— Blogs —
—Products—
Consumer hotline +8618073152920 WhatsApp:+8615367865107
Address:Room 102, District D, Houhu Industrial Park, Yuelu District, Changsha City, Hunan Province, China
Product knowledge
Time:2021-12-23 21:54:32 Popularity:2367
Quick Answer: Edge computing for industrial IoT means running a defined processing task near the instruments or machines that generate the data. Useful tasks include local aggregation, event evaluation and outage-tolerant recording. Choose local processing when it solves a specific timing, traffic or availability problem. Installing a gateway with network ports does not, on its own, provide those application functions or replace a safety controller.

Good edge tasks have a clear reason to run locally. Examples include converting Modbus registers into engineering units, calculating a rolling average, checking a threshold, compressing repetitive data or buffering records before upload. A PLC should still handle deterministic machine control and safety functions for which it was designed.
| Processing location | Suitable work | Main dependency |
|---|---|---|
| Sensor or instrument | Measurement conversion, compensation and diagnostics | Sensor firmware and calibration |
| PLC, RTU or data logger | Local acquisition, control, alarms and timestamped storage | I/O capacity and control program |
| Edge gateway | Protocol translation, filtering, buffering and selected analytics | Compute resources, software maintenance and interface support |
| Central server or cloud | Fleet comparison, long-term records, reporting and model management | Wide-area network and service availability |
For remote environmental monitoring, the reason may be to retain an event through a cellular interruption. For industrial equipment monitoring, it may be to calculate a local operating statistic without waiting for a central server. In either case, identify the input, calculation, output and response to invalid data. That description is more useful than specifying edge capability without a workload.
A mean is not a compressed copy of every sample. Once the raw samples have been discarded, a central analyst cannot reconstruct a short peak from the average alone. Specify which statistics need to be kept and whether a short local raw-data buffer is necessary for investigating events. The calculation should follow the measurement meaning: wind direction needs circular treatment, and logarithmic sound levels are not generally combined as ordinary arithmetic means.
| Illustrative one-minute input | Local result | What is lost if only that result is retained |
|---|---|---|
| Six valid ten-second values: 20, 20, 20, 20, 20, 80 | Arithmetic mean 30; maximum 80 | Sending the mean alone hides the short maximum. |
| Only five of the six expected samples are valid | A result with valid count 5 of 6 | Without coverage, the server cannot distinguish a complete interval from a partial one. |
| A cumulative counter resets during the interval | A reset event plus a result computed under the agreed rule | Subtracting counters blindly can create a false negative increment. |
These values illustrate calculation behavior, not a sensor range or a universal sampling schedule. For a threshold of 70 in this example, a rule applied to the minute mean would not trigger, while a rule applied to the valid ten-second samples would. Neither rule is correct until the project defines what duration and statistic the threshold is intended to represent.
For data reduction, compare the encoded records rather than raw sample counts alone. Packet headers, timestamps, quality flags, encryption and retransmission add traffic. A reduction from six samples to one aggregate does not automatically mean one sixth of the cellular bill. It may still be worthwhile, but measure the complete message stream during a trial.
Local processing can reduce upstream traffic when raw samples are converted to averages or events. It can shorten alarm response where the rule runs on site. Store-and-forward operation can improve data completeness when a cellular link is intermittent. These benefits depend on configuration; merely installing an “edge” device does not create them.

A field gateway can forward serial bytes without understanding them. To compute an engineering statistic, an application must first decode registers, data types, scaling and units. RS485 is the electrical interface, Modbus RTU supplies a message structure, and the chosen application supplies the meaning of the points. Routing traffic and calculating a temperature average are different functions.
A NiuBoL field data logger is one possible acquisition component. Confirm the actual installed software, supported calculations, record format and connection to the receiving system. A local display, Ethernet socket or relay terminal does not prove that a custom algorithm can be deployed or that a particular alarm workflow is included.
| Processing requirement | What must be specified | How to demonstrate it |
|---|---|---|
| Input acquisition | Point count, polling schedule, supported driver and invalid-value handling | Replay known valid, missing and out-of-range inputs. |
| Calculation | Formula, window, units, minimum coverage and configuration version | Compare results with a separately calculated test set. |
| Timing | Acquisition time, event time, processing delay and clock behavior | Trace a controlled input change through the full path. |
| Local records | Encoded record size, usable capacity and overwrite policy | Test the configured retention and a storage-limit condition. |
| Hardware environment | Ordered DC input, measured peak and normal demand, temperature and enclosure limits | Check the actual assembly under the specified operating load and cabinet conditions. |
CPU speed and memory matter only in relation to that workload. A device that is adequate for several slow environmental channels may be unsuitable for high-rate vibration analysis. Conversely, a general-purpose industrial computer can add power consumption, operating-system maintenance and deployment work without improving a simple logger task. Demonstrate the required application before buying computing capacity on a generic performance label.
Separate measurement time from arrival time. A recovered message from an hour earlier should not become a current alarm simply because it has just reached the server. Locally, a last-known value needs an age limit before it is used for a decision. The appropriate limit depends on the process; do not label a frozen input normal merely because its last value was within range.
Specify persistence, hysteresis and clearing rules where they are needed. Test a brief excursion, a sustained excursion, recovery and loss of input. If a rule requires several consecutive valid observations, decide how a missing observation affects that sequence. Quietly substituting zero can suppress one alarm and create another.
Local execution removes the wide-area journey for a local decision, but not sensor response, bus scheduling, computation or output delay. Measure each relevant part under realistic load. A gateway is not a deterministic safety system merely because the event rule runs on site. Keep protective functions in the approved controller and architecture for that duty.
During a network outage, field acquisition can continue only while power, the local application and storage remain available. Distinguish a network outage from a power interruption. If the clock loses its time source, determine how the application labels records and restores timing without producing an apparently continuous but incorrect history.
For an illustrative retention calculation, a 160-byte record every ten seconds produces 8640 records, or 1,382,400 bytes, per day. Thirty days requires about 41.5 MB before filesystem overhead, indexes, logs and retry information. This is an application-data calculation, not a NiuBoL hardware retention guarantee. Compare it with the usable allocation rather than the memory device's nominal capacity.
On reconnection, preserve acquisition timestamps and record identifiers. The receiving application should distinguish recovered records from new observations and handle a retry without counting an observation twice. Define whether current alarms and historical backlog use separate priorities. Test recovery while acquisition continues, not only after stopping the field application.
Storage eventually fills. Decide whether the application overwrites the oldest records, stops collecting or prioritizes events. An equipment specification that mentions storage capacity without this behavior leaves the operational requirement incomplete. Include a full-storage test using an approved simulation rather than waiting for an unplanned long outage.
More software at remote sites creates maintenance work. Define operating-system updates, application versions, configuration backups and user access. If remote commands are allowed, separate monitoring rights from control rights and keep an audit trail. Sensitive credentials should not be hard-coded in exported configuration files.
Edge devices also need clock synchronization, clear units and quality flags. Otherwise the central platform may receive calculated values without enough context to interpret them.
Give each configuration and calculation a version. If the averaging window or sensor scale changes, retain the time of change with the records. Otherwise a new algorithm can create an apparent environmental trend. For a fleet, test a change on a representative unit before wider deployment and keep a working rollback package that local staff can actually use.
Network segmentation and controlled remote access should follow the site's operational requirements. NIST's Guide to Operational Technology Security treats OT security in the context of performance, reliability and safety needs. Apply the relevant safeguards to the chosen architecture; the presence of an edge device is not a cybersecurity certification.
Keep long-term comparison, fleet reporting and calculations needing many sites centrally when there is no specific local requirement. If an existing PLC already performs the necessary acquisition, a communication device may be enough to carry its records. Adding a second application solely to claim edge computing can create conflicting point maps and unnecessary maintenance.
Before specifying a separate processing platform, run a short proof of function: known inputs, independently calculated outputs, invalid data, an interrupted connection and a restart. Record execution timing and power demand under the actual workload. Include who maintains the application after commissioning. That evidence determines whether edge computing is useful for this project more reliably than a long feature list.

A logger may perform useful local processing, but the label is less important than the supported task. Check whether it can execute the required calculation, preserve quality information and export the agreed records. Local logging alone does not establish a general-purpose application platform.
Use it when a defined task benefits from local timing, data reduction or operation during a network outage. Compare those benefits against power, software maintenance and support requirements. If the central system already meets the need, another local application may add no useful capability.
It can avoid a wide-area round trip for a locally evaluated rule. Total latency still includes sensing, acquisition, scheduling and the final action. Test the complete path and keep safety-related actions within an approved control design.
Only if the project no longer needs it. Averages hide peaks and sequence information. Retain appropriate maxima, coverage and event records, and consider a limited raw-data buffer when later diagnosis requires it. State the retention choice explicitly.
No. The configuration still needs a defined payload, authentication, delivery handling and receiving application. A retry can create duplicate records unless identifiers and receiver behavior are designed for it. Test the chosen end-to-end exchange rather than inferring completeness from protocol support.
Prove calculation results, input validity handling, record timing, network interruption, storage limits, restart and rollback. Verify the installed application version and workload on the ordered hardware. A successful normal-condition upload is only one part of that test.
Edge computing for industrial IoT is justified by a useful local task, not by a hardware label. Preserve measurement meaning when aggregating, evaluate alarms against valid and current inputs, and demonstrate outage recovery. NiuBoL acquisition and communication equipment should be selected for the functions actually configured, with software ownership and acceptance defined before deployment.
Related recommendations
Sensors & Weather Stations Catalog
Agriculture Sensors and Weather Stations Catalog-NiuBoL.pdf
Weather Stations Catalog-NiuBoL.pdf
Agriculture Sensors Catalog-NiuBoL.pdf
Water Quality Sensor Catalog-NiuBoL.pdf
Related products
Combined air temperature and relative humidity sensor
TDR Soil Moisture & Temperature Sensor for irrigation| RS485 |4-20mA| NBL-S-THR
Soil pH sensor RS485 soil Testing instrument soil ph meter for agriculture |NBL-S-PH
Wind Speed sensor Output Modbus/RS485/Analog/0-5V/4-20mA
Tipping bucket rain gauge for weather monitoring auto rainfall sensor RS485/Outdoor/stainless steel
Pyranometer Solar Radiation Sensor 4-20mA/RS485
Screenshot, WhatsApp to identify the QR code
WhatsApp number:+8615367865107
(Click on WhatsApp to copy and add friends)