River sand-mining management platform
An IoT management platform for river sand extraction and aggregate production: eight modules and 24 ...
View Details →The industry is not short of rules; it is short of the traces rules leave behind. The working face is on the water, the office is kilometres away, and between them sit three breaks.
One decision shaped the rest of the system: the core of sand-mining management is not managing people, it is managing a single haul. People, vessels, contracts and finance are all derived from “this truck crossed the weighbridge once”. Get that right and every module has something to stand on; get it wrong and more reports only amplify the error.
The manual was written menu by menu, so it reads as eight parallel blocks. Follow the data instead and they share one key chain: pit → contract → truck / vessel → weighing record → finance. Remove one link and the far end stops adding up.
| Module | What is inside | Who uses it |
|---|---|---|
| Dashboard | Six headline figures — pits, contracts, trucks, vessels, cameras, contract value — plus an output trend curve | Owner and site manager: the first screen opened every morning |
| Pits | Pit register, vessel register, permits, geofence records — four screens covering where extraction happens, who may do it, and how far | Site administrator: new pits, permits, fence maintenance |
| Contracts | Contract number, unit price, volume, contract value, amount paid; customer register; balance adjustment and reporting sync | Sales and finance: signing, quota checks, customer records |
| Devices | Video recorders with arming, device exception records, pit weighbridges, pit barriers | Operations: uptime, exceptions, arming state |
| Trucks | Plate, home pit, rated load, status — filterable and maintained in bulk | Gate and weighing staff: an unknown plate should never raise the barrier |
| Vessels | Trip records and position messages, searchable by vessel, pit and period | Site management: where is the dredger, how long has it been there |
| Finance | Sand sale records (query and backfill), contract balance adjustment records | Finance: one truck one entry, quota deducted line by line |
| Administration | Users, roles, departments, menus, IP allowlist, login history, audit log, exceptions — eight items | System owner: accounts, and who changed which number |
One deliberate split: the device module holds both the asset register and the exception log. The register answers “which pit does this weighbridge belong to, what model, which IP”; the exception log answers “it was offline for 47 minutes today”. One is an asset, the other is operations — merged, neither works well.
Six steps from gate to gate, each doing exactly one thing, each leaving a raw record that cannot be edited afterwards. This is the part of the system we refused to compromise on.
Why refuse manual entry for net weight? Because it is the cheapest thing to negotiate. Once the net figure can be typed, the scale in front of it and the contract behind it become decoration: the scale is accurate, the number was written by a person. The system therefore accepts only two things — an indicator reading and a weighing event bound to that reading. Net is gross minus tare, two decimals, one rounding rule across the whole system. If it truly must be corrected, it goes through backfill, and it shows up in reports as backfill.
Wiring the barrier is what gives the software any authority on site. Without it the system only records history: the truck has already left, the entry is typed afterwards, blocking is meaningless. With it, an unregistered plate or an exhausted quota simply does not raise the barrier. But a hard block creates a new problem — a queue — so the rule is configurable: block by default, allow an authorised override with a record. The site keeps moving, and the record shows who let the truck through.
Contract, weighbridge and finance can each defend themselves and each be doubted. What turns them into evidence is not a richer field list but whether three records point at the same event.
A trade-off we argued about: block or warn when the quota runs out. Blocking is the compliant answer, but one queue at the gate and drivers simply take another exit — the system loses the data it was built to capture. Warning keeps traffic moving, but by month end sales and finance are arguing about an overdrawn contract. We shipped block by default, authorised override: the barrier stays down, an administrator can post a balance adjustment (reason mandatory) which lands in the adjustment log, and only then does the truck leave. Compliance and throughput both hold, at the cost of one human confirmation — and that confirmation is exactly where responsibility should sit.
The second trade-off: volume or value, whichever runs out first. A contract for 6,000 t at 55 CNY/t implies 330,000 CNY, but rates change by batch, so “tonnage exhausted, value left” and the reverse both happen. The system tracks both available quantities and treats either reaching zero as exhaustion, showing the two side by side. We deliberately do not auto-convert one into the other: once the system decides which measure closes the contract, reconciliation gets harder, not easier.
A permit carries coordinates, depth, permitted volume and a validity period. On paper, the site can only hope. The first thing the system does is turn those coordinates into a fence.
Positioning is a trade-off between accuracy and cost. Reporting every minute is precise, but it drains the terminal, consumes data plans and writes millions of rows a year across a fleet. We made reporting frequency situational: a low base rate normally, a higher rate inside the zone or when a deviation is suspected, and a lower rate when docked. The cost is a delay of seconds to minutes before an excursion is flagged — acceptable when the business is measured in hours, not seconds.
Cameras are for checking, not for watching. Nobody can watch dozens of streams all day. Video earns its place after the fact: when a sale record looks odd, replay the gate at that timestamp and compare plate, load and time. This is also why video recorders sit in the device module — a recorder offline means a gap in the evidence chain, and that “offline for 47 minutes” entry matters more than any explanation when a case has to be defended.
Field operations will meet a dead network, a broken scale, an ad-hoc truck. Remove backfill and the business routes around the system: paper ledgers return and the software is dead within a quarter. So the path must exist — just make it narrow.
Every backfilled record is flagged, the sale list can filter to backfill only, and reports carry a separate column. The effect is practical: backfill stops being a quiet way to change numbers and becomes an exception that has to be explained. Used rarely, it is a safety valve; used often, management sees it immediately. That feedback is what a process is supposed to have.
The balance we aimed for: normal business should be smooth enough to need no training; exceptional business should be awkward enough to demand an explanation. Achieve both and the system survives its first year.
The real threat to a sand-mining platform is not an outside attacker but an insider editing one number. All eight administration screens exist to answer a single question: who changed this figure, when, and from which IP.
| Function | What it solves | How it lands |
|---|---|---|
| Users | One person, one account | Created by an administrator, never shared; disabling revokes access but keeps history |
| Roles | What can be seen and changed | Roles bind menus and actions: a weighing clerk cannot edit contract rates; sales cannot see raw scale readings |
| Departments | Data scoping | Visible pits are scoped by department, so several operations share one system without leaking data |
| Menus | Navigation matches permission | Menus are issued per role; the front end hides what is not permitted and the server checks again |
| IP allowlist | Accounts are not lent out | Only site and office ranges may sign in; off-network attempts are refused |
| Login history | Who came in, and when | Account, IP and time in three columns; odd-hour logins stand out |
| Audit log | Who changed what | Contract adjustments, deletions and device parameter changes, each with before and after values |
| Exceptions | The system's own faults | Interface errors, failed jobs and device timeouts collected in one list, so operations need not read server logs |
One detail worth naming: the audit log stores the value before and after, not merely the action. A log of clicks is useless in a dispute; a log that restores the exact number is what auditing means.
As agreed with the client, this page shows no client name, no site location, no permit numbers and no screenshots of production screens. Amounts and tonnages are de-identified examples used only to explain the calculation.
Projects like this — site outdoors, ledger in the office — rarely fail on interface design. They fail on three questions: how the field devices connect, how exceptions leave a trace, and how one set of numbers is agreed. Bring your current paperwork to the first call and we will tell you which of the three is your bottleneck, then scope it. Related: smart factory & IoT and how we build.
MORE CASES
An IoT management platform for river sand extraction and aggregate production: eight modules and 24 ...
View Details →Start with our pricing logic, then let's talk requirements — cost structure, technology comparison and acceptance criteria, all laid out in the open.