormalized total in one line: The GC unnormalized total is the sum of all component mole percents as measured, before the analyzer scales them to add up to exactly one hundred percent. In a healthy custody GC this raw sum should sit very close to one hundred, so a drift away from one hundred is a direct sign of a calibration, response-factor, or column problem. It is the single best real-time health check because normalization forces the reported composition to look correct even when the underlying measurement has gone wrong, and SCADA limit alarms on the unnormalized total catch that before it affects billing.
A gas chromatograph measures each component independently, applying that component's response factor to its peak area to get a mole percent, and the raw results almost never add up to exactly one hundred percent. Small errors in every component's measurement accumulate, so the analyzer normalizes: it takes the raw sum, computes a normalization factor that scales that sum to exactly one hundred, and multiplies every component by it. The reported composition then adds up perfectly, which is convenient for the downstream energy calculation that expects a complete composition summing to one hundred percent.
The problem is that normalization is a cosmetic operation as much as a mathematical one. If the raw sum came out to only ninety-eight percent because the whole analysis was reading low, normalization scales everything up by roughly two percent and hands you a composition that sums perfectly and looks entirely plausible. The individual component percentages are all slightly wrong, but they are wrong in a way that is invisible in the final numbers, because the sum-to-one-hundred check that a person would naturally apply has been satisfied by force. Normalization guarantees the reported composition looks healthy regardless of whether the measurement actually was.
This is exactly why the unnormalized total is so valuable: it is the number normalization was about to erase. Before the scaling is applied, the raw sum carries the honest signal of how far off the measurement was. A raw sum near one hundred means every component was measured accurately and only tiny errors needed correcting. A raw sum that has drifted to ninety-seven or a hundred and three means something is systematically wrong with the analysis, and that information is destroyed the instant normalization runs. Watching the unnormalized total is watching the health signal before it is hidden.
The unnormalized total earns its status as the single best real-time health check because it responds to the broadest range of analyzer faults with one number. A response factor that has drifted, a calibration bottle that was wrong, a carrier flow that has shifted, a column starting to degrade, or a sample that is not being delivered cleanly will all tend to push the raw component measurements up or down together, and that shows up immediately as the unnormalized total moving away from one hundred. You do not have to know in advance which fault occurred; the total flags that some fault occurred, and it does so in real time on every analysis rather than only at the next scheduled calibration.
Compare that to trying to catch these problems in the normalized composition, where by design they are hidden. A component-level check on the final composition can catch a gross error where one peak vanishes, but it cannot catch the whole-analysis biases that normalization absorbs, because those leave a composition that still sums correctly and still looks reasonable. The unnormalized total sees precisely the class of error the normalized numbers cannot, which is the systematic, all-components-together drift that most often accompanies a degrading analyzer. That complementary blind spot is what makes the raw total uniquely informative.
There is also a diagnostic subtlety worth using. The direction and speed of the drift narrow the cause. A total that creeps slowly away from one hundred over many analyses suggests gradual detector aging or a slow flow change. A total that steps abruptly at a calibration suggests a bad calibration bottle or a discrete event such as a leak or a plugged sample path. A total that jumps and stays put after maintenance suggests something changed during that maintenance. Reading the pattern of the unnormalized total, not just whether it is out of band, points the measurement technician toward the specific problem to chase.
Because the unnormalized total is computed on every analysis, it is straightforward to bring into a monitoring system as a live tag and put limit alarms around it. The natural configuration is a band centered on one hundred with a warning limit for a modest deviation and an alarm limit for a larger one, so a small excursion prompts a look while a significant drift raises a real alert. The exact limits depend on the analyzer and the tolerances the operator is comfortable with, but the principle is simple: as long as the raw total sits inside the band, the analyzer is proving on every cycle that its measurement is internally consistent, and when it leaves the band, it is announcing that it is not.
A cloud SCADA platform such as Merobix is well suited to this because it can read the unnormalized total from the custody GC on each analysis, trend it over time, and evaluate the limit alarms continuously across many analyzers at once. The trend is as valuable as the alarm, because a total that is slowly walking toward its limit gives days or weeks of warning that maintenance is due, and having those trends from a fleet of sites in one place lets a measurement team triage which analyzers need attention first. An alarm on the raw total reaches someone while the error is still small, which is the whole point of monitoring it rather than discovering the problem in a monthly reconciliation.
The operational payoff mirrors that of watching response factors: it moves the detection of a bad analyzer upstream of billing. The composition a custody GC reports drives the energy that gas is bought and sold on, and a normalized composition that looks fine while the analysis is biased can quietly mismeasure gas for an entire billing period. An alarm on the unnormalized total intercepts that at the source, flagging the analyzer as unhealthy before its numbers corrupt the energy total, so the measurement team can recalibrate, change a carrier bottle, or clear a sample path while the impact is negligible. For a custody meter, that is the difference between a maintenance ticket and a billing dispute.
It should sit very close to one hundred percent on its own, before any normalization is applied. Small, unavoidable errors in each component measurement keep it from being exactly one hundred, but a healthy custody GC produces a raw sum only slightly off. A total that has drifted noticeably away from one hundred is a direct signal that the measurement has a systematic problem, even if the final normalized composition still looks correct.
Normalization forces the reported composition to add up to exactly one hundred percent by scaling every component. If the whole analysis is biased so the raw sum comes out low, normalization scales it back up and hands you a composition that sums perfectly and looks plausible, even though every component is slightly wrong. The natural sum-to-one-hundred sanity check is satisfied by force, so the underlying bias becomes invisible in the final numbers.
Bring the raw total in as a live tag and put a band around one hundred percent, with a warning limit for a modest deviation and an alarm limit for a larger drift. The exact limits depend on the analyzer and the operator's tolerances. Trending the value alongside the alarm is valuable because a total slowly walking toward its limit gives days of warning that maintenance is due before it ever affects the reported composition.
Merobix reads your field devices into a cloud SCADA - the real thing behind these terms, live in days from any browser.