Ask three suppliers what makes a sensor smart and you will get three answers, none of which mentions a standard. That is the whole difficulty with the question in the title. Quality control is one of the most tightly specified activities in manufacturing, with auditable requirements attached to it, and “smart” is a marketing adjective. The useful version of the question is narrower: can an instrumented, self-diagnosing device take over work that a calibrated measurement system currently does, and under what conditions.
Key takeaways
- IO-Link, standardised as IEC 61131-9, is the closest thing to a real definition of a smart sensor.
- NAMUR NE 107 gives device self-diagnosis four status signals instead of a vague alarm.
- ISO 9001 requires measuring equipment to be calibrated against traceable standards and safeguarded.
- Under AIAG MSA, a measurement system above 30 percent gauge R&R is unacceptable, whatever it costs.
What actually separates a smart sensor from an instrumented one
Two published specifications do the work that the marketing word does not.
The first is IO-Link, standardised as IEC 61131-9 under the name single-drop digital communication interface for small sensors and actuators. It turns a device that could only send an on or off signal into one that holds a bidirectional point-to-point link over ordinary unshielded three-wire or four-wire sensor cable. That matters for two reasons: parameters can be pushed down to the device, and diagnostic information can come back up. Replacing a failed unit stops being a manual reconfiguration job, because the controller writes the settings.
The second is NAMUR NE 107, the recommendation on self-monitoring and diagnosis of field devices. It condenses whatever a device knows about itself into four standardised status signals: Failure, Function check, Out of specification and Maintenance required. The value is not the categories themselves but the fact that they are the same categories across vendors, with defined visual representation, so an operator reads one convention rather than eleven.
A device that offers both is meaningfully smarter than a switch. A device sold as smart because it has a wireless module and a dashboard is a switch with a subscription.
Why the sensor is rarely the reason a quality project fails
Quality control has requirements that predate any of this. Under ISO 9001:2015, clause 7.1.5, monitoring and measuring resources have to be suitable and maintained. Where measurement traceability is required, or where the organisation treats it as essential to confidence in results, equipment must be calibrated or verified at specified intervals or before use, against standards traceable to international or national measurement standards, identifiable as to its status, and safeguarded against adjustments that would invalidate it. Records of all of that are retained.
Read that against a networked sensor whose parameters can be rewritten remotely by a controller, and the tension is obvious. The same feature that makes IO-Link efficient, remote parameterisation, is a change-control problem for a device inside a quality loop. It is solvable. It is not solved by buying the sensor.
The second constraint is statistical. Before a measurement is allowed to make an accept or reject decision, it should survive a measurement system analysis. Under the AIAG reference manual, a gauge repeatability and reproducibility result under 10 percent is generally adequate, 10 to 30 percent may be acceptable depending on the application and the criticality of the characteristic, and above 30 percent is unacceptable. The number of distinct categories should also reach 5 or more, and both criteria have to be met rather than either one. A sensor sampling a thousand times a second that fails this test is producing a thousand unreliable numbers a second.
Where continuous sensing genuinely changes the answer
The honest case for these devices is not that they inspect better. It is that they see the gaps between inspections, which no sampling plan can.
- Drift, before it becomes a defect. A characteristic trending toward a tolerance limit is visible hours before a sampled part fails, and that is a process correction rather than a scrap decision.
- Conditions that cannot be sampled retrospectively. Temperature, humidity and pressure in food and pharmaceutical processing are the clearest case, because the evidence of a deviation disappears with the batch.
- Traceability records that write themselves. Continuous logs tied to batch identifiers narrow the scope of a recall from a production week to a shift, which is usually where the financial case actually sits.
- Device health as a leading indicator. An NE 107 maintenance request on an instrument is a reason to distrust yesterday’s readings, not just to schedule a visit.
None of that replaces final inspection. All of it reduces how often final inspection is the first place a problem appears.
The two numbers that should temper the pitch
Scale first. The World Economic Forum’s Global Lighthouse Network, its register of advanced manufacturing sites, counted 238 sites in June 2026 across more than 30 countries. Set that against a global manufacturing base measured in millions of plants and the picture is clear: this is still the exception being documented, not the norm being described. McKinsey’s own work on digital manufacturing put fewer than 30 percent of industrial pilots as making it to scale, and the share stuck in pilot rose between surveys rather than falling.
Security second, because every smart sensor is an edge device. Verizon’s 2025 Data Breach Investigations Report found edge devices and VPNs accounted for 22 percent of targets in breaches involving exploitation of a vulnerability, against 3 percent the year before. Only around 54 percent of those vulnerabilities were fully remediated over the year, with a median of 32 days to do it. If your quality data now travels over the plant network, IEC 62443 is part of the quality conversation whether or not it was in the business case.
Three questions we would ask before signing
- What does the measurement system analysis say? Not the datasheet accuracy, the gauge R&R result on your parts, on your line, with your operators.
- Who can change a device parameter, and where is that recorded? If the answer is anyone with controller access and nowhere, the device is not ready for a quality loop.
- What happens to the decision when the sensor reports out of specification? A defined fallback is the difference between a diagnostic feature and an unplanned line stop.
Two questions that come up in every review
Do smart sensors reduce quality costs?
They shift them. Inspection labour and scrap tend to fall; calibration, network security and data engineering rise. Whether the net is positive depends almost entirely on how expensive your current failure mode is, which is a question about your product rather than about the technology.
Can a smart sensor replace a human inspector?
For a continuously measurable physical quantity, often yes, once the measurement system is validated. For judgement about a defect nobody defined in advance, no, and that residual category is larger on most lines than the business case assumes.
So, are smart sensors the future of quality control? They are already the present of process monitoring, and they become part of quality control at the point where calibration, change control and measurement system analysis are handled with the same seriousness as the purchase order. That is the unglamorous part, and it is the part that decides whether the investment shows up in the defect rate. The same pattern runs through the wider question of what makes a plant genuinely smart rather than merely connected.
Same sensors, different question
The data that flags a drifting process is the data that predicts a failing machine, and the economics there are far easier to prove.
Sources: IEC 61131-9, Programmable Controllers, Part 9, Single-drop digital communication interface for small sensors and actuators, for the technical basis of IO-Link, its point-to-point bidirectional link over standard three-wire and four-wire sensor cable, and its parameterisation and diagnostic capabilities. NAMUR NE 107, Self-Monitoring and Diagnosis of Field Devices, for the four standardised status signals Failure, Function check, Out of specification and Maintenance required, and their standardised visual representation. ISO 9001:2015, clause 7.1.5 and 7.1.5.2, for the requirements on monitoring and measuring resources, calibration or verification at specified intervals against traceable measurement standards, identification of status, safeguarding against invalidating adjustment and retention of records. AIAG Measurement Systems Analysis reference manual, fourth edition, for the gauge R&R acceptance bands of under 10 percent, 10 to 30 percent and above 30 percent, and for the number of distinct categories of 5 or more. World Economic Forum Global Lighthouse Network, June 2026, for the figure of 238 designated sites across more than 30 countries. McKinsey, “How digital manufacturing can escape pilot purgatory”, for fewer than 30 percent of industrial pilots reaching scale. Verizon 2025 Data Breach Investigations Report, for edge devices and VPNs representing 22 percent of targets in breaches involving vulnerability exploitation, against 3 percent in the prior edition, with approximately 54 percent fully remediated and a median remediation time of 32 days. IEC 62443 for industrial automation and control system security. Updated August 2026.
