Leaving machine data off the job, lot, or asset it belongs to leads to downtime and rework, because the stored value does not say which order to release, hold, or repair. Digital manufacturing on the factory floor means that record is captured, tied to the job, handed to a named person, and written back when that person acts.
On the factory floor, digital manufacturing is the job-linked record and the decision it supports. NIST’s Manufacturing Extension Partnership describes tools that include robotics, additive manufacturing, augmented reality, big data, simulation, and cybersecurity, and says use among small and medium-sized manufacturers varies (infographic published 6 November 2020, page updated 22 August 2025). The page names platform choice, the initial investment, and weak coordination among departments as the challenges. ISA-95, also published as IEC 62264, places records in five activity levels, with manufacturing execution at level 3 and business planning, including ERP, at level 4. Volume is a weak test. NIST’s Extended Digital Thread page says manufacturing generates over 2 exabytes of data a year, citing a 2010 McKinsey report, and says smaller firms have limited resources to integrate lifecycle information.
| Item | Value or source | Why it matters |
|---|---|---|
| Tool family | Robotics, additive manufacturing, augmented reality, big data, simulation, cybersecurity (NIST MEP, updated 22 August 2025) | The label covers many tools. The floor decision names which one changes a record. |
| Activity levels | ISA-95 public model: levels 0 through 4 (ISA standards page, October 2026) | A signal, an MES transaction, and an ERP order are different levels until an interface joins them. |
| Level 3 to level 4 interface | Part 2 objects and Part 5 transactions, between manufacturing systems and business systems | Two copies of one order appear when this interface is missing. |
| Maintenance types | Predictive (condition-based), preventive (schedule or cycle), reactive (after stoppage). NIST Applied Economics Office, page updated 23 September 2026 | A stored vibration reading is not yet one of these types. |
| U.S. discrete manufacturing, 2016 | Maintenance activity $74.5 billion. Preventable losses $119.1 billion. NAICS 321-339 excluding 324 and 325. NIST AMS 100-34, as summarized on that page | National ledger. The plant-budget cell below is blank. |
| Plant budget, payback, or rollout time | None of the pages read state a typical dollar budget, payback, or rollout length for one workflow. |
Other plant-level explainers are in the Industry section.
What has to be true on the floor
NIST MEP describes the digital thread as the link among machines, suppliers, shippers, distributors, and end users. Its rewards list, once the challenges above are handled, is shorter time to market, less machine downtime, and lower inventory carrying cost. That page does not attach a percentage to any of the three. NIST’s Extended Digital Thread page, updated 26 March 2025, uses a second label for a harder target: smart manufacturing research there aims for fully integrated, collaborative systems that respond in near real time to changing conditions in the factory, the supply network, and customer demand. The two pages do not draw a bright line between the words. A floor project still has to name the decision the record will change.
Takeaway: Ask which job decision the proposal changes. The tool list does not answer that.
Where the record sits
ISA’s public ISA-95 page defines five levels by activity, and it says the standard deals primarily with the interface between levels 3 and 4.
| Level | What the ISA-95 page assigns to it |
|---|---|
| 0 | The physical process: machinery and other assets on the floor. |
| 1 | Sensing and manipulating: sensors, smart devices, valves, and other devices that sense or affect production. |
| 2 | Monitoring and supervising control: PLCs, distributed control systems, and other control devices. |
| 3 | Manufacturing operations management: manufacturing execution systems and, on this page, other systems such as SCADA. |
| 4 | Business planning and logistics, including ERP. |
ANSI/ISA-95.00.01-2025 (IEC 62264-1 Mod), announced by ISA on 10 April 2025, is the current Part 1 on models and terminology. ISA says the update highlights the boundary between the enterprise domain and the manufacturing and control domains. Committee chair Christian Monchinski, in that announcement, ties the revision to data-centric architectures in which system data is enriched by metadata from standards-based contextualization.
Part 2 specifies the conceptual content exchanged between level 3 manufacturing systems and level 4 business systems, and it states a goal of reducing the risk, cost, and errors of implementing that interface. Part 5 defines the transactions that collect, retrieve, transfer, and store information between applications at those two levels. The public page does not price the interface. The paid text of the standard is not quoted here.
A manufacturing execution system, in ISA’s 26 February 2020 announcement of Manufacturing Execution Systems: An Operations Management Approach, is an application inside manufacturing operations management. Thomas Seubert, quoted there, says the industrial internet of things enables an MES and does not replace it. Grant Vokey, quoted in the same announcement, says an IT-only definition of MES, centered on databases and servers, limits what the application can do for operations.
Charlie Gifford and David Daff, in ISA InTech for November-December 2017, put the same point on the message itself: exchange of data without context is valueless, and corrupted data makes that worse. Their examples of operations events include a job order started or completed and a schedule created, released, or dispatched.
A new connection can report a measurement or it can command equipment. Those are different paths. How a plant-floor sensor report differs from a command to an actuator separates them. NIST SP 800-82 Rev. 3, final in September 2023, is the guide for securing operational technology while addressing performance, reliability, and safety requirements. Its public abstract says OT detects or causes a direct change through monitoring, control, or both, and it names industrial control systems among the examples. A planning note dated 21 September 2026 points to an initial public draft of Revision 4, with comments due 30 November 2026. This page uses that abstract, not a control setting from the full PDF.
Takeaway: Attach the job, lot, or asset before the value is shown on a screen, and name the person who acts.
Dominant failure mode
Each row is a floor failure the sources describe, and a configuration that blocks it. The prevention column is a record or an owner, not a product name.
| Practice | Dominant failure mode | Prevention |
|---|---|---|
| Machine or operator entry with no job, lot, or asset | The message has no operations context. Gifford and Daff call that exchange valueless. | Add the job, lot, or asset before anyone treats the value as a production fact. |
| Work instruction stored as a loose file | The file on the tablet is not the document set dispatched with the job. | Stage the SOPs and checklists the job requires, with the tools and parts, at the work location. |
| Missing inspection or disposition link | The batch cannot be traced, and a wrong file version can correspond to a weaker or different part. | Hold for a person when a required link is missing. |
| Sensor history labeled as predictive maintenance | Temperature, noise, or vibration is only observed data until a failure prediction uses it. | Keep the schedule or cycle until that prediction exists and someone is assigned to act. |
| Level 3 execution left apart from level 4 planning | One order exists in two versions. | Carry the Part 2 objects and Part 5 transactions across the level 3-4 boundary. |
| A network path that can change the process | OT performance, reliability, and safety constraints differ from an office network (SP 800-82 Rev. 3 abstract). | Decide whether the path only reports or can also command, before it is connected. |
Work instructions at the point of work
Gifford and Daff describe the staging step in the same 2017 article. When the assigned person accepts the task, collaborating applications dispatch the job order and stage the required documents (SOPs and checklists), tools, and parts at that location. They give the reason as capacity: stage the documents, tools, and parts so the specialist can perform the corrective action and limit the capacity loss of production.
MxD’s project page for the Digital Twin Testbed for Process Manufacturing (project 19-04-04, completion date 28 April 2022) describes a related pattern in its proposed solution: a mobile-worker proof of concept in which operators receive tasks on a device, open procedures and alerts, do the task, and report status. The outcome section says a process demonstration skid was built at the MxD Future Factory, with a software stack that includes manufacturing operations management, a process historian, and asset and process health monitoring. The page does not report a measured result for the mobile-worker functions. From the bottom of that stack, the page lists pressure, level, and temperature instruments, wireless temperature and vibration sensors, drive controls, and valve actuators.
Read together, those pages support a narrower test than a generic “paperless” claim. The documents, tools, and parts staged at the location are the set dispatched with the job, and the assigned person accepts the task. Gifford and Daff do not describe a separate electronic signature. NIST’s trustworthiness page, cited below, is the source for requiring the correct type and version from the expected sender. That requirement is stated for manufacturing data, especially in regulated industries. A work-instruction file is one of those data objects. If the actor is an AI agent rather than an MES transaction, which actions should wait for a person is a separate decision.
Common mistake: photographing a paper SOP onto a tablet and treating the photograph as the document dispatched with the job. The InTech staging step is the document set sent with the order, at the work location.
Takeaway: Stage the job’s SOPs and checklists at the work location, with the tools and parts.
Traceability is the links
Traceability here means starting from a job or product identity and finding the material lot, the instruction revision, the processing events, the inspection result, and the disposition. For inline camera stations that verify lot codes, GS1 barcodes, and physical diversion into a locked reject bin before batch release, see machine vision inspection line handoff. The figure below draws those links as an example. The image states that the fields are an example, not a standard schema.
NIST’s page on supply-chain data trustworthiness says manufacturers, especially in regulated industries, need data of the correct type and version, authorized for the intended use, from the expected sender. The page says tampering, whether from a cyberattack or from human or system error, can produce structurally weaker or functionally different parts. NIST labels that page as no longer being updated. The footer shows 2026-05-27. The sentences above are the ones still printed there. The same page describes a certificate toolkit that signs files such as STEP, G-code, PDF, and QIF. That is a signature on a file. It is a different object from the job links in the figure.
Aerospace, drug, and food rules on retention and signatures sit outside this page. Moving a record from paper to a screen does not, by itself, meet those rules. The floor test here is whether the links exist and what happens when one is missing.
On the example checklist in the figure, completeness = 4 / 5 when four of the five links are present. The quotient is 0.8. That line is arithmetic on the example fields drawn in the figure. It is a count of links, not a yield and not a measured accuracy. The missing inspection is a hold.
Takeaway: Check the missing-link case before adding more fields.
Maintenance, and the 2016 ledger
NIST’s machinery-maintenance page, updated 23 September 2026, uses three categories. Predictive maintenance, which it treats as analogous to condition-based maintenance, starts from predictions of failure made with observed data such as temperature, noise, and vibration. Preventive maintenance is scheduled, timed, or based on a cycle. Reactive maintenance, also called run-to-failure or breakdown maintenance, happens after equipment has failed or stopped.
For a machine-level budget, compare predictive versus preventive maintenance costs by failure mode, warning time, and production loss before funding condition monitoring.
The MxD skid’s pressure, level, temperature, and vibration sensors feed a historian and health-monitoring tools. Under NIST’s definitions, those inputs support predictive maintenance only after a failure prediction is made from them and someone is assigned to act. Until then they are observed data, and the schedule or cycle still stands.
Rule of thumb. If reactive maintenance stays under 50 percent of machinery maintenance, NIST’s 2016 survey treated the establishment as one that primarily uses preventive and predictive maintenance. Inside that group, the half using more predictive maintenance was associated with 15 percent less downtime, an 87 percent lower defect rate, and 66 percent less inventory increase from unplanned maintenance. The page also says that, on average, 45.7 percent of machinery maintenance was reactive, and that the top 25 percent of establishments relying on reactive maintenance had 3.3 times more downtime, 16.0 times more defects, 2.4 times more lost sales from delays, and 4.9 times more inventory increases than the bottom 25 percent. All of those figures are associations for U.S. discrete manufacturing in 2016 (NAICS 321-339 excluding 324 and 325). They are not a forecast that one machine will move by those percentages.
NIST says some other estimates put machinery-maintenance cost between 15 percent and 70 percent of the cost of goods produced, and it calls that range wide. The figures it then publishes from its own manufacturer survey are the ones in the arithmetic below.
Maintenance cost = 57.3 + 16.3 + 0.9
Worked example, arithmetic on the three 2016 expenditure lines NIST prints for that NAICS range. 57.3 + 16.3 = 73.6, and 73.6 + 0.9 = 74.5. NIST prints $74.5 billion: $57.3 billion of machinery-maintenance expenditures, $16.3 billion of additional fault and failure expenditures, and $0.9 billion of inventory held to buffer maintenance issues. The sum checks. It is not a budget for one factory.
Preventable loss = 18.1 + 0.8 + 100.2
Worked example, arithmetic again. 18.1 + 0.8 = 18.9, and 18.9 + 100.2 = 119.1. NIST prints $119.1 billion of 2016 losses from preventable maintenance issues: $18.1 billion downtime, $0.8 billion defects, and $100.2 billion lost sales from delays and defects. Lost-sales share = 100.2 / 119.1 = 0.841 when rounded to three decimals, so 84.1 percent of that loss total is the lost-sales line. The page also associates an estimated 16.03 injuries and 0.05 deaths per million employees with these maintenance issues. Perceived benefit of adopting some additional predictive maintenance, as NIST labels it, was $6.5 billion from downtime reduction and $67.3 billion in increased sales, $73.8 billion together (6.5 + 67.3 = 73.8). NIST says other perceived benefits, including reduced defects, were not given a dollar figure in that pair of lines.
| Cost question | Figure the pages support | What stays blank |
|---|---|---|
| National maintenance activity, 2016, NAICS range above | $74.5 billion | |
| National preventable losses, same scope and year | $119.1 billion | |
| Perceived benefit of some additional predictive maintenance | $73.8 billion, of which $6.5 billion is downtime reduction | Dollar value of reduced defects |
| This plant’s interface, device, training, and disruption cost | No page read states it | |
| This plant’s payback or rollout length | No page read states it |
Copy the questions into the project sheet and fill the plant rows from supplier quotes and from the plant’s own time logs. The national totals stay in the national rows.
Before treating a training course as a reason to expand the workflow, record signup, completion, and observed work tasks separately. Observed use of the instruction is its own record.
Takeaway: Use the national totals as a ledger of where the money sits. Do not paste them into a single-plant savings claim.
A first workflow
The path below matches the decision figure. The image calls it a suggested path, not a standard method.
- Name the decision that is hard to make today.
- Name the record that would inform it.
- If that record is unreliable or unavailable, fix the definition, the collection, or the ownership before connecting anything else.
- Name who acts, and who supports the instruction, the data, and the devices.
- Trial it against a baseline. The figure’s three checks are cost, operator training, and an operational-technology review.
- Review the result before expanding.
Items to price in that trial, with no dollar amounts taken from the sources: the level 3-4 interface, access to older equipment, cleanup of job and lot identifiers, upkeep of instruction revisions, devices at the station, operator time in training, lost production during the trial, and the operational-technology review. NIST MEP already names platform choice, initial investment, and cross-department coordination as the practical challenges. The Extended Digital Thread page adds that smaller firms have limited resources for lifecycle integration.
Takeaway: Expand after the trial shows a reliable record and a person acting on it.
Where this does not hold
The path above stops being useful in a few concrete cases:
- The signal cannot be tied to a job, lot, or asset.
- The workflow itself is still disputed. Digitizing it preserves the dispute.
- The alert has no owner, or the owner cannot act during the shift.
- Level 4 planning and level 3 execution stay disconnected, so the order has two versions.
- The problem is rare and already corrected by a short manual step, and the sources give no basis for a larger scope.
- Operational-technology timing, safety, or security limits rule out the connection. The SP 800-82 Rev. 3 abstract is the boundary statement used here; the sibling explainer linked above is where a measurement path and a command path are separated.
Takeaway: More software does not supply a missing record, a missing owner, or a missing response.
Method and sources
Public pages from NIST, ISA, and MxD, fetched on 18 October 2026, are the basis for every figure in the tables. National maintenance totals are NIST’s 2016 estimates for the NAICS range named on its machinery-maintenance page. Sums and the lost-sales share are arithmetic on those printed components, marked as arithmetic above. The paid text of ISA-95 and the full PDF of SP 800-82 Rev. 3 are not quoted. How this publication handles sources is on the editorial policy.
