A medium-sized refinery or power station contains thousands of pressure and differential-pressure transmitters. They measure not only pressure but, through orifice plates and wet legs, most of the flows and levels as well. Each is normally removed from service at a fixed interval, isolated, connected to a pressure calibrator and compared with it at several points across the range. Experience in the nuclear industry, where records are kept with particular care, is that the large majority of these checks find the instrument within tolerance and lead to no adjustment (Hashemian, 2006). The work is expensive, it exposes technicians and plant to the hazards of breaking into process connections, and it occasionally introduces faults that were not there before.

Against this background the appeal of an instrument that checks itself is obvious. Transmitters built around a microprocessor have offered digital compensation and communication since the 1980s, and the trend towards embedding more intelligence in the sensor has been steady (Kanoun and Tränkler, 2004). Recent devices add functions described by their makers as self-diagnosis, self-verification or self-calibration, sometimes with a claim that verification intervals can be lengthened. The claim deserves scrutiny, because the interval is not only an economic choice. It is a statement about how long the measurement can be trusted.

This article reviews what the published literature and the relevant standards say about that claim. It asks what a transmitter can and cannot learn about its own accuracy, what evidence other than the device itself can supply, and what the rules that govern intervals would require before an extension is accepted.

The review is narrative. It draws on the academic literature on self-validating sensors and sensor fault detection, on monographs and regulatory reports dealing with on-line monitoring of instrument channels, and on international standards and guidance documents. Manufacturer literature is not cited, since its claims are specific to products and are not independently reviewed. No formal search protocol was applied and the coverage is selective.

Words matter here. The international vocabulary of metrology defines calibration as an operation that establishes a relation between the values provided by measurement standards and the indications of the instrument, each with its uncertainty (JCGM, 2012). Verification is the provision of objective evidence that an item fulfils specified requirements. Adjustment is what brings the indications into agreement with the standard. A device with no pressure reference inside it cannot calibrate itself in this sense. What it can do is compensate for known influences, test parts of itself, and report evidence that bears on whether its last calibration still holds. The term self-calibrating is retained in the title because it is the term in commercial use, but the functions it covers are better described as self-compensation and self-diagnosis.

Most industrial transmitters sense pressure through the deflection of a thin diaphragm, read out as a change in capacitance, in the resistance of diffused piezoresistors, or in the frequency of a resonant silicon element (Bentley, 2005). The sensing element is usually isolated from the process by a second diaphragm and a fill fluid. Raw output depends on temperature and, in differential devices, on static line pressure. Smart transmitters remove most of this dependence by measuring sensor temperature and applying a correction characterised individually at the factory, and standard test methods exist for evaluating how well the correction works (IEC, 2008; IEC, 2014).

Drift is the slow departure of the output from its calibrated value under constant conditions. Its causes are mechanical and chemical more often than electronic: relaxation of stresses locked into the diaphragm and its mounting, fatigue under pressure cycling, the after-effects of overpressure, gradual loss or degradation of fill fluid, and permeation of hydrogen through the isolating diaphragm in some services. Outside the transmitter, sensing lines block, leak, freeze or accumulate gas or condensate, and the resulting errors are indistinguishable at the output from a fault in the device. Hashemian and Jiang (2009) examine these contributions to transmitter accuracy in nuclear plants, the sensing lines among them. The important feature of most drift mechanisms is that they change zero or span without leaving any other trace in the signal.

The sources of evidence about a measurement can be arranged by how much metrological weight they carry, as in Figure 1. At the bottom are checks of the electronics, which are comprehensive and say nothing about the diaphragm. At the top is traceable calibration, which says everything and requires access. The interesting ground lies between. Table 1 summarises, for each kind of check, what it detects and what it leaves undetected.

The strongest case for extending intervals has been made in the nuclear power industry, and it rests mainly on evidence from outside the transmitter. Safety-related parameters there are measured by three or four redundant channels. On-line monitoring compares each channel with an estimate of the true process value formed from its peers, by simple or weighted averaging, and tracks the deviation over the operating cycle. A channel whose deviation stays within a defined band is judged not to have drifted and is not calibrated at the outage (Hashemian, 2011). The technical basis, including the treatment of uncertainty in the process estimate, was examined for the United States regulator by Hines and Seibert (2006).

Where redundancy is absent, an estimate can be built from other measurements that are physically related to the one in question. Principal component analysis was applied to the identification of faulty sensors by Dunia et al. (1996), and it sits within a much larger family of analytical redundancy methods surveyed by Venkatasubramanian et al. (2003) and treated at length by Isermann (2006). Their common limitation is that the estimate is only as good as the model and the other sensors, and its uncertainty is usually larger than the tolerance of a good transmitter. Such methods detect a channel that has gone wrong. They are less able to confirm that a channel is right to within a tight specification.

Two weaknesses of comparison methods should be stated plainly. Common-mode drift, in which all redundant transmitters of the same model shift in the same direction, is invisible to them, and the nuclear programmes retain a periodic calibration of at least one channel on a rotating basis for that reason. In addition, on-line monitoring observes the instrument only at the pressures the process visits, typically one operating point, and cannot by itself verify linearity across the range.

No general standard fixes a calibration interval for process transmitters. The accreditation standard for laboratories leaves the interval to the owner of the instrument (ISO/IEC, 2017), and the standard on measurement management systems requires that intervals of metrological confirmation be reviewed and adjusted when necessary to ensure continuing compliance (ISO, 2003). The joint guidance of the international laboratory accreditation and legal metrology bodies describes several methods for doing so: stepwise adjustment according to whether the instrument was found in tolerance, control charts of calibration results, intervals based on time in use, in-service checking with a portable standard or built-in check, and statistical analysis of populations (ILAC and OIML, 2007). A recommended practice from the measurement science community gives the statistical methods in detail (NCSL International, 2010).

Two points follow. First, these documents already allow an interval to be lengthened on evidence, and the evidence they chiefly contemplate is the history of as-found results, which any plant with a calibration database possesses. Secondly, the in-service check they describe is close to what self-verification offers, provided the check is understood as a means of detecting change between calibrations and not as a substitute for them.

Transmitters in safety instrumented functions fall under a separate regime. The functional safety standards require periodic proof tests to reveal dangerous failures that automatic diagnostics do not detect, and the achievable test interval depends on the diagnostic coverage credited to the device (IEC, 2010; IEC, 2016). Proof testing and calibration are often carried out together but answer different questions: one asks whether the function will operate on demand, the other whether the reading is accurate. Better diagnostics can support a longer proof-test interval through the reliability calculation, yet that argument says nothing by itself about measurement uncertainty.

Taken as a whole, the literature supports a moderate position. Built-in diagnostics shorten the time to detection of many faults from months to minutes, and that alone weakens the case for frequent intrusive checks whose main yield was the discovery of such faults. For slow drift within the diagnostic threshold the device remains largely silent, and here the justification for a longer interval has to come from drift history, redundancy or process models. A defensible extension therefore combines sources: device status, an internal zero or dual-element check where available, comparison with related measurements, and a statistical record of past calibrations.

Several questions remain open. Independent, published evidence on the drift-detection capability of commercial self-verification functions is scarce, and the test standards for intelligent transmitters do not yet define how such a capability should be measured and declared. There is no agreed way to express the result of a self-check as a contribution to measurement uncertainty, which is what the self-validating sensor concept originally proposed. The treatment of sensing lines, which lie outside the device but inside the measurement, is uneven. And regulators in custody transfer, emissions reporting and pharmaceutical manufacture have their own expectations of periodic calibration that device diagnostics do not automatically satisfy. Each of these is a matter for standards committees and independent test laboratories as much as for designers.

A smart pressure transmitter does not calibrate itself, and the phrase should be read as shorthand for compensation and diagnosis. Those functions are nonetheless useful evidence about the state of the measurement, and existing guidance on calibration intervals already provides the means to give such evidence weight. The prospects for extended verification intervals are good where three conditions are met: the plant holds a drift history showing the transmitter population to be stable, there is some independent check on each measurement through redundancy or a process relation, and the residual faults that no automatic check can reveal, common-mode drift and span error above all, are named and covered by a reduced programme of traceable calibration. Where those conditions are absent, diagnostics improve availability but do not by themselves justify calibrating less often.

To be declared by the author(s) before publication.

To be declared by the author(s) before publication.