Direct Answer

Every current data point you see on the platform does not come from the sensor "snapping a picture" — it passes through a seven-step chain: sensing and sampling, A/D conversion, feature calculation, protocol packing, edge caching, uplink forwarding, and in-database cleaning. Each step can introduce deviation: the sampling interval decides how fast you see things, the A/D precision decides how accurately you measure, the feature calculation decides whether what is calculated is the value you want, and the protocol and database steps decide whether the fields match. If any of the seven steps goes wrong, the data point you finally see is "looks right, but actually biased."

A scene: a park found that the current curve on the platform did not match the actual load and suspected the sensor was broken. The troubleshooting found: the sensor was fine; the problem was in the chain — the device sampling interval was set to 15 minutes, so the load fluctuations were averaged away; at the A/D conversion, the ratio was configured wrong, so the values were overall too large; the feature calculation computed the "average", while the platform displayed the "instantaneous value"; at the protocol packing, the fields were misaligned, and the temperature data landed in the current field. With the same sensor and four of the seven steps biased, the resulting data is naturally unusable.

1. Step 1: Sensing and Sampling — "What You See" Depends on the Sampling Interval

The sensor grabs the physical quantity at a fixed interval; this step decides the "granularity of the data." The shorter the interval, the more transients you see; the longer, the more the fluctuations are averaged away.

Deviation this step can introduce: the sampling interval does not match the business need — transient current monitoring needs the millisecond level, trend monitoring uses the minute level; use the wrong one and you either store a sea of useless data or lose the key fluctuations. Check action: confirm the sampling interval matches the business scenario — "what judgement is this data supporting? is this interval enough?"

2. Step 2: A/D Conversion — "How Accurately You Measure" Is Decided by Precision

The conversion of the analogue signal to digital (A/D conversion) decides the numerical precision: the more bits and the better the range match, the smaller the error; with insufficient bits or a mismatched range, small signals are eaten by quantisation noise.

Deviation this step can introduce: insufficient A/D bits make small-current measurement inaccurate; an improperly set range clips signals exceeding it, or small signals fall in a low-resolution interval. Check action: verify the A/D bits and range match the measurement object — "can this acquisition precision resolve a 0.1 A signal?"

3. Step 3: Feature Calculation — "Whether What Is Calculated Is What You Want" Is the Key

The raw data must be processed into "feature values" — average, RMS, peak, frequency, etc. This step is the most prone to the "looks right, but actually not right" problem.

Deviation this step can introduce: the calculation calibre does not match the display calibre — the platform labels "instantaneous current" but displays a 15-minute average; a feature algorithm error — the three-phase imbalance uses the wrong formula; a ratio conversion error — it shows 125 A but the actual is 1,250 A. Check action: at the same moment, compare "the on-site meter reading vs the platform data point" — if they match, the feature calculation is credible.

4. Step 4: Protocol Packing — "Whether the Fields Are Right" Decides Whether Data Gets Misaligned

The device packs the data for upload according to a protocol (Modbus, IEC 104, MQTT, etc.). This step decides "whether the structure of the data is right during transmission."

Deviation this step can introduce: a register address mapping error, with device A's data landing in device B's field; a byte-order (endian) problem, with the value flipped; a protocol version mismatch, with the field parsing misaligned. Check action: spot-check the raw frames and verify the "register address ↔ platform field" mapping — if every one matches, the protocol layer is reliable.

5. Step 5: Edge Caching — "Whether Data Is Lost During Outages" Is Covered by the Cache

The data is cached locally at the edge gateway, not lost during network outages, and back-filled after recovery. This step decides the "continuity of the data."

Deviation this step can introduce: insufficient cache capacity — if the outage exceeds the capacity, data is lost; a wrong cache strategy — overwrite-style writing flushes out the old data. Check action: run a network-outage drill — unplug the cable for 30 minutes, then check whether the data is completely back-filled and the timestamps are aligned after recovery.

6. Step 6: Uplink Forwarding — "Whether It Arrives" Is Guaranteed by the Network

The cached data is forwarded uplink to the platform over the network. This step decides the "timeliness and completeness of the data."

Deviation this step can introduce: network jitter causes retransmission and out-of-order data; insufficient bandwidth causes high-density data to be dropped or delayed; an uplink interruption is unperceived and the data silently disappears. Check action: look at the uplink channel's status monitoring and packet-loss statistics — "was yesterday's data complete? are there any gaps?"

7. Step 7: In-Database Cleaning — "Whether It Can Be Used" Is Covered by Data Governance

After arriving at the platform, the data passes through validation, cleaning and calibration before being stored. This step decides "whether the data can be directly consumed in the database."

Deviation this step can introduce: a wrong cleaning rule filters out real anomalies as noise; inconsistent units (kW and MW mixed); misaligned timestamps, with the data time not matching; a missing object model, with the measurement points not bound to devices. Check action: export the raw data and check — "what is the difference before and after cleaning? is the cleaning rule documented?"

8. A "Courier Chain" Analogy: A Data Point Is Also a Package

Think of the seven-step chain as sending a package: sensing and sampling is packing (the sampling interval decides what is in the package); A/D conversion is weighing (the precision decides how accurately it is weighed); feature calculation is filling in the form (whether what is calculated is the value you want, like a wrong waybill number); protocol packing is sealing (field misalignment is like sticking the address on wrong); edge caching is temporary storage (a full warehouse means lost items); uplink forwarding is transport (if the road is jammed or the item is lost, you check); in-database cleaning is sorting into the warehouse (wrong shelf or wrong label, and the customer cannot find it).

If any link goes wrong, the recipient (the platform) does not receive what you sent (the real data). When troubleshooting data problems, do not only stare at the two ends of the sensor and the platform — of the five links in the middle, each one may have "lost the package."

One layer deeper: every link of a courier has a "waybill number" to trace; the data chain is the same — every piece of data should carry a timestamp and an object identifier, and it should be fully traceable which step the chain went through, whether it was processed, and what it became. A data chain without traceability is like a courier without a waybill number — when a package is lost, you cannot find out which link lost it.

9. Where the Seven Steps Belong: How Many Each of Device-Edge-Cloud Manages

Putting the seven-step chain into the device-edge-cloud architecture, the ownership is clear and echoes the iron rule of the architectural division:

  • The device side manages the first 3 steps (sensing and sampling, A/D conversion, feature calculation): the physical quantity is captured and processed on site, latency-sensitive, not depending on the network — these 3 steps are the source of "whether the measurement is right";
  • The edge side manages the middle 3 steps (protocol packing, edge caching, uplink forwarding): the gateway does protocol conversion, outage caching and reliable uplink — these 3 steps are the guarantee of "whether the transmission is complete";
  • The cloud side manages the last 1 step (in-database cleaning): the platform validates, cleans and calibrates so the data can be directly consumed — this step is the fallback of "whether it can be used."

When troubleshooting, follow the ownership: values wrong, check the device side (the first 3 steps); gaps and lost data, check the edge side (the middle 3 steps); messy in the database, check the cloud side (the last 1 step) — locate the layer first, then locate the step; the efficiency doubles.

10. The Troubleshooting Route: Reverse-Deduce the Seven Steps from the Symptom

When the data has a problem, do not guess; check against the "symptom → suspicious step" comparison to converge quickly:

  • Values overall too large/small: first check Step 2 A/D ratio and Step 3 feature conversion — a wrong ratio or wrong unit conversion is the prime suspect;
  • The curve is flat, the fluctuations gone: check Step 1 sampling interval — too long and the transients are averaged away; then check Step 7 cleaning — the cleaning rule may have filtered the real fluctuations as noise;
  • Fields misaligned, temperature shown as current: check Step 4 protocol packing — register mapping misaligned, byte order flipped;
  • Data gaps, timestamp jumps: check Step 5 edge caching and Step 6 uplink forwarding — insufficient cache capacity loses data, network jitter drops packets;
  • Platform reading does not match the on-site meter: check from Step 1 to Step 7 link by link — the most common root cause is Step 3 feature calibre inconsistency (the platform displays the instantaneous value but calculates the average).

The troubleshooting iron rule: change one variable at a time — after changing one step, check the data again; only after the problem is located, change the next step; do not change three places at once.

11. The Three Questions of a Credible Data Point: Any Data Point Can Pass These Three Gates

No matter where a data point comes from, gate it with three questions before go-live:

Q1: Is the timestamp right — which moment is this data point from? Does the timestamp align with the on-site event? Are the timestamps of multiple channels consistent? (corresponds to Steps 1/5/6/7) Q2: Is the value right — is this number the same as on the on-site meter? Are the precision, ratio and unit matched? (corresponds to Steps 2/3) Q3: Is the object right — which device and circuit does this data point belong to? Is the object model bound? (corresponds to Steps 4/7)

With all three questions passed, the data point dares to enter the database and dares to be used for alarms and decisions; if any question fails, this point is "to be checked" and cannot enter the analysis.

12. The Seven-Step Chain Iron Rule

For a data point to be credible, every link of the seven-step chain must stand up to verification — the sampling interval matches the scenario, the A/D precision can resolve the signal, the feature calibre is consistent, the protocol mapping is correct, the cache does not lose data during outages, the uplink channel is monitorable, and the cleaning rules are documented.

Four sentences to land it:

  • A data point is not "snapped" but "processed" — every step of the seven-step chain can introduce deviation;
  • The sampling interval matching the scenario and the A/D precision resolving the signal are the foundation of "measuring right";
  • The feature calibre consistent and the protocol mapping correct are the guarantee of "describing clearly";
  • The cache not losing data during outages, the uplink monitorable and the cleaning documented are the fallback of "being usable."

Next time the platform data does not match, do not rush to blame the sensor — check the seven-step chain link by link: is the interval right? is the precision enough? is the calibre consistent? are the fields right? is data lost during outages? is the uplink complete? is the cleaning right? After the seven questions, the problem is located.

The seven-step self-check card: make these seven questions into a card and go through it with the equipment inspection every quarter — does Step 1's interval match the scenario? is Step 2's A/D precision and range enough? is Step 3's feature calibre consistent with the display? is Step 4's register mapping right? is Step 5's cache capacity enough for the outage buffer? is Step 6's uplink packet-loss statistics normal? is Step 7's cleaning rule documented? With the self-check card for regular check-ups, data quality changes from "checking after an incident" to "prevention in advance."

13. Back to FEXLINK: How the Seven-Step Chain Lands

To land this chain on products: FEXLINK's ESF-22110-R electrical fire monitoring controller handles the front segment — sensing and sampling configured by scenario, A/D conversion guaranteeing precision, and feature calculation calibre consistent with the platform; the ESX-0223-GR edge gateway handles the middle segment — protocol conversion, edge caching (not lost during outages, back-filled after recovery) and local fast judgement; the platform handles the last segment — in-database cleaning, object model and data-quality report. With every link of the seven-step chain having a clear responsibility and a verifiable action, the data points are truly credible. During selection you can also require the supplier to provide a "seven-step chain description document" — how each link is implemented and how each step is verified; once written clearly, the data quality has a basis to check.

Scope and Limitations

  • The conclusions in this article are intended for LV distribution and electrical-safety sites. Criteria are compiled from public codes and FEXLINK product documentation, and product parameters follow the latest selection handbook and product knowledge base.
  • Quantitative criteria (protection zone, discharge and residual voltage, earth resistance, thresholds and grading) hold under the preconditions stated in this article. Where the site voltage level, earthing arrangement, environmental category or load profile differs, the figures must be recalculated. Specific configurations involving ESF-22110-R、ESX-0223-GR should be verified against actual circuit parameters.
  • Specific product models and parameters are subject to the latest selection handbook and product knowledge base; this article does not constitute selection or procurement advice.
  • Standard numbers cited here refer to the currently effective editions; when a standard is revised, the related quantitative conclusions must be re-checked.
  • Deployment and implementation for specific projects must be carried out by qualified engineers based on site survey and design drawings.

Frequently Asked Questions

From Current Waveform to Cloud Data Point

Every current data point you see on the platform does not come from the sensor "snapping a picture" — it passes through a seven-step chain: sensing and