Pressure Trend Diagnostic Workbook

Pressure TrendDiagnostic Workbook

Trace pressure from generation to the critical consumer. Separate real restrictions, supply response and measurement artefacts before choosing the remedy.

1. Engineering answer: diagnose the pressure path, not one gauge

A pressure trend diagnostic workbook compares time-aligned pressures at the source, through treatment and distribution, and at the affected consumer. It connects those observations to demand, compressor state, operating events and a defined process requirement. Its useful output is a supported investigation direction: where the pressure problem develops, which explanation fits the evidence, what remains uncertain and which test can distinguish the competing causes.

Low pressure does not automatically mean insufficient compressor capacity. The engineering approach identifies pipework restrictions, treatment losses, leakage, sequencing, storage, simultaneous demand and point-of-use components as alternative explanations. A larger compressor may be necessary after these possibilities are evaluated, but it is not a diagnosis. A high compressor-room reading can coexist with an inadequate machine-inlet pressure.

The workbook therefore separates four layers: recorded observations, calculated indicators, engineering hypotheses and authorised corrective actions. A correlation between two signals belongs to the evidence layer; it does not automatically establish the cause. A calculation showing a shortfall does not authorise a pressure increase, treatment bypass or machinery-control change.

Use the record to answer a practical question: did generated pressure fall, did an intervening segment consume more pressure, or did the measurement chain create a misleading indication? The answer may involve more than one mechanism. Retain that complexity where it affects the decision rather than forcing every incident into a single fault label.

2. What this workbook owns

It provides a campaign definition, aligned observation structure, event metrics, hypothesis record and repeat-test comparison. This workbook turns those foundations into reusable analysis records instead of reproducing either complete article.

Use it for intermittent machine starvation, repeating header dips, unexpected pressure oscillation, a rising treatment differential, commissioning observations or an unexplained change after maintenance. It can also structure evidence for a controls, receiver or distribution review. The appropriate observation window depends on the event and operating pattern, not a universal logging duration.

The workbook is not a transient network simulator, a compressor controller, a vessel-sizing approval or a machine-safety assessment. Pressure alone does not establish delivered airflow, compressor efficiency, contamination class or actual energy savings. Those questions require their own compatible measurements and methods. The word diagnostic means that the evidence narrows the investigation; it does not imply an automatic final diagnosis.

The editable Word records are manual working templates. The later developer section defines calculation and interface behaviour for an implementation. Neither the templates nor the synthetic examples represent a connected plant dashboard or an executed customer test.

3. Define the service failure before collecting more data

Start with the affected asset, the exact pressure location and the event in which service is inadequate. Record whether the symptom is a low-pressure alarm, slow movement, incomplete cycle, process instability or another observed consequence. A pressure complaint without an associated service condition can lead to logging the wrong location or testing during a period when the fault cannot occur.

Obtain the relevant minimum, maximum or permitted operating band from the applicable equipment or process record. Identify whether the requirement applies while flowing, during actuation, after a regulator or at a specific connection. An operator's customary setting may be useful context, but it is not automatically the equipment's minimum requirement. Keep an unconfirmed limit marked unconfirmed.

State how duration matters. A process may respond differently to a brief fluctuation and a sustained depression; a protection device may have its own validated delay. Preserve the process requirement separately from any display smoothing or alarm debounce used by the diagnostic tool. A software filter must not redefine an equipment requirement.

The campaign should identify its decision owner, permitted activities, normal configuration and relevant restrictions. Record any emergency adjustment made before logging begins. That changed setting becomes part of the measured configuration, not proof of the original condition. Where safety or production continuity requires immediate intervention, preserve whatever evidence can be obtained safely and describe the resulting limitation.

4. Place sensors where they divide the problem

Select locations that divide the actual supply path into interpretable segments. More sensors are not inherently better; each should answer a distinct question. Confirm the as-installed route, including parallel connections, receivers, non-return valves, regulators, temporary hoses and isolation valves. A simplified sketch must not hide a second feed that changes the meaning of a pressure difference.

These labels are a proposed workbook convention, not a mandatory plant layout. A site may need fewer points, more points or a different topology. Name each sensor and its physical tapping independently of its chart colour. Record whether the measured pressure is upstream or downstream of a local regulator.

Across a regulator, part of the pressure difference is intentional regulation. Across a charging receiver or a ring segment, flow direction may change. Across a long path, intermediate users may change the flow along that path. Retain these facts before treating every positive difference as one component's irreversible loss.

5. Preserve pressure units and reference conventions

Every channel needs a quantity, unit and pressure reference. Gauge pressure is referenced to its local atmospheric pressure; absolute pressure is referenced to vacuum. Converting bar to kilopascals changes units, not the pressure reference: 1 bar equals 100 kPa. To reconcile an absolute reading with a gauge reading, the applicable atmospheric reference must also be known.

For compatible references, segment difference is upstream pressure minus downstream pressure. Do not subtract bar absolute from bar gauge merely because both columns contain plausible numbers. Keep the raw values, original units and transformation record. Where a differential-pressure transmitter directly reports the difference, identify its high and low connections and its polarity.

A negative difference is not automatically an invalid observation. It may arise from reverse flow, local storage discharge, another supply connection, instrument bias, a clock mismatch or a wrongly identified tapping. Flag and investigate it rather than clipping it to zero. Clipping would remove information needed to distinguish the possibilities.

Record the pressure range and overload limits of each instrument, but keep those separate from the process operating band. A sensor that can measure a pressure does not establish that the machine or pipework may safely operate at that pressure. Pressure-system limits, protective devices and authorised operating instructions remain controlling. The workbook does not issue a replacement setpoint.

6. Review the complete measurement chain

The chain includes the pressure tapping, connecting line, transducer, electrical signal, logger, historian and export. A calibration record supports instrument performance under its stated conditions; it does not establish that a blocked sensing connection or a slow historian captured the plant transient correctly. Record sensor response, logger sampling, any internal filtering and the export aggregation interval.

A long or restrictive impulse connection can delay or damp the apparent response. A sensor fitted behind an unintentionally closed valve may preserve an old pressure. A logger can report a flat line because the plant is stable or because communication stopped. These cases require different responses even when the chart looks similar.

Select the range and measurement quality for the question. Subtracting two wide-range instruments may not resolve a small filter differential. A suitable differential instrument may offer a better measurement arrangement, but it still needs correct installation and an appropriate operating envelope. Preserve instrument identity and applicable verification information so later comparisons can account for replacements.

Installation, disconnection and any reference checks belong to competent personnel following site and manufacturer procedures. Do not loosen a pressurised connection to improve a trace, defeat an interlock for a convenient event, or obstruct a protective device. The diagnostic worksheet records an approved measurement plan; it does not instruct improvised live work.

7. Align clocks without manufacturing a result

Keep original timestamps, timezone or offset, clock source and any known logger latency. Define the analysis time reference before pairing channels. A timestamp may denote the beginning, end or midpoint of an averaging interval; two identical labels can therefore represent different physical periods. Preserve that convention with each imported series.

For a documented constant offset, define delta as logger time minus reference time. The corrected timestamp is logger time minus delta. A logger one second fast therefore has one second subtracted. If drift is established, record a supported time-dependent correction rather than assuming one offset remains valid throughout the campaign.

Pair observations only under an explicit rule. Exact common timestamps are preferable where the instruments actually share a timebase. Nearest-neighbour pairing requires a stated tolerance, no reuse that distorts duration, and disclosure of the residual mismatch. Interpolation requires an explicit estimate status and a justified event-resolution limit.

Never move one curve until the pressure difference becomes positive or until correlation is maximised, then describe the result as corrected measurement. Cross-correlation may suggest a timing question, but a physical lag and a clock error are not the same thing. A documented common event or synchronization record is stronger evidence for changing timestamps. Preserve both original and corrected views.

8. Match sampling and coverage to the event

A trend intended to investigate a two-second dip needs a measurement chain capable of resolving that event. A one-minute average cannot reconstruct the missing minimum, even when its displayed decimals look precise. Conversely, rapid acquisition without sufficient sensor response or correct tapping can produce many samples of the same filtered signal.

Document whether rows represent instantaneous observations, interval averages or values interpreted under a declared sample-and-hold model. These representations support different claims. An interval-average pressure below a limit means the average was below the limit; it does not prove that every instant of that interval was below it. Label duration metrics accordingly.

Treat missing data as missing, not zero pressure. Retain outages, rejected rows, gaps, duplicates and manual exclusions. Use the union of valid time intervals to calculate coverage; adding overlapping rows can produce an impossible coverage above 100%. A downstream channel with complete coverage cannot make an upstream/downstream differential complete when the upstream channel is missing.

Do not bridge a gap across a compressor trip, valve action or machine event with an apparently smooth line and then interpret that line as observed recovery. Display the gap and preserve its effect on the event conclusion. A high overall coverage percentage may still leave the decisive ten seconds unobserved. Report both period coverage and coverage of the actual complaint window.

9. Prepare an analysis dataset without altering the source

Retain the original logger exports and field notes. Create a separate analysis table containing the source record identifier, channel identity, raw timestamp, corrected timestamp where applicable, raw value, converted value, quality state and transformation reference. Sorting or renaming the file must not break the connection to the original observation.

Distinguish duplicate imports from genuine repeated measurements. A second instrument at the same timestamp is not a duplicate merely because the time matches. A repeated export of the same instrument sample should not receive a second time weight. Resolve conflicting values using evidence, not by automatically choosing the smaller differential.

Mark outliers for review. A brief extreme may be a real pressure event, electrical noise, an out-of-range sensor or an import error. Automatic deletion can remove the event being investigated. Similarly, clipping negative readings or replacing blanks with a preceding value can conceal instrumentation problems.

Use one controlled set of units and timestamp conventions in the analysis view. Record decimal separators and identifiers that may lose leading zeros during spreadsheet import. The same requirements apply when data arrive through email, Drive, a contractor workbook or a historian. The communication channel does not establish measurement quality; the observation and its metadata do.

10. Classify operating states before calculating averages

Separate normal production, peak combinations, start-up, cleaning, standby, maintenance and authorised abnormal operation where those states materially differ. Add compressor operating state, relevant speed or load command, online machine combination, treatment switching and known demand events when the records support them. Do not infer a precise operating state from pressure alone when no corroborating signal exists.

Keep control command and actual machine response distinct. A start command is not proof that the machine reached useful output. A running indication is not necessarily loaded operation. A stationary pressure band may reflect effective control, little demand or an isolated tapping; the surrounding evidence determines which interpretation is plausible.

State classification must serve the question. Comparing one post-repair quiet shift with a pre-repair production peak can falsely attribute a lower pressure loss to maintenance. Compare similar states or disclose the difference. The same caution applies to changes in pressure setpoint, treatment path, active well-defined air users or temporary rental connections.

Unknown is a valid state when the supporting record is absent. Keep unknown periods visible and prevent a calculation from silently treating them as normal operation. A carefully qualified comparison is more useful than an apparently exact before-and-after percentage built from incompatible duty.

11. Calculate transparent pressure and coverage indicators

Use metrics that retain their denominator and measurement meaning. Let w_i be an accepted time weight, p_i the pressure represented by that weight, T_valid their non-overlapping total, and T_window the analysis-window duration. All times in a calculation must use the same unit. Derived duration and deficit results require a declared interval representation.

Show the actual limit and its source next to the indicator. When no approved minimum exists, the tool may calculate against a labelled investigation threshold but must not call the result equipment non-compliance.

Do not report a row-count average as a time average for irregular records. Preserve raw extrema when plotting smoothed trends. Display rounding should not drive decisions: compare using retained numerical precision and then format the displayed result. A chart with fewer points may be useful for navigation, but the event calculation must use the accepted underlying records.

12. Define events and recovery explicitly

A low-pressure event needs an entry condition, a time basis and an exit condition. State whether one below-limit observation opens an event or whether a configured duration is used for a separate alarm view. Keep the unfiltered threshold-crossing record available even when an operational alarm includes debounce.

Hysteresis can prevent rapid repeated event opening and closing near a boundary. For example, a report can distinguish entry below a selected limit from recovery above a separately approved recovery threshold for a stated dwell time. Those settings are project inputs. The workbook does not prescribe one pressure margin or recovery delay for every application.

Record event start, lowest recorded pressure, maximum observed deficit, represented below-limit duration, associated operating change and first sustained recovery. State whether the start or end was observed or censored by the beginning of the file, a data gap or the end of logging. A censored event must not receive an invented complete duration.

Recovery time can be measured from a documented demand-end event, from the recorded minimum, or from another agreed origin. These are different metrics. Preserve the selected definition with the number. If the demand-end time is uncertain, the uncertainty belongs in the recovery result rather than being hidden by a precise timestamp.

13. Build simultaneous segment differences

For a confirmed serial path, calculate each adjacent difference at the same accepted timestamp. Their sum should telescope to the end-to-end pressure difference when they use the same readings and references. This is an arithmetic reconciliation, not independent validation of the sensors: a shared intermediate sensor can cancel algebraically while remaining biased.

Do not add each segment's separate daily maximum and present that sum as a measured peak end-to-end loss. Those maxima may have occurred at different times, demand states or flow directions. Record the timestamp of each maximum and also report the largest simultaneous end-to-end difference.

State whether the segment includes intentional regulation, a change in cross-section, active storage or a parallel feed. A measured static-pressure difference is useful evidence, but assigning all of it to one frictional resistance can require additional analysis. Pressure along an accelerating or decelerating flow path is not always a simple monotonic chain.

When a treatment differential rises, compare flow, pressure, path and equipment condition before calling it a blocked element. When a branch differential rises only during one machine cycle, investigate that cycle and the local path. These are applications of the supplied segmentation method; neither establishes a universal acceptable percentage loss or component replacement threshold.

14. Use pattern recognition as a hypothesis table

Read several channels together and connect the event to equipment state. The following matrix is an investigation aid, not an automatic fault code. More than one row can describe the same event, particularly when a demand peak interacts with both a restriction and a delayed supply response.

Document evidence against the favoured explanation as well as evidence for it. A persistent common pressure decline supports a supply-demand investigation, but controls, unavailable capacity and measurement quality still need checking.

The action record should identify the next observation that would change the decision. This prevents repeated general surveys when one additional tapping, a verified timestamp or a specific machine-state record would resolve the uncertainty.

15. Distinguish capacity, control response and availability

A station can have adequate installed capacity but insufficient available capacity during an event. A machine may be out of service, restricted, unloading unexpectedly or unable to receive a valid start command. Conversely, all available machines may reach their supported duty while demand exceeds what the station can deliver. These situations should not share one undifferentiated capacity-shortage label.

Compare actual operating states with the pressure event. Record command time, confirmed start, loaded response where available, pressure trajectory and demand duration. Review whether the controller's sensing point represents the users it is intended to protect. A control sensor separated from the critical branch by a large dynamic loss can report satisfactory local pressure while the machine inlet is inadequate.

A repeating sawtooth pressure trace is not automatically hunting. It may reflect a legitimate load/unload band. Diagnose unwanted interaction using timing, machine-state evidence, unnecessary simultaneous unloading, starts and the effect on useful service. Do not infer a control defect simply because the graph is not flat.

Any trial change needs an authorised plan, equipment limits, monitoring, a stop condition and a reversible configuration record. Pressure trend analysis does not replace manufacturer control logic or approve changes to protective settings. After an intervention, verify relevant normal and contingency states rather than only the quiet interval that produced the most favourable trace.

16. Interpret storage and transient response without a sizing shortcut

A receiver can support a transient only through the usable pressure window and the actual connection to the load. A large vessel behind a restrictive valve or long small-bore line may not deliver air quickly enough to protect the consumer. Simultaneous receiver and machine-inlet trends can reveal whether stored pressure remains available while local service collapses.

For a deliberately simplified, fixed-volume, isothermal control volume, reference-volume net outflow can be screened from pressure decline: Q_out,ref − Q_in,ref approximately equals −(V/p_ref) × dp_abs/dt when gas temperature equals the stated reference temperature and compressibility effects are neglected. The sign makes a falling pressure correspond to net withdrawal. Use absolute pressure in the gas relation and compatible time units.

This equation is an explanatory mass-balance model, not a leak calculation or a final receiver design. Temperature change, connected pipe volume, changing inflow, active regulators, uncertain vessel isolation and spatial pressure differences can invalidate a simple estimate. Where temperature differs but is approximately constant, the appropriate reference-temperature ratio is required; where temperature varies, the fuller mass balance is needed.

Avoid extending storage conclusions beyond the evidence. A receiver that bridges one observed short event cannot correct a sustained supply deficit indefinitely. Refill time, repeated-event spacing and all critical consumers matter. Pass the defined event, pressure window, connection and assumptions to the receiver transient-storage method rather than selecting a vessel from one pressure slope.

17. Quantify uncertainty and timing sensitivity

A difference inherits the uncertainties of both inputs. For delta_p = p_up − p_down, a first-order standard-uncertainty calculation is u_delta = sqrt(u_up² + u_down² − 2 × rho × u_up × u_down), where rho represents the supported correlation between the input uncertainties. Use standard uncertainties in this expression, not unrelated instrument maximum-error limits. [R2]

An expanded uncertainty is U = k × u_delta when an explicit coverage factor and its interpretation are provided. The reported interval is not automatically a guaranteed physical bound, and k = 2 does not by itself establish every statistical assumption needed for a particular confidence claim. Known systematic effects should be addressed through an appropriate correction and residual uncertainty, not hidden inside unexplained percentages.

Timing error is another mechanism. On a locally smooth trace, |dp/dt| × |delta_t| can screen the apparent pressure error caused by a time mismatch. It is a first-order sensitivity estimate, not a universal uncertainty model for discontinuities. Near a sharp event, inspect the actual signals and synchronization evidence.

Do not automatically combine that bound with statistical sensor uncertainty by root-sum-square. A defensible combination needs defined distributions, correlations and relevance to the same result. Where the basis is missing, show the sensitivity separately. A small pressure difference that cannot be resolved by the available instruments should remain unresolved, not become a precise restriction diagnosis.

18. Convert evidence into a controlled next action

Summarise each incident in a short evidence chain: observed condition, affected boundary, calculated result, candidate cause, competing explanation, next test and responsible owner. Rank the next action by its ability to resolve the decision and by safety or production consequence, not by how easily it produces a completed work order.

Possible actions include correcting a timestamp mapping, inspecting a sensing connection, verifying a valve position, obtaining a regulator curve, separating treatment losses, collecting a faster event record or testing an approved operating-state change. A proposed mechanical modification should identify the pressure segment it is expected to improve and the requirement it must preserve.

Retest the same relevant boundary after correction. Record the changed configuration, duty, instruments and criteria. When several changes occur together, attribute the observed result to the combined intervention unless separate evidence supports individual contributions. Retain failed trials; replacing them with the final favourable trace removes valuable diagnostic history.

Report pressure-service improvement separately from energy outcome. Lower differential, fewer low-pressure events and faster recovery are meaningful results on their own terms. They become verified electricity savings only through compatible power and operating evidence. A fixed assumed energy saving per bar is not a substitute for that assessment.

19. Worked example A: a local path dominates the event

All four worked examples are synthetic teaching cases. Their pressures, limits, times and uncertainties are not customer measurements, universal design targets or authorised settings. This first case uses the P0–P4 serial layout from Section 4, with no active regulator or parallel feed inside the final P3–P4 segment.

A machine has a fictional documented minimum inlet requirement of 6.00 bar(g) during the selected production event. Synchronized observations show the following values. The readings describe two operating snapshots, not an annual capacity assessment.

During the event, adjacent differences are 0.30, 0.20, 0.10 and 1.50 bar. Their simultaneous sum is 2.10 bar, equal to 7.55 − 5.45. The final segment contributes 1.50/2.10 = 71.43% of that observed end-to-end difference. The inlet shortfall is 6.00 − 5.45 = 0.55 bar.

Between the snapshots, the source falls only 0.05 bar while the machine inlet falls 1.35 bar. The final-segment difference rises from 0.30 to 1.50 bar. This makes the local path and its event demand a strong investigation priority. It does not establish whether the cause is hose diameter, coupling geometry, a valve, contamination, a changed duty or several interacting factors.

The next test should divide the final segment and confirm its configuration and flow context. A proposal to raise the whole station by 0.55 bar is not justified by the arithmetic alone. Preserve the machine requirement and air quality while investigating the concentrated difference. The 71.43% result describes one event and must not be published as a general proportion of plant losses.

20. Worked example B: duration weighting and missing time

A ten-minute investigation window uses a declared piecewise-constant analysis representation. The accepted pressure values below are assumed to represent their listed intervals for this example. Real interval-average data would instead be labelled as duration of below-limit interval averages, without claiming continuous pointwise pressure.

Valid represented time is 570 seconds, so coverage is 570/600 = 95.00%. The time-weighted mean is (6.40×120 + 5.80×60 + 6.10×90 + 6.30×300)/570 = 6.236842 bar(g). Averaging the four accepted row values gives 6.15 bar(g), which is not the time-weighted mean because their durations differ.

For a fictional investigation limit of 6.00 bar(g), represented below-limit time is 60 seconds. Its share of valid represented time is 10.5263%. The pressure-deficit integral is (6.00 − 5.80)×60 = 12.00 bar·s, equivalent to 0.20 bar·min. None of these quantities is lost air volume or electricity consumption.

Under this example's representation, the missing 30 seconds could add zero to 30 seconds to below-limit time. The scheduled-window share is therefore bounded between 10.00% and 15.00% on that limited missing-interval assumption. This is not a statistical confidence interval and does not account for unresolved faster variations within the represented intervals.

The gap interrupts the event record. Its true recovery time is not established merely because the next accepted value is above the limit. The report should preserve an incomplete event rather than silently joining the two sides of the gap.

21. Worked example C: a clock offset imitates reverse pressure loss

Consider simultaneous physical readings with a constant true upstream/downstream difference of 0.20 bar. The upstream logger uses reference time. A documented synchronization check shows the downstream logger is exactly one second fast throughout this short example.

A naive merge on raw timestamp pairs values from different physical times. At timestamp 1 it yields 7.20 − 7.20 = 0.00 bar; at timestamp 2 it yields 6.90 − 7.00 = −0.10 bar; at timestamp 3 it yields 7.00 − 6.70 = 0.30 bar. The apparent negative difference could trigger an unnecessary reverse-flow investigation.

With delta defined as raw logger time minus reference time, delta is +1 second. Subtracting that supported offset restores the four simultaneous pairs, each with a 0.20 bar difference. The correction is justified by the stated synchronization evidence, not by a desire to make every difference identical.

Preserve the raw timestamps, correction record and corrected results. If the offset were only approximate, the residual timing uncertainty would still need attention. If an alternative supply or receiver could produce genuine reverse flow, correcting the clock would not remove the need to understand the topology. The example demonstrates one mechanism; it does not make every negative differential an instrumentation error.

22. Worked example D: a small difference is not resolved confidently

Two synchronized, compatible measurements give upstream pressure 7.20 bar(g) and downstream pressure 7.12 bar(g). The observed difference is 0.08 bar. For this synthetic case, the applicable standard uncertainties are 0.03 and 0.04 bar, with independent input uncertainties assumed and documented for the example.

The combined standard uncertainty is sqrt(0.03² + 0.04²) = 0.05 bar. With an explicitly chosen illustrative factor k = 2, expanded uncertainty is 0.10 bar. The displayed expanded-uncertainty interval is therefore −0.02 to +0.18 bar. This wide interval does not mean that the physical component necessarily produces negative loss; it shows the limited resolution of the inferred small difference.

Suppose a fictional project decision uses a maximum difference of 0.10 bar and a pre-agreed conservative interval rule. A below-limit central estimate alone is insufficient under that rule because the interval overlaps the limit. The outcome is inconclusive under the selected rule, not an automatic pass or a proven failure. Other agreed decision rules may treat the evidence differently; retain the actual rule.

A separate timing sensitivity check finds a locally smooth pressure rate of 0.25 bar/s and a possible inter-channel mismatch of 0.20 seconds. Their product is 0.05 bar of first-order apparent-pressure sensitivity. It is reported separately, not added to the statistical uncertainty as though its probability model were already established.

The practical response is to improve the measurement arrangement or obtain enough additional evidence for the decision. Replacing the component solely because an unresolved difference looks large on an expanded chart would not address the quality of the evidence.

23. Editable worksheet A: campaign and measurement definition

Complete this record before the analysis is issued. Enter actual project values in the blank column; explanatory text is guidance, not a default requirement. Use a separate issue when the boundary, sensor arrangement or event definition changes materially. The original campaign record remains the reference for the earlier analysis.

A missing process limit does not prevent describing the observed pressure. It does prevent calling that observation a confirmed breach of an unknown requirement. Similarly, a missing flow record does not erase a pressure event, but it can limit comparisons across different demand states.

The campaign owner should verify the point identities against the installation. A correct formula applied to wrongly identified tappings remains a wrong diagnostic result. Keep photographs or a marked diagram with the internal evidence package where they help future users find the same boundary.

24. Editable worksheet B: observations and event record

Use one observation identity for each retained source record and a separate event identity for each investigation event. A row number is not a stable identifier after sorting. Large time series should remain in an attached machine-readable dataset; the Word record captures the selected evidence and how it is interpreted.

Do not fill missing cells with nominal operating values to make the record look complete. A manufacturer rating can support a hypothesis or planning assumption, but it is not an observation from the incident. Keep such information in a separate source field with its applicability.

If a corrected dataset changes an earlier conclusion, identify the affected report and issue the revision explicitly. Preserve the original event record, correction basis and revised conclusion so the next engineer can understand why the interpretation changed.

25. Editable worksheet C: hypothesis, action and retest

Use the evidence record to choose a discriminating next action. The same symptom can require an instrumentation correction, a local repair, a control review or a capacity study. Keep these possibilities separate until evidence supports a disposition. Closing a work order does not automatically close the diagnostic question.

Where a trial improves pressure, verify that required air quality, process performance and other consumers were not compromised. A bypassed filter may create a favourable pressure trace while making the installation unsuitable for its intended process. An isolated branch may improve the header by removing productive demand rather than by correcting the original constraint.

The final record should be concise enough for action but traceable enough for review. Include the key event, evidence and unresolved limitation instead of exporting only a coloured verdict or a smoothed trend image.

26. Questions engineers and maintenance teams ask

Does low pressure prove that the compressor is too small?

No. It identifies inadequate pressure at the observed point and time. Distribution, treatment, local connections, control response, unavailable equipment and demand combinations may be involved. Compare source and downstream trends with the operating record before selecting a capacity remedy. A demonstrated capacity shortfall is a possible conclusion, not the starting assumption.

Can one sensor diagnose the complete problem?

One sensor can identify a local trend, timing and recorded extrema. It usually cannot separate source-pressure changes from losses between the source and the consumer. Additional synchronized points should be chosen to divide the uncertain path. The required number follows the physical arrangement and diagnostic question, not a fixed universal sensor count.

Why is the compressor gauge satisfactory while the machine stops?

The readings may describe different locations, times or response speeds. The machine may experience a local restriction or short transient not visible on the compressor display. Measure the relevant machine inlet during the actual event and compare it with synchronized upstream points. Also check that the complaint genuinely relates to pressure rather than another machine condition.

Should every negative differential be rejected?

No. Check reference convention, clock alignment, tapping identity, sensor uncertainty and topology. Stored-air discharge or a second feed can create a real reversal, while an offset clock can create an apparent one. Preserve the original reading with a review flag until the evidence supports its treatment. Do not force a preferred sign by clipping the data.

What sampling interval should the workbook prescribe?

It should require an interval and sensor response appropriate to the event rather than prescribe one for every system. Start with the event duration and the process question, then confirm the whole acquisition chain can resolve it. Long averages cannot recover a shorter unseen minimum, regardless of the number of decimals in the exported result.

Is time below a limit always measured directly?

No. It may be reconstructed from discrete observations under a declared hold or interpolation assumption. With interval averages, the direct result is the duration represented by averages below the limit, not exact instantaneous exposure. The report must name its representation and distinguish observed values, estimated crossings and missing periods.

Does a large regulator pressure difference prove a fault?

Not by itself. Regulation intentionally reduces pressure. Compare the required outlet pressure, flow demand, upstream margin and applicable regulator behaviour. An outlet sag during demand may justify checking capacity, installation or supply margin, but the static difference between a higher inlet setting and a lower controlled outlet is not automatically wasted pipe pressure.

Can correlation or an AI classifier establish the cause?

It can identify patterns that deserve investigation, but shared demand, clock offsets, filtering and control interaction can produce similar traces. Retain source records and competing explanations. The workbook should produce a proposed hypothesis and the next confirming observation, not an unsupported final diagnosis or automatic instruction to change equipment settings.

Does lower pressure loss prove an energy saving?

It proves a difference at the measured boundary under the compared conditions, subject to measurement limitations. Electricity savings need compatible power, production and operating evidence. Controls or other loads may change at the same time. Keep verified pressure-service improvement separate from an estimated energy opportunity until the energy comparison is supported.

What should accompany a diagnostic enquiry?

Provide the affected machine, documented pressure requirement, incident times, system diagram, sensor and logger details, raw traces, compressor-state records, known configuration changes and relevant operator observations. Identify missing channels or intervals. These inputs help scope a focused investigation rather than restarting with an equipment quotation unsupported by the actual pressure path.